betaFeedback

WorksMathematics

A seven-element interval of power-semiring varieties

Abstract

We determine the interval between the varieties generated by the nonempty and full power semirings of a semigroup satisfying xy≈x2xy\approx x^2. It has two elements for a one-element semigroup, three for a nontrivial left-zero or null semigroup, and seven otherwise, without any finiteness assumption. In the seven-element case the interval is [\Rvar,\Rvar^0], where \Rvar=\Mod_{\rm ai}\{xy\approx xz\} and \Rvar^0 is its zero-adjunction variety. Every word absorption valid in \Rvar but not in \Rvar^0 is equivalent over \Rvar^0 to one of six relative identities. We give exact absorption criteria and finite bases for all seven varieties. Three have basis rank two and four have basis rank three; their two-generated free algebras have orders 1515, 2424, or 3535. We also count the finite-rank free algebras of both endpoints.

Introduction

We work with additively idempotent semirings in the signature (+,⋅)(+,\cdot), without named constants. For a semigroup SS, let P+(S)P^+(S) and P(S)P(S) denote its nonempty and full power semirings, with union and setwise multiplication. Thus P(S)P(S) is obtained from P+(S)P^+(S) by adjoining the empty set as an additively least multiplicative zero. Put

I(S)=[V(P+(S)),V(P(S))].I(S)=[\mathcal{V}(P^+(S)),\mathcal{V}(P(S))].

Yue, Ren, and Gao [9, Problem 1.3] ask for the cardinality of I(S)I(S). They prove that it is 2ℵ02^{\aleph_0} when SS is the multiplicative reduct of S7S_7 [9 Theorem 5.2].

We determine I(S)I(S) for every semigroup satisfying

xy≈x2(1)xy \approx x^2 \tag*{(1)}

Such a semigroup has multiplication xy=f(x)xy=f(x), where f2=ff^2=f. The interval has two elements if ∣S∣=1|S|=1, three if SS is a nontrivial left-zero or null semigroup, and seven if f≠id⁡Sf\ne\operatorname{id}_S and ∣f(S)∣≥2|f(S)|\ge2. In the last case both endpoint varieties are generated by the power semirings of the three-element subsemigroup

⋅\cdotaabbcc
aaaaaaaa
bbbbbbbb
ccaaaaaa

Table 1.2.

The lower endpoint in this case is

R=Mod⁡ai{xy≈xz}.\mathrm{R}=\operatorname{Mod}_{\mathrm{ai}}\{xy\approx xz\}.

Yue, Ren, Zeng, and Shao established its finite generation [10 Proposition 4.5 and Remark 4.6]; Yue and Ren determined its ten-element subvariety lattice and proved that it is a Cross variety [8 Lemma 2.1, Theorem 2.7 and Corollary 2.8]. Here we determine the interval [R,R0][\mathrm{R},\mathrm{R}^0], where R0\mathrm{R}^0 is generated by the zero extensions of members of R\mathrm{R}.

The proof classifies the relative identities of these endpoints. Every word absorption valid in R\mathrm{R} but not in R0\mathrm{R}^0 reduces, over R0\mathrm{R}^0, to one of six identities. Exact absorption criteria distinguish the resulting varieties and establish their inclusion relations. The criteria follow from word normal forms and invariants under substitutions and term contexts; explicit substitutions give the six-type reduction. All seven varieties are finitely based.

2020 Mathematics Subject Classification. 08B05, 16Y60, 20M07.

Key words and phrases. power semiring, additively idempotent semiring, variety interval, zero adjunction, equational basis, basis rank.

Corresponding author: Aifa Wang. We also determine their basis ranks. The seven varieties have only three distinct two-variable equational theories, with two-generated free algebras of orders 15, 24, and 35. Three varieties are generated by these free algebras; the other four require three free generators. Counting formulas for the endpoint free algebras follow from their normal forms.

The upper-endpoint absorption criterion is a specialization of the zero-adjunction criterion of Wu, Ren, and Zhao [7 Proposition 1.5], restated in Gao, Ren, and Zhao [2 Proposition 2.1]. The latter paper also shows that zero adjunction preserves equality of generated varieties. Together with the known four-element generator of R\mathrm{R}, this gives a five-element generator of R0\mathrm{R}^{0} (Proposition 3.4). For finite groups of order at least three, Shao and Gao [6 Corollary 1.2] prove that every variety in the corresponding power-semiring interval has infinite axiomatic rank and is nonfinitely based; their P(G)\mathcal{P}(G) denotes the nonempty power semiring. Related classifications for semilattice-ordered semigroups and normal bands appear in Kuřil and Polák [4] and Kuřil [3]. Ren and Zhao [5] determine a nine-element variety lattice under x3≈xx^{3} \approx x and xy≈yxxy \approx yx. For general variety theory we refer to Burris and Sankappanavar [1].

Preliminaries and the main theorem

Let AI denote the variety of additively idempotent semirings in the signature (+,⋅)(+, \cdot). For terms uu, vv, write

u≤vfor the identityu+v≈v.u \leq v \quad\text{for the identity} \quad u+v \approx v.

Every term is, by distributivity and additive idempotence, a finite nonempty sum of nonempty words. We identify such a polynomial pp with its finite set of word summands and write w∈pw \in p for membership in that set. For a word ww, let ∣w∣|w|, h(w)h(w), and c(w)c(w) denote its length, first variable, and content, respectively. A word of length one is called a one-letter word; a word of length at least two is called a long word. For h∈Ch \in C, where CC is a finite nonempty variable set, fix a long word wh,Cw_{h,C} with first variable hh and content CC. For C={h}C=\{h\}, take wh,C=h2w_{h,C}=h^{2}; otherwise take hh followed by the remaining elements of CC in some fixed order. For a polynomial pp, set

L(p)={x:x occurs as a one-letter summand of p},L(p)=\{x:x\text{ occurs as a one-letter summand of }p\},
E(p)={h(w):w is a long summand of p}.(2)E(p)=\{h(w):w\text{ is a long summand of }p\}. \tag*{(2)}

Theorem 2.1 (classification). Let S≠∅S \ne\varnothing be a semigroup satisfying xy≈x2xy \approx x^{2}. Then there is an idempotent map f:S→Sf:S \to S such that xy=f(x)xy=f(x), and

∣I(S)∣={2,∣S∣=1,3,∣S∣≥2 and (f=id⁡S or ∣f(S)∣=1),7,f≠id⁡S and ∣f(S)∣≥2.(3)|I(S)| = \begin{cases} 2, & |S|=1,\\ 3, & |S|\geq2\text{ and }(f=\operatorname{id}_{S}\text{ or }|f(S)|=1),\\ 7, & f\ne\operatorname{id}_{S}\text{ and }|f(S)|\geq2. \end{cases} \tag*{(3)}

In the last case the interval is, up to isomorphism, the seven-element lattice described in Theorem 6.1 below.

Proof. If xy=f(x)xy=f(x) with f2=ff^{2}=f, then

(xy)z=f(f(x))=f(x)=x(yz),(xy)z=f(f(x))=f(x)=x(yz),

so the operation is associative and x2=f(x)x^{2}=f(x). Conversely, if xy≈x2xy\approx x^{2}, define f(x)=x2f(x)=x^{2}. Taking y=x2y=x^{2} gives x3=x2x^{3}=x^{2}, whence x4=x2x^{4}=x^{2} and f2=ff^{2}=f. This proves the structural assertion. The nondegenerate case of the cardinality assertion is proved in Theorem 6.1; all remaining cases are proved in Section 7.

For the remainder of the paper put

A=V(P+(S)),U=V(P(S)).(4)A=\mathsf{V}(\mathcal{P}^{+}(S)), \qquad U=\mathsf{V}(\mathcal{P}(S)). \tag*{(4)}

The symbol V\mathsf{V} denotes the variety generated by an algebra, and Mod⁡aiΣ\operatorname{Mod}_{\mathsf{ai}}\Sigma denotes the class of algebras in AI satisfying Σ\Sigma. All inclusions of varieties are in this fixed signature. The cardinality of an interval includes its endpoints. For an ai-semiring TT, let T0T^{0} be obtained by adjoining a new additively least multiplicative zero, with the original operations on TT unchanged. For a variety WW, put

W0=V{T0:T∈W}.W^{0}=\mathsf{V}\{T^{0}:T\in W\}.

The zero is an element of these algebras, not a named constant in the signature. In particular, P(S)≅(P+(S))0\mathcal{P}(S) \cong(\mathcal{P}^{+}(S))^{0}.

Endpoint theories and a basis for the upper endpoint

Assume throughout this section and Sections 4–6 that

f≠id⁡S,∣f(S)∣≥2.(5)f \ne\operatorname{id}_{S}, \qquad|f(S)| \ge2. \tag*{(5)}

Since an idempotent surjective map is the identity, choose c∉f(S)c \notin f(S), put a=f(c)a=f(c), and choose b∈f(S)∖{a}b \in f(S) \setminus\{a\}. The set S∗:={a,b,c}S_{*}:=\{a,b,c\} is a subsemigroup, with multiplication table (1.2). In particular, the endpoint theories of every such SS are already visible on this three-element subsemigroup.

Lemma 3.1 (endpoint absorption criterion). For every polynomial pp,

A⊨x≤p⟺U⊨x≤p⟺x∈L(p).(6)A \models x \le p \quad\Longleftrightarrow\quad U \models x \le p \quad\Longleftrightarrow\quad x \in L(p). \tag*{(6)}

If u=wh,Cu=w_{h,C} is a long word with first variable hh and content CC, then

A⊨u≤p⟺h∈E(p),(7)A \models u \le p \quad\Longleftrightarrow\quad h \in E(p), \tag*{(7)}

whereas

U⊨u≤p⟺∃v∈p (∣v∣≥2, h(v)=h, c(v)⊆C).(8)U \models u \le p \quad\Longleftrightarrow\quad\exists v \in p\ \bigl(|v| \ge2,\ h(v)=h,\ c(v) \subseteq C\bigr). \tag*{(8)}

Consequently V(P+(S))=V(P+(S∗))\mathsf{V}(\mathcal{P}^{+}(S))=\mathsf{V}(\mathcal{P}^{+}(S_{*})) and V(P(S))=V(P(S∗))\mathsf{V}(\mathcal{P}(S))=\mathsf{V}(\mathcal{P}(S_{*})).

Proof. Write f(X)={f(s):s∈X}f(X)=\{f(s):s\in X\} for the direct image of a subset. For nonempty variable sets Xy⊆SX_y\subseteq S, every long word ww evaluates in P+(S)\mathcal{P}^{+}(S) to

w(X1,…,Xn)=f(Xh(w)).w(X_1,\ldots,X_n)=f(X_{h(w)}).

In the full power semiring the same formula holds when Xy≠∅X_y\ne\varnothing for every y∈c(w)y\in c(w). Its value is empty if Xy=∅X_y=\varnothing for some y∈c(w)y\in c(w). This proves the sufficiency of (3.2)–(3.4).

For necessity it is enough to use the three-element table (1.2). If x∉L(p)x\notin L(p), assign Xx={c}X_x=\{c\} and every other variable the value {b}\{b\}. Every right-hand long summand has value {a}\{a\} or {b}\{b\}, and every right-hand one-letter summand has value {b}\{b\}. Thus pp does not contain cc, whereas xx does. If h∉E(p)h\notin E(p), assign Xh={c}X_h=\{c\} and all other variables {b}\{b\}. The long left-hand word has value {a}\{a\}, while the right-hand side is contained in {b,c}\{b,c\}. Finally, if the condition in (3.4) fails, assign variables outside CC the empty set, assign hh the set {c}\{c\}, and assign the remaining variables of CC the set {b}\{b\}. The left-hand word has value {a}\{a\}; surviving right-hand long words cannot start with hh, so no right-hand summand contains aa. Every identity is equivalent to word absorptions in both directions. The criteria therefore describe the entire endpoint theories, both for SS and for S∗S_{*}, proving the final assertion. □

Proposition 3.2 (finite basis for UU). Relative to the axioms of AI, the upper endpoint UU has the basis

B={xy≈x2y, xy≈xy2, xyz≈xzy, xyz≤xy}.(9)\mathcal{B}=\{xy\approx x^{2}y,\ xy\approx xy^{2},\ xyz\approx xzy,\ xyz\le xy\}. \tag*{(9)}

Proof. The four identities hold in P(S)\mathcal{P}(S) by the endpoint computation. The identity xyz≈xzyxyz\approx xzy, applied to prefixes and in right contexts, permutes any adjacent pair of letters after the first position. The identity xy≈xy2xy\approx xy^{2} removes repetitions in those positions. If a word contains a variable other than its first variable, tail permutation followed by x2y≈xyx^{2}y\approx xy removes extra occurrences of the first variable. A word whose content is {h}\{h\} reduces to h2h^{2}, since x2≈x3x^{2}\approx x^{3} follows from the first identity. Thus every long word reduces to wh,Cw_{h,C}.

Write a long word vv as v=v1v2v=v_{1}v_{2} with both factors nonempty. Substituting x=v1,y=v2,z=tx=v_{1},y=v_{2},z=t in the last identity gives vt≤vvt\le v. If long words u,vu,v have the same first variable and c(v)⊆c(u)c(v)\subseteq c(u), append the missing variables to vv and normalize to obtain u≤vu\le v. Lemma 3.1 now derives every word absorption valid in UU. An identity between sums is equivalent to absorbing every summand of each side into the other side, proving completeness. □

We shall repeatedly use two consequences: every long word uu satisfies u2≈uu^{2}\approx u in UU, and uv≈uuv\approx u whenever c(v)⊆c(u)c(v)\subseteq c(u). These follow by the same normal forms.

Remark 3.3 (the lower endpoint). The lower endpoint in the nondegenerate case is the already studied variety

A=Mod⁡ai{xy≈xz}.(10)A = \operatorname{Mod}_{\mathrm{ai}}\{xy \approx xz\}. \tag*{(10)}

Indeed, xy≈xzxy \approx xz implies xy≈x2xy \approx x^{2} by putting z=xz=x, and then every long word reduces to the square of its first variable. The criterion (6)–(7) shows that no additional identity identifies distinct LL- or EE-data. Thus (10) follows. This endpoint is the variety R (denoted by R) studied by Yue and Ren [8]; its finite generation is due to Yue, Ren, Zeng, and Shao [10], and its ten-element subvariety lattice and Cross property are given in [8 Theorem 2.7 and Corollary 2.8].

Proposition 3.4 (zero-adjunction description and a small generator). Under (5),

A=R,U=R0,I(S)=[R,R0].A = \mathrm{R}, \qquad U = \mathrm{R}^{0}, \qquad I(S) = [\mathrm{R},\mathrm{R}^{0}].

If K=S(4,475)K=S_{(4,475)} is the four-element ai-semiring in [8 Table 4 and Lemma 2.1], then

A=V(K),U=V(K0).A = \mathsf{V}(K), \qquad U = \mathsf{V}(K^{0}).

In particular, the upper endpoint is generated by a five-element algebra.

Proof. Remark 3.3 gives A=RA=\mathrm{R}, and [8 Lemma 2.1] gives R=V(K)\mathrm{R}=\mathsf{V}(K). By the general zero-adjunction result [2 Corollaries 2.2 and 2.3], V(T)0=V(T0)\mathsf{V}(T)^{0}=\mathsf{V}(T^{0}) for every ai-semiring TT. Applying it to T=P+(S)T=\mathcal{P}^{+}(S) and to T=KT=K yields

U=V((P+(S))0)=A0=R0=V(K0).U=\mathsf{V}\left((\mathcal{P}^{+}(S))^{0}\right)=A^{0}=\mathrm{R}^{0}=\mathsf{V}(K^{0}).

Since zero adjunction adds one new element, ∣K0∣=5\lvert K^{0}\rvert=5. □

Remark 3.5. The five-element generator in Proposition 3.4 is not asserted to be free. The orders of free algebras and the least number of free generators needed to generate each interval variety are determined in Sections 6 and 8.

Six relative absorption rules

Define the following six identities, all written as absorptions:

α:xy≤x+y2+xz,β:xy≤y2+xz,η:x2≤x+xy,δ:xy≤y+xz+yz,γ:xy≤y+xz,σ:x2≤xy.(11)\begin{aligned} \alpha&: xy \leq x+y^{2}+xz, & \beta&: xy \leq y^{2}+xz,\\ \eta&: x^{2} \leq x+xy, & \delta&: xy \leq y+xz+yz,\\ \gamma&: xy \leq y+xz, & \sigma&: x^{2} \leq xy. \tag*{(11)} \end{aligned}

Let Uθ=U∩Mod⁡{θ}U_{\theta}=U\cap\operatorname{Mod}\{\theta\} for θ\theta among these six rules. Every rule in (11) holds in AA.

Fix a finite nonempty set CC of variables and h∈Ch\in C. For a polynomial pp, define

Hh,C(p)⟺∃v∈p (h(v)=h, c(v)⊆C),H_{h,C}(p)\Longleftrightarrow\exists v\in p\ (h(v)=h,\ c(v)\subseteq C),
NC(p)⟺∃v∈p (∣v∣≥2, c(v)⊆C),N_{C}(p)\Longleftrightarrow\exists v\in p\ (|v|\geq2,\ c(v)\subseteq C),
TC(p)⟺C∩L(p)∩E(p)≠∅,T_{C}(p)\Longleftrightarrow C\cap L(p)\cap E(p)\ne\varnothing,
JC(p)⟺∃v∈p (c(v)⊆C),J_{C}(p)\Longleftrightarrow\exists v\in p\ (c(v)\subseteq C),
QC(p)⟺NC(p)∨TC(p).(12)Q_{C}(p)\Longleftrightarrow N_{C}(p)\lor T_{C}(p). \tag*{(12)}

Thus HH and JJ may be witnessed by a one-letter word, while NN must be witnessed by a long word.

Lemma 4.1 (contextual invariants). Every identity valid in UU preserves the seven data LL, EE, HH, NN, TT, JJ, QQ. After arbitrary polynomial substitutions and insertion into an arbitrary one-hole term context, the two sides of one of the rules in (11) preserve the data in the

following table:

ruledata preserved for every (h,C)αL,E,H,NβL,E,NηL,E,HδL,E,QγL,E,JσL,E(13)\begin{array}{c|l} \textit{rule} & \textit{data preserved for every }(h,C) \\ \alpha& L,E,H,N \\ \beta& L,E,N \\ \eta& L,E,H \\ \delta& L,E,Q \\ \gamma& L,E,J \\ \sigma& L,E \tag*{(13)} \end{array}

Proof. If p≈qp \approx q holds in UU, (3.2)–(3.4) say that the one-letter summands agree and each long summand of either side is covered by a long summand on the other side with the same first variable and smaller content. Thus L,E,H,N,JL,E,H,N,J agree, and so do T,QT,Q. This applies in particular to any substituted and contextual instance of an identity of UU.

Consider a rule l≤rl \le r, regarded as l+r≈rl+r \approx r, and a substitution φ\varphi. Expanding a one-hole context by distributivity expresses its polynomial as a common part together with finitely many pieces λφ(l+r)ρ\lambda\varphi(l+r)\rho or λφ(r)ρ\lambda\varphi(r)\rho. Here λ,ρ\lambda,\rho are possibly empty words; this notation does not introduce constants into the signature. Since the second expansion is contained in the first, it suffices to transfer every witness supplied by a deleted summand to a summand common to both sides.

For α,β,γ,δ\alpha,\beta,\gamma,\delta, a deleted summand has the form λabρ\lambda ab\rho, with a∈φ(x)a \in\varphi(x), b∈φ(y)b \in\varphi(y). For any d∈φ(z)d \in\varphi(z), the common expansion contains λadρ\lambda ad\rho. These are long words with the same first variable. For η,σ\eta,\sigma, a deleted summand has the same form, but now a,b∈φ(x)a,b \in\varphi(x) are chosen independently; choose d∈φ(y)d \in\varphi(y) to get the same protection. All substituted polynomials are nonempty, so dd exists. Thus all six rules preserve EE, and they preserve LL because every deleted summand is long.

For α\alpha, the common summand λaρ\lambda a\rho has the same first variable and smaller content than λabρ\lambda ab\rho, so it protects HH. The common long summand λb2ρ\lambda b^{2}\rho protects NN, since its content is contained in that of the deleted summand. This also proves NN-preservation for β\beta. For η\eta, λaρ\lambda a\rho protects HH, even when the two selected words a,b∈φ(x)a,b \in\varphi(x) differ. For γ\gamma, λbρ\lambda b\rho protects JJ.

For δ\delta, suppose the deleted summand witnesses NCN_C. The common summand λbρ\lambda b\rho also has content in CC. If it is long, it witnesses NCN_C. Otherwise λ=ρ=ε\lambda=\rho=\varepsilon and b=kb=k is a variable in CC. The common right-hand side contains both kk, from φ(y)\varphi(y), and kdkd, from φ(yz)\varphi(yz). Hence k∈C∩L∩Ek \in C \cap L \cap E, which witnesses TCT_C. Already existing TCT_C-witnesses are preserved because L,EL,E are. Therefore QC=NC∨TCQ_C=N_C \lor T_C is preserved.

Applying these transfers to every deleted summand proves equality of the specified data on the two sides of each elementary replacement. It applies to replacements in either direction. Induction on an equational derivation proves the assertion. □

Theorem 4.2 (exact relative criteria). For every θ\theta in (4.1), absorption of a one-letter word in UθU_\theta is characterized by (3.2). For a long word u=wh,Cu=w_{h,C}, the necessary and sufficient conditions for Uθ⊨u≤pU_\theta\models u \le p are

varietyconditionUαh∈E(p), Hh,C(p), NC(p)Uβh∈E(p), NC(p)Uηh∈E(p), Hh,C(p)Uδh∈E(p), NC(p)∨TC(p)Uγh∈E(p), JC(p)Uσh∈E(p)(14)\begin{array}{c|l} \textit{variety} & \textit{condition} \\ U_{\alpha} & h \in E(p),\ H_{h,C}(p),\ N_C(p) \\ U_{\beta} & h \in E(p),\ N_C(p) \\ U_{\eta} & h \in E(p),\ H_{h,C}(p) \\ U_{\delta} & h \in E(p),\ N_C(p) \lor T_C(p) \\ U_{\gamma} & h \in E(p),\ J_C(p) \\ U_{\sigma} & h \in E(p) \tag*{(14)} \end{array}

In particular, Uσ=AU_\sigma=A.

Proof. The necessity of x∈L(p)x \in L(p) is (3.2), and the necessity of h∈E(p)h \in E(p) is (3.3). The remaining necessities follow from Lemma 4.1: the left-hand word itself witnesses the relevant data, and every derivation from the defining rule must preserve that data. Equational completeness turns this non-derivability statement into failure in the corresponding variety.

For sufficiency, choose a long summand g∈pg \in p with first variable hh, and choose an auxiliary word tt whose content contains all variables needed to ensure ht≤ght \le g. For σ\sigma, substitute x=x=

uu, y=ty=t and use u2≈uu^{2}\approx u to obtain u≤ut≤pu\le ut\le p. If JC(p)J_C(p) holds, choose a witness vv and use x=u,y=v,z=tx=u,y=v,z=t in γ\gamma; since c(v)⊆Cc(v)\subseteq C, uv≈uuv\approx u, and hence u≤v+ut≤pu\le v+ut\le p. The same substitution with an internal long witness and v2≈vv^{2}\approx v proves the β\beta-case.

For η\eta, if HH is witnessed by a long word, (8) already applies. Otherwise h∈L(p)h\in L(p); with x=h,y=tx=h,y=t, the identity x2≤x+xyx^{2}\le x+xy, together with u≤h2u\le h^{2} and ht≤ght\le g, gives the result. For α\alpha, the only additional case is again h∈L(p)h\in L(p). Choose an internal long witness vv for NN, use x=h,y=v,z=tx=h,y=v,z=t, and obtain

u≤hv≤h+v2+ht≈h+v+ht≤p.u\le hv\le h+v^{2}+ht\approx h+v+ht\le p.

Finally suppose that δ\delta applies. If NN holds, use an internal long witness vv, substitute x=u,y=v,z=tx=u,y=v,z=t, and use vt≤vvt\le v to get

u≈uv≤v+ut+vt≤p.u\approx uv\le v+ut+vt\le p.

If TT holds, choose k∈C∩L(p)∩E(p)k\in C\cap L(p)\cap E(p), and choose a long right-hand word gkg_k with first variable kk. Taking tt to contain the contents of both gg and gkg_k, the substitution x=u,y=k,z=tx=u,y=k,z=t gives

u≈uk≤k+ut+kt≤p.u\approx uk\le k+ut+kt\le p.

This proves all rows of (14). The last row is exactly the endpoint criterion (7), so Uσ=AU_{\sigma}=A.

The six-type reduction

Theorem 5.1 (six types). Let uu be a word and pp a polynomial. Suppose that A⊨u≤pA\models u\le p but U⊭u≤pU\not\models u\le p. Then the absorption is equivalent, over UU, to exactly one of the six rules in (11). More precisely, with u=wh,Cu=w_{h,C}, the six mutually exclusive cases are

conditionequivalent ruleH∧NαH∧¬Nη¬H∧Nβ¬H∧¬N∧Tδ¬H∧¬N∧¬T∧Jγ¬Jσ(15)\begin{array}{c|c} \text{condition} & \text{equivalent rule}\\ H\land N & \alpha\\ H\land\neg N & \eta\\ \neg H\land N & \beta\\ \neg H\land\neg N\land T & \delta\\ \neg H\land\neg N\land\neg T\land J & \gamma\\ \neg J & \sigma \tag*{(15)} \end{array}

Here all predicates refer to (h,C,p)(h,C,p).

Proof. By Lemma 3.1, uu is long and h∈E(p)h\in E(p). Call a word internal if its content is contained in CC, and external otherwise. The failure of (8) says that no internal long right-hand word starts with hh. Hence, in the strict situation, HH is equivalent to h∈L(p)h\in L(p). In particular HH implies TT, and NN or TT implies JJ, proving exhaustiveness and mutual exclusivity of the cases in (15).

In each case below, a substitution φ\varphi gives φ(u)≤φ(p)≤R\varphi(u)\le\varphi(p)\le R; the second absorption is derived using only the identities of UU.

The cases with NN. An internal long right-hand word starts with a variable other than hh, so ∣C∣≥2|C|\ge2. Use

φ(k)={x,k=h,y2,k∈C∖{h},xz,k∉C.\varphi(k)= \begin{cases} x, & k=h,\\ y^{2}, & k\in C\setminus\{h\},\\ xz, & k\notin C. \end{cases}

The left side normalizes to xyxy. Internal long words and internal one-letter words other than hh are absorbed by y2y^{2}. An external word whose image starts with xx contains zz and is absorbed by xzxz; if its image starts with yy, it is absorbed by y2y^{2}. The possible one-letter summand hh contributes xx. Thus R=x+y2+xzR=x+y^{2}+xz if HH holds and R=y2+xzR=y^{2}+xz otherwise. These are α\alpha and β\beta, respectively.

The case H∧¬NH\land\neg N. Map internal variables to xx and external variables to xyxy. The left side becomes x2x^{2}. Every internal right-hand word is one-letter and becomes xx, while every external word becomes a long word with first variable xx and content {x,y}\{x,y\}. Thus R=x+xyR=x+xy, giving η\eta.

The case ¬H∧¬N∧T\neg H \land\neg N \land T. Put M=C∩L(p)M = C \cap L(p). Here M≠∅M \ne\varnothing and h∉Mh \notin M. Use

φ(k)={y,k∈M,x,k∈C∖M,xz,k∉C.\varphi(k) = \begin{cases} y, & k \in M,\\ x, & k \in C \setminus M,\\ xz, & k \notin C. \end{cases}

The left side becomes xyxy, since its content includes hh and some element of MM. Internal right-hand words are one-letter summands from MM, so they become yy. External words whose first variable is in MM are absorbed by yzyz; all other external words are absorbed by xzxz. Thus R=y+xz+yzR = y + xz + yz, giving δ\delta.

The case ¬H∧¬N∧¬T∧J\neg H \land\neg N \land\neg T \land J. Since JJ holds and NN fails, M=C∩L(p)M = C \cap L(p) is again nonempty. Use the preceding substitution. Now M∩E(p)=∅M \cap E(p) = \varnothing, since TT fails. No external long word starts in MM, and an external one-letter word cannot lie in MM. Consequently R=y+xzR = y + xz, giving γ\gamma.

The case ¬J\neg J. Map internal variables to xx and external variables to xyxy. Every right-hand word contains an external variable and normalizes to xyxy, whereas the left side normalizes to x2x^2. This gives σ\sigma.

In each row, the indicated rule implies the original absorption by Theorem 4.2. The rules are pairwise inequivalent, as will follow from the separation argument in Theorem 6.1. □\square

The seven-element interval

Theorem 6.1 (the lattice). Under the nondegeneracy assumption (3.1),

I(S)={U,Uα,Uβ,Uη,Uδ,Uγ,Uσ=A},(16)I(S) = \{U, U_{\alpha}, U_{\beta}, U_{\eta}, U_{\delta}, U_{\gamma}, U_{\sigma} = A\}, \tag*{(16)}

and the covering relations are

U≻Uα,Uα≻Uβ,Uη,U \succ U_{\alpha}, \qquad U_{\alpha} \succ U_{\beta}, U_{\eta},
Uβ,Uη≻Uδ,Uδ≻Uγ≻A.(17)U_{\beta}, U_{\eta} \succ U_{\delta}, \qquad U_{\delta} \succ U_{\gamma} \succ A. \tag*{(17)}

The varieties UβU_{\beta} and UηU_{\eta} are incomparable. The Hasse diagram is shown in Figure 1.

Hasse diagram of the seven-element interval

Figure 1. The seven-element interval, ordered by inclusion.

Proof. The implications between the six rules are

σ⟹γ⟹δ,δ⟹β,η,β,η⟹α.(18)\sigma\Longrightarrow\gamma\Longrightarrow\delta,\qquad\delta\Longrightarrow\beta,\eta,\qquad\beta,\eta\Longrightarrow\alpha. \tag*{(18)}

For example, in δ\delta put y=xy = x, z=yz = y to obtain η\eta; put y=y2y = y^2 and use the upper identities to obtain β\beta. Conversely, β\beta and η\eta give

xy≤y2+xz≤y+yz+xz,xy \le y^2 + xz \le y + yz + xz,

so together they are equivalent to δ\delta. Also η\eta implies xy≤x2≤x+xzxy \le x^2 \le x + xz, hence α\alpha.

Let WW be any variety in the interval. An identity between sums is equivalent to its word absorptions. Since A⊆WA \subseteq W, every absorption valid in WW is valid in AA. Remove the absorptions already valid in UU. Theorem 5.1 replaces each remaining absorption by one of the six rules, and (6.3) reduces every set of rules to no rule or to exactly one of the six displayed nodes. Thus there are at most seven varieties. All seven candidates are distinct. The rule α\alpha fails in UU under x={c}x=\{c\}, y={b}y=\{b\}, z=∅z=\varnothing in (1.2): the left side contains aa and the right side does not. For the other separations, the predicates associated with the left-hand word and right-hand polynomial of each rule are

HHNNTTJJ
α\alpha1111
β\beta0101
η\eta1011
δ\delta0011
γ\gamma0001
σ\sigma0000

Table 1 (PDF p. 8).

The first-variable condition h∈E(p)h \in E(p) holds in every row. Thus (14) shows that β\beta fails in both UαU_\alpha and UηU_\eta, while η\eta fails in both UαU_\alpha and UβU_\beta. This proves incomparability of Uβ,UηU_\beta,U_\eta. The rule δ\delta fails in each of these two varieties, γ\gamma fails in UδU_\delta, and σ\sigma fails in UγU_\gamma. All relations in (17) are therefore strict. Since the seven candidates exhaust the interval, they are covering relations. □

Corollary 6.2 (finite bases). Relative to AI, the seven varieties have the finite bases

B,B∪{θ}(θ∈{α,β,η,δ,γ,σ}),(19)\mathcal{B}, \qquad\mathcal{B} \cup\{\theta\}\quad(\theta\in\{\alpha,\beta,\eta,\delta,\gamma,\sigma\}), \tag*{(19)}

where Uσ=AU_\sigma=A. Every displayed basis uses at most three variables.

Proof. Combine Proposition 3.2 with Theorem 6.1. The standard axioms for AI also use at most three variables. □

Proposition 6.3 (free-algebra counts). For n≥1n \ge1, let FW(n)F_W(n) be the free nn-generated algebra in a variety WW in the seven-element interval. Let dmd_m be the number of antichains, including the empty antichain, in the subset lattice of an mm-element set. Then

∣FU(n)∣=(2dn−1−1)n−1,∣FA(n)∣=4n−1.(20)\lvert F_U(n)\rvert=(2d_{n-1}-1)^n-1,\qquad\lvert F_A(n)\rvert=4^n-1. \tag*{(20)}

In particular,

∣FU(3)∣=1727,∣FA(3)∣=63.\lvert F_U(3)\rvert=1727,\qquad\lvert F_A(3)\rvert=63.

Proof. For a fixed first variable hh, the contents of long words all contain hh. Removing hh, the inclusion-minimal contents that remain form an antichain in the subset lattice on the other n−1n-1 variables. The upper criterion (8) shows that this antichain uniquely determines the long-word contribution with first variable hh, up to the identities of UU; the one-letter summand hh is an independent choice. Thus there are 2dn−12d_{n-1} choices for each hh. The criteria distinguish all these choices, and every choice is represented by a polynomial unless no summand at all is selected. This proves the first count. In AA, only the independent choices of membership in LL and in EE remain, giving 4n−14^n-1. For n=3n=3, the two-element subset lattice has six antichains, so (6.5) gives 123−1=172712^3-1=1727 and 43−1=634^3-1=63. □

Corollary 6.4 (finite generation by a three-generated algebra). For every W∈I(S)W \in I(S),

W=V(FW(3)),∣FW(3)∣≤1727.(21)W=V(F_W(3)),\qquad\lvert F_W(3)\rvert\le1727. \tag*{(21)}

Proof. The algebra FW(3)F_W(3) is a quotient of FU(3)F_U(3), hence finite. The subsets {b}\{b\} and {c}\{c\} generate P+(S∗)P^+(S^*), since {c}2={a}\{c\}^2=\{a\} and addition is union. Since P+(S∗)∈A⊆WP^+(S^*) \in A \subseteq W, freeness makes it a homomorphic image of FW(3)F_W(3). Lemma 3.1 therefore gives A⊆V(FW(3))⊆W⊆UA \subseteq V(F_W(3)) \subseteq W \subseteq U. By freeness, FW(3)F_W(3) and WW have the same identities in at most three variables. The rules in (4.1) separate all seven interval elements, so no two distinct nodes can occur between AA and UU with the same three-variable theory. Thus V(FW(3))=WV(F_W(3))=W. □

The degenerate cases

Let 2={0,1}\mathbf{2}=\{0,1\}, ∨\vee, ∧\wedge and D=V(2)D=V(\mathbf{2}). In 2\mathbf{2}, a word is the meet of its variables, and hence

D⊨u≤p⟺∃v∈p  c(v)⊆c(u).(22)D \models u \le p \quad\Longleftrightarrow\quad\exists v \in p\; c(v)\subseteq c(u). \tag*{(22)}

This criterion follows by assigning variables in c(u)c(u) the value one and all other variables zero.

The one-element semigroup

If ∣S∣=1|S| = 1, then P+(S)\mathcal{P}^{+}(S) is the one-element semiring and P(S)≅2\mathcal{P}(S) \cong2. Every nontrivial distributive lattice contains a two-element sublattice: from two distinct elements, take their meet if necessary and retain a comparable pair. Thus every nontrivial subvariety of D contains 22, and the interval has exactly two elements.

Nontrivial left-zero semigroups

Suppose f=id⁡Sf = \operatorname{id}_{S} and ∣S∣≥2|S| \ge2. Write AL=V(P+(S))A_{L} = \mathsf{V}(\mathcal{P}^{+}(S)) and UL=V(P(S))U_{L} = \mathsf{V}(\mathcal{P}(S)). Every word has the value of its first variable when its content variables are nonempty, and is empty otherwise. Consequently,

AL⊨u≤p⟺h(u)∈{h(v):v∈p},A_{L} \models u \le p \Longleftrightarrow h(u) \in\{h(v) : v \in p\},
UL⊨u≤p⟺∃v∈p [h(v)=h(u), c(v)⊆c(u)].(23)U_{L} \models u \le p \Longleftrightarrow\exists v \in p\,[h(v) = h(u),\ c(v) \subseteq c(u)]. \tag*{(23)}

For necessity of the second criterion, take distinct a,b∈Sa, b \in S; assign the first variable of uu the set {b}\{b\}, the other content variables {a}\{a\}, and outside variables the empty set. If no covering word exists, the right side avoids bb. For the lower criterion use nonempty singleton assignments throughout. The sufficiency follows from the word evaluation. In particular AL=Mod⁡ai{xy≈x}A_{L} = \operatorname{Mod}_{\mathrm{ai}}\{xy \approx x\}: all words reduce to their first variables, and the criterion distinguishes the remaining sums. The upper endpoint has the relative basis

x2≈x,xyz≈xzy,xy≤x.(24)x^{2} \approx x,\qquad xyz \approx xzy,\qquad xy \le x. \tag*{(24)}

Put ρL:xz≤xy+z\rho_{L} : xz \le xy + z. Then

I(S)={AL, AL∨D, UL},I(S) = \{A_{L},\ A_{L} \vee D,\ U_{L}\},
AL∨D=UL∩Mod⁡{ρL}.(25)A_{L} \vee D = U_{L} \cap\operatorname{Mod}\{\rho_{L}\}. \tag*{(25)}

To justify the basis (7.3), tail permutation and idempotence reduce each word to its first variable and content. The inequality xy≤xxy \le x absorbs larger content into smaller content with the same first variable. The upper criterion in (7.2) proves completeness.

We now prove (7.4). The subsemiring {∅,{a}}\{\varnothing, \{a\}\} gives D⊆ULD \subseteq U_{L}, and ρL\rho_{L} holds in both ALA_{L} and DD. If an absorption u≤pu \le p holds in these two varieties, choose a right-hand word gg starting with h=h(u)h = h(u) and a right-hand word vv with c(v)⊆c(u)c(v) \subseteq c(u), using (7.2) and (7.1). If g=hg = h, the upper criterion already proves the absorption. Otherwise write g=htg = ht and use ρL\rho_{L} with x=h,y=t,z=vx = h, y = t, z = v:

u≤hv≤ht+v=g+v≤p.u \le hv \le ht + v = g + v \le p.

It follows that all identities of AL∨DA_{L} \vee D follow from (7.3) and ρL\rho_{L}, proving the equality in (7.4).

For exhaustiveness, take any strict absorption valid in ALA_{L} but not ULU_{L}. Put h=h(u)h = h(u) and C=c(u)C = c(u). Map hh to xx, variables in C∖{h}C \setminus\{h\} to zz, and outside variables to xyxy. A right-hand word starting with hh must be external, so its image is absorbed by xyxy. Words starting with other internal variables are absorbed by zz; those starting with outside variables are absorbed by xyxy. The left side becomes xzxz if C≠{h}C \ne\{h\}, and xx otherwise. In the latter case xz≤xxz \le x still yields ρL\rho_{L}. Thus every proper subvariety of ULU_{L} containing ALA_{L} is contained in AL∨DA_{L} \vee D.

If an absorption valid in ALA_{L} fails in AL∨DA_{L} \vee D, it fails in DD; by (7.1) every right-hand word is external. Mapping internal variables to xx and external variables to xyxy gives x≤xyx \le xy. Together with (7.3) this yields xy≈xxy \approx x, and hence ALA_{L}. Any proper subvariety of AL∨DA_{L} \vee D containing ALA_{L} has such a separating absorption, so must equal ALA_{L}. Finally D⊈ALD \not\subseteq A_{L}, because 22 fails xy≈xxy \approx x; and ρL\rho_{L} fails in ULU_{L} at x={a},y=∅,z={b}x = \{a\}, y = \varnothing, z = \{b\}. This proves the strict three-element chain.

Nontrivial null semigroups

Suppose ∣f(S)∣=1|f(S)| = 1, write oo for the common product, and assume ∣S∣≥2|S| \ge2. Let AN,UNA_{N}, U_{N} denote the two endpoints. A long word evaluates to {o}\{o\} when all its variables are nonempty and to the empty set otherwise. Hence

AN⊨u≤p⟺p has a long summand,A_{N} \models u \le p \Longleftrightarrow p\text{ has a long summand},
UN⊨u≤p⟺∃v∈p [∣v∣≥2, c(v)⊆c(u)](26)U_{N} \models u \le p \Longleftrightarrow\exists v \in p\,[|v| \ge2,\ c(v) \subseteq c(u)] \tag*{(26)}

for every long word uu. For both endpoints a one-letter absorption x≤px \le p holds exactly when x∈L(p)x \in L(p). To see its necessity, choose a≠oa \ne o, assign xx the value {a}\{a\}, and all other variables {o}\{o\}. For necessity in (7.5), assign every internal variable {a}\{a\}; use {a}\{a\} also outside c(u)c(u) for the lower endpoint, and the empty set outside c(u)c(u) for the upper endpoint. When the corresponding condition fails, the right side avoids oo. The upper endpoint has the relative basis

xy≈yx,x2≈x3,xy≈x2y2,x2y≤x2.(27)xy \approx yx,\quad x^2 \approx x^3,\quad xy \approx x^2y^2,\quad x^2y \le x^2. \tag*{(27)}

Let ρN:x2≤x+y2\rho_N: x^2 \le x+y^2. Then

I(S)={AN,AN∨D,UN},(28)I(S)=\{A_N,A_N\vee D,U_N\}, \tag*{(28)}
AN∨D=UN∩Mod⁡{ρN}.A_N\vee D=U_N\cap\operatorname{Mod}\{\rho_N\}.

For completeness of (27), commutativity and the third identity let us square all letters in a long word; the second identity removes higher powers. Thus a long word with content CC reduces to ∏x∈Cx2\prod_{x\in C}x^2. The last identity absorbs larger content into smaller content. These reductions derive exactly (26), and the one-letter criterion then proves completeness.

The subsemiring {∅,{o}}\{\varnothing,\{o\}\} gives D⊆UND\subseteq U_N, and ρN\rho_N holds in both ANA_N and DD. If a long absorption holds in both, (26) supplies a long right-hand word and (22) supplies an internal right-hand word. An internal long word already gives the upper criterion. Otherwise choose an internal one-letter summand kk and a long right-hand summand vv. Substitution x=k,y=vx=k,y=v in ρN\rho_N gives

u≤k2≤k+v2≈k+v≤p.u\le k^2\le k+v^2\approx k+v\le p.

The one-letter absorptions are already upper identities, so this proves the relative basis in (28).

For a strict absorption valid in ANA_N but not UNU_N, the left side must be long and there is no internal long right-hand word. Map variables in its content to xx and all outside variables to y2y^2. The left side becomes x2x^2. Internal one-letter summands become xx, and every external summand is absorbed by y2y^2. Thus the absorption implies ρN\rho_N, so every proper intermediate variety is contained in AN∨DA_N\vee D.

If an absorption valid in ANA_N fails in AN∨DA_N\vee D, it fails in DD, and therefore has no internal right-hand word. The same substitution gives x2≤y2x^2\le y^2. Swapping variables gives equality of all squares. By (27), for any x,yx,y we have xy≈x2y2≈x4≈x2xy\approx x^2y^2\approx x^4\approx x^2; all long products consequently coincide. Every identity of ANA_N now follows from (26) and the one-letter criterion, so this additional identity defines ANA_N over UNU_N. This excludes any further variety below AN∨DA_N\vee D.

Finally D⊈AND\not\subseteq A_N, since x2≈y2x^2\approx y^2 holds in ANA_N but fails in 2\mathbf{2}; and ρN\rho_N fails in UNU_N at x={a},y=∅x=\{a\},y=\varnothing. This proves (28) and completes the proof of Theorem 2.1.

Two-generated free algebras and basis ranks

Throughout this section assume (5). Let LL and NN be a two-element left-zero semigroup and a two-element null semigroup, respectively. Write UL=V(P(L))U_L=V(\mathcal{P}(L)) and UN=V(P(N))U_N=V(\mathcal{P}(N)), and retain D=V(2)D=V(\mathbf{2}) from Section 7.

Proposition 8.1 (join descriptions). The intermediate varieties admit the descriptions

Uγ=A∨D,Uβ=A∨UN,Uη=A∨UL,Uα=A∨UN∨UL,Uδ=(A∨UN)∩(A∨UL).(29)\begin{aligned} U_\gamma&=A\vee D, & U_\beta&=A\vee U_N,\\ U_\eta&=A\vee U_L, & U_\alpha&=A\vee U_N\vee U_L,\\ U_\delta&=(A\vee U_N)\cap(A\vee U_L). \tag*{(29)} \end{aligned}

Proof. An identity holds in a join of varieties if and only if it holds in each of them. For a long word u=wh,Cu=w_{h,C}, the lower endpoint requires h∈E(p)h\in E(p) for the absorption u≤pu\le p. The additional requirements from D,UN,ULD,U_N,U_L are, respectively, JC(p),NC(p),Hh,C(p)J_C(p),N_C(p),H_{h,C}(p), by (22), (26), and (23). These are exactly the rows for Uγ,Uβ,UηU_\gamma,U_\beta,U_\eta in (14). For a one-letter word, the condition from AA already requires that word to occur as a summand, which suffices in every variety. This proves the first three descriptions. The last two follow from Theorem 6.1.

Lemma 8.2 (two-variable theories). There are natural isomorphisms

FU(2)≅FUα(2)≅FUβ(2),F_U(2)\cong F_{U_\alpha}(2)\cong F_{U_\beta}(2),
FUη(2)≅FUδ(2)≅FUγ(2).(30)F_{U_\eta}(2)\cong F_{U_\delta}(2)\cong F_{U_\gamma}(2). \tag*{(30)}

The orders are 35 and 24, respectively, while ∣FA(2)∣=15|F_A(2)|=15.

Proof. Consider an absorption u≤pu \le p in at most two variables that holds in AA but fails in UU. By Lemma 3.1, uu is long and some long summand of pp starts with h=h(u)h = h(u). If c(u)c(u) contained every variable occurring in the absorption, that summand would give the upper criterion, a contradiction. Thus c(u)={h}c(u) = \{h\}. There is no internal long right-hand word, and the predicates HH, TT, JJ all reduce to the condition h∈L(p)h \in L(p). Theorem 5.1 therefore assigns such an absorption to η\eta or σ\sigma, according as that condition holds or fails. The varieties UU, UαU_{\alpha}, UβU_{\beta} satisfy neither of these rules, and UηU_{\eta}, UδU_{\delta}, UγU_{\gamma} satisfy η\eta but not σ\sigma. They consequently have the same identities in at most two variables within each of these two groups. Their free algebras on two named generators have identical defining congruences, proving (8.2).

The values 35 and 15 follow from Proposition 6.3, since d1=3d_1 = 3. To obtain 24, write the two generators as hh, kk. The six possible contributions with first variable hh in an upper normal form are: no summand, hh, h2h^2, hkhk, h+h2h + h^2, and h+hkh + hk. The rule η\eta identifies precisely the last two contributions. Indeed hk≤h2hk \le h^2 holds in UU, and h2≤h+hkh^2 \le h + hk gives the reverse absorption after adding hh. The criteria in Theorem 4.2 distinguish the other choices independently for each first variable: LL, EE distinguish all but h2h^2 and hkhk, and these remain distinct when hh is absent as a one-letter summand. There are therefore five choices for each of the two first variables. Excluding the choice with no summand at all gives 52−1=245^2 - 1 = 24.

For a variety WW generated by one of its finite-rank free algebras, define its basis rank by

br⁡(W)=min⁡{n≥1:W=V(FW(n))}.\operatorname{br}(W) = \min\{n \ge1 : W = \mathsf{V}(F_W(n))\}.

This measures free generation of a variety; it is distinct from the number of variables required in an identity basis.

Theorem 8.3 (exact basis ranks). The two-generated free algebras and basis ranks of the seven interval elements are as follows:

WW∣FW(2)∣\lvert F_W(2)\rvertV(FW(2))\mathsf{V}(F_W(2))br⁡(W)\operatorname{br}(W)
UU35UβU_{\beta}3
UαU_{\alpha}35UβU_{\beta}3
UβU_{\beta}35UβU_{\beta}2
UηU_{\eta}24UγU_{\gamma}3
UδU_{\delta}24UγU_{\gamma}3
UγU_{\gamma}24UγU_{\gamma}2
AA15AA2

Table 8.3.

Proof. As observed in the proof of Corollary 6.4, the semiring P+(S∗)\mathcal{P}^{+}(S_*) is generated by {b}\{b\}, {c}\{c\}. Consequently AA is generated by a two-generated algebra. Write N={o,d}N = \{o, d\}, where every product equals oo. The semiring P(N)\mathcal{P}(N) is generated by ∅\varnothing, {d}\{d\}, since {d}2={o}\{d\}^2 = \{o\}. The semiring 2\mathbf{2} is generated by its two elements.

Every two-generated member of WW is a homomorphic image of FW(2)F_W(2). The descriptions of UβU_{\beta} and UγU_{\gamma} in Proposition 8.1 therefore give

V(FA(2))=A,V(FUβ(2))=Uβ,V(FUγ(2))=Uγ.\mathsf{V}(F_A(2)) = A,\qquad\mathsf{V}(F_{U_{\beta}}(2)) = U_{\beta},\qquad\mathsf{V}(F_{U_{\gamma}}(2)) = U_{\gamma}.

Lemma 8.2 now proves the two middle columns of (8.3) for all seven varieties. For completeness, basis rank one is impossible here. Proposition 6.3 gives ∣FU(1)∣=∣FA(1)∣=3\lvert F_U(1)\rvert= \lvert F_A(1)\rvert= 3. Every FW(1)F_W(1) between them is thus the same three-element algebra, represented by xx, x2x^2, x+x2x + x^2; every product in this algebra equals x2x^2. In particular it is commutative, while A⊆WA \subseteq W contains the noncommutative semiring P+(S∗)\mathcal{P}^{+}(S_*). Thus br⁡(W)≥2\operatorname{br}(W) \ge2. The strict inclusions in Theorem 6.1 and (8.3) exclude rank two for exactly four nodes. Corollary 6.4 supplies rank at most three for each, completing the table.

The opposite semigroup

The classification also applies to semigroups satisfying xy≈y2xy \approx y^2. If Sop⁡S^{\operatorname{op}} denotes the opposite semigroup, then

P+(Sop⁡)≅P+(S)op⁡,P(Sop⁡)≅P(S)op⁡.\mathcal{P}^{+}(S^{\operatorname{op}}) \cong\mathcal{P}^{+}(S)^{\operatorname{op}}, \qquad\mathcal{P}(S^{\operatorname{op}}) \cong\mathcal{P}(S)^{\operatorname{op}}.

Reversing multiplication induces a lattice isomorphism I(S)≅I(Sop⁡)I(S) \cong I(S^{\operatorname{op}}). Theorem 2.1 therefore gives the same 2, 3, 7 classification, with right-zero semigroups in place of left-zero semigroups.

Acknowledgements

This work was supported by the Science and Technology Research Program of the Chongqing Municipal Education Commission (Grant No. KJZD-K2024011102) and the Joint Fund for Innovation and Development of the Chongqing Natural Science Foundation (Chongqing Municipal Education Commission) (Grant No. CSTB2025NSCQ-LZX0067).

References

(Aifa Wang) School of Mathematical Sciences, Chongqing University of Technology, Chongqing 400054, P.R. China

Email address: [email protected]

(Lili Wang) School of Mathematical Sciences, Chongqing University of Technology, Chongqing 400054, P.R. China

Email address: [email protected]

(Qingrui Yin) School of Mathematical Sciences, Chongqing University of Technology, Chongqing 400054, P.R. China

Email address: [email protected]

References

  1. [1]Stanley Burris and H. P. Sankappanavar, A course in universal algebra, Graduate Texts in Mathematics, vol. 78, Springer-Verlag, New York, 1981.
  2. [2]Zidong Gao, Miaomiao Ren, and Xianzhong Zhao, A continuum in the lattice of semiring varieties: the interval [V(S), V(S⁰)], Preprint, 2026, arXiv:2609.24514.arxiv.org/abs/2609.24514
  3. [3]Martin Kuřil, Admissible closure operators and varieties of semilattice-ordered normal bands, Acta Sci. Math. (Szeged) 83 (2017), no. 1–2, 35–50, doi:10.1432/actasm-016-777-4.DOI
  4. [4]Martin Kuřil and Libor Polák, On varieties of semilattice-ordered semigroups, Semigroup Forum 71 (2005), no. 1, 27–48.DOI
  5. [5]Miaomiao Ren and Xianzhong Zhao, The varieties of semilattice-ordered semigroups satisfying x³ ≈ x and xy ≈ yx, Period. Math. Hungar. 72 (2016), no. 2, 158–170.DOI
  6. [6]Xiaolei Shao and Zidong Gao, Nonfinitely based intervals of power semiring varieties, Preprint, 2026, arXiv:2609.26019.arxiv.org/abs/2609.26019
  7. [7]Y. N. Wu, Miaomiao Ren, and Xianzhong Zhao, The additively idempotent semiring S₇⁰ is nonfinitely based, Semigroup Forum 108 (2024), no. 2, 479–487.
  8. [8]Mengya Yue and Miaomiao Ren, Every additively idempotent semiring satisfying xy ≈ xz is finitely based, Preprint, 2025, arXiv:2509.16711.arxiv.org/abs/2509.16711
  9. [9]Mengya Yue, Miaomiao Ren, and Zidong Gao, Explicit equational bases for the power semirings of S₇, Preprint, 2026, arXiv:2609.19957.arxiv.org/abs/2609.19957
  10. [10]Mengya Yue, Miaomiao Ren, L. L. Zeng, and Y. Shao, The finite basis problem for additively idempotent semirings of order four, II, Algebra Universalis (2025), doi:10.1007/s00012-025-00908-5.

Paper details

vlnc:W_fgsde7px
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents