Arxiv:1602.00353V6 [Math.RA] 6 May 2021

Home , Magma (algebra)

arXiv:1602.00353v6 [math.RA] 6 May 2021 lolk otakteUiest fVrii o t support its for Virginia of University the thank grea to like a provided also Maria who to referee, involved, the been to first has especially the point and he of clarified, versions, which reading various with careful in papers a L inaccuracies interesting for Sergio several Shachar and finding Guy hyperfield,” for to “sign due the are of thanks role the for Knebusch especially Max and examples, interesting for and metrization 13A18. o icsin ntoiaiain aineAinadSte and Akian Marianne tropicalization, on discussions for al semiring Lie module, matrix, monoid, algebra, semigroup, algebra, cong linear pertropical symmetrization, product, system, tensor triple, polynomial, map, negation gible, 2010 Date 1 1 h uhrsrsac a upre yIre cec Found Science Israel by supported was research author’s The ..Sprrpclsmrnsadsprrpcldomains Symmetrization supertropical and semirings semirings 3.3. bipotent Supertropical and systems algebra and max-plus 3.2. triples The oriented tropically main 3.1. The multiplication 3. paper Redefining this of results 2.8. Main paper this of 2.7. Organization 2.6. Systems relations 2.5. Surpassing triples and 2.4. Pseudo-triples maps 2.3. Negation 2.2. .Amr ealdoverview detailed more 2.1. notation A ongoing and definitions 2. Basic overview 1.2. Short 1.1. Introduction 1. e od n phrases. and words Key h uhrwudlk otaktefloigrsacesfor researchers following the thank to like would author The hspproiial a nild“ymtisi tropical in “Symmetries entitled was originally paper This a ,2021. 7, May : ahmtc ujc Classification. Subject Mathematics Abstract. rpclzto uco,a ela rvdn ikbetwee link a analogs. providing hyperfield as and well via as studied functor, be tropicalization can which structure studie i underlying systems the and The are examples, results. they structural main as of the host a all provide includes and theory algebraic whose epn oepansmlrte nteetere.Special theories. these in similarities symmetrizatio explain as to well helping as versions, (super)tropical viable esuyagbacsrcue ihngto as called furthe maps, pursued negation and with ingre group structures crucial Max-Plus for algebraic the a to study and is trying we negative when Gaubert, the negation Dress, which of of for lack concepts the by algebraic challenged is often T -modules yfraiigtesrcue ecnitouemrhss M morphisms. introduce can we structure, the formalizing By ntoia ahmtc,a ela te ahmtcltheo mathematical other as well as mathematics, tropical In rpclagba rpclgoer,toiaiain Pu tropicalization, geometry, tropical algebra, tropical LERSWT EAINMAP NEGATION A WITH ALGEBRAS rmr 66,1K0 87,2N0 80;Scnay06F05 Secondary 08A05; 20N20, 08A72, 12K10, 16Y60, Primary OI AL ROWEN HALLE LOUIS Contents n u h oncint uz ig,Hnysppr n oth and paper, Henry’s rings, fuzzy to connection the out ing eied upsigrelation. surpassing semifield, , hn abr,tgte ihAiNv o icsin nsym on discussions for Niv, Adi with together Gaubert, phane une yefil,fzyrn,epoe ler,ETalgeb ELT algebra, exploded ring, fuzzy hyperfield, ruence, ayimprovements. many t er,sprler,Gasanagba xeiralgebra exterior algebra, Grassmann superalgebra, gebra, uigteiiilpeaaino hswr n2015. in work this of preparation initial the during nihso bsmrns atBkrfrepann hyperfie explaining for Baker Matt semirings, ub on insights 1 pzfrhlflcmet ntefis ri eso.Special version. arXiv first the on comments helpful for opez n ka o onigothwvrosterm hudbe should theorems various how out pointing for Akian nne ler”i ri:62033[math.RA] arXiv:1602.00353 in algebra” to rnsN.10/2ad19/0 h uhrwould author The 1994/20. and 1207/12 No. grants ation epu ovrain:Oie oshi n u Izhakian Zur and Lorscheid Oliver conversations: helpful ri eso hc e osvrlcretos oJ Jun J. to corrections, several to led which version arXiv ,hprru hoyadfzyrns thereby rings, fuzzy and theory hypergroup n, lsia leri eut n hi tropical their and results algebraic classical n teto spi to paid is attention systems nti ae r gon”sses insofar systems, “ground” are paper this in d in.Floigiesoiiaigi work in originating ideas Following dient. iheog ofcltt computations facilitate to enough rich s te ssei modules”. “systemic other uaetetoia esoso classical of versions tropical the mulate yAin abr,adGuterman, and Gaubert, Akian, by r hwn o hs nf h more the unify these how showing , rhsseal st eciethe describe to us enable orphisms su eis auto,tnil,metatan- tangible, valuation, series, iseux isivligsmrns one semirings, involving ries metatangible systems, 14T05, , su- , lds, ra, 16 15 15 14 14 11 11 10 er 9 8 6 5 5 4 3 2 - 2 LOUIS ROWEN

3.4. *Hypergroups 17 4. Exploring triples and systems 19 4.1. Layered semirings† 19 4.2. Major examples of systems, by height 20 4.3. Tropically related examples viewed in terms of algebraic varieties and model theory 21 4.4. Properties of e, and the characteristic of a system A 22 4.5. T -Strongly negated systems 23 5. Related notions 23 5.1. Matrices over triples and systems 23 5.2. Polynomial triples 24 5.3. Involutions 25 5.4. Localization of semiring triples 26 5.5. Tensor products with a negation map, and their semialgebras 26 6. Metatangible triples and their systems 27 6.1. Basic properties of metatangible triples and systems 27 6.2. Metatangible versus (−)-bipotent 27 6.3. Computations in a metatangible triple (A, T , (−)) via T 29 6.4. Uniform elements and height 31 6.5. Surpassing relations on a metatangible triple 33 6.6. T -classical metatangible triples 34 6.7. Squares and sums of squares 35 6.8. Classifying metatangible systems 37 6.9. Speciﬁc applications of metatangible systems 38 7. Categorical properties of systems 39 7.1. -Morphisms of systems 39 7.2. Valuations on systems 40 7.3. The transfer principle 41 8. Linear algebra over a triple 41 8.1. ◦-identities for matrices 42 8.2. Ranks of matrices 43 9. More applications 44 9.1. Tropicalization of Puiseux series 44 9.2. Exterior (Grassmann) semialgebras 45 10. Nonassociative semialgebras with a negation map 46 10.1. Super-semialgebras 47 10.2. Lie -semialgebras and Lie -super-semialgebras, and their triples 47 10.3. Poisson semialgebras and their module congruences 48 11. Areas for further research 49 12. *Appendix A: Major examples of hypergroups and hyperﬁelds and their power sets 49 13. **Appendix B: Fuzzy rings as systems 51 13.1. Fuzzy rings versus pre-fuzzy triples and systems 52 References 52

1. Introduction This paper, a trimmed and rearranged version of [61], which was in turn based on [60], was born from the desire to understand a mysterious parallel between structural results in what we will call the “classical algebraic theory” and theorems formulated directly in varied aspects of tropical algebra, despite the former dealing with fields and the latter with the max-plus algebra and related semifields. It is designed to lay the foundation for a unified algebraic theory, which also encompasses diverse recent research in hyperfields, fuzzy rings, and especially tropical mathematics. Our objective in this project is three-fold: laying out the underlying foundations, providing applications to the literature, and then developing the structure with an emphasis on representation theory and ALGEBRAS WITH A NEGATION MAP 3 geometry. This paper focuses on the first two parts; the third has involved several additional papers in the last two years, including [62, 4, 5, 23, 48, 49]. We need considerable background to set up these foundations, so this paper has two introductions. The first introduction (§1.1) gives an idea of our goals, but without full definitions, and the second elaborates on them more fully in §2, together with an outline of the organization in §2.6 and a list of main results in §2.7. The main ideas behind the theory are explained in a sequence of 17 notes interspersed throughout the paper, each labeled as a “MAJOR NOTE.”

1.1. Short overview. Tropicalization originally was viewed as a process of taking logarithms and passing in the limiting case to the max-plus algebra, which is a semiring. Thus, tropical algebra customarily has relied on the theory of semirings which goes back to Costa [18] and Eilhauer [22], and for which we use Golan [28] as our standard reference. As the field of Puiseux series came into play, the underlying semiring was viewed as the target of the “Puiseux valuation,” which differs somewhat from the max-plus algebra. Towards this end, in [34, 37, 41, 44] a “supertropical” theory was initiated over a semiring by means of a “ghost map” (where the negation map actually is the identity map), with various applications to affine varieties, matrices, linear algebra, and quadratic forms. Classical algebraic results were transferred to the tropical theory by means of a somewhat mysterious “surpassing” relation on semirings, which satisfies many properties of equality, and replaces equality in many generalizations of classical theorems, especially for polynomials and matrix theory. Thus we are motivated to ask how exactly this surpassing relation fits into the algebraic theory. Viro [68] cast the supertropical theory and other mathematical ideas in terms of hyperfields, where the sum of two elements is a set, rather than a single element. In all of these instances, there is a set T of main interest which we call the set of tangible elements, (e.g., the max-plus semialgebra but without a+a = a, the symmetrization, or a hyperfield), together with a special operation resembling negation (the switch map or hypernegation in the latter two cases). Its intrinsic algebraic structure is not sufficient for a full investigation, necessitating many algebraic results to be formulated and proved on an ad hoc basis. The situation is clarified significantly by tying T to a T -module (A, +) with a richer structure, often via an embedding, studied in accordance with well-known techniques from universal algebra. Due to the fact that T is almost never closed under addition, much of its theory depends on understanding (A, +), even though our ultimate interest lies in T . In many instances, T is a multiplicative monoid, although we keep open the option that T has different structure, possibly nonassociative, for example with Lie multiplication. To give an indication of where we are heading, let us start by listing some classical examples.

Example 1.1. (1) A is an integral domain and T = A\{0}. (2) A is a graded algebra, and T is the monoid of homogeneous elements. (3) A is a vector space with base T . More specifically, A could be an algebra with a multiplicative base T . For example, A could be the group algebra of a group T . (4) A is a Hopf algebra and T is a special subset (such as the group-like elements or primitive elements). For example, A could be the enveloping algebra of a Lie algebra T . (5) A is the set of class functions from a finite group to a field F ; T is the irreducible characters. (6) T is the set of reduced words from an algebra A with a reduction procedure.

On the other hand, our main motivation is from tropical mathematics, whose parent structure in tropical algebra is the well-known max-plus algebra, described thoroughly in [1], and which can be viewed as a semiring, whose zero element (written 0 here) is −∞. But its algebraic structure is inadequate. The most basic trouble with the max-plus algebra is its lack of negation, especially for studying the determinant of matrices. Other shortcomings of the max-plus algebra include failure of unique factorization in polynomials, conflicting definitions in linear algebra, and lack of a multiplicative determinant for matrices. Various researchers, going back to Kuntzman [52] in 1972, tackled this last issue, with decisive results [57, p.352, end of proof of (a)]. As noted earlier, Gaubert [24] offered a remedy in his dissertation, which was continued together with the M. Plus group and Akian and Guterman, using a “symmetry” 4 LOUIS ROWEN

([56], [24], [6, §3.4], [25], [2], [32], [3], and [46, Appendix A]), leading to a general “transfer principle” to generate semiring identities. Here we offer a general “negation map.” Other predecessors of this study include the following, where T is as in this paper: Dress [20] introduced fuzzy rings to study valuated matroids (where T is a multiplicative subgroup). Gaubert [24] doubled T , using “symmetry,” to provide a viable negation. Lorscheid [53] introduced the rather general framework of “blueprints” in terms of semigroup rings. Our goal here is to simplify and unify these approaches with an axiomatic theory which is robust enough to provide features lacking in the max-plus algebra. MAJOR NOTE 1.2. The negation map (−) is the key to our approach, pervading all of the concepts. When T ⊂ A, for every element a ∈ T there is a unique element (−)a for which a + (−)a is a special element of A, called a “quasi-zero” since it replaces the classical element 0. When 1 ∈ T , then (−)b = ((−)1)b for all a ∈ A. The other main notion is a “surpassing relation,” to replace equality in most theorems. Altogether, our structure of choice, a system (Definition 2.36), is a quadruple (A, T , (−), ), where A is a T -module, (−) is a negation map, and is a surpassing relation. Although systems were motivated by tropical algebra, our purpose here is to show that they are justi- fied in having a theory of their own. Systems include the classical case, as well as two key applications: the “standard” supertropical semiring given in Definition 3.4, and the symmetrized semiring (Remark 3.7 and Example 3.11). Other applications include hypergroups (§3.4, Theorem 3.21), fuzzy rings (§6.54), matrices (§8, Theorem 8.6), tensor semialgebras (§5.5), polynomials (§5.2), Grassmann semialgebras (Ex- ample 9.9), the “layered” semiring of [36] (§4.1), bilinear and quadratic forms (Example 4.7), and Poisson semialgebras (§9). Nonassociative examples include Lie semialgebras given in §10.2. Even when multiplication is associative, one might need to relax distributivity, to accommodate application to hyperfields; distributivity of A over elements of T is enough to run the theory. Once we have an organized structure, we talk of the appropriate category and its morphisms in 7.1. Some of the traditional maps, such as tropicalization, fit into this context, and the Lie context in Lemma 10.4 also fits in. But this set-up still requires more hypotheses to obtain strong results. MAJOR NOTE 1.3. In all of our applications, we have “unique quasi-negation” as well as the property, called being metatangible, which says that for a,a′ ∈ T , a′ + a ∈ T iff a′ 6= (−)a. Most of this paper explores the ramifications of metatangibility, which encompasses many areas related to tropical mathematics. 1.2. Basic definitions and ongoing notation. As customary, N denotes the positive natural numbers, N0 denotes N ∪{0}, Q denotes the rational numbers, and R denotes the real numbers, all ordered with respect to addition. Recall that a monoid is a semigroup with a two-sided identity element, denoted as 0 for addition, and as 1 for multiplication. We customarily write ab for a · b. The power set of a set S is denoted P(S). † Definition 1.4. A semiring is a semiring (A, +, ·, 1R) without 0, i.e., an additive Abelian semigroup 1 (A, +) and multiplicative monoid (A, ·, 1R) satisfying the usual distributive laws. A semiring† (A, + , · , 1) is a semifield† if (A, ·) is an Abelian group. Definition 1.5. A semigroup (A, +) has characteristic k> 0 if (k + 1)a = a for all a ∈ A, with k ≥ 1 minimal. A has characteristic 0 if A does not have characteristic k for any k ≥ 1. When 1 ∈T ⊆A it is enough to check that (k + 1)1 = 1. For example, the max-plus algebra has “characteristic 1.” This leads to the notion of “F1 geometry.” Definition 1.6. A pre-order is a transitive and reflexive relation. A partial order (PO) is an antisymmetric pre-order. An order is a total PO. A (pre-)ordered monoid is a monoid M with a (pre-)order satisfying a ≤ b implies ca ≤ cb, ac ≤ bc, ∀a,b,c ∈M. (1.1)

1The reason that we do not always require A to have the element 0 is that 0 just distracts from our true goal, negation maps and quasi-zeros, not negatives. ALGEBRAS WITH A NEGATION MAP 5

1.2.1. Bimagmas. We want to maintain the ability to handle the Lie case. Even when lacking associativity and/or distributivity, our algebraic structures will have addition and multiplication of various sorts, so we should weaken the notion of semiring. A magma, sometimes called groupoid, is a set with an operation having no prescribed identities. Towards this end (and not knowing of pre-existing terminology) we call a semigroup (A, +) a bimagma if it has multiplication as well as addition. A homomorphism of bimagmas is a map preserving addition and multiplication. A bimagma is unital if (A, ·) possesses a multiplicative unit 1A. Although the algebraic theory can be carried out without commutativity, many applications are commutative, so we often work with a commutative bimagma, denoted cbimagma. Note 1.7. The reader who feels uncomfortable with cbimagmas can read instead “commutative semiring,” which provides the main tropical setting. An ideal of a cbimagma A is an additive sub-semigroup I such that aI ⊆ I for all a ∈ A. {0} is an ideal when 0 ∈ A, but it will be replaced by other ideals arising naturally in the structure. Definition 1.8. For any unital bimagma A, we want the image of N inside A. Towards this end, we define 1 = 1A, and inductively n + 1 = n + 1A, and N(A) to be {n : n ∈ N} ⊆ A. When A is understood we write N for N(A). When A is a bimagma we still require the following special case of distributivity: ma = ma, ∀m ∈ N, a ∈ A. One tricky point is that we may not be able to identify n with n, even for semirings of characteristic 0, for example with truncated numbers (Example 4.2(vii) below). Nevertheless, we have: Remark 1.9. There is a homomorphism N → A given by n 7→ n. 1.2.2. Connection to universal algebra and model theory. In many of our examples, T is a multiplicative monoid, even a group, and this structure is to be viewed intrinsically, as is the negation map. Other properties (such as nonassociative Lie multiplication, or nondistributive operations for hyperfields) also can come into play. Furthermore, one might want to consider the more esoteric bimagmas. Because of the varied algebraic structures involved, the appropriate setting for the investigation would be universal algebra, for which [45] serves as our reference (also cf. [15, 10]), but to avoid digressions we leave that aspect of the theory to the reader’s discretion, and make do with the terminology “algebraic structure” instead of the more technical “Ω-algebra” of [45, Definition 2.1]. Roughly speaking, we deal with given algebraic structures, each equipped with given n-ary operations for n = 0, 1, 2, which satisfy given identities, which are simple universal sentences (i.e., of the form f(x1,...,xm) = g(x1,...,xn)). A homomorphism by definition preserves all of the operations in the structure. We recall from [45, p. 61] that a congruence of an algebraic structure A is an equivalence relation Φ on A preserving the operations of A, or in other words which is a subalgebra of A × A. Ideals do not work well in the absence of negatives, but fortunately congruences serve as a replacement (although technically more complicated). We also denote Φ as ≡, or, equivalently, as {(a, a′): a ≡ a′}⊆A×A. [45, p. 62] provides the Noether correspondence between homomorphic images and congruences: any ′ ′ homomorphism ψ : A→B gives rise to a congruence Φψ on A given by (a, a ) ∈ Φψ iff ψ(a) = ψ(a ); conversely, any congruence Φ gives rise to the corresponding structure A/Φ on the equivalence classes, together with a natural homomorphism ψ : A → A/Φ defined by a 7→ [a].

2. A more detailed overview We provide more details on the main concepts: T -modules, negation maps, and surpassing relations. 2.1. T -modules.

Deﬁnition 2.1. A (left) T -module over a set T is an additive monoid (A, +, 0A) together with scalar multiplication T ×A→A satisfying distributivity over T in the sense that

a(b1 + b2)= ab1 + ab2, ∀a ∈ T , bi ∈ A, 6 LOUIS ROWEN

2 also stipulating that a0A = 0A for all a in T . In this case, we also say that the set T acts on (A, +). A T -module A is cancelative if ab = ab′ for a in T implies b = b′. Our relevant structures all are T -modules. Often T itself has extra structure, passed on to A.

Definition 2.2. When (T , ·, 1) is a monoid, we say there is a monoid action when 1T b = b and (a1a2)b = a1(a2b), for all ai ∈ T and b ∈ A, and call A a monoid module. Cancelation is clear for a monoid action over a group. Example 2.3. We say that T is trivial if |T | =1. Here are some instances: • A = T = {0}. • (The Boolean semifield) A = {0, 1} and T = {1} where 1 + 1 = 12 = 1. • (The Boolean “supertropical semifield”) A = {0, 1, 1ν := 1 + 1}, where T = {1}, 12 = 1, and 1ν + 1 = 1ν . Definition 2.4. A T -bimagma (resp. T -cbimagma) is a bimagma (resp. cbimagma) A which is a unital T -module satisfying 3 a(b1b2) = (ab1)b2, ∀a ∈ T , bi ∈ A. A T -semiring† is a semiring† which is a T -bimagma. A T -semifield† is a commutative T -semiring for which T is a group. (This is called a demifield in [7, Definition 4.1] when T is a hyperfield generating (A, +).) A semialgebra over a commutative (associative) semiring† C is a C-bimagma (i.e., T = C) which also distributes over C. Note 2.5 (Ongoing hypotheses and notation). (A, +) is a T -module, whose zero element 0 is absorbing in the sense that a0 = 0 for all a ∈ T . To simplify the exposition we assume that T ⊆ A. (See [5] for an elaboration of this point.) We normally designate elements of T as “a”, and elements of A as “b.” For simplicity, we assume T ⊆ A, and write T0 for T ∪{0}. Note that {0}∪T ∪ (T + T )= T0 + T0. Definition 2.6. A (pre-)ordered T -module is a T -module A with a (pre-)order relation ≤, satisfying ′ ′ the following, for a,a ∈ T and b,bi,bi ∈ A: ′ ′ ′ (i) If bi ≤ bi for i =1, 2, then b1 + b2 ≤ b1 + b2. ′ ′ (ii) If b1 ≤ b1 then ab1 ≤ ab1. (iii) (When T is pre-ordered) If a ≤ a′, then ab ≤ a′b. 2.1.1. Height. Definition 2.7. When T additively generates A, we define the height of an element c ∈ A as the mc minimal number mc such that c = i=1 ai with each ai ∈ T . (We say that 0 has height 0.) The height of A is the maximal height of its elements (which is said to be ∞ if these heights are not bounded). P Thus A has height 1 iff A = T or T0. A has height 2 iff A = T ∪(T +T ) or T0 +T0, which also will play an important role. The tropical theory falls largely into height 2. Height 3 involves extra subtlety, such as various hyperfields and “quasi-periodicity” as indicated for example in Theorem 6.25 and Example 12.1. 2.2. Negation maps. Since semigroups lack negation, we introduce a negation map: Definition 2.8. A negation map on a T -module (A, +) is an additive homomorphism (−): A → A of order ≤ 2, written b 7→ (−)b, restricting to a map (−): T → T , compatible in the sense that ((−)a)b = a((−)b) = (−)(ab), ∀a ∈ T , b ∈ A. (2.1) A negation map on a T -bimagma (A, ·, +) is a negation map also satisfying (2.1) for all a,b ∈ A.

Remark 2.9. As Gaubert and Knebusch both observed, when T contains a multiplicative unit 1T , the negation map on A is given simply by b 7→ ((−)1T )b. But one has to be careful in choosing (−)1 when it is not speciﬁed a priori, cf. [61, Lemma 2.18].

2 Modules A over a semiring T are called semimodules in the literature, where one stipulates that 0T b = 0A, ∀b ∈ A. 3We do not require distributivity over all of A a priori, which could fail for hypergroups, one of our motivating examples. ALGEBRAS WITH A NEGATION MAP 7

Thus a negation map is a formal map a 7→ (−)a, viewed as part of our structure, that satisfies all of the properties of negation except a + ((−)a)= 0, which comes automatically for classical algebra. Initially, negation is notably absent in tropical situations, but is circumvented in two main ways: the identity itself is a negation map, paving the way to the “supertropical theory,” or else one can introduce a negation map through the process of “symmetrization” (cf. §3.3 and [3]), which passes to A × A. To simplify notation, we write b(−)b′ for b + ((−)b′). Then we put b◦ := b(−)b, called a quasi-zero. To avoid ambiguity, we then write the product of b and (−)b′ as b((−)b′), which occurs much more rarely. Also we write (±)b for “b or (−)b,” and b(±)b′ for “b + b′ or b(−)b′.” We write A◦ for {b◦ : b ∈ A}, a T -submodule of (A, +), and T ◦ for {a◦ : a ∈T}. In the supertropical theory (Definition 3.4 below), the quasi-zeros are the “ghost” elements. Under symmetrization (Defini- tion 3.6) the quasi-zeros have the form (b,b). (In [3, Definition 2.6] the quasi-zeros are called “balanced elements.”) MAJOR NOTE 2.10. The quasi-zero takes the role customarily assigned to the zero element in classical algebra, where the only quasi-zero is 0 itself. A◦ plays a fundamental role. Expressed in these terms, one of the challenges in tropical structure theories has been to describe A◦ accurately. Lemma 2.11. (−)0 = 0 and 0◦ = 0. Proof. (−)0 = (−)0 + 0 = (−)0 + ((−)(−)0) = (−)(0(−)0) = (−)((−)0)= 0. Hence 0◦ = 0 + 0 = 0. Note that b◦ = ((−)b)◦, and (mb)◦ = mb◦ for all m ∈ N. Definition 2.12. We designate several important elements of a unital T -module A, for future reference:

e = 1(−)1, e′ = e + 1, e◦ = e(−)e = e + e =2e. (2.2) The most important quasi-zero is e, which acts similarly to 0. But e need not absorb in multiplication; rather ae = a(1(−)1)= a1(−)a1 = (a1)◦ for any a ∈ T . Deﬁnition 2.13. A quasi-negative of a ∈ T is an element b ∈ T such that a + b ∈ A◦. T has unique quasi-negatives if a + b ∈ A◦ for a,b ∈ T implies b = (−)a. In this situation we say A and (−) are uniquely quasi-negated. By deﬁnition, (−)a is a quasi-negative of a. MAJOR NOTE 2.14. The introduction of quasi-negatives to replace negatives enables us to develop the analogs of some of the most basic structures of algebra. The natural condition of unique quasi-negatives is surprising powerful, showing that a′(−)a ∈ A◦ implies a′ = a. Lemma 2.15. Suppose that (−) is uniquely quasi-negated.

(i) For a1 6= a2 in T and a1(−)a2 ∈ T , a1 + a2 = a2 implies a1(−)a2 = (−)a2 (or equivalently, a2(−)a1 = a2). (ii) If a3 = a1 + a2, and a3(−)a2 is tangible then a3(−)a2 = a1. ◦ Proof. (i) (a1(−)a2)+ a2 = (a1 + a2)(−)a2 = a2(−)a2 = a2, implying a1(−)a2 = (−)a2. ◦ (ii) (a3(−)a2)(−)a1 = a3(−)a3 = a3, so a1 = a3(−)a2.

(As M. Akian has pointed out, the conclusion of (i) fails when 2 = 1 6= e for a1 = a2 = 1, since a1 + a2 = a2 = 1 whereas a1(−)a2 = e.) Other crucial consequences are to be given in Proposition 2.31 and Proposition 2.40. 2.2.1. Negation maps of the first and second kinds. Definition 2.16. The negation map (−) is of the first kind if (−)b = b for all b ∈ A. The negation map is of the second kind if (−)a 6= a for all a ∈ T . This distinguishes between “supertropical” and “symmetrized” algebra, helping to explain why theorems holding for supertropical semirings might fail for symmetrized semirings. 8 LOUIS ROWEN

Lemma 2.17. When A is a T -cancelative unital T -module with 1 ∈ T , the kind of the negation map is determined by whether or not (−)1 = 1. Proof. If a1 = a = (−)a = a((−)1) for some a ∈ T , then 1 = (−)1. If 1 = (−)1, then b = 1b = ((−)1)b = 1((−)b) = (−)b.

As an indication of where we are headed, in the literature, a semigroup (A, +) is called idempotent (resp. bipotent) if a + a = a (resp. a + b ∈{a,b}) for all a,b ∈ A. 2.3. Pseudo-triples and triples. Definition 2.18. A pseudo-triple is a collection (A, T , (−)), where A is a T -module and (−) is a negation map on A. A triple is a pseudo-triple satisfying: (i) T ∩ A◦ = ∅, (ii) T generates (A, +). The triple is unital if A is unital with 1 ∈ T . MAJOR NOTE 2.19. The triple is the fundamental structure in this paper, taking the role akin to that of the base ring for a module. T ∩ A◦ = ∅ enables us to distinguish tangible elements from quasi-zeroes. The fact that T generates (A, +) enables us to reduce assertions about A to T . Example 2.20. The only triple with T trivial must be of the form Na where T = {a}. Proposition 2.21. For any uniquely quasi-negated T -cancelative pseudo-triple (A, T , (−)) with T closed under multiplication (other than the triple with T trivial of Example 2.20), T ∩ A◦ = ∅. Proof. Suppose a ∈T ∩A◦. Then a + a ∈ A◦, so a = (−)a. Moreover, for any a′ ∈ T , aa′ ∈ T and aa′(−)a ∈ A◦, so aa′ = a. Likewise a2 = a, so a′ = a by cancellation. Thus T = {a}. Any unital triple has the subtriple (h1i, (±)1, (−)) generated by 1, which plays a useful role parallel to F1-geometry. The pseudo-triple is needed in linear algebra, to describe linear combinations of vectors. Lemma 2.22 ([2, Remark 4.5]). A◦ is a T -submodule of A fixed under (−), for any pseudo-triple (A, T , (−)). Proof. 0 = 0◦ ∈ A◦, and a(b◦)= a(b(−)b) = (ab)(−)(ab) = (ab)◦. We express extra structure of T and A with the following terminology. Definition 2.23. A bimagma (resp. cbimagma) triple is a triple (A, T , (−)), where A is also a bimagma (resp. cbimagma). A semiring triple is a bimagma triple which is also a semiring (i.e., associative and distributive). Although the max-plus algebra is bipotent, idempotence fails in many other examples. It is too restrictive for our needs, so instead, we settle for a slightly weaker version.

Definition 2.24. A triple (A, T , (−)) is (−)-bipotent if a1 + a2 ∈ {a1,a2} whenever a1,a2 ∈ T with ◦ a2 6= (−)a1. In other words, a1 + a2 ∈{a1,a2,a1} for all a1,a2 ∈ T . (−)-bipotence turns out to hold in the main variants of the max-plus algebra that arise in the tropical literature, but the following slightly weaker condition suffices to develop the theory. Definition 2.25. A uniquely quasi-negated triple (A, T , (−)) is metatangible if a + b ∈ T whenever a,b ∈ T with b 6= (−)a. All (−)-bipotent triples clearly are metatangible. Remark 2.26. Often for metatangible triples, negation maps of second kind are preferable to those of first kind, since then a 6= (−)a for a ∈ T implies a + a ∈ T . If (A, T , (−)) is (−)-bipotent with (−) of second kind, then e + 1 = 1(−)1 + 1 = 1(−)1 = e = e(−)1. ALGEBRAS WITH A NEGATION MAP 9

In Theorem 6.7 we shall see that “most” metatangible triples are (−)-bipotent. MAJOR NOTE 2.27. Metatangibility is the main engine for this paper, encompassing a wide range of examples, as seen in Theorem 6.50. Tensor products of triples are triples, as described below in §5.5. Although they lose (−)-bipotence, they provide a powerful tool in the theory.

2.4. Surpassing relations. Many subtleties cannot be explained properly until we bring in another notion. Deﬁnition 2.28. A surpassing relation on a triple (A, T , (−)), denoted , is a partial pre-order satisfying the following, for elements of A: (i) b b′ whenever b + c◦ = b′ for some c ∈ A. (ii) If b b′ then (−)b (−)b′. (iii) If b c and b′ c′ then b + b′ c + c′. (iv) If a ∈ T and b b′ then ab ab′. (v) If a a′ for a,a′ ∈ T , then a = a′. We also write b′ b to indicate that b b′. A strong surpassing relation on a triple A is a surpassing relation also satisfying b◦ 6 a for any a ∈ T . A surpassing ◦-PO on A is a surpassing relation which restricts to a PO on A◦. One other property that one often wants is that a a◦, which holds whenever e′ = e, since then a◦ = a + a◦. It holds in all of the tropical examples except the layered ones (when e′ 6= e), but fails miserably in the classical case. Here is our major tropical example.

′ ′ ◦ Definition 2.29. The ◦-relation ◦ is the relation b ◦ b iff b = b + c for some c ∈ A. In case (−) is of the first kind, this says b′ = b +2c.

Lemma 2.30. The relation ◦ is a partial pre-order satisfying properties (i)–(iv) of Definition 2.28. Proof. (i) Condition (i) of Definition 2.28 is by definition. To see Condition (ii), note that a + c◦ = b implies (−)a + c◦ = (−)(a + c◦) = (−)b. Condition (iii) is immediate. For Condition (iv), if b′ = b + c◦ then ab′ = ab + ac◦ = ab + (ac)◦.

Proposition 2.31. The relation ◦ on a triple with unique quasi-negatives is a strong surpassing relation. Proof. By Lemma 2.30. If a = a′ + c◦ for a,a′ ∈ T , then a(−)a′ = (a′ + c)◦, implying a′ = a, yielding Condition (v). If a′ = b + c◦ for a′ ∈ T ,b ∈ A◦, then a′ ∈ T ∩ A◦, a contradiction. The surpassing relation provides an important sub-T -module of A.

Definition 2.32. ANull = {b ∈ A : b 0}. ◦ Remark 2.33. A ⊆ ANull, by Definition 2.28(i). MAJOR NOTE 2.34. One main idea promoted here is that the surpassing relation lies at the crux of the theory, replacing equality when generalizing classical theorems, and {0} should be replaced by the ◦ ideals A and ANull. Let us see why the conditions of Definition 2.28 are desired.

(i) shows that reﬁnes ◦, and shows how the quasi-zeros behave like 0 under . (ii), (iii), and (iv) are needed to preserve the operations. By (v), the surpassing relation restricted T is equality, and provides a way of recovering theorems from classical algebra. Ironically, instead of being symmetric (and thus an equivalence), is antisymmetric. In introducing the surpassing relation, which is not an identity in universal algebra, we are crossing over from universal algebra (with identities) to a more general version of one-sided relations in model theory (with subtleties which we shall not discuss here). 10 LOUIS ROWEN

Here is an enlightening example of how ◦ generalizes classical algebra. In any semiring with a negation map (−), we write [a,b] for the Lie commutator ab(−)ba. Lemma 2.35 (Leibniz -identities). [a,b]c + b[a,c] = [a,bc] + (bac)◦. In particular,

[a,bc] ◦ [a,b]c + b[a,c]; [ab,c] ◦ a[b,c] + [a,c]b. Proof. [a,b]c + b[a,c] = (ab(−)ba)c + b(ac(−)ca) = (abc(−)bca) + (bac(−)bac) = [a,bc] + (bac)◦. The second assertion is analogous. Note that when T is a group, it satisfies the condition of “thin elements” of [3, Definition 2.7], and thus [3, Property 4.6] is relevant. We return to this issue in §5.1, §8.1 and in [5]. 2.5. Systems. We put everything together. Definition 2.36. A system (A, T , (−), ) is a triple (A, T , (−)) together with a surpassing relation , satisfying the following properties: (i) If a + b 0 for a,b ∈ T then b = (−)a. (This is a bit stronger than “unique quasi-negatives” for triples, but is the same when =◦.) (ii) T ∩ ANull = ∅. (This is already given when =◦.) The system (A, T , (−), ) is of the resp. first, second kind if (−) is a negation map of the resp. first, second kind. A semiring system (resp. T -semifield system) is a system where (A, T , (−)) is a semiring triple (resp. T -semifield triple). A bimagma (resp. cbimagma) system is a system where (A, T , (−)) is a bimagma triple (A, T , (−)), where A is also a bimagma (resp. cbimagma). The system is strong if the surpassing relation is strong. The role of is sublime. It often comes naturally with the triple, being equality for classical algebra, ◦ in tropically-oriented situations (see Theorem 6.30), and ⊆ for hypergroups (Definition 3.19). Definition 2.37. A system (A, T , (−), ) is of hypergroup type if: • A⊆P(T ), • T is identified with the set of singletons of A, • =⊆, • 0 ∈ a(−)a for all a ∈ T , ′ ′ ′ ′ ′ • b + b = ∪{ai + aj : ai ∈ b, aj ∈ b , ∀b,b ∈ A}. Lemma 2.38. Any system (A, T , (−), ⊆) of hypergroup type is strong. Proof. Otherwise suppose b◦ ⊆{a}. Then b◦ = {a}∈T ∩A◦ = ∅.

MAJOR NOTE 2.39. Our two main sorts of systems either have =◦ or are of hypergroup type, and thus are strong. Systems include the classical case, the “standard” supertropical semiring, and symmetrized algebras of [3], as well as examples given in §4.2: hyperfields [68], fuzzy rings [20], tracts [8], the “layered” semiring of [36], the “exploded” algebra [55], and the ELT-algebra of Sheiner [12, 64]. Once the system is established, it provides a mechanism for obtaining effective definitions of new tropical algebraic structures, and also provides a guide for applying classical algebraic techniques in situations such as in [5, 23, 48]. The next observation is the key to the relationship between A and T in a system. Proposition 2.40. Suppose (A, T , (−), ) is a system.

(i) If a + c = b for a,b ∈ T and c ∈ ANull, then b = a. (ii) b 6 a for any a ∈ T and b ∈ ANull. (iii) Conversely, suppose that (A, T , (−)) has a sub-T -module I ⊇ A◦ with T ∩ I = ∅, having the ′ ′ ′ property that for each a ∈ T there is a unique a ∈ T such that a + a ∈ I. Deﬁne I by b I b ′ if b = b + c for c ∈ I. Then I is a strong surpassing relation, and (A, T , (−), I) is a system with I = ANull. ALGEBRAS WITH A NEGATION MAP 11

◦ Proof. (i) b(−)a = a + c ∈ ANull, so b = a. (ii) a b 0 contradicts a ∈ T . ′ ′ (iii) 0 I b iff b ∈ I, proving I = ANull. The rest follows easily from (i) and (ii). There are two ways of approaching systems — one is in terms of the basic (expanded) algebraic structure, for instance a ring or semiring, and the other, as in representation theory, is in terms of a secondary structure (such as a module). Our emphasis in this paper is on the former, covering the basic tropical algebraic structures, hypergroups, and fuzzy rings. One can perform standard algebraic constructions, such as matrices, formal traces, bilinear forms, quadratic forms (all in Example 4.7). The theory of semiring systems can be viewed in the context of Lorscheid’s “blueprints” [53], but also their specific extra information permits us to hone in on other applications, which are not necessarily associative. The second approach is taken in [5, 48, 49]. 2.6. Organization of this paper. See §2.7 for a list of the main results. This paper is structured as follows: (i) A brief survey of the main concepts (tangible, negation map, triple, and system) has been given in §1. The main objective is to describe the set T in terms of a better structured T -module (A, +) which also inherits whatever structure comes with T . (ii) Metatangible triples, characterized by the property that a1 + a2 ∈ T for all a1 6= (−)a2 ∈ T , are the focus of this paper and include all of the tropical applications, as well as many hyperfields and fuzzy rings. Surpassing relations, an extension of equality on T , replace equality in much of the theory. These ingredients are combined in §2.5 to yield the system. (iii) The major applications (max-plus algebra, supertropical semirings, symmetrized semialgebras, layered semirings†, “classical” semialgebras, hypergroups, and fuzzy rings) are described in §3. In particular, the important technique of symmetrization is presented in §3.3, to provide a negation map when one is lacking. Also we we bring ordered monoids into the picture. (iv) In §4 we describe other major examples, and bring in other key properties of triples and systems. (v) Additional structure is considered in §5, where we also describe matrices, including the determinant, involutions, polynomials and their roots (to pave the way for affine geometry in Remark 5.7), localization, and tensor products (§5.5). (vi) Metatangible systems are studied in depth and largely classified in §6. (vii) Seeing that systems have a robust algebraic theory, we proceed to view them categorically in §7, utilizing the surpassing relation as an essential ingredient in the definition of morphism. (viii) Linear algebra is treated in §8, with emphasis on varying notions of matrix rank. (ix) Tropicalization, which provides the connection with classical mathematics via valuations, is studied in §9 in terms of morphisms of systems. This provides the framework of defining and inves- tigating tropical analogs of classical algebraic structures. (x) The Lie point of view is given in §10. (xi) Directions suggested for further research are given in §11. (xii) The two main non-tropical applications are hypergroups (marked by ∗) and fuzzy rings (marked by ∗∗). Since hypergroups provide a rich source of examples and motivation, they are treated throughout the main text, although their main examples are put into Appendix A in order not to interrupt the flow of this study. Likewise, fuzzy rings are viewed as systems in Appendix B.

2.7. Main results of this paper. Proposition A (Proposition 2.40). A system is described explicitly in terms of a triple (A, T , (−)) and a “null” T -module ANull. Although distributivity can fail (e.g., over hyperﬁelds), there is a way of recovering distributivity: Theorem B (Theorem 2.45). Any left and right T -module A generated additively by a commutative monoid (T , ·) can be made (uniquely) into a semiring via the multiplication rule

a b = a b i  j  i j i ! j i,j X X X   12 LOUIS ROWEN

for ai,bj ∈ T .

Theorem C (Theorem 3.21, Proposition 4.16). For any hypergroup T with negation (−), let T0 be the sub-semigroup of the power set additively spanned by T ; then (T0, T , (−)) is a triple. ⊆ is a surpassing PO, and (T0, T , (−), ⊆) is a T -strictly negated system. The hypergroup T is metatangible (resp. closed), iff its hypersystem (T0, T , (−), ⊆) is metatangible (resp. (−)-bipotent). The assortment of hypergroup examples, given in Appendix A, sheds considerable light on the theory. Our main results on metatangible systems are of special interest, encompassing all of the tropical algebraic theories, as well as many other examples. Theorem D (Theorem 6.7). Any cancelative metatangible unital triple (A, T , (−)) satisfies one of the following cases: (i) A is (−)-bipotent. (ii) 1(−)1 + 1 = 1, with one of the following two possibilities. • (−) is of the first kind, of characteristic 2. (In other words 3 = 1 = 1.) In this case, A has height ≤ 2. • (−) is of the second kind, either of finite characteristic or with {m : m ∈ Z} all distinct.

Theorem E (Theorem 6.25). Any element c of height mc ∈ N in a metatangible triple has a uniform ◦ presentation, c = mccT for some element cT ∈ T and m 6=2, or c = cT . Theorem F (Theorem 6.28). The uniform presentation is unique for any element of height > 2 in a cancelative (−)-bipotent unital system. Theorem G (Theorem 6.30). For any cancelative metatangible unital triple (A, T , (−)), we have the ′ ′ metatangible system (A, T , (−), ◦). Conversely, if there are elements b b but b 6◦ b , the triple (A, T , (−)) is either of first kind, of height > 2, or of height 2 satisfying b + b′ = b. Despite the large assortment of examples given in §6.8.1 and [61], metatangible systems are described in Theorem 6.50, as belonging to one of the classes reducing to the familiar examples from tropical theory, or satisfying specific properties called “exceptional.” Theorem H (Theorem 7.4). There is a faithful functor Ψ from the category of canonical hypergroups into the category of T -reversible systems, whose morphisms are the -morphisms, sending a hypergroup T to its hypersystem (T0, T , (−), ⊆). [61, Example 6.46] discusses weakening distributivity for hypersystems. Symmetrization (§3.3, §3.3.1) is a powerful tool that enables us to move from arbitrary T -semirings to systems, leading to a major application, the transfer principle which passes identities of rings to T -semiring systems, and obtained for matrices in [2, Theorem 3.4] following an idea originating in [57, p. 352, end of proof of (a)] carried on in [24], as described in §7.3. To state the transfer principle precisely, one needs free objects. Jacobson [45, §2.7] provides a unified free construction, but we only work with the following special cases. The parent structure in tropical algebra is the well-known max-plus algebra, described thoroughly in [1]. We append the subscript max to indicate the corresponding max-plus algebra, e.g., Nmax or Qmax, but to emphasize the algebraic structure theory we still use the usual algebraic notation of · and + throughout (rather than ⊙ and ⊕). Definition 2.41. Throughout this definition, M denotes the free monoid, which is the monoid in formal indeterminates, with multiplication given by concatenation.

(i) The free monoid with negation map M± is the monoid in formal indeterminates and their formal negations, with multiplication given by concatenation, together with the relation (−ei)h = ei((−)h) deﬁned inductively. (For example, ((−)e1)((−)e2)= e1e2.) (ii) The free semiring† also is N[M] with the semiring† structure. (iii) The construction of the free T -monoid semialgebra is similar, where we take A to be the free module over a monoid T . (iv) The free associative Nmax-semialgebra, is Nmax[M]. Here, nx evaluates as x in Nmax[M]. ALGEBRAS WITH A NEGATION MAP 13

(v) The free associative N-semialgebra is N[M]. (vi) The free N-semialgebra with negation map N[x, (−)x] is N[M±], which we denote as N[x, (−)x], writing xi instead of ei. A typical element of N[x, (−)x] has the form i(ai(−)bi)xi, where i = (i1,...,it). (vii) We can do this all with magmas. The free magma is the magma in formal indeterminates,P with multiplication given by concatenation, written with parentheses since we lack associativity. The free magma with negation map M± is the magma in formal indeterminates and their formal negations. The N-magma semialgebra is N[M] where M is the free magma. The free bimagma is given the natural bimagma structure. The free (nonassociative) Nmax-semialgebra, which we denote N{{X}}, is the magma semialgebra Nmax[M]. The free N-semialgebra is N[M] where M is the free magma. The free N-semialgebra with negation map N[x, (−)x] is N[M±], which we denote as N[x, (−)x], writing xi instead of ei.

Theorem I (Transfer principle, Theorem 7.13). Suppose P = i aixi,Q = i bixi ∈ N{x, (−)x}, where ai ≥ bi for each i. If the free semiring satisfies the identity P¯ = Q¯, then P ◦ Q in N{x, (−)x}. P P 2.7.1. The free module. Definition 2.42. The free module A(I) is the usual direct sum of copies of A (i.e., with almost all entries 0), where we identify the i-th base element with the vector having 1 in the i component. If A is a (I) (I) (I) T -module then so is A , under the diagonal action a(bi) = (abi). We can define T ⊆ A . We write (I) T(j) for the j-th component of T , and put TA(I) := ∪j T(j). Note that for |I| infinite, T (I) is not a monoid even when T is a monoid. Example 2.43. We can formally obtain a “generic” negation map in the free A-module. Given a set I, we formally define (−)I to be another copy of I, indexed formally by (−)i : i ∈ I, and put I to be the disjoint union I ∪(−)I. The free A-module with negation map over a semiring A is the free A-module (I) A whose base is formally denoted as {ei, (−)ei = e(−)i : i ∈ I}, with negation map given by ei 7→ (−)ei and (−)ei 7→ ei. Remark 2.44. Notation as in Definition 2.42. (i) If (A, T , (−)) is a pseudo-triple, then (A(I), T , (−)) is a pseudo-triple, where the negation map is defined diagonally. (I) (ii) If (A, T , (−)) is a triple, then (A , TA(I) , (−)) is a triple over TA(I) , defining negation and multiplication componentwise. Likewise for a system, the surpassing relation is defined componentwise. (I) (iii) If a triple (A, T , (−)) has unique quasi-negatives, then so does the triple (A , TA(I) , (−)), which will play an important role in Sections 5.5, 8, and 9. Special cases are matrix semirings over semiring triples and polynomial semirings over semiring triples, to be treated separately in §5.1 and §5.2. 2.7.2. Linear algebra. Linear algebra over systems is particularly intriguing, since some of the supertropical results go over, but others have counterexamples, as discussed in §8. The negation map (−) is used to define the (−)-determinant and adjoint in Equation (5.1).

Theorem J (Theorem 8.6, reformulation of [65]). |A||B|◦ |AB|, for any matrices A, B ∈ Mn(A). The main result unifying different notions of matrix rank in [43, Theorem 3.4] are formulated rather transparently in this more general context (for tangible vectors) in §8, but only one direction holds: Theorem K (Theorem 8.11). If the rows of a tangible n × n matrix A over a cancelative (−)-bipotent triple are dependent, then |A| ∈ A◦. A wide-ranging counterexample for the converse was presented in [2]. This flavor of the theory seems to depend on whether negation is of the first or second kind, since we have more positive results for systems of the first kind in [5], which delves more deeply into linear algebra over systems. 14 LOUIS ROWEN

Having the basic theory in place, we turn in §9 to the mainstay of tropical mathematics, which is tropicalization. Tropicalization is explained in §9.1 as a -morphism of systems. 2.7.3. Applications. Here is a sample illustration of how the systemic theory can be given a classical flavor. Proposition L (Proposition 10.6). If L is a Lie semialgebra (over a commutative semiring C) with a negation map, then ad L is a Lie sub-semialgebra of EndC L, and there is a Lie -morphism L → ad L, given by a 7→ ada. The main examples of hyperfields (supertropical hyperfield, Krasner hyperfield, hyperfield of signs, phase hyperfield, and triangle hyperfield) are described explicitly in Appendix A as systems, some of which are bipotent. The application to fuzzy rings is given in Appendix B: Theorem M (Theorem 13.8). Any T -coherent fuzzy triple A gives rise to a system (A′, T , (−), ) with unique quasi-negatives, where (−)a = εa. The converse is given in Proposition 13.10. 2.8. Redefining multiplication. This discussion is intended for those readers who would like to see how hypergroups fit into the theory. The motivation grew out of a conversation with Baker. Since the “tropical hyperfield” of [7] and [68, §5.2] is isomorphic to the “extended” tropical arithmetic in Izhakian’s Ph.D. dissertation (Tel-Aviv University) of 2005, also cf. [34], given more formally in [42], one would like to see how other major hyperrings also can be studied in terms of the more amenable semiring theory. The tricky aspect is to obtain distributivity for all of A, which can be written down as follows, when we assume that T generates (A, +):

a b = a b . (2.3) i  j i j i ! j i,j X X X for ai,bi ∈ T . For instance, in the study of hyperfields  it might seem at first glance that we must forego distributivity in P(T ), since the multiplication in the power set of certain hyperfields need not distribute over addition, as to be seen in Examples 12.1. However, this difficulty is bypassed by the following surprising result, which we call a theorem because of its significance, despite its being almost trivial. Theorem 2.45. Any left and right T -module A generated additively by a commutative monoid (T , ·) can be made (uniquely) into a T -semiring with absorbing 0, via the multiplication

a b = a b , (2.4) i  j i j i ! j i,j X X X  

ai 0 = 0 ai = 0, i ! i ! X X for ai,bj ∈ T . ′ ′ Proof. It suﬃces to show that this is well-deﬁned, i.e., if i ai = i ai then i,j aibj = i,j aibj (and likewise for b ,b′ ) for a′ ,b′ ∈ T . But j j i j P P P P

′ ′ aibj = aibj = ai bj = ai bj = aibj . i,j j i ! j i ! j i ! i,j X X X X X X X X Going the other direction, distributivity in the semiring forces (2.4) to hold.

3. The main tropically oriented triples and systems

We review some of the concepts that have played a major role in tropical algebra. ALGEBRAS WITH A NEGATION MAP 15

3.1. The max-plus algebra and bipotent semirings. The max-plus algebra really concerns ordered groups, such as (Q, +) or (R, +), which are viewed at once as max-plus semiﬁelds†, generalizing Remark 3.2.

Remark 3.1. The identity is the only negation map on the max-plus algebra, by [3, Proposition 2.11]). Then a = (−)a = a◦, so the max-plus algebra cannot be a triple (and unique quasi-negatives fail).

3.1.1. Green’s partial order. We recall the following elegant observation of Green:

Remark 3.2. (i) Any ordered monoid (M, · ) gives rise to a bipotent semiring†, where we deﬁne a + b to be max{a,b}. Indeed, associativity is clear, and distributivity follows from the inequalities (1.1). (ii) Conversely, any set M with a partial addition and 0 has a natural partial pre-order given by a1 ≥ a2 in M if a1 = a2 + b for some b ∈M. It is a pre-order when M is bipotent. Any such relation becomes trivial on a T -module when T contains −1, the classical negative of 1, since then a2 = a1 + (a2 + (−1)a1).

Remark 3.2 is tied in with the following property:

Deﬁnition 3.3. An additive semigroup S ⊆ A is ub (for upper bound) if a + b + c = a always implies a + b = a.

For example, the max-plus algebra is a ub semiﬁeld†. This criterion abounds in tropical algebra, as noted in [38]. By [40, Proposition 0.5], the Green partial pre-order is a PO iﬀ the semigroup A is ub.

3.2. Supertropical semirings and supertropical domains. The diﬃculty arising in Remark 3.1 is remedied by turning to supertropical algebra [34, 36, 42].

Deﬁnition 3.4. A supertropical semiring is a quadruple A := (A, T0, G0,ν) where A = T0 ∪ G0 is a commutative semiring, T0 is a submonoid, and G0 ⊂ A is a bipotent ideal given Green’s order of Remark 3.2(ii), together with an onto multiplicative monoid homomorphism ν : A → G0 satisfying 2 ν = ν, with ν|T being 1:1. Addition is given by

ν(b) whenever ν(b)= ν(b′), ′ ′ b + b = b whenever ν(b) >ν(b ), b′ whenever ν(b) <ν(b′).

A supertropical domain is a supertropical semiring, where T = T0 \ 0 (in case 0 ∈ T0) is a multiplicative monoid.

The elements of G0 are called ghost elements and ν : A → G0 is called the ghost map. The monoid T encapsulates the tropical aspect. To get a triple, we take (−)a = a, a negation map of the ﬁrst kind. The standard supertropical semiﬁeld is A := T ∪ G0 where G := (G0 \{0}, ·)= ν(T ) is an ordered group. (Customarily G0 = Qmax or Rmax, rewriting + as ·).

Deﬁnition 3.5. The standard supertropical semiﬁeld A := T ∪ G0 yields the standard supertropical ◦ ′ ′ ◦ system (A, T , 1A, ◦), where b is b + b = ν(b), and b ◦ b when b = b + c for some c ∈ A. (The relation ◦ is “ghost surpasses,” written as:

a1 |= a2 in A if a1 = a2 + b + b for some b ∈ T , G

◦ and ν of Deﬁnition 3.4 is a special case of ◦.) Now e = 1A + 1A = 1A is the multiplicative unit of G, and ◦ G0 = eA = A .

Classical algebraic results were transferred to the tropical theory by means of |= in [2, 42, 44]. G 16 LOUIS ROWEN

3.3. Symmetrization. Although the max-plus algebra and its modules initially lack negation, one obtains negation maps for them through the next main idea, the symmetrization process, extracted from [2, 24, 29], where a T -module is embedded into a super-module. Deﬁnition 3.6. Given any T -module A, not necessarily with a negation map, deﬁne A = A × A with componentwise addition, and T = (T ×{0}) ∪ ({0} × T ) with multiplication T × A → A given by the twist action b

(a0,a1)(bb0,b1) = (a0b0 + a1b1,a0b1 + a1b0), ai ∈ T ,bi ∈b A. b b

When A is a bimagma we extend multiplication to A by taking also ai ∈ A. We also deﬁne the negation map on A as the switch map (−)sw given by (−)sw(b0,b1) = (b1,b0). Then (b0,b1)(−)sw(b0,b1) = (b0 + b1,b1 + b0), so the quasi-zeroesb all have the form (c,c), and the surpassing relation ◦ is given by: b

′ ′ ′ (b0,b1) ◦ (b0,b1) iﬀ bi = bi + c for some c ∈ A, i =0, 1. (3.1) (The same c is used for both components.) (A, T , (−)sw, ◦) is the symmetrized system. Remark 3.7. This is the structure given at the beginning of [25, §3.8] in the case A = R , and is b b max the venue for [6, §3.4], [2], [3, Example 2.21], and [9, 47]), rather than what is called the “symmetrized algebra” in [25]. But we prefer the terminology “symmetrized” for this version, which is appropriate to the general structure theory. Lemma 3.8. T ∪{0, 0} is a monoid whenever T is. T is a union of two subgroups whenever T is a group. b b Proof. It is closed under multiplication, and inverses exist when T is a group ((0,a)−1 = (0,a−1)). \ In particular, N ∪{0} is itself a semiring which we call Z, with negation given by (−)(m, n) = (n, m). The construction of Z from N takes the place of the familiar construction of Z from N, with the diﬀerence that here we distinguish (m, n) from (m + k, n + k).

Example 3.9. For any semigroup (A, +), Aˆ is naturally a module over N. \ Lemma 3.10. N[M] is isomorphic to N[M]. b Proof. We send (m, n)a to (ma , na) for a ∈M, and check that addition and multiplication are preserved. b

The free module with negation map can be viewed as the symmetrization of the free module (without (I) negation) A , where we identify ei with (ei, 0) and (−)ei with (0,ei).

3.3.1. The (−)-bipotent symmetrized system: the version according to [3, Proposition-Definition 2.12]. Definition 3.6 fails to be metatangible. The following modification was introduced in [2, Proposition 5.1] and studied further under the name of “symmetrized max-plus semiring” in [3, Proposition-Definition 2.12] in the context of tropical constructions.

Example 3.11. One starts with an ordered semigroup G, putting G0 = G∪{0}, and deﬁnes Tsym = (G×{0}) ∪ ({0} × G),

Gsym := Tsym ∪{(a,a): a ∈ G0}⊆ A.

Thus, viewing G0 as a bipotent semiring, addition on Gsym is deﬁned componentwise on G0 ×{0}, {0}×G0, b and {(a,a): a ∈ G0}, whereas “mixed” addition satisﬁes:

(a0, 0) if a0 >a1; (a0, 0) + (0,a1)= (0,a1) if a0

 ALGEBRAS WITH A NEGATION MAP 17

(a0, 0) if a0 >a1; (a0, 0) + (a1,a1)= ((a1,a1) if a0 ≤ a1;

(0,a0) if a0 >a1; (0,a0) + (a1,a1)= ((a1,a1) if a0 ≤ a1.

Multiplication in Gsym is the twist action as in Definition 3.6. The (−)-bipotent symmetrized system is (Gsym, Tsym, (−)sw, ◦). MAJOR NOTE 3.12. The symmetrized system of Example 3.11 is a system which is (−)-bipotent of the second kind, but its construction requires that G be ordered. To obtain an ordered semigroup G we could apply the following modification of Remark 3.2(ii) to a bipotent semiring: a1 ≤ a2 iff a1 + a2 = a2. However, more in the spirit of this study is to apply this condition to the tangible elements of a (−)- bipotent pseudo-triple, also putting a ≤ a. This works whenever (−) is of the first kind, but when (−) is of the second kind we might have a and (−)a incomparable, leading to Definition 6.40 below. To bypass these considerations we often use the symmetrization of Definition 3.6 anyway. 3.4. *Hypergroups. Our other major example of a system is over a hypergroup T0, defined below, which has inspired much of our material in systems. We follow the treatments of Baker and Bowler [7] and Jun [46], discussing examples from [7] in Example 12.1; also see [16, 68]. One would like to formulate the structure of T0 in terms of addition (as well as other possible operations such as multiplication) on T0. But this is not feasible since T0 itself need not be closed under addition. Intuitively, a hyper-semigroup should be a structure (T0, ⊞, 0) with ⊞ : T0 × T0 → P(T0), for which the analog of associativity holds:

(a1 ⊞ a2) ⊞ a3 = a1 ⊞ (a2 ⊞ a3), ∀a ∈ T0.

T0 as defined in [7, 68] is injected naturally into P(T0), identifying T0 with the singletons in P(T0). There is some difficulty in the details: a1 ⊞ a2 need not be a singleton, so technically (a1 ⊞ a2) ⊞ a3 is not defined. This difficulty is exacerbated with generalized associativity; for example, what does (a1 ⊞ a2) ⊞ (a3 ⊞ a4) mean?

Deﬁnition 3.13. A hyper-semigroup is a structure (T0, ⊞, 0), where (i) ⊞ is a commutative binary operation T0 × T0 →P(T0), which also is associative in the sense that if we deﬁne a ⊞ S = ∪s∈S a ⊞ s,

then (a1 ⊞ a2) ⊞ a3 = a1 ⊞ (a2 ⊞ a3) for all ai in T0. (ii) 0 is the neutral element.

We transfer addition to P(T0) by deﬁning S1 + S2 = { (a1 ⊞ a2): ai ∈ Si} . Note that repeated addition in the hyper-semigroup needS not be deﬁned until one passes to its power set, which hampers checking basic identities such as associativity. Associativity could hold at the level of elements but fail at the level of sets.

Definition 3.14. A hyperzero of a hyper-semigroup T0 is a subset of P(T0) containing 0. A hypernegative of an element a in a hyper-semigroup (T0, ⊞, 0) is an element “−a” for which 0 ∈ a ⊞ (−a). (Following [46, Definition 2.1])A hypergroup is a hyper-semigroup (T0, ⊞, 0) for which every element a has a unique hypernegative −a. Hypernegation is the map a 7→ −a. A canonical hypergroup is a hypergroup satisfying the extra property: • (Reversibility) a ∈ b ⊞ c for a,b,c ∈ T0 iff b ∈ a ⊞ (−c). We need to translate this into triples. 18 LOUIS ROWEN

Remark 3.15. Henry [32, §2] shows that the reversibility condition holds if associativity holds and hypernegation distributes over addition, in the sense that −(a ⊞ b) = (−a) ⊞ (−b). M. Akian indicated the reverse direction to me, cf. [5, Lemma 3.7]. Proposition 3.16. A hypergroup is canonical if and only if hypernegation distributes over addition. Proof. (⇐) By Henry [32, §2]. (⇒) If c ∈−(a ⊞ b) then −c ∈ a ⊞ b, so a ∈ (−c) ⊞ (−b), and 0 ∈ a ⊞ (−a) ⊆ ((−c) ⊞ (−b)) ⊞ (−a) = (−c) ⊞ ((−b) ⊞ (−a)), i.e., c ∈ (−b) ⊞ (−a). This proves that −(a ⊞ b) ⊆ ((−b) ⊞ (−a)). On the other hand, if c ∈ (−b) ⊞ (−a), then −b ∈ c ⊞ a, implying 0 ∈ (−b) ⊞ b ⊆ c ⊞ (a ⊞ b), so −c ∈ a ⊞ b.

Deﬁnition 3.17. (T0, ⊞, ·, 0, 1) is a hyperring if T0 is a monoid and P(T0) is both a bimagma (with multiplication S1S2 = {a1a2 : ai ∈ Si}) and a T0-module. We put T = T0 \{0}. A hyperring (T0, ⊞, ·, 0, 1) is a hyperﬁeld if (T , ·, 1) is a group.

Lemma 3.18. Hypernegation on a canonical hypergroup (or hyperring) T0 is a negation map, and induces a negation map on P(T0), viewed as a pseudo-triple via (−)S = {−a : a ∈ S}.

Proof. We see that −(a1 ⊞ a2) = (−a1) ⊞ (−a2), by Proposition 3.16. Likewise, 0 ∈ a1 ⊞ (−a1) implies 0 ∈ a(a1 ⊞ (−a1)) = aa1 ⊞ (−aa1). In case T0 is a hyperring we note from the previous paragraph that −(a1a2) = (−a1)a2 = a1(−a2), and thus (−a1)(−a2)= −(−a1)a2 = a1a2.

Our ﬁrst candidate for a system might be (P(T0), T , (−), ⊆). But T0 might not generate (P(T0), ⊞), as seen below for the phase hyperﬁeld (Example 12.1). Accordingly, we restrict (P(T0), ⊞).

Deﬁnition 3.19. Given a hypergroup T0, we deﬁne T0 to be the sub-semigroup of (P(T0)\∅, ⊞) generated by the singletons, which we identify with T0.

Remark 3.20. (T0, T , (−)), with (−) as in Lemma 3.18, is a triple having unique quasi-negatives. In [7, Examples 2.8, 2.9, 2.12] the negation map is the identity map, whereas in [7, Examples 2.10, 2.11] it is the usual hypernegative.

Theorem 3.21. (T0, T , (−), ⊆) is a system of hypergroup type, for any hypergroup T0. Proof. We need to verify the conditions of Definition 2.28. To see Condition (i), suppose b = a ⊞ c◦. Since 0 ∈ c◦, we have a ⊆ a ⊞ 0 ⊆ a ⊞ c◦ = b. The other conditions are clear (since any nonempty subset of a singleton is that singleton). The relation ⊆ clearly satisfies S ⊆{a} iff S = {a}.

We call (T0, T , (−), ⊆) a hypersystem. MAJOR NOTE 3.22. Thus the theory of canonical hypergroups and hyperrings embeds into the theory of hypersystems. We make this more formal in Theorem 7.4. The recent surge in research in hypergroups provides further motivation and intuition for the study of systems.

Definition 3.23. A canonical hypergroup T0 is (−)-closed if a ⊞ b ∈ T whenever a 6= −b; T is (−)- bipotent if a ⊞ b ∈{a,b} whenever a 6= −b. ^ Lemma 3.24. T0 is (−)-closed, resp. (−)-bipotent, iff the hypersystem (P(T0), T , (−), ⊆) is metatangible, resp. (−)-bipotent. Proof. The definitions match. Example 3.25. (i) (−)-bipotent hypergroups include Viro’s “tropical hyperfield,” which is isomorphic to the tangible part of the supertropical algebra, the Krasner hypergroup (of the first kind), and the sign hypergroup (of the second kind), all of height 2. ALGEBRAS WITH A NEGATION MAP 19

(ii) A natural system that is not metatangible, the phase hyperfield, will be presented in Example 12.1, taken from [3]. It is idempotent of height 3. (iii) Viro’s “triangle” hyperfield of Example 12.1, is of the first kind and is neither idempotent nor metatangible. Distributivity also fails in its system. In order to accommodate hypergroups such as the “triangle” hyperfield, we relax the semiring assumption (i.e., distributivity) for A. Recently ties have been found in [27] between hyperfields and fuzzy rings, which also are described in terms of systems in Appendix B.

4. Exploring triples and systems

4.1. Layered semirings†. “Layered semirings” were introduced in [36], and called “extensions” in [3, Proposition-Definition 2.12]. They are of the form L × G, where L is a “layering semiring” and (G, ·) is an ordered monoid. In fact, associativity of multiplication in G is irrelevant, so we could also define “layered bimagmas.” Example 4.1. We assume that the “layering semiring” L is an arbitrary semiring with 1, but with a negation map that we designate as −. We can define the layered semiring as follows: A = L × G. The layers are the subsets {ℓ} × G for ℓ ∈ L. Multiplication is defined componentwise. Addition is given by: (ℓ1,b1) if b1 >b2; (ℓ1,b1) + (ℓ2,b2)= (ℓ2,b2) if b1 0, taking L = Zn, identify each layer modulo n. (This has height n and characteristic n.) (v) (The truncated semiring) A weird example, which leads to counterexamples in linear algebra in [5] and must be confronted. Fixing n> 1, we say that L = {1,...,n} is truncated at n if addition and multiplication are given by identifying every number greater than n with n. In other words,

k1 + k2 = n in L if k1 + k2 ≥ n in N; k1k2 = n in L if k1k2 ≥ n in N. The negation map is the identity. This triple has characteristic 0, since m 6= 1 for all m > 1, but it has height n. (vi) L itself is a classical algebraic structure, such as a ring, or an exterior algebra, or a Lie algebra. (vii) (exploded-ELT – special case of (vi)) L is the residue ring of a valuation with value group G, where now T = {(ℓ,a) ∈ L × G : ℓ 6=0}. Example 4.1 can be modiﬁed, viewing an ordered monoid as a bipotent semiring as in Remark 3.2(i). Example 4.3. Suppose G is a semiring with a negation map (−), whose addition yields a PO on G◦, via ◦ ◦ ◦ ◦ ◦ Remark 3.2(ii), with (b ) = b for all b. We write b1 <◦ b2 for bi ∈ G if b1

(ℓ1,b1) if b1 >◦ b2; (ℓ1,b1) + (ℓ2,b2)= (ℓ2,b2) if b1 <◦ b2;  ◦ ◦ ◦ (ℓ1 + ℓ2, b1) if b1 = b2.

 20 LOUIS ROWEN

◦ (ℓ1,b1) if b1 >◦ b2; ◦ (ℓ1,b1) + (ℓ2,b2)= (ℓ2,b2) if b1 <◦ b2;  ◦ ◦ ◦ (ℓ1 + ℓ2, b1) if b1 = b2.

 ◦ (ℓ1,b1) if b1 >◦ b2; ◦ ◦ ◦ (ℓ1,b1) + (ℓ2,b2)= (ℓ2,b2) if b1 <◦ b2;  ◦ ◦ ◦ (ℓ1 + ℓ2, b1) if b1 = b2.

This has a negation map given by (−)(ℓ,b) = (−ℓ, (−)b). The requirement (b◦)◦ = b◦ gives this example a supertropical flavor, but if we try to delete this assumption we have some difficulty defining the connection between (b◦)◦ and b◦.

4.2. Major examples of systems, by height. Before delving further into the theory, we describe some of the main examples in terms of height.

Example 4.4. Semiring and semifield systems, where =◦. (i) Height 1. This makes T0 = A. • Classical algebra, for example an integral domain. (There are many other examples in classical algebra, including graded algebras, cf. §9.2.) Here the quasi-negative is the usual negative, ◦ which is unique, and A = {0}. b ◦ a iff a = b + 0 = b, so we have the metatangible system (A, T , −, =). When T is a group, a classical system is just a partial field in the terminology of [7, Definition 4.2], also cf. [63]. In some ways we want the general theory of metatangible systems to mimic classical algebra. The negation map is of the second kind unless A has characteristic 2, in which case (−) is of the first kind. This helps to “explain” why the theory of metatangible systems of the first kind often has the flavor of characteristic 2. • The max-plus algebra T yields a pseudo-triple, taking A = T and (−) the identity map, so a◦ = a and A◦ = A, but quasi-negatives are far from unique, since whenever b

Remark 4.5. Triples of the first kind behave quite differently from those of the second kind. Triples of the first kind that contain 1 satisfy e′ = 1 + 1 + 1 = 3. • If e′ = 1, then we are in characteristic 2. • If e′ = e, i.e. 3 = e = 2, the system often has height 2, such as in the first two examples of Examples 4.4(iii), and when (−)-bipotent it is isomorphic to the supertropical domain: T is the set of tangible elements, and T ◦ is the set of “ghost” elements. • Otherwise we are usually in the more esoteric region of height ≥ 3, occurring for layered semirings, cf. Examples 4.4(iii), as well as certain hyperfields (Examples 3.25(ii,iii)). ALGEBRAS WITH A NEGATION MAP 21

T -Triples of the second kind often have either the flavor of classical algebra or of the symmetrized algebra. (−)-bipotent triples of the second kind are all idempotent since a + a ∈ max{a,a} = a. (The converse also holds for metatangible triples, as to be seen in Corollary 6.8.) 4.3. Tropically related examples viewed in terms of algebraic varieties and model theory. Let us see how well notions related to T -modules mesh with universal algebra. Definition 4.6 ([45]). A variety is a class V of algebraic structures (in universal algebra) which are systems closed in the following sense: (i) Any substructure of a structure of V is itself in V; (ii) If ϕ : A→B is a homomorphism with A∈V, then ϕ(A) ∈V; (iii) The Cartesian product of systems in V is in V. T -modules fit in the general framework of universal algebra once we have the algebraic structures A and T . Triples fit in, viewing the negation map as a unary operator on each of A and T . However, the surpassing relation is not an identity, since it is not symmetric. Varieties arising naturally in tropical mathematics include semirings†, semirings, idempotent semirings†, modules over semirings†, bimagmas and semialgebras, and super-semialgebras. The defining identities are the familiar ones. Here are some subtler instances. Example 4.7.

(i) Matrix semirings. The n × n matrix structure can be obtained in terms of matrix units {eij : 1 ≤ i, j ≤ n}, viewed as constants (0-are operators) satisfying the identities n eii = 1; eij ekℓ = δj,keiℓ, 1 ≤ i,j,k,ℓ ≤ n. i=1 X We get a variety by viewing the matrix units as part of the structure in this way. Namely, given a matrix semiring A = Mn(R) with matrix units {eij :1 ≤ i, j ≤ n} and a semiring homomorphism ′ ′ ϕ : A → A , the set {ϕ(eij ):1 ≤ i, j ≤ n} is a set of n×n matrix units for A . Note that the base ring R for the matrices can be recovered as ei,1Ae1,i, so does not have to be notated separately. One could make the base ring commutative by means of a suitable identity. One major point is that the standard proofP given for example in [59, Proposition 13.9] does not use negation. ◦ Matrices gives rise to the trace operator tr(a) = eiiaeii. Note, for n ≥ 2, that tr(I) = 1 over the supertropical semiring†. The determinant is more problematic since the classical formula involves negatives; we shall return to this issue in §5.1.P (ii) Formal traces. Since much of linear algebra involves the trace bilinear form, let us formalize the trace from the previous example. For a semialgebra A over a commutative associative ring C deﬁne a trace operator tr : A → C satisfying the identities

tr(x1x2) = tr(x2x1); tr(1A)= n1C. The trace operator can be viewed as a unary operator, so we can define the variety of semialgebras with traces. (iii) Bilinear forms on a module M. Linear algebra, which has played a significant role in tropical mathematics, can be taken over systems. Although one can define modules M over semirings† (and systems) the same way as modules over rings, [39, 48], much research focuses on the free module M = A(n). A bilinear form is an operator b : M × M → A satisfying the classical identities defining bilinearity; an example is the inner product. (iv) Quadratic forms. The general definition of quadratic form over a semiring is given in [39], also cf. [14]. Continuing (iii), we introduce a (quadratic) operator Q : M → A satisfying the identity Q(x + y)= Q(x)+ Q(y)+ b(x, y), where the binary operator b(x, y) is the accompanying bilinear form of Q. We get a variety by incorporating the quadratic form into the structure. (v) Blueprints. Lorscheid [53, Definition 1.1] has put tropical geometry in a rather general framework, which we review. 22 LOUIS ROWEN

A blueprint B is a monoid A with zero, together with an equivalence relation Φ on the † monoid semiring N[A]= { ai : ai ∈ A} (of finite formal sums of elements of A) that satisfies ′ ′ the following axioms (where we write ai ≡ aj whenever ( ai, aj ) ∈ Φ): (a) The equivalence Φ is additiveP and multiplicative. (Thus Φ is a congruence.) (b) The absorbing element 0 of A isP compatibleP with the zeroP of N[AP]; i.e., 0 ≡ empty sum. (c) If a,b ∈ A and a ≡ b, then a = b (as elements in A). When the monoid A has a given negation map (−), we extend the congruence Φ to Z[A] via the identity generated by the relation (−m)x ≡ m((−)x), and then the A-blueprint B has the negation map given by (−)a = ((−)1)a. Indeed, we verify the extra relation: if ai ≡ bj and a′ ≡ b′ , then i j P P P P ′ ′ ′ ′ ai − ai = ai(−) ai ≡ bi(−) bi = bi − bi. 4.3.1. Structures ofX tropicalX mathematicsX whichX do notX compriseX varieties.X X Several important concepts fail to correspond to varieties, because at least one of the key ingredients of Definition 4.6, either homomorphic images or direct products, is missing. Example 4.8. (i) Ordered semirings† versus bipotence. One takes Green’s order a + b = max{a,b}, using the sentence

x1 + x2 = x1 ∨ x1 + x2 = x2, ∀x1, x2, which passes to subalgebras and homomorphic images, but not to direct products, since for example (componentwise) (1, 2)+(2, 1) = (2, 2). (ii) Supertropical semirings†. One can describe supertropical semirings† (Deﬁnition 3.4), by declaring the constant e := 1ν to be both an additive and multiplicative idempotent, i.e., e + e = e and e2 = e. Then re = rν , so the map r 7→ re corresponds to the ghost map, cf. [37, Remark 2.1]. Supertropicality passes to subalgebras but not to direct products, just as in (i), and is ruined in images when ϕ(a)= ϕ(a′) with a

x1 + x2 + x3 = x1 ⇒ x1 + x2 = x1, ∀x1, x2, x3. This property passes to sub-semigroups and direct products, but not to homomorphic images, for example N → Z2, taken modulo 2. (It is an example of a “quasi-identity” in mathematical logic.) (iv) * Hypergroups. The power set construction fails to satisfy the conditions of Deﬁnition 4.6. 4.4. Properties of e, and the characteristic of a system A. Lemma 4.9. In a triple, if e′ ∈ T , then e′ = 1. Proof. If e′ ∈ T then e′ = e + 1 implies e′(−)1 = e + e ∈ A◦, implying e′ = 1 by unique negation.

When e′ = 1 the characteristic behaves in the familiar manner. Proposition 4.10. Suppose e′ = 1. If m′ = m, for m

Remark 4.11. When (−) is of the first kind, then 2 = e ∈ A◦, implying inductively 2m ∈ A◦ for each m. Consequently, if k ∈ A◦ for some odd k, then m ∈ A◦ for all m ≥ k. Example 4.12. (i) A has characteristic 1 iff it is idempotent. (ii) For (−) of the first kind, A has characteristic 1 or 2 iff e′ ∈ T , by Lemma 4.9, since in this case 1 = e′ = 3. ALGEBRAS WITH A NEGATION MAP 23

(iii) We say that A is quasi-periodic if m = m′ for some mm the j comprise a cycle with period m′ − m. (When m =1, this is precisely the deﬁnition of characteristic m′ −m. But one could have characteristic 0 with m> 1, as illustrated in the truncated algebra of Example 4.2(vi).) Proposition 4.13. If A has characteristic k and m + 1 = 1, then k divides m. Proof. A standard Euclidean algorithm argument. Write m = qk + r, where 0 ≤ r < k. By deﬁnition m ≥ k, and r + 1 = qk + r + 1 = m + 1 = 1. But r < k, so we must have r =0.

4.5. T -Strongly negated systems. The next definition becomes relevant when metatangibility is absent, as for the phase hyperfield. Definition 4.14. A system (A, T , (−), ) is strictly negated if, for any c, d ∈ A,

d(−)c 0 implies either c, d 0 or c d or d c. (4.1) A system (A, T , (−), ) is T -strictly negated, if for c ∈ T , d(−)c 0 implies c d.

Lemma 4.15. A strictly negated system (A, T , (−), ) is T -strictly negated whenever A = T ∪ A◦. Proof. If d ∈ T then d = c. If d ∈ A◦ then we cannot have d c, by Deﬁnition 2.28.

Although our emphasis in this paper is on metatangible systems, “T -strictly negated” is a reasonably broad substitute since it has several nice consequences and holds for hypersystems:

Proposition 4.16. Any hypersystem (T0, T , (−), ⊆) is T -strictly negated. Proof. 0 d(−)c means 0 ∈ d ⊞ (−c) so 0 ∈ (−c) ⊞ a for some tangible a ∈ d; thus a = c, i.e., c ∈ d.

Also we will see in Proposition 6.18 that every cancelative metatangible system of height 2 is strictly negated.

4.5.1. T -reversibility. Deﬁnition 4.17. A surpassing relation in a system is T -reversible if a b + c implies b a(−)c for a,b ∈ T . A T -reversible system is a system (A, T , (−), ) where is T -reversible.

Proposition 4.18. Any T -strictly negated system (A, T , (−), ) is T -reversible. Proof. Suppose that a b + c for a,b ∈ T , i.e., b + c(−)a 0. Then (a(−)c)(−)b = (−)(b + c(−)a) 0, so b a(−)c.

5. Related notions In this section we apply extra algebraic structure to triples and systems.

5.1. Matrices over triples and systems. Matrices were deﬁned in Example 4.7(i). (Mn(T0), ·) needs no longer be a monoid even when (T , ·) is a

monoid, because the matrix product involves addition. Thus, we deﬁne TMn(A) to be ∪T ei,j , yielding the

triple (Mn(A), TMn(A), (−)). Matrices over a T -system are a TMn(A)-module, as with classical algebra,

and we get a system over TMn(A), deﬁning (−) and componentwise. Summarizing, we have:

Remark 5.1. If (A, T , (−), ) is a bimagma system, then (Mn(A), TMn(A), (−), ) is a bimagma system, in view of Remark 2.44.

We also have a pseudo-triple (Mn(A), T , (−)), where T is identiﬁed with the scalar matrices, which works with linear algebra in §8. 24 LOUIS ROWEN

5.1.1. (−)-Determinants and singularity. We formulate some standard concepts from matrix algebra over semirings, largely translated from [2]. Deﬁnition 5.2. Suppose a commutative semiring A has a negation map (−). For a permutation π, write b : π even; (−)πb = ((−)b : π odd. The (−)-determinant |A| of a matrix A is

π (−) ai,π(i) . (5.1) i ! πX∈Sn Y even part a , and the odd part is a . The is even π∈Sn i i,π(i) odd π∈Sn i i,π(i) The matrix A is nonsingular if |A| ∈ T . A is singular if |A| 0. A is ◦-singular if |A| ∈ A◦. P Q P Q Lemma 5.3. (i) The (−)-determinant is linear in any given row or column.

(ii) If (ai,j ) (bi,j ) then |(ai,j )| |(bi,j )|. This yields the TMn(A)-system (Mn(A), TMn(A), (−), ). Proof. (i) The same as for the classical situation. (ii) Match the sums and products in the formula. Lemma 5.4. If two rows or columns of a matrix A are the same, then A is ◦-singular. Proof. The formula for the (−)-determinant partitions into pairs of opposite (−).

Proposition 5.5. If the first row v1 of A ◦-surpasses a linear combination of the other rows v2,...,vn, then A is ◦-singular. ◦ Proof. Breaking up the first row v1 = w + i≥2 αivi for some vector w, we write Aj for the matrix whose first row is v and A for the matrix whose first row is w, and see that |A| = |A|◦ + α |A |. j w P w j j Hence, by Lemma 5.4, |A| is a sum of elements of A◦. P 5.1.2. The special linear monoid. † In order for this system theory to be at our disposal for Mn(A) for a general commutative semiring A, we pass to Mn(A), with the switch negation map, in which case, for a matrix A , the (−)-determinant matches [2]. Namely, we define |A|◦ = |(A, (0))|. This is an ordered pair (a0,a1), where a0 is the even part of the determinantb and a1 is the odd part. These considerations lead us to the -special linear monoid SLn(Aˆ) := {A ∈ Mn(A): |A|◦ ◦ (1, 0)}. This is essentially the definition used in [41]. It contains all the elementary matrices, but is not generated by them multiplicatively, cf. [54]. Mimicking the classical algebraic groups, we can define PSLn(A) by taking SLn(A) modulo the congruence {(A, αA): α ∈ A}. Versions of the other algebraic groups will be obtained presently by utilizing involutions. 5.2. Polynomial triples. To set up affine geometry over systems, one defines the polynomial cbimagma A[λ] in the usual way over a cbimagma A, as a monoid cbimagma. We write λ for 1λ. A[λ1,...,λn] is defined inductively. T [λ] is not closed under multiplication, since for example (λ + 1)(λ(−)1) = λ2 + eλ(−)1. Instead, one takes i1 in TA[λ1,...,λn] to be the set of monomials {aλ1 ··· λn : a ∈ T , ij ∈ N0}.

When (A, T , (−)) is a triple, (A[λ1,...,λn], TA[λ1,...,λn], (−)) is a triple, under the negation map

i1 in i1 in (−)(aλ1 ··· λn ) = ((−)a)λ1 ··· λn , and has unique quasi-negatives if (A, T , (−)) has unique quasi-negatives. A surpassing relation on A is extended to A[λ1,...,λn] by comparing respective components. (One does the same for Laurent polynomials, rational functions, etc.) The analogy for cbimagmas also works. \ Remark 5.6. Aˆ[λ1,...,λn] =∼ A[λ1,...,λn], under the map

(α , α )λi1 ··· λin 7→ α λi1 ··· λin , α λi1 ··· λin , 0,˜i 1,˜i 1 n  0,˜i 1 n 1,˜i 1 n  ˜ ˜ ˜ i=(iX1,...,in) i=(iX1,...,in) i=(iX1,...,in)   ALGEBRAS WITH A NEGATION MAP 25 seen by matching components, and noting that the map is 1:1 and onto.

Polynomial systems are important in affine geometry. One often identifies polynomials in terms of ({1,...,n}) their values as functions, by mapping A[λ1,...,λn] to Fun(A , A), as given in Definition 4.6(iii).

(n) Remark 5.7. Generalizing [42, 46], a ∈ T is a systemic root of f ∈ A[λ1,...,λn] if f(a) 0. This yields an analog of the Zariski topology, in which the closed sets are the sets of systemic roots.

5.3. Involutions.

Deﬁnition 5.8. An involution on a system (A, T , (−), ) is an anti-isomorphism of degree 2, i.e., an additive homomorphism (∗) satisfying (∀ b,bi ∈ A): (i) (b∗)∗ = b, ∗ ∗ ∗ (ii) (b1b2) = b2b1, (iii) ((−)b)∗ = (−)b∗. ∗ ∗ (iv) If b1 b2 in A, then b1 b2. (v) T ∗ = T .

◦ ∗ ∗ ∗ ◦ (iv) is automatic for ◦, since (b1 + c ) = b1 + (c ) . Lemma 5.9. (c◦)∗ = (c∗)◦.

Proof. (c◦)∗ = (c(−)c)∗ = c∗(−)c∗ = (c∗)◦.

Example 5.10. Examples of involutions on the matrix semialgebra Mn(A) over A: t (i) The transpose map on Mn(A) is an involution denoted by A 7→ A . (ii) When n = 2m and A has a negation map, there is another involution, called the symplectic s A A At (−)At involution (s), given by 11 12 = 22 12 , where the A ∈ M (A). A A (−)At At ij m 21 22 21 11 Remark 5.11. We can deﬁne -orthogonal matrices via the condition (I, (0)) AAt, AtA, and thereby deﬁne the -orthogonal monoids, special -orthogonal monoids, and -symplectic monoids.

Involutions are studied in terms of symmetric and antisymmetric elements, given respectively as

{a ∈ A : a∗ = a}, {a ∈ A : a∗ = (−)a}.

Lemma 5.12. Deﬁne (A, ∗)+ := {a∗ + a : a ∈ A} and (A, ∗)− := {a∗(−)a : a ∈ A}. These sets respectively are symmetric and antisymmetric.

Proof. (a∗ + a)∗ = a + a∗ = a∗ + a, and (a∗(−)a)∗ = a + ((−)a)∗ = a(−)a∗ = (−)(a∗(−)a).

5.3.1. Involutions under symmetrization.

Proposition 5.13. If A has an involution (∗), then the symmetrization Aˆ has an involution given by

∗ ∗ ∗ (a0,a1) = (a0,a1).

The symmetric elements are {(a0,a1): each ai ∈ A is symmetric}. The antisymmetric elements are ∗ {(a0,a0): a0 ∈ A}. Proof. The ﬁrst two assertions are seen by matching components. For the last assertion, we have ∗ ∗ ∗ (−)(a0,a1) = (a1,a0), whereas (a0,a1) = (a0,a1), so matching components in the antisymmetric case ∗ shows that a1 = a0.

Corollary 5.14. The sets of symmetric and antisymmetric elements of Aˆ are precisely (Aˆ, ∗)+ and (Aˆ, ∗)− respectively. 26 LOUIS ROWEN

5.4. Localization of semiring triples. We use the standard technique of commutative localization, cf. [13], to pass from the case that T is a commutative monoid to the case that T is an Abelian group. (We defer noncommutative localization for future work.) Assume that S is a submonoid of T . One deﬁnes the equivalence (s1,b1) ≡ (s2,b2) when s(s1b2) = −1 s(s2b1) for some s ∈ S. In the cancelative case, we can dispose of s. We write s b for the equivalence class of (s,b). We localize a triple (A, T , (−)) with respect to S, via

−1 −1 −1 −1 −1 −1 s1 b1s2 b2 = (s1s2) b1b2, s1 b1 + s2 b2 = (s1s2) (s2b1 + s1b2). The ensuing triple is denoted (S−1A,S−1T , (−)), where (−)(s−1b) := s−1((−)b), s ∈ S. Lemma 5.15. (i) s−1((−)b) = ((−)s)−1b. (ii) (s−1b)◦ = s−1b◦. Proof. (i) Cross multiply to get (−)sb = ((−)s)b. (ii) (s−1b)◦ = s−1b(−)s−1b = s−1(b(−)b)= s−1(b◦). Proposition 5.16. If (A, T , (−)) is a triple with unique quasi-negatives, and S is a multiplicative submonoid of T , then the triple (S−1A, S−1T , (−)) also has unique quasi-negatives. −1 −1 Proof. Suppose s1 a1 is a quasi-negative of s a, for a ∈ T . Then −1 −1 −1 −1 ◦ (ss1) (sa1 + s1a)= s1 a1 + s a ∈ (S A) , ◦ −1 −1 implying sa1 + s1a ∈ A , and thus sa1 = (−)s1a = s1((−)a), and then s1 a1 = s ((−)a). In particular, if (A, T , (−)) is cancelative over a monoid T , then taking S = T we can localize T to the −1 group T T . For instance, one might localize (A[λ1,...,λn], TA[λ1,...,λn], (−)) (where A is a T -semiﬁeld)

at the monomials, to get the Laurent triple (A[[λ1,...,λn]], TA[[λ1,...,λn]], (−)). 5.5. Tensor products with a negation map, and their semialgebras. The tensor product is a very well-known process in general category theory, [30, 50, 51, 66], and has been studied in the context of monoidal categories. Here we consider the tensor product of module triples and semialgebra triples over commutative semirings†. These are described in terms of congruences, as given for example in [51, Definition 3]. Let us work with a module V over a commutative associative semiring† C. If V has a negation map (−), then we can incorporate the negation map into the tensor product, defining a negation map on V ⊗C W by (−)(v ⊗ w) = ((−)v) ⊗ w. When W also has a negation map we define a negated tensor product by imposing the extra axiom ((−)v) ⊗ w = v ⊗ ((−)w). (This is done by modding out the usual congruence defining the tensor product by the congruence generated by all elements ((−)v ⊗ w, v ⊗ (−)w) .) From now on, the notation V ⊗ W includes this negated tensor product stipulation, and C is understood. Remark 5.17. One can easily prove the following facts, modifying for example [59, Chapter 18]:

(i) If fi : Vi → Wi are module maps then there is a unique map f1 ⊗ f2 : V1 ⊗ V2 → W1 ⊗ W2 satisfying (f1 ⊗ f2)(v1 ⊗ v2)= f1(v1) ⊗ f2(v2). (ii) The tensor product (A, T , (−)) ⊗ (A′, T ′, (−)′) of triples is a triple ′ ′ ′ ′ (A ⊗ A , {a ⊗ a : a ∈ T ,a ∈ T }, (−) ⊗ 1A′ ).

• This deﬁnition is suited towards “multilinear” algebra, where TA⊗A′ is a set of rank 1 tensors. • We call a ⊗ a′ a simple tensor. • If F is a commutative associative semialgebra over C, then ⊗C F yields a functor from semialgebras over C to semialgebras over F , which extends to triples. (In particular, this holds when F is the symmetrization of C.) ALGEBRAS WITH A NEGATION MAP 27

• The tensor product is a semiring when A and A′ are semirings. This enables us to view matrices and polynomials in terms of tensor products. Next, as usual, given a module V over C, one deﬁnes V ⊗(1) = V, and inductively V ⊗(k) = V ⊗ V ⊗(k−1). From what we just described, if V has a negation map (−) then V ⊗(k) also has a natural negation map. ⊗(n) Now deﬁne the negated tensor semialgebra T (V )= n V (adjoining a copy of C if we want to have a unit element), with the usual multiplication. If V has a negation map then so does T (V ), induced ⊗(k) L from the negation maps on V ; writinga ˜k = ak,1 ⊗···⊗ ak,k, we put

(−)(˜ak) = ((−)ak,1) ⊗···⊗ ak,k. One can view the polynomial semialgebra of §5.2 as a negated tensor semialgebra, where V = Aλ.

6. Metatangible triples and their systems

This section explores metatangible triples (A, T , (−)) (Definition 2.25) and metatangible systems. Decisive results for metatangible triples are available, which cover the tropical applications, but only hold for certain hyperfields. Eventually we show in Theorem 6.30 that ◦ finishes out the system (although there are other possible surpassing relations). Many of our arguments involve the height of an element, from §2.1.1. Here are some of the main results. Every (−)-bipotent triple clearly is metatangible. Perhaps as a surprise, conversely, by Theorem 6.7, a metatangible triple either is (−)-bipotent (with the ensuing tropical flavor) or satisfies e′ = 1 (in which case e◦ = e), which happens in classical algebra. The elements of metatangible triples have a surprisingly nice form to be given in Theorem 6.25, which enables us to prove various nice properties although there also are some annoying counterexamples. The heights of elements tie in with the characteristic of a triple, in describing surpassing relations in Theorem 6.30. Theorem 6.41 enables us to describe the symmetrized semialgebra in terms of classical considerations about sums of squares. This pertains to “real” groups of tangible elements, in Proposi- tion 6.46. Theorem 6.50 largely classifies metatangible systems, in the sense that it describes the properties of all of the major examples. But of course a full classification requires all of the examples, some of which are rather esoteric and not present here. Some more examples are given in [61]. 6.1. Basic properties of metatangible triples and systems. The key property is: Lemma 6.1. A triple (A, T , (−)) with unique quasi-negatives is metatangible iff T + T ⊆T ∪A◦. Proof. (⇒) Immediate from the definition. (⇐) If a′ + a∈ / T , then a′ + a ∈ A◦, so a′ = (−)a by uniqueness of quasi-negatives.

Remark 6.2. By Lemma 2.15(i), in a metatangible triple, if a1 6= a2 in T and a1 + a2 = a2, then a1(−)a2 = (−)a2 and a2(−)a1 = a2. 6.2. Metatangible versus (−)-bipotent. One may ask whether metatangible systems are necessarily (−)-bipotent. Here is a surprisingly strong observation. Lemma 6.3. One of the following must hold, for a,b ∈ T in a metatangible triple: (i) a = (−)b. (ii) a + b = a (and thus a◦ + b = a◦). (iii) a◦ + b = b. We cannot have both (ii) and (iii) simultaneously. Proof. Assume that a 6= (−)b, implying a + b ∈ T . If a + b 6= a, then a◦ + b = (a + b)(−)a ∈ T . Hence b = a◦ + b by Deﬁnition 2.28(i,v). For the last assertion, if a + b = a then a◦ + b = a(−)a = a◦, and if also a◦ + b = b then a◦ = a◦ + b = b ∈T ∩T ◦ = {∅}. 28 LOUIS ROWEN

Examples of a non-bipotent, metatangible triple are the classical (Z, +), or (Z/n, +), taking T = A\{0}, but they all have a classical ﬂavor, because of the next observation.

Lemma 6.4. Suppose in a metatangible triple (A, T , (−)) that ai ∈ T with a2 6= (−)a1 and a3 := ′ a1 + a2 ∈/ {a1,a2}. Then a1 = a3(−)a2 and a2 = a3(−)a1. Let A be the sub-semigroup generated by ′ ′ ′ ′ ◦ {(±)a1, (±)a2, (±)a3}, and T = A ∩ T . Then (A , T , (−)) is a triple in which (a1 + a2 + a3) acts as ′ the 0 element, and ai = aie . The unique quasi-negative of ai is ai+1 + ai+2, subscripts modulo 3.

Proof. a1 = a3(−)a2 and a2 = a3(−)a1 by Lemma 2.15(ii). Hence ◦ a3 = a1 + a2 = a3(−)a2 + a2 = a3 + a2, ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ and likewise a3 = a3 + a1, implying a3 = a3 + a2 = a1 and likewise a3 = a2. Now a3 + a3 = a3 + a1 = a3, ◦ so all the ai act as the zero element for a3, and symmetrically also for a1 and a2. Finally, ◦ ◦ ◦ ◦ (a1 + a2 + a3) = a1 + a2 + a3(−)a3 = a3 + (−)a3 = a3, ′ and ai = ai(−)ai + ai = aie . ◦ The last assertion holds since (a1 + a2 + a3) is the only quasi-zero.

Remark 6.5. We say that {a1,a2,a3} satisfy the trio property when a2 6= a1 in Lemma 6.4. When ′ ′ ′ ◦ (−) has the first kind, the trio property is automatic and T = {a1,a2,a3} and A = T ∪{(a1 +a2 +a3) }. So suppose for the rest of this remark that (−) has the second kind in Lemma 6.4. We could have the ′ degenerate condition a2 = a1 in Lemma 6.4. Now T consists of multiples of a1, but could conceivably be infinite. Bypassing the previous paragraph, assuming that a + a = a for all a ∈ T , in other words that A is ′ ′ ′ ′ ◦ idempotent. Then T = {(±)a1, (±)a2, (±)a3} since ai = aie , and again A = T ∪{(a1 + a2 + a3) }. We do have a non-classical example for pseudo-triples: Example 6.6. In Example 4.2(v), take L = {0, 1,ℓ,ℓ+1} to be the finite field of 4 elements. Although only a pseudo-triple since 1 does not generate L, the layered algebra A = L × G does satisfy the metatangibility condition and is of first kind, but not (−)-bipotent (since (1,a)+ (ℓ,a) = (ℓ +1,a)), satisfying e′ = 3 = 1. Summarizing, we have: Theorem 6.7. Any cancelative metatangible unital triple (A, T , (−)) satisfies one of the following cases: (i) A is (−)-bipotent. (ii) e′ = 1, with one of the following two possibilities. (a) (−) is of the first kind, with (A, +) of characteristic 2. (In other words e′ = 3 = 1 = 1.) In this case, A has height ≤ 2. (b) (−) is of the second either with (A, +) of finite characteristic or with {m : m ∈ Z} all distinct. Proof. If A is not (−)-bipotent, we have a,a′ ∈ T with a′ 6= (−)a and a + a′ ∈/ {a,a′}. By Lemma 6.4, a = a + a(−)a; canceling a yields e′ = 1. If (−) is of the first kind, then 1 = e′ = 3, so A has characteristic 2. If (−) is of the second kind, then we conclude with Proposition 4.10. Corollary 6.8. Suppose (A, T , (−)) is a cancelative unital metatangible T -semifield triple of the second kind. Then A is (−)-bipotent iff A is idempotent. Proof. (⇒) Since (−)1 6= 1, we have 1 + 1 ∈{1, 1} = {1}, so a + a = a(1 + 1)= a1 = a. (⇐) e′ = 1(−)1 = e, so e′ 6= 1, implying A is (−)-bipotent by Theorem 6.7. MAJOR NOTE 6.9. This dichotomy of (−)-bipotence and e′ = 1 flavors our entire discussion. There are examples for each of these conclusions. • The classical triple satisfies e′ = 1, but is not (−)-bipotent. • The triple of the standard supertropical algebra is (−)-bipotent of first kind, satisfying e′ = e, and the same holds for the symmetrized algebra. • The triple of the modified symmetrized algebra of Example 3.11 is (−)-bipotent of second kind. • The ELT triple satisfies both conditions and is of the second kind. ALGEBRAS WITH A NEGATION MAP 29

• Example 4.2(i) is (−)-bipotent of the ﬁrst kind, even failing e′ ∈T ∪T ◦.

6.3. Computations in a metatangible triple (A, T , (−)) via T . k−1 k k−1 Lemma 6.10. Suppose ai ∈ T with i=1 ai ∈ T but i=1 ai ∈/ T . Then i=1 ai = (−)ak. k k−1 Proof. i=1 ai ∈ T unless i=1 ai =P (−)ak. P P t k−1 PropositionP 6.11. SupposeP i=1 ai 0. Then for some k 1, with k ∈ T . (iii) k> 1, with k − 1 = (−)1, and k = e. (iv) (−) is of the ﬁrst kind and k is even. Moreover, if (−) is of the second kind, then either A is idempotent or k + 1 = 1, implying the characteristic of A divides k. Proof. First assume that n 6= 1 for each n> 1. By induction, we may assume that n ∈ T , and then each (n + 1)= n + 1 ∈ T and we conclude that Z ⊆ T by negating. Thus n + 1 = 1 for some n, implying A has some characteristic k. We are done unless k> 1. If k ∈ T then we have (ii), so assume that k ∈/ T . Assume presently that (−) is of the second kind. If T is also (−)-bipotent then 1 + 1 = 1, so we have (i). Otherwise 2(−)1 = e′ = 1 by Theorem 6.7, implying by induction that j(−)1 = j − 1 for all j > 1. Take k′ minimal such that k′ ∈/ T . (Clearly k′ > 1, and k′ ≤ k.) Then by Lemma 6.10, k′ − 1 = (−)1, k′ = e, and k′ + 1 = e′ = 1, so k′ = k and we have (iii). k+1 ◦ So we may assume that (−) is of the ﬁrst kind. If k is odd then 1 = k + 1 = 2 e ∈ A , which is a contradiction, so k is even. 6.3.2. Comparing A to T . Lemma 6.13. If (A, T , (−)) is a metatangible triple, then T + A◦ = A.

t ◦ Proof. Write b = i=1 ai + c with ai ∈ T , and t minimal. (Clearly t is at most the height of b, since one could take c = 0.) Then at = (−)at−1, since otherwise we could take at + at−1 ∈ T and lower t. But ◦ ◦P t−2 then b = at−1 + c + i=1 ai and we apply induction on t, reducing to the case t = 1. Actually we are aimingP for the condition of whether T ∪ A◦ = A. Lemma 6.14. Suppose the unital triple (A, T , (−)) is metatangible. (i) If (−) is of the second kind, then T ∪ A◦ = A. (ii) If (−) is of the ﬁrst kind, then A = ∪{mT : m ∈ N}. t Proof. Write b = i=1 ai with each ai ∈ T , t minimal. This implies ai = (−)aj for each i, j since otherwise we could combine them to an element of T and reduce t. For t > 2, when e′ = 1, we can P replace a1(−)a1 +a1 by a1, so again reduce t. Hence we are done by Theorem 6.7 unless A is (−)-bipotent. ◦ But then for second kind we get b = (±a1) or b = a1, yielding (i). Thus we may assume that a1 = (−)a1. we obtain (ii) using induction on height. Let us consider the possibilities for e′. Proposition 6.15. One of the following must hold in a metatangible unital triple A: (i) e′ ∈ T , and then e′ = 1. 30 LOUIS ROWEN

(ii) e′ ∈ T ◦, and then e′ = e. (iii) e′ has height ≥ 3, i.e., e′ ∈T/ ∪T ◦, and then (−)1 = 1, with e′ = 3. Proof. By Lemma 4.9 we may assume that e′ = 2(−)1 is not tangible, or else we have (i) by unique quasi-negation. But then if 2 is tangible then 2 = 1, implying e′ = e. Thus we may assume that 2 is not tangible, so (−)1 = 1. If e′ = a◦ = 2a for a ∈ T with a 6= 1, then writing b = 1 + a ∈ T , we have b + a = 1 +2a = 1 + e′ = e + e ∈ A◦, and thus b = a. But then 1 + b = 1 + a = b, so 3a = e′ + a = 1 + 1 + b = 1 + b = b = a, implying 1 = 3 = e′. We are left with (iii). 6.3.3. Triples of height 2. Our earlier considerations give decisive results for height 2, which include both the supertropical and symmetrized constructions. Proposition 6.16. The following assertions are equivalent for a unital triple (A, T , (−)) (not presumed a priori to be metatangible): (i) T ∪ T ◦ = A, (ii) A is metatangible of height 2, (iii) A is metatangible with e′ ∈{1,e}. Proof. ((i) ⇒ (ii)) A fortiori. ((ii) ⇒ (iii)) We exclude (iii) in Proposition 6.15. ((iii) ⇒ (i)) The assertion is that A has height 2. We need to show that any sum of T ∪ T ◦ and elements of T remains in T ∪T ◦, and it is enough to show that if a ∈ T and b ∈T∪T ◦, then a+b ∈T∪T ◦. The assertion is clear if b ∈ T , and follows from Lemma 6.3 unless b = a◦, in which case a + b = ae′ ∈{a,ae}.

Corollary 6.17. Any metatangible triple of second kind has height 2. Proof. By Proposition 6.16 since e′ = e. An example of a non-metatangible hypersystem of height 2 is given in Example 12.1.

Proposition 6.18. Every unital metatangible system of height 2 is strictly negated, when =◦ . Proof. Suppose d(−)c 0. The result is immediate when c, d ∈ T ◦, and clear by unique quasi-negation for c, d ∈ T (since then c = d), so by Proposition 6.16 we may assume that c ∈ T and d, d(−)c ∈ T ◦. Write d = a◦ for a ∈ T . We cannot have d(−)c = c since T ∩ A◦ = ∅. We are done if d = d(−)c = c + a◦. Thus, applying Lemma 6.3 for b = c, we may assume that c = (±)a. Now e′ 6= 1 since otherwise c = c+c◦ = c+a◦ = c+d. Hence e′ = e by Proposition 6.16(iii), implying c ce′ = ce = a◦ = d. Proposition 6.19. Any metatangible triple (A, T , (−), ) over a group T is either (−)-bipotent or of height 2. Proof. We are done by Theorem 6.7 unless e′ = 1, so Proposition 6.15 implies (A, T , (−), ) has height 2.

6.3.4. The natural pre-order on A◦. ◦ ◦ ◦ ◦ ◦ ◦ ◦ Lemma 6.20. Any triple has the partial pre-order < , given by a1 ≤ a2 iff a1 = a2 or a1 + a2 = a2. ◦ ◦ ◦ a1 ≤ a2 iff a1 ≤ (−)a2, iff (−)a1 ≤ a2. ◦ ◦ ◦ ◦ ◦ ◦ Proof. Suppose a1 + a2 = a2 and a2 + a3 = a3. Then ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ a1 + a3 = a1 + (a2 + a3) = (a1 + a2)+ a3 = a2 + a3 = a3. The other verifications are patent.

Proposition 6.21. For any metatangible triple, with ai ∈ T , ◦ ◦ ◦ ◦ ◦ (i) a1 + a2 ∈{a1,a2, 2a1}, the last possibility occurring when a1 = (±)a2. ALGEBRAS WITH A NEGATION MAP 31

(ii) In particular, ≤◦ is a surpassing ◦-PO.

Proof. (i) by Lemma 6.3, applied to both a2 and (−)a2. (The ﬁrst two possibilities arise when a1 6= (±)a2, and the last, 2(a1(−)a1), when a1 = (±)a2). (ii) Follows at once from Deﬁnition 2.28.

Proposition 6.22. Any metatangible system (A, T , (−), ) has the congruence ◦ ◦ Φ= {(b1,b2): b1 = b2}, and the corresponding system of A/Φ is (−)-bipotent of ﬁrst kind.

′ ′ Proof. If (b1,b2), (b1,b2) ∈ Φ and a ∈ T then ◦ ◦ ◦ ◦ ′ ◦ ◦ ′ ◦ ◦ ′ ◦ (ab1) = ab1 = ab2 = (ab2) , (b1 + b1) = b1 + b1 = b2 + b2 , implying Φ is a congruence, modulo which (−)b becomes b since b◦ = ((−)b)◦. Furthermore, Lemma 6.3 yields (−)-bipotence since a◦ + b = b implies (a + b)◦ = a◦ + b◦ = b◦.

6.4. Uniform elements and height. A cancelative metatangible triple A is called exceptional if it is of ﬁrst kind, of height > 2. Case (ii) of Theorem 6.7 is ruled out, so A is (−)-bipotent. The main example is the layered triple (Example 4.1). Exceptional triples also are a source of weird counterexamples, as in [61, Examples 7.40 and 7.42].

Deﬁnition 6.23. A nonzero element c ∈ A of height mc ∈ N is uniform if for some element cT ∈ T , one of the following three possibilities occurs:

(i) Type 1: mc =1, i.e., c = cT ∈ T , ◦ (ii) Type 2: c = cT . (iii) Type 3: mc ≥ 3, and the triple (A, T , (−)) is exceptional, with 3 6= 1. We call this the uniform presentation of c. MAJOR NOTE 6.24. The uniform presentation enables us to reduce many proofs to the tangible case. Type 2 is the only case for which we do not have c = mcc, and requires separate treatment. If we can show that some element has Type 3, the negation map is of ﬁrst kind, and the notation becomes simpler. The uniform presentation need not be unique. For example, in a ring, all quasi-zeros are equal (to 0). But we will see in Theorem 6.28 that quasi-zeros are the only elements with non-unique presentation in (−)-bipotent triples. Theorem 6.25. Every element of a cancelative metatangible triple (A, T , (−)) is uniform.

Proof. Let m be the height of c. We may assume that m ≥ 2. If m = 2 and a1 6= (−)a2, then c ∈ T , so ◦ a1 = (−)a2 and c = a1 . For m ≥ 3, if some ai 6= aj then ai + aj ∈ T , contradicting definition of height. Hence all ai = a1. If 3 = 1 then we replace a1 + a1 + a1 by a1, again contradicting the definition of height. If (−) is of second kind, then a1 + a1 is tangible, again contradicting the definition of height. Hence (−) is of first kind, so A is exceptional.

Corollary 6.26. There are the following possibilities for c + d in a cancelative metatangible triple (A, T , (−)), where mc ≤ md:

(i) c, with cT + dT = cT . (ii) c, where mc =1 and md =2. (iii) d, with cT + dT = dT . (iv) d, where md ≥ 2 and c + dT = dT . (v) cT + dT ∈ T , where c = cT and d = dT but c 6= (−)d. ◦ ◦ ◦ ◦ (vi) (cT + dT ) ∈ T , where c = cT and d = dT but cT 6= (−)dT . ◦ (vii) 2cT , where c = d and mc =2. ◦ (viii) cT , where c = cT = (−)dT = (−)d. (ix) (mc + md)cT , where cT = dT . 32 LOUIS ROWEN

Proof. If cT + dT ∈{cT , dT } we get (i), (iii) by iterating, noting by Remark 6.2 to handle Type 2, that for tangible elements a + b = b also implies (−)a + b = b. Thus, we assume that cT + dT ∈{/ cT , dT }. ◦ For the next three paragraphs we suppose that mc =1. If md = 2 then d = dT and then by Lemma 6.3, we have three cases. Either c + dT = dT , in which case c + d = d, which is (iv), or c + dT = c, which ◦ is (i), or excluding these yields c = (−)dT , so c + d = c(−)c + c. Hence c + d(−)c ∈ A . If c + d ∈ T then c + d = c by unique quasi-negation, yielding (ii). That leaves us with c = (−)d ∈T ∩A◦ = ∅, a contradiction. ◦ If d ∈ T then c + d ∈ A implies d = (−)c, yielding (viii); if cT = dT we have (v). If md ≥ 3 then we write d = 2d + (md − 2) and apply the previous two paragraphs to get c which is (i), or d which is (iv), or we have cT = dT , yielding (ix). ◦ ◦ ◦ Hence we may assume mc ≥ 2. If c = cT and d = dT then c + d = (cT + dT ) , yielding (v) and (viii). ◦ ◦ If c = cT and md ≥ 3, then we again apply Lemma 6.3 to cT and dT , yielding (i), (iv), or (ix). For mc ≥ 3, if we do not have (i) or (iii), then cT = (−)cT = dT , yielding (ix). Corollary 6.27. c + d ∈ A◦ in a cancelative unital metatangible triple (A, T , (−)) precisely in the following situations, again according to the Type of c: (i) Type 1. (a) d = (−)c and c + d = c◦. ◦ (b) d = dT has Type 2 and: ◦ ′ ′ (1) dT = (±)c so d = c , and c + d = ce (when e = e). (2) dT + c = dT , so c + d = d. (c) d = mddT has Type 3 and: ◦ (1) dT = c so c + d = (md + 1)c (when this is in A ). ◦ (2) (for d ∈ A ) dT + c = dT , so c + d = d. (ii) Type 2. (a) d has Type 1 with c + d = c. ◦ ◦ (b) d = dT has Type 2 with c + d = (cT + dT ) . (c) d has Type 3. (1) dT = (−)cT . If (−) is of second kind, then c + d = c. ◦ ◦ If (−) is of first kind, then c =2dT and c + d = (u + 2)b, which is in A if u + 2 ∈ A . ◦ ◦ (2) (for d ∈ A ) cT + dT = dT . Then c + d = d. (3) cT + dT = cT . Then c + d = c. (iii) Type 3. (Then (−) is of first kind.) (a) d has Type 1, i.e., md =1. ◦ (1) dT = cT , with (mc + 1)d ∈ A . ◦ (2) dT + c = c, with mcc ∈ A . (b) d has Type 2. (1) dT = cT . Then d =2cT and c + d = (m + 2)cT . (2) cT + dT = dT . Then c + d = d. ◦ (3) (for c ∈ A ) cT + d = cT . Then c + d = c. (c) d has Type 3. ◦ (1) dT = cT . Then c + d = (mc + md)cT , which could be in A . ◦ (2) (for d ∈ A ) cT + dT = dT . Then c + d = d. ◦ (3) (for c ∈ A ) cT + d = cT . Then c + d = c. Proof. We go through the cases of Corollary 6.26. (v) is impossible. If c has Type 1 then (i), (ii), (vi), and (vii) are impossible; (iii), (iv) yield (b)(2) or (c)(2); (viii) yields (a); (ix) yields (b)(1) or (c)(1). If c has Type 2 then (ii) and (viii) are impossible; (i) yields (a)(2) or (c)(3); (iii) yields (b) or (c)(2); (iv) yields (b) or (c)(2); (vi), (vii) yield (b) ; (ix) yields (c)(1). If c has Type 3 then (−) must be of first kind and (ii), (vi), (vii), and (viii) are impossible; (i) yields (a)(2); (iii), (iv) yield (b)(2) or (c)(2); (ix) yields (c)(1).

The cases (i)(c)(1), (ii)(c)(1), (iii)(b)(1), and (iii)(c)(1) are particularly intriguing. A suﬃcient condition clearly is for mc + md to be even, and we will consider necessity in Corollary 6.29 below. ALGEBRAS WITH A NEGATION MAP 33

Distributivity for cancelative A can be obtained from metatangibility, cf. [61, Theorem 7.34]. We turn to uniqueness of the uniform presentation. ◦ ′ ′ ◦ Theorem 6.28. Suppose c = mccT or c = cT has another presentation mc′ cT or cT in a cancelative unital metatangible triple (A, T , (−)). Then one of the following occurs taking mc′ minimal with respect to c′: ′ (i) cT = cT , and thus mc = mc′ . ◦ ′◦ (ii) mc = mc′ =2, and cT = cT . (iii) mc =1, (−) is of the second kind, and the triple is not (−)-bipotent. Consequently, the uniform presentation of any element of height > 2 is unique. ′ ′ ′ ′ ′ Proof. Write m = mc and m = mc′ . If m = m = 1 then cT = c = c = cT , yielding (i). If m = m =2 then we have (ii), so we may assume m 6= 2. If m′ = 1 then m 6= 2 since otherwise c ∈T ∩A◦ = ∅. If m = 1 then we have (iii) unless (−) is of the first kind or , in which case either 3 = 1 or (A, T , (−)) is (−)-bipotent, yielding (i) or (ii). Thus we may assume that m ≥ 3, so (−) is of the first kind, and height 2 (m = 2) is no longer relevent. If 3 = 1, we can reduce m,m′ mod 2 and get (i), with m = m′ = 1. Hence, by Theorem 6.7, we may assume that (A, T , (−)) is (−)-bipotent. ′ ′ ′ ′ ′ Assume that cT 6= cT . If cT + cT = cT then cT + m cT = cT by iteration, so for k ≥ m +1, ′ ′ (m + 1)cT = cT + mcT = cT + m cT = cT ; (6.1) ′ ′ ′ ′ ′ ′ ′ ′ ′ (k + 1)cT = (k +1 − m )cT + m cT = (k +1 − m )cT + mcT = mcT = m cT , (6.2) ′ ′ which likewise is (k + 2)cT . By cancelation we get m + 1 = 1 and k + 2 = k + 1 for all k ≥ m + 1. Putting these two equalities together shows that 1 = m + 1 = 2m + 1 = ··· = m′m + 1 = m′m + 2 = m + 2 = 1 + 1, ′ ′ ′ ′ so cT = m cT = mcT = cT . Hence minimality yields m = m , yielding (i). ′ ′ ′ The analogous argument holds if cT + cT = cT . Hence we may assume that cT = cT , and again we have (i). The second assertion follows from the first. The case (ii) is truly exceptional, since the classical triple has a(−)a = a − a = 0 for every a. Likewise the ambiguity in (iii) can arise there since 3 = 3 · 1. However, when the triple is (−)-bipotent, one has cT = (±)cT , cf. [4, Proposition 2.14], and (iii) does not arise. Corollary 6.29. Suppose A is as in the theorem, and c ∈ A◦ has height m. Then m =2k for some k. ◦ Proof. We write c = b . There is nothing to prove unless mc ≥ 3, in which case c = 2mbbT = 2mbcT , implying m =2mb. 6.5. Surpassing relations on a metatangible triple. Let us see how surpassing relations can arise on metatangible triples to yield systems. Theorem 6.30. Suppose (A, T , (−)) is a metatangible triple.

(i) ◦ is a surpassing relation, so (A, T , (−), ◦) is a metatangible system. ′ ′ ′ (ii) We say that a pair (b,b ) with b b is usual if b ◦ b . A pair which is not usual is unusual. One of the following holds for any cancelative unital metatangible triple, with surpassing ◦-PO : (a) = ◦, i.e., all pairs are usual; (b) The triple (A, T , (−)) is exceptional. (c) The triple (A, T , (−)) has height 2, and there is an unusual pair satisfying b + b′ = b. In this case, if is a surpassing ◦-PO, then (−) is of the ﬁrst kind, with A of of characteristic 2. Proof. (i) By Proposition 2.40 and Proposition 2.21. (ii) If b or b′ has height ≥ 3, Theorem 6.25 says that the triple is exceptional. Thus we may assume that both have height ≤ 2. We shall show that an unusual pair (b,b′) yields (c), relying heavily on Lemma 6.3 and Remark 6.2. If b,b′ ∈ T then b = b′ and the assertion is trivial. If b′ ∈ T we cannot have b of height 2, by Proposition 2.21. 34 LOUIS ROWEN

Thus we may assume that b′ ∈/ T , so b′ = d◦ for d ∈ T . We cannot have b + d = d, for then the pair would be usual (since b + d◦ = d◦ = b′), so we may assume by Lemma 6.3 and Deﬁnition 2.28(i),(v) that b = b + d◦ = b + b′, (6.3) and we have (c). Furthermore, in (c) assume that is a surpassing ◦-PO. We have b◦ = b◦ + b′, so b′◦ b◦ and b◦ b′◦, implying b◦ = b′◦. If(−) is of the second kind then A is idempotent so b∈ / T (since otherwise b = b + b′◦ = b + b◦ = b◦, a contradiction), and then b ∈ A◦ implying b = b2 = b◦ = b′◦ = (b′)2 = b′, contradicting unusuality. Hence (−) is of the ﬁrst kind. If e′ = 1 then 3 = 1, and we are done. ′ We are left with e 6= 1, implying that the system is (−)-bipotent. Write b = mbT . If bT = d then ′ ′ ′ mbT = (m + 2)bT , implying 3 = 1; if b1 + d = d then b + b = b implying (b,b ) is usual. Thus we may ◦ ◦ ′◦ ◦ ′◦ ◦ ◦ ′◦ assume that bT + d = bT . But now b + d = b, implying b + b = b , i.e. b b , as well as b b , and thus b◦ = b′◦. Hence b = b + b′◦ = b + b◦ =3b, so again 3 = 1 after all. Example 6.31. Here are some unexpected surpassing relations. Take the N-layered system of Example 4.2(i) (of characteristic 0), with L = N.

(i) Write (ℓ1,a1) (ℓ2,a2) if a1

but 1 6◦ 2. (ii) Write (ℓ1,a1) (ℓ2,a2) if a1 < a2 with ℓ1 > 1, or if a1 = a2 with ℓ1 ≤ ℓ2. Here 1 6 2 = (2, 1), whereas 2 = (2, 1) (3, 1) = 3 but 2 6◦ 3. One also has unusual pairs arising from Example 4.12. 6.6. T -classical metatangible triples. The usual classical triple satisfies a◦ = b◦ = 0 for all a,b, leading us to the following weaker definition. Definition 6.32. A triple (A, T , (−)) is T -classical if a◦ = b◦ for some a 6= (±)b in T . (A, T , (−)) is T -nonclassical if a◦ 6= b◦ for all a 6= (±)b in T . Corollary 6.33. If a cancelative unital metatangible triple (A, T , (−), ) is T -classical, then e′ = 1. Proof. Suppose a◦ = b◦ with a 6= (±)b. If e′ 6= 1 then by (−)-bipotence (Theorem 6.7) we may assume that a + b = b. But then by Remark 6.2, b = a + b = a + (b(−)a)= a◦ + b = b◦ + b = e′b, implying e′ = 1 after all. There is a nice partial converse. Lemma 6.34. Any non-(−)-bipotent metatangible triple (A, T , (−), ) is T -classical. Proof. Suppose a 6= (±)b and a + b∈ / {a,b}. Then a◦ + b = (a + b)(−)a ∈ T , implying a◦ + b = b by unique quasi-negation. Hence (a + b)◦ = a◦ + b(−)b = b(−)b = b◦, with a + b 6= b. [61, Example 3.12] provides a way to merge classical and nonclassical in a metatangible triple. 6.6.1. Anti-negated triples. Next we view the opposite situation, coming from tropical considerations, in which sums are rarely 0. Definition 6.35. A triple is anti-negated if a◦ 6= 0 for all a ∈ T .

It follows that if a + b = 0 for a,b ∈ T0 then a = 0. This property has diﬀerent names in the literature: “antiring” in [19, 67], “zero-sum free” in [28], and “lacking zero sums” in [40]. t Lemma 6.36. In a cancelative metatangible triple (A, T , (−)), if some sum of tangible elements i=1 ai is 0, with t ≥ 2 minimal, then one of the following holds: P (i) t =2 with a2 = −a1 (the classical negative). (ii) t ≥ 3, (−) is of the ﬁrst kind, with all ai equal, and t = 0. ALGEBRAS WITH A NEGATION MAP 35

Proof. Otherwise, if ai 6= (−)aj , then we could replace ai + aj by their sum in T and reduce t. Thus, we may assume that all of the ai are quasi-negatives of each other. If t ≥ 3, then all of the ai are equal with (−)ai = ai. (If say a1 6= (−)a1 then a2 = (−)a1 and (−)a1 = a3 = (−)a2 = a1, contradiction.) t Hence att = i=1 at1 = at0. Canceling at yields t = 0. We are left with the case t = 2, in which case a1 + a2 = 0, so a2 = −a1. P Lemma 6.37. Any ub metatangible unital triple is anti-negated, and satisﬁes e′ 6= 1.

′ ◦ Proof. The ﬁrst assertion is immediate. If e = 1, the ub property implies 1(−)1 = 1 ∈ T0 ∩ T0 = 0, a contradiction.

Dolzan and Oblak [19] develop the tie between anti-negated semirings with matrix theory, by showing that the only invertible matrices over multiplicatively cancelative anti-negated semirings are generalized permutation matrices. Let us formulate this key feature from tropical algebra, in terms of triples.

Proposition 6.38. Over an anti-negated metatangible triple with n 6= 0 for each n ∈ N, the only invertible matrices are the generalized permutation matrices.

Proof. In view of Lemma 6.36, the proof in [67] goes through.

6.7. Squares and sums of squares. One example is analogous to the classical theory of real closed ﬁelds, in which squares are always positive.

Lemma 6.39. Suppose (T , ·, (−)) is an Abelian group with a negation map, and N is a subgroup of T , containing the set of elements of T which are squares, with N maximal with respect to the property that (−)1 ∈/ N. (Such N exists by Zorn’s lemma.) Then for any a ∈T \ N we have t ≥ 0 for which t a2 ∈ (−)N.

Proof. N ∩ (−)N = ∅, since if a, (−)a ∈ N then (−)1 = ((−))aa−1 ∈ N. For a ∈T \ N, we could adjoin a to N (i.e. expand N with all elements aib for b ∈ N) contrary to maximality of N, unless aib = (−)aj b′ for some i, j ∈ Z and b,b′ ∈ N. Now

ai−j = (−)b′b−1 ∈ (−)N.

Take m> 0 minimal such that am ∈ (−)N, and write m =2tq for q odd. t Let c = a2 ∈ T ; then cq ∈ (−)N. But c2 ∈ N since N includes all squares of T , so reducing the power t q modulo 2 must yield 1, i.e., a2 = c ∈ (−)N.

Deﬁnition 6.40. An element a of a monoid T with negation map (−) is positive if (−)a is not a square in T . A submonoid T ′ of T is positive if each element of T ′ is positive. (In particular (−)1 cannot be a square.)

Note that even when (T , ·) is a group, a subgroup of T could be positive, for example the subgroup Q+ of (Q×, ·), where (−) is classical negation.

Theorem 6.41. Suppose (N, ·) is a positive subgroup of an Abelian group T with a negation map, containing all squares in T , which is maximal with respect to the property that N ∩ (−)N = ∅, and suppose that (−)1 is not a square in T . Then T = N ∪ (−)N. Furthermore, suppose that (A, T , (−), ) is a (−)-bipotent triple, and let N be the symmetrization of N by Deﬁnition 3.6. Then there is an additive homomorphism ϕ : N → A given by (a,a′) 7→ a(−)a′. b 2t 2t 2t−1 2 Proof. Suppose a ∈Tb \ N. Then by Lemma 6.39, (−)a is in N, but a = (a ) for t ≥ 1 so t the element (−)a2 ∈ N would not be positive, a contradiction unless t = 0, i.e., (−)a ∈ N. Hence T = N ∪ (−)N. We write a>b in T if a + b = a. The map ϕ is a homomorphism, since for a>a′ ∈ N we have ϕ(a,a′)= a and ϕ(a,a)= a(−)a, so the greater term dominates in each veriﬁcation. 36 LOUIS ROWEN

6.7.1. Sign maps. The following system ties in with triples over a multiplicative group.

Example 6.42. The sign system is (Asgn, Tsgn, (−), ◦), where Asgn = {−1, 0, 1, ∞} endowed with the obvious multiplication, and with idempotent addition also satisfying a+0 = 0+a = a, a+∞ = ∞+a = ∞, ◦ −1 + 1 = ∞. Thus Asgn = {0, ∞}. This can be identified with the hyperfield of signs in [7], described in Example12.1, where ∞ is identified with {0, 1, −1}. Definition 6.43. A sign map on a monoid T is a multiplicative homomorphism sgn : T → ({−1, 0, 1, ∞}, ·) When T has a negation map (−), we require furthermore that sgn((−)a)= −sgn(a). (This is very close to the minus sign used in [25, §3.1].) T + := sgn−1(1) is a submonoid of T . Example 6.44. (i) R has the classical sign map. (ii) The semiring A of Example 3.11 has a sign map, given by sgn(0, 0)=0, sgn(a, 0)=1, sgn(0,a)= −1, sgn(a,a)= ∞, ∀a ∈ G. The monoid A+ = G× 0 is positive. Indeed, suppose that 2 2 2 (−)1 = (0, 1) = (a0,a1) = (a0 + a1,a0a1 + a1a0).

If a0 = 0 or a1 = 0 the second component is 0, and if a0 = a1 then both components are equal, both of which are contradictions. Conversely, we have: Lemma 6.45. Suppose N is a subgroup of a group T , containing all squares, which is maximal with respect to the property that N ∩ (−)N = ∅, and suppose that (−)1 is not a square in T . Then there is a sign map sgn on T such that N = T +. Proof. Using Theorem 6.41, we define sgn(a) = 1 iff a ∈ N. Proposition 6.46. Suppose (A, T , (−), ) is a metatangible system, with T a group, and suppose that (−)1 is not a square in T . Then there is a sign map sgn on A given by Lemma 6.45 on T , and sgn(0)=0, and sgn(a◦)= ∞ for each a 6= 0. Proof. Take the sign map of Lemma 6.45, formally defining sgn(a◦)= ∞. 6.7.2. Other basic properties of metatangible triples. Reversibility holds for tangible elements.

Lemma 6.47. (Compare with Lemma 2.15(ii).) In a metatangible system, if a ◦ b + c for a,b,c ∈ T , then b ◦ a(−)c. Proof. Write a + d◦ = b + c. Then (a + d)◦ = b + c(−)a. If c 6= a then c(−)a is tangible, and thus equals (−)b (since otherwise b + c(−)a ∈T ∩T ◦ = ∅). ◦ Thus we may assume c = a. If b+c is tangible then a = b+c, implying b ◦ b+c = (b+c)(−)c = a(−)c. So we are done unless c = (−)b, so b = (−)a ◦ (−)b(−)c = a(−)c. In general, we have a weird situation. Example 6.48. In the truncated layered system of Example 4.2(v), for n =5, take a = (1, 1), b = (1, 2), and c = (4, 2). Then a + (5, 1)◦ = (5, 2) = b + c, but we cannot write b + d◦ = a(−)c = a + c = c because the parities in L do not match. This is the only sort of counterexample: Theorem 6.49. In a cancelative metatangible unital system, if a b + c for a,b ∈ T , then b a(−)c, except in the following situation (as in Example 6.48): there are 1 < m′ ≤ m such that c = mb, with m′ = m but m′ − 2 6= m − 2, and a + c = c. The proof, omitted here but given in [61, Theorem 7.44], is by uniform presentation to reduce the assertion to tangible elements; this reduction fails to pass through the above exception. ALGEBRAS WITH A NEGATION MAP 37

6.8. Classifying metatangible systems. In a cancelative unital semiring A, 2 = 1 precisely when A is idempotent. In other words, characteristic 1 of the first kind gives the max-plus algebra, but quasi-negatives are far from unique since T = T ◦. The following result shows how metatangible systems naturally lead us to the other main tropical structures. The theorem is quite comprehensive, except for glossing over the exceptional systems. Theorem 6.50. Any cancelative metatangible unital system (A, T , (−), ) must satisfy one of the following: ′ (1) (−) is of the first kind. A = ∪m∈N mT , and e = 3. (a) 3 6= 1. Then T is (−)-bipotent, and (A, T , (−), ) is isomorphic to a layered system (either layered by N or quasi-periodic in characteristic 0 (Example 4.12), and layered by Z/k in characteristic k> 0). In particular, when 3 = 2, we have m = 2 for all m ≥ 2, and A has height 2. (b) 3 = 1. Then (A, T , (−), ) has characteristic 2 and height 2. The semiring† A◦ is bipotent, and the conditions of Proposition 6.16 hold. (2) (−) is of the second kind. There are two possibilities: (a) T is (−)-bipotent, and T (and thus A) is idempotent, and A has height 2. (b) T is not (−)-bipotent. Then the system is T -classical, and the semiring† A◦ is bipotent. Fur- thermore e′ = 1. Hence A has height 2. Either N ⊆ T , or (A, T , (−), ) has characteristic k for some k ≥ 1. In the latter case, (A, T , (−), ) is layered by Z/k. Proof. We start with Theorem 6.7, which says that T is (−)-bipotent or e′ = 1. This enables us to subdivide parts (1) and (2) (although in the reverse order). Also, by Theorem 6.25, every element of A is uniform. (1) If (−) is of the first kind, this means that (−)a = a and all elements have the form ma for a ∈ T . If T is (−)-bipotent, and a + b = b, we get ma + m′b = m′a, ma + m′a = (m + m′)b, (ma)(m′b) = (mm′)ab, which are precisely the rules for layered addition and multiplication, so A is layered by N. Eventually the numbers m may cycle modulo k, in which case one can identify subsequent layers modulo k. When 3 = 2, we clearly have m = 1◦ = 2 for all m ≥ 2. The last assertion is by Theorem 6.25. When 3 = 1, every element has height ≤ 2 by Theorem 6.25, and we conclude with Proposition 6.16, noting that a◦ + a◦ = e′a + a = a + a = a◦. (2) First assume that T is (−)-bipotent, so T (and thus A) is idempotent. In particular, A has height 2, cf. Corollary 6.17. Now assume that T is not (−)-bipotent, so e′ = 1 by Theorem 6.7. The system is T -classical by Lemma 6.34. Again Proposition 6.21 shows that A◦ is bipotent, noting that a◦ + a◦ = e′a(−)a = a(−)a = a◦. By Proposition 6.12, either N ⊆ T or (A, T , (−), ) has characteristic k for some k ≥ 1. In the latter case, 1,..., k − 1, are distinct, since if m = m′ for 1 ≤ m

• Case (2a) includes another example [61, Definition 3.6]. • “Layered semirings” (which come up in Cases (1a) and (2a)) were reviewed in Example 4.1, also see [61, Example 7.58]. • Case (2a) leads to the following approach. If T is positive, then (A, T , (−), ) comes from the ′ symmetrized system (A )sym of Theorem 6.41. (This is layered over A.) • For Case (2b), in characteristic 6=2, the classical system (A, A, −, =) is of second kind, satisfying e′ = 1. 6.9. Specific applications of metatangible systems. We turn now to specific applications, i.e., hypergroups and fuzzy rings, to be elaborated in the appen- dices.

6.9.1. * Metatangible systems versus metatangible hypergroups. We have embedded the theory of hypergroups into that of systems. We can go the other direction for metatangible systems.

Proposition 6.52. Any metatangible (resp. (−)-bipotent) system (T0, T , (−), ⊆) gives rise to a (−)- closed (resp. (−)-bipotent) hypergroup structure on the set T , as follows: Deﬁne [a]= {a′ ∈ T : a′ + a = a}. Then deﬁne addition on T by

a + b : a 6= (−)b, a ⊞ b = ([a]: a = (−)b. Proof. We verify the conditions of Deﬁnitions 3.13 and 3.14. Recall from Remark 6.2 that a ∈ [a′] iﬀ a′(−)a = (−)a, so [a] = [(−)a]. Hence a⊞(−a) = [a] = (−a)⊞a, implying addition is commutative. Next write b

b if b>a; ⊞ [a] b = [a] if b = a; [b] ∪ (b,a] = [a] if b < a. ⊞ ⊞ ⊞ ⊞ We need to check associativity. (a1 a2) a3 = a1 (a2 a3) = max{a1,a2,a3} unless we get equality at some intermediate stage, i.e., one of the following holds:

(i) a1 = (−)a2, (ii) a2 = (−)a3, (iii) a1 + a2 = (−)a3, (iv) (−)a1 = a2 + a3, which we check in course.

(i) If a1 + a3 = a3 then also a2 + a3 = a3 and (a1 ⊞ a2) ⊞ a3 = [a1]+ a3 = a3 = a1 ⊞ (a2 ⊞ a3). If a1 + a3 = a1 then also a2 + a3 = a2 and (a1 ⊞ a2) ⊞ a3 = [a1]+ a3 = [a1] = a1 ⊞ a2 = a1 ⊞ (a2 ⊞ a3). If a1 = (−)a3 then (a1 ⊞ a2) ⊞ a3 = [a1] ⊞ a3 = [a3]= a1 ⊞ (a2 ⊞ a3). (ii) Symmetric argument to (i). (iii) Suppose a1 6= (−)a2 and a1 + a2 = (−)a3. Then, by (−)-bipotence, a1 = (−)a3 or a2 = (−)a3, so either way (a1 ⊞a2)⊞a3 = [a3]; a1 ⊞(a2 ⊞a3) is respectively (−)a3 +(a2 +a3) = (−)a3 +a3 = [a3] or a1 + [a3] = [a3]. (iv) Symmetric argument to (iii). In each veriﬁcation we applied brackets to each side whenever there is equality. Deﬁne −a = (−)a. Then the quasi-zeroes are exactly the sets [a], which are the hyperzeros, and a1 +a2 is a hyperzero precisely when a2 = −a1.

A more encompassing result, for hypersystems that are not necessarily metatangible, is given in [4, Theorem 3.27]. ALGEBRAS WITH A NEGATION MAP 39

6.9.2. **Metatangible triples and the fuzzy property. We can also recover the key property of fuzzy ring, [20, Definitions 2.1,2.8]. Definition 6.53. The fuzzy property for a triple (A, T , (−)) is: ′ ◦ ′ ◦ ′ ′ ◦ ′ b1(−)b1 ∈ A and b2(−)b2 ∈ A imply b1b2(−)b1b2 ∈ A , ∀bi,bi ∈ A. Theorem 6.54. Metatangible triples satisfy the fuzzy property. ′ ◦ ′ ◦ Proof. Suppose that b1(−)b1 ∈ A and b2(−)b2 ∈ A . We appeal to the uniform presentation (Defini- tion 6.23), Theorem 6.25, and Corollary 6.27. ′ ′ ◦ Case I b1 + b1 = b1. By Remark 6.2, b1 = b1(−)b1 ∈ A , so ′ ′ ′ ′ ′ ′ ′ ′ ◦ b1b2(−)b1b2 = (b1 + b1)b2(−)b1b2 = b1b2 + (b1(−)b1)b2 ∈ A . ′ ′ Likewise if b2 + b2 = b2. ′ ′ ′ ′ Case II b1 ∈ T and b1 ∈ T . By unique quasi-negatives b1 = b1, in which case b1b2(−)b1b2 = ′ ◦ b1(b2(−)b2) ∈ A and we are done. ′ ′ ◦ Case III b1 ∈ T and b1 has Type 2. Writing b1 = a we have by Case (i)(b) of Corollary 6.27 that either a + b = a so we conclude by Case I, or a◦ = (±)b with e′ = e, in which case ′ ′ ′ ′ ′ ′ ′ ′ ◦ b1b2(−)b1b2 = b1b2(−)b1b2(−)b1b2 = b1(b2(−)b2)(−)b1b2 ∈ A and we are done. ′ Thus we may assume that if b1 has Type 1 then b1 has Type 3, and the analogous assertions hold for ′ ′ b1,b2,b2. ′ ′ We are done if b1 and b1 are both of Type 2, so we may assume b1 has Type 6= 2. But then b1 or b1 has Type 3, so (−) is of the first kind. ′ ′ ′ ′ Case IV (−) is of the first kind, bi = miai and bi = miai for i = 1, 2 and a,a ∈ T . If m1 − m1 is ′ ′ ′ ′ ′ ′ ′ ◦ even, then m1m2 − m1m2 ≡ m1(m2 − m2) (mod 2), implying b1b2(−)b1b2 = (m1m2 − m1m2)a1a2 ∈ A ′ ′ ◦ ′ since (m2 − m2)a2 = b2(−)b2 ∈ A , and we are done. A similar proof applies if m2 − m2 is even. ′ ′ ′ ′ We write bi = mibiT and bi = mibiT for biT ,biT ∈ T . In view of Case I, we may assume that ′ biT = biT . ′ ′ Hence, by Case IV we may assume that m1 −m1 and m2 −m2 are odd, and may assume that m1 ≤ m2. ◦ ′ But then mai ∈ A for all m ≥ |m1 − m1|, by Remark 4.11, and in particular ′ ′ ′ ′ ◦ b1b2(−)b1b2 = m1m2a1a2(−)m1m2a1a2 ∈ A .

We elaborate in Appendix B.

7. Categorical properties of systems Having established the ubiquity of systems, let us view systems in categorical terms in order to relate different systems. We make systems of a given algebraic structure into a category, whose objects are systems. The question is how to define morphisms. The customary way to do so in universal algebra would be via a homomorphism, as defined in §1.2.2. However, in the context of systems, it is often preferable to bring into the picture. 7.1. -Morphisms of systems. Definition 7.1. A -morphism of systems ϕ : (A, T , (−), ) → (A′, T ′, (−)′, ′) ′ ′ is a map ϕ : A → A satisfying the following properties for ai ∈ T and b b in A: ′ ′ (i) ϕ(a1) ∈ T , and ϕ((−)a1) = (−) ϕ(a1); ′ (ii) ϕ(a1 + a2) ϕ(a1)+ ϕ(a2); (iii) ϕ(a1b)= ϕ(a1)ϕ(b). (iv) ϕ(b) ′ ϕ(b′). (v) ϕ(ANull) ⊆ ϕ(A)Null. (vi) When 0A ∈ A, we also require ϕ(0A)= 0A′ . 40 LOUIS ROWEN

MAJOR NOTE 7.2. The key difference from homomorphism is (ii), which plays a major role in the structure (as in [48, 49]) and often holds when equality may fail, for example with tropicalization. It is in line with hypergroup morphisms, which stipulate that ϕ(a1 ⊞ a2) ⊆ ϕ(a1) ⊞ ϕ(a2). These conditions arise naturally in the cases of hypergroups and also for Lie semialgebras. For example, although for any R-module M, the left multiplication map ℓr : M → M is a homomorphism iff left multiplication by r distributes over M, the map ℓr is a -morphism iff r(a1 + a2) ra1 + ra2 for each ai ∈ M; this is described in [68, §4.1], and treated in [61, §6.9].

Lemma 7.3. Any homomorphism ϕ of systems is a ◦-morphism.

◦ ◦ Proof. If a2 = a1 + c , then ϕ(a2)= ϕ(a1)+ ϕ(c) .

7.1.1. *Embedding hypergroups into the category of systems. In Definition 3.19 we presented the system of a hypergroup. This can be made more explicit using the formalism of categories. Theorem 7.4. There is a faithful functor Ψ from the category of canonical hypergroups into the category of T -reversible systems, whose morphisms are the -morphisms, sending a hypergroup T to its hypersystem (T0, T , (−), ⊆). Furthermore, the hypergroup T is metatangible (resp. closed), iff its hypersystem (T0, T , (−), ⊆) is metatangible (resp. (−)-bipotent). Proof. The first assertion is by Proposition 4.18 and Theorem 3.21, and the second by Lemma 3.24.

In particular, closed hypergroups can be studied in terms of §6.

7.2. Valuations on systems. Usually valuations have been studied in terms of multiplication in monoids, cf. [31, 35], but addition plays the main role here. Suppose G is an ordered semigroup. We view it as a max-plus algebra where (−) is the identity map 1G, and build its supertropical semiring system of Deﬁnition 3.5, written here as

(AG , TAG , 1AG , ◦), where TAG , a disjoint copy of G, is the target of the valuation, and AG := TAG ∪ G0. Deﬁnition 7.5. A valuation of a cbimagma system (A, T , (−), ) is a -morphism

v : (A, T , (−), ) → (AG , TAG , 1AG , ◦) satisfying

v(a1a2)= v(a1)v(a2) ∀ai ∈ T . (7.1)

(Recall that multiplication in TAG is really addition.)

This resembles the “modulus” in [5], except that v normally sends elements to TAG instead of G. Note that v((−)a)= v((−)1)v(a)=1AG v(a)= v(a). MAJOR NOTE 7.6. We use the supertropical algebra rather than the max-plus algebra, in order to take into account the uncertainty of v(a1 + a2) in regard to v(a1)+ v(a2). Remark 7.7.

(i) By definition, v(a1 + a2) ◦ v(a1)+ v(a2) for ai ∈ T . This is the opposite direction from the custom for valuations, but we could just reverse our definition of inequality. (ii) Zelinsky [69] also defines valuations over (classical) nonassociative algebras. (iii) Condition (7.1) is not needed in our next result, modeled after a well-known proof!

Proposition 7.8. If v : (A, T , (−), ) → (AG , TAG , 1AG , ◦) is a -morphism and v(a1) ◦ v(a2) for ai ∈ T , then v(a1 + a2)= v(a1).

Proof. We are given v(a1 + a2) ◦ v(a1)+ v(a2)= v(a1). But a1 a1 + a2(−)a2 and v((−)a2)= v(a2), so v(a1) ◦ v(a1 + a2(−)a2) ◦ v(a1 + a2)+ v((−)a2) = v(a1 + a2)+ v(a2), a contradiction unless v(a1 + a2)= v(a1). ALGEBRAS WITH A NEGATION MAP 41

7.3. The transfer principle.

Deﬁnition 7.9. A ◦-identity of a system is a universal elementary sentence involving the surpassing relation .

Although not strictly identities in the sense of universal algebra since they do not pass to homomorphic images, ◦-identities take the place of identities in the systemic theory. The transfer principle, whose roots are in [57], is a method of obtaining ◦-identities, introduced formally in [24] and made explicit in [2]. This treatment essentially is a reformulation of [2, Corollary 4.18], expressed through morphisms in order to increase its applicability. It is based on a way of passing from [ semirings† to rings, by means of the symmetrization N{x} of the free semiring† N{x}, with the switch negation map. We start with an easy but enlightening special case. Take N = N(A) as in Deﬁnition 1.8. Given P = i(ai(−)bi)xi ∈ N{x, (−)x}, we deﬁne the corresponding classical polynomial P¯ = i(ai − bi)xi ∈ N{x, (−)x}. P

Proposition 7.10. Suppose P = Pi(ai(−)bi)xi ∈ N{x, (−)x}. If the free T -semiring N{x;Ω} (under the usual operations of N) satisﬁes the identity P¯ =0, then P ◦ 0 in N{x, (−)x}. P

Proof. P¯ = i(ai − bi)xi. For this to be 0, we must have each ai = bi, so P = i(ai(−)ai)xi ∈ N{x, (−)x}◦. P P The same ideas give the full transfer principle (strong form) of [2]. Let Z be as in Example 3.9. Extending Remark 1.9 we have:

Lemma 7.11. There is a -morphism ϕ : Z → Z given by (i) n 7→ (n, 0); (ii) −n 7→ (0, n); (iii) 0 7→ (0, 0).

Proof. ϕ(m)+ ϕ(−n) = (m, n), which for m ≥ n is (n, n) + (m − n, 0) ϕ(m − n), and the other veriﬁcations are analogous.

Lemma 7.12. The -morphism ϕ : Z → Z of Lemma 7.11 extends to a -morphism

ϕ : Z{x} → Z{x, (−)x} by xi 7→ (xi, 0) and (−)xi 7→ (0, xi).

Proof. The same proof as in Proposition 7.10, since any presentation of 0 must be sent to N{x, (−)x}◦.

Theorem 7.13. Suppose P = i aixi,Q = i bixi ∈ N{x, (−)x}, where ai ≥ bi for each i. If the free (classical) semiring N{x} satisﬁes the identity P¯ = Q¯, then P ◦ Q in N{x, (−)x}. P P

Proof. P¯ − Q¯ = i(ai − bi)xi. Now applying Lemma 7.12 yields the assertion. P Remark 7.14. Because of the ambiguity involved with n, it is misleading to deal with identities over N whose coeﬃcients are not (±)1.

The same ideas apply to arbitrary varieties, even nonassociative, and can provide powerful intuition.

8. Linear algebra over a triple

Here we tackle the various notions of linear algebra over a system. Only the foundation is presented here; deeper theorems and their subtleties involved are given in [5]. 42 LOUIS ROWEN

8.1. ◦-identities for matrices. Identities of n × n matrices can be translated (matching the matrix entries) into n2 identities in commuting indeterminates. Using the transfer principle, we see that many identities of matrices over rings translate to ◦-identities of Nmax[Λ], and thus of semirings. (These results hold more generally over T -semirings†.)

Lemma 8.1. Suppose A is a square matrix whose entries are all 0 and ±1. If the determinant of A (taken in Z) is 0, then A is ◦-singular in the sense of Deﬁnition 5.2.

Proof. Immediate from Proposition 7.10.

Although other results also are consequences of the transfer principle, we indicate their easy direct proofs.

Deﬁnition 8.2. Write Mi,j for the i, j minor, which is the (−)-determinant of (ak,ℓ)k6=i,ℓ6=j of a given ′ matrix A. The (−)-adjoint matrix adj(A) is (Mj,i). When A is ambiguous we will write ai,j for Mi,j .

We inductively write (−)k for (−)((−)k−1), where (−)1 is (−).

n i+j Remark 8.3. |A| = j=1(−) Mi,j ai,j , for any given i.

Proposition 8.4. adj(PB) adj(A) ◦ adj(AB) for n × n matrices A and B.

′ Proof. Write AB = (ci,j ), we see that adj(AB) = (cj,i) whereas the (i, j)-entry of adj(B) adj(A) is n ′ ′ ′ ′ ′ ′ k=1 bk,iaj,k. Since aj,kbk,i appears in cj,i, we need only check that the other terms in cj,i occur in matching pairs with opposite signs. This kind of computation goes back to [65]. These are sums of P products of the form

dk1,π(k1)dk2,π(k2) ··· dkn−1,π(kn−1),

where kt 6= j, π(kt) 6= i for all 1 ≤ t ≤ n − 1, and

dkt,π(kt) = akt,ℓbℓ,π(kt), for suitable ℓ.

If the ℓ do not repeat, we have a term from adj(B) adj(A). But if some ℓ repeats, i.e., if we have

dkt,π(kt) = akt,ℓbℓ,π(kt), dku,π(ku) = aku,ℓbℓ,π(ku),

′ then in computing cj,i we also have a contribution from another permutation σ where σ(kt)= π(ku) and σ(ku)= π(kt) (and σ = π on all other indices), whereby we get

akt,ℓbℓ,σ(kt)aku,ℓbℓ,σ(ku) = akt,ℓbℓ,π(ku)aku,ℓbℓ,π(kt) = akt,ℓbℓ,π(kt)aku,ℓbℓ,π(ku),

as desired.

Lemma 8.5. |A|I ◦ A adj(A) over any T -semiring triple A.

Proof. The diagonal terms are equal, by deﬁnition, and the extra terms oﬀ the diagonal are known to match, by rewording [57, Lemma 2].

Theorem 8.6. |A||B|◦ |AB|, for any matrices A, B ∈ Mn(A).

Proof. By the semiring argument in [57], matching terms in the products, since any term in det(AB) not in det(A) det(B) occurs twice, with opposing signs. This follows from [57, p.352, end of proof of (a)].

Note however that the determinant is not a morphism since it reverses . In this sense the determinant could be considered an “-antimorphism.” We do get equality when |AB| ∈ T . ALGEBRAS WITH A NEGATION MAP 43

8.2. Ranks of matrices. Our next task is to compare different notions of rank of matrices, in terms of its row vectors and its column vectors. A vector v ∈ A(n) is called tangible if each of its entries is in T0. A matrix is tangible if each of its rows is tangible. We only consider tangible matrices A, mostly for metatangible systems. This is a small step back from [43], but the tangible case is a compelling one, since one can recover the full supertropical result from it. Modules over triples are studied in subsequent papers, such as [48, 49, 23, 5]. Here, to stay on track, (n) we just consider the “free module” M := (A , TM , (−)) where TM is the set of vectors with a single nonzero component, which is in T , and (−) is defined componentwise. In this case, M is also a T -module, with multiplication defined componentwise.

(n) Definition 8.7. Suppose that (M := A , TM, (−), ) is a systemic module over A. A set S ⊆M is ◦-dependent if there are v1,...,vm ∈ S and αj ∈ T such that m ◦ αj vj ∈M . j=1 X Otherwise S is ◦-independent. An element v ∈ M is ◦-dependent on a ◦-independent set S ⊆ M, written v ∈dep S, if S ∪{v} is ◦-dependent. Definition 8.8. The (surpassing) row rank of a matrix A is the maximal number of ◦-independent rows of A. The column rank of the matrix A is the maximal number of ◦-independent columns of A. The submatrix rank of the matrix A is the maximal k such that A has a nonsingular k×k submatrix. Let us consider the following assertions: (i) Condition A1: The submatrix rank is less than or equal to the row rank and the column rank. (ii) Condition A2: The three definitions of rank are equal for any tangible matrix, when T is a multiplicative group. MAJOR NOTE 8.9. We are about to see that Condition A1 holds for cancelative (−)-bipotent triples. Condition A2 is considerably more delicate, usually also requiring height 2 even in the n × n case; a thorough treatment is given in [5], linked to [3]. An easy induction argument enables one to reduce Condition A1 to proving that a square matrix A is ◦-singular if its rows are ◦-dependent, which is our next result.

Lemma 8.10. Suppose (A, T , (−)) is a semiring triple, with ai ∈ T and bi ∈ A. If a1 + a2 = a1 and b1 + b2 = b1, then a1b1 + a2b2 = a1b1.

Proof. a1b1 + a2b2 = (a1 + a2)b1 + a2b2 = a1b1 + a2(b1 + b2)= a1b1 + a2b1 = (a1 + a2)b1 = a1b1. Theorem 8.11. Suppose that (A, T , (−)) is a cancelative (−)-bipotent triple over a monoid T . If the rows v1,...,vn of a tangible n × n matrix A over a cancelative (−)-bipotent triple are ◦-dependent, then |A| ∈ A◦. Proof. Localizing, we may assume that (T , ·) is a group. We start by mimicking the proof in [44]. Suppose m ◦ j=1 αj vj ∈M for αj ∈ T . Replacing vj by αj vj , we may assume that the sum of the rows is a vector in (A(n))◦. Write A = (a ). P i,j Recall the uniform presentation of Deﬁnition 6.23. We say that an element c dominates c′ if either ′ ′ cT = (−)cT or cT + cT = cT . (Then if c = bi for bi ∈ A, some bi must dominate c.) From Equation (5.1), there are k1,...,kn such that |A| is dominated by ak ,1 ··· ak ,n ∈ T . Inter- P 1 n changing rows we may assume that ki = i for each i, i.e., |A| is dominated by a1,1 ··· an,n. We say that ai,j is (column) critical if ai,j dominates all other entries in the j column of A. Any critical diagonal entry of A must be matched by another critical entry in the same column; i.e., ai,j = (−)ai′,j.

Now starting with some critical nondiagonal entry, say ai2,a1 with i1 =1, we take ai3,i2 with i3 6= i2, and continue in this way until we return to i1. This gives us

ai1,ik aik ,ik−1 ··· ai2,i1 44 LOUIS ROWEN

where each entry is critical. Deﬁning the permutation π by π(i1) = ik,...,π(i2) = i1 and the identity

elsewhere, it is clear that ai1,i1 ··· aik−1,ik−1 aik ,ik is dominated by ai1,ik aik ,ik−1 ··· ai2,i1 , and thus |A| is also attained by π. This means that for each of these i, ai,j = (−)aj,j , and dividing the j-th column through by aj,j enables us to assume that every dominant entry is (±)1. But now the non-dominant entries do not play a role either in the hypothesis (that the rows are ◦-dependent) or the conclusion (that |A| ∈ A◦), so we replace them by 0, and assume that every entry of A is in {0, −1, 1}. In order to cancel 1 on the diagonal, we may assume that each row has some non-diagonal entry −1, so picking

these entries when building π of the previous paragraph, we have all aiℓ,iℓ+1 = (−)1. Renumbering the rows and columns (since interchanging both the i and j rows and columns does not aﬀect the hypothesis 1 (−)1 0 ... 0 0 1 (−)1 ... 0

or conclusion) we may assume that iℓ = ℓ for all ℓ ≤ k. Since . . . . = 1(−)1 = e, . . .. .

(−)1 0 0 ... 1

we see that e is a summand of |A|.

When (−) is of the second kind, idempotence implies that |A| = e, and we are done. Thus we may assume that (−) is of the ﬁrst kind, and every entry of A is in {0, 1}. First assume that in each column the number of entries that are 1 is even. Then reading A as a classical matrix, the sum of each column is 0 (mod 2), so its classical determinant is 0 (mod 2), i.e., |A| ∈ A◦. Thus one may assume that in some column the number of entries that are 1 is odd, and take ◦ ◦ ◦ the column with the smallest such number of entries k0. Then k0 ∈ A ; since 2 ∈ A , we have k ∈ A for all k ≥ k0. But the argument of the fourth paragraph gives us at least k0 summands (each equal to 1) of |A|, so |A| must be some k ∈ A◦. Corollary 8.12. Condition A1 holds over a cancelative (−)-bipotent triple. Proof. If m is the row rank of a matrix A, then any m +1 rows are ◦-dependent, implying that every (m + 1) × (m + 1) minor is ◦-singular, so the submatrix rank is at most m. The same argument holds for the column rank. [3, Theorem 4.18] provides a stronger conclusion, called “Cramer’s rule,” under extra hypotheses. Cramer’s rule is obtained rather generally in [5, Theorem 4.8] when assuming a Noetherian-type property.

9. More applications

We focus on three major examples — tropicalization, exterior semialgebras and Lie semialgebras.

9.1. Tropicalization of Puiseux series. Tropicalization, perhaps the main tool in tropical mathematics, has been studied in various contexts. Originally “standard” tropicalization was a map to the max-plus algebra, obtained by applying logarithms to real or complex varieties, as exposed in [33]. Most of the recent research on tropicalization has focused on the Puiseux series valuation. For any cbimagma K, one can define the set K{{t}} of Puiseux series on the variable t, which is the set of formal ∞ j/N series of the form f = j=ℓ cj t where N ∈ N, ℓ ∈ Z, and cj ∈ K. (One could use any subgroup of (R, +) for the exponents in the series, but the definition becomes more complicated without enhancing P the theory, since (Q, +) is model complete in the elementary theory of ordered groups.) Then we have the Puiseux valuation val : K{{t}}\{0} → Qmax ⊂ Rmax defined by val(f)= − min{j/N}, (9.1) cj 6=0 and formally val(0) = 0 (= −∞). (We put in the negative to pass from minimum to maximum.) We also call val tropicalization, now viewed as a -morphism, cf. Definition 7.5. Customarily one takes K to be the field of complex numbers, so that K{{t}} is an algebraically closed field, even though we find it convenient to consider tropicalization over other semirings, especially N0. We would want val to be a -morphism. But this does not quite work since Qmax does not have negatives, so we consider several related versions of tropicalization which are more amenable to algebraic methods. ALGEBRAS WITH A NEGATION MAP 45

Example 9.1. The Puiseux series valuation comes in various forms:

(i) The usual Puiseux series valuation val to the max-plus algebra Qmax from the Puiseux series algebra K{{t}} on the variable t, again as exposed in [33], and to be reviewed presently. (ii) The Puiseux series valuation from the Puiseux series algebra K{{t}} to the supertropical semiring, [42]. (iii) The Puiseux series valuation from K{{t}} to the “exploded algebra,” [55] or, more generally, to Example 4.2(viii). (iv) The Puiseux series valuation from K{{t}} to the semiring layered by N, [36]. Each version has its specific motivation. Supertropical algebra is compatible with the value group of the Puiseux series valuation. If one wants to take the residue field into account one would pass to the exploded algebra. Even so, this only utilizes the lowest term of the Puiseux series. When this is lost through cancelation, one would need to dig deeper into the Puiseux series, taking an infinite direct sum ⊕i∈NGi of target systems. This would be the tropicalization of the associated valuation ring, but has not yet been utilized in the literature. These various approaches are unified in terms of -morphisms of systems of the relevant categories. Proposition 9.2. In each of the cases taken from Example 9.1 (in the same order), val(f) provides a - morphism v from the Puiseux series K{{t}} (viewed as a classical system) to T in one of the metatangible systems that we have described earlier: (i) v(f)= − val(f), taking values in the max-plus algebra, cf. Remark 3.1. (ii) v(f)= − val(f), taking values in the supertropical algebra. (iii) v(f)=(1, − val(f)), taking values in the layered algebra. ∞ k/N (iv) For a Puiseux series f = k=ℓ ckt with cℓ 6=0, take v(f) = (cℓ, −ℓ/N) in the ELT algebra. This can be viewed more generally, in analogy to viewing tropicalization as passing to the target of a valuation v : R → Q,P where R is a ring. Suppose that the valuation v has a uniformizer π such that v(π)=1. Thus, for any element r, taking v(r) = m/n we have v(π−m/nr)=0. Now one can also take into account the residue ring R/P where P is the valuation ideal and, letting L = R/P, consider the map R → L × G, the ELT-algebra, given by a 7→ (π−m/nr, w(r)). Proof. In each case, we verify that v(−f) = (−)v(f), and v preserves addition (with respect to ). This process indicates a way of tropicalizing standard algebraic definitions in this setting, where one expects that some case of Proposition 9.2 to be used, according to the context. 9.2. Exterior (Grassmann) semialgebras. Paralleling the classical case, for free modules, the tensor semialgebra yields a construction of the Grassmann semialgebra whose base is the union of even elements and odd elements. The definition given in [26] (which goes on to treat the Plücker equations) is a semialgebra G generated by a free module V G G G 2 with a base {ei : i ∈ I}, together with a product × → satisfying ei = 0 for each i ∈ I. These could 2 be constructed by means of the tensor semialgebra, modulo the relations xi = 0. As noted in [26], such a definition relies on the presentation in terms of the base, since in general v2 6= 0 for v ∈ V. This would mean that a sub-semialgebra of a Grassmann algebra need not be Grassmann, such as the semialgebra generated by e1 and e1 + e2. Definition 9.3. A (faithful) Grassmann, or exterior, semialgebra, over a C-module V with a negation map, is a semialgebra G generated by V , together with a negation map extending (−) and a product G × G → G satisfying (i) v1v2 = (−)v2v1 for vi ∈ V, (9.2) (ii) (−)(v1 ··· vt) = ((−)v1)v2 ··· vt. (9.3) sgn(π) Thus vπ(1) ··· vπ(t) = (−) v1 ··· vt. When V is the free module, this definition, which is independent of the base, covers the one in [26], in which (−) is the identity map. Definition 9.3 maps onto [26], where ei(−)ei is sent to 0. 46 LOUIS ROWEN

The appropriate triple is (G, T , (−)), where T = {v1 ··· vt : vi ∈ T , t ∈ N}, the submonoid generated by V . Lemma 9.4. v2 ∈{b ∈ G : b = (−)b}. 2 2 2 2 2 Proof. ( αiei) = αi ei + i

P G P P G 1 (This set {b ∈ : b = (−)b} is just when 2 ∈ T .) Lemma 9.5. If G = A(I), then it is enough to check that

eiej = (−)ej ei, ∀i, j ∈ I, extended via distributivity.

Proof. ( αiei)( βj ej )= αiβj eiej = (−) αiβjej ei = ( βj ej)( αiei), yielding (i). (ii) is also by linearity. P P P P P P

Lemma 9.6. v1v2 = (−)v2v1 is central in G, for all v1, v2 ∈ V.

Proof. v1v2v3 = (−)v1v3v2 = v3v1v2, implying that v1v2 is central.

Deﬁnition 9.7. Given a Grassmann semialgebra G over a module V with a negation map (−), we deﬁne + + T to be the set of all even products of elements of V , G0 to be the submodule of G generated by T , − − T to be the set of all odd products of elements of V , and G1 to be the submodule of G generated by T .

Lemma 9.8. G = G0 + G1. G0 is in the center of G, and G1 = G0V. When V is a free module with negation, then G = G0 ⊕ G1 is a super-semialgebra. Proof. The ﬁrst assertion is an immediate induction based on Lemma 9.6. For the free module with negation, we match components.

Example 9.9. When V is a free A-module with negation, with base {ei, (−)ei : i ∈ I}, the tensor semialgebra T (V ) becomes a Grassmann semialgebra G when we impose the extra relations that ejei = ((−)ei)ej = ei((−)ej ) for all i, j ∈ I. T is the set of simple tensors in which one does not have both ei and (−)ei. Every term of even degree in the ei is central, so G satisﬁes the ◦-surpassing identity [x1, [x2, x3]] 0. G Lemma 9.10. In Example 9.9, any nonzero element of is of the form (±)aiei1 ··· eik , summed over i1 < ···

Proof. Take an element (±)αei1 ··· eik . Rearrange the ei appearing in each summand, since any time ◦ an ei repeats, the product is in G .

MAJOR NOTE 9.11. Ironically, even when V = A(n) does not have a negation map, we can still deﬁne a negation map on the ideal T (V )≥2 of T (V ) comprised of tensors of length ≥ 2, given by (−)vivj = vj vi.

Letting TT (V )≥2 be the simple tensors of length ≥ 2, we have the triple (T (V )≥2, TT (V )≥2 , (−)), and then the theory of triples is applicable! Various Grassmann semialgebras are studied in detail in [23], which reformulates identities for classical Grassmann semialgebras.

10. Nonassociative semialgebras with a negation map

In this section we bring in nonassociative semialgebras, especially Lie semialgebras, since Lie algebras are so important in classical representation theory. Now we might want T to have Lie multiplication instead of being a monoid. Whereas the Jacobi identity on a Lie algebra L is equivalent to the adjoint representation ad : L → ad L being a Lie homomorphism, the correspondence in tropical algebra is more delicate. ALGEBRAS WITH A NEGATION MAP 47

10.1. Super-semialgebras. As in the classical case, one can “superize,” to make a theory super.

Deﬁnition 10.1. The Grassmann envelope of a super-semialgebra A = A0 ⊕A1 is the sub-semialgebra (A0 ⊗ G0) ⊕ (A1 ⊗ G1) of A⊗ G, with G as in Lemma 9.8. Suppose V is a variety. A super-V semialgebra is a super-semialgebra A whose Grassmann envelope is in V. ij For example, A is super-commutative if aiaj = (−) aj ai whenever ai ∈ Ai,aj ∈ Aj for i, j ∈{0, 1}.

The Grassmann envelope of a Grassmann super-semialgebra G itself is (G0 ⊗ G0) ⊕ (G1 ⊗ G1) which is commutative, so G is super-commutative. Conceptually, Definition 10.1 is just an elegant form of book-keeping, where in evaluating multilinear operations on a superalgebra we put in (−)k, where k is the number of odd occurrences of the entries. 10.2. Lie -semialgebras and Lie -super-semialgebras, and their triples. Definition 10.2. A bimagma A with negation map is (−)-anticommutative if it satisfies the following conditions for all b,b′ ∈ A: (i) b2 ∈ A◦; (ii) b′b = (−)(bb′)= b((−)b′) = ((−)b)b′. (In classical mathematics, (ii) is derived from (i) by multilinearization, but this argument requires a genuine negative, and so is inapplicable here.)

Deﬁnition 10.3. For a bimagma A with negation map, given b ∈ A, we deﬁne adb ∈ Hom(A, A) by ′ ′ adb(b )= bb , and adA = {adb : b ∈ A}.

adA is a semialgebra and T -submodule of Hom(A, A), with a natural negation map (−)adb = ad(−)b. Definition 10.4. A Lie -semialgebra with a negation map (over a semifield F ) is an F - module L with a negation map (−), endowed with (−)-anticommutative multiplication L×L → L, written (b,b′) 7→ [bb′], called a Lie bracket (in view of the standard notation [ab] for Lie multiplication), satisfying ′ ′ ad[bb′] [adb, adb] ∀b,b ∈ L, (10.1) where the right bracket is the Lie commutator. (Note that we do not require a negation map on F .) Lemma 10.5. [[bb′]v] [b[bv]](−)[b′[bv]] for all b,b′, v ∈ L. ′ ′ Proof. [[bb ]v]=ad[bb′](v) adb(adb′ (v))(−)adb′ (adb(v)) = [b[bv]](−)[b [bv]]. Lemma 10.5 can be viewed as the -surpassing version of Jacobi’s identity. Proposition 10.6. If L is a Lie semialgebra with a negation map, then there is a Lie -morphism ad : L → ad L, given by b 7→ adb. (In fact ad preserves addition.) Proof. By Lemma 10.5. Definition 10.4 is a bit stronger than the analog of Blachar’s definition [11].

Proposition 10.7. Any associative semialgebra R with negation map becomes a Lie ◦-semialgebra under the Lie product [bb′] = [b,b′]. Proof. Using Lemma 2.35, one veriﬁes for any v ∈ L that ′ ′ ′ ′ ′ ′ ad[bb′](v) = [[bb ]v] = [[b,b ], v] = [bb (−)b b, v] = [bb , v] + [v,b b] ′ ′ ′ ′ ′ (10.2) ◦ b[b , v] + [b, v]b + [v,b ]b + b [v,b] = (adbadb′ (−)adb′ adb)v = [adb, adb]v, for all b,b′ ∈ L.

(−) We call this Lie ◦-semialgebra R . Corollary 10.8. For any associative semialgebra (R, ∗) with involution and negation map, (R, ∗)− is a (−) Lie ◦-sub-semialgebra of R . 48 LOUIS ROWEN

Proof. It is closed under the Lie product.

Remark 10.9. We are now in a position to deﬁne the symmetrized analogs of the classical Lie algebras, † ˆ over a semiring A. Namely, we take sln(A) = {((ai,j ), (bi,j )) ∈ Mn(A): i ai,i = i bi,i}, the symmetrized analog of the classical Lie algebra An−1. To obtain the analogs of Bn, Cn, and Dn, one applies P ∗ P Corollary 10.8 to the transpose and symplecticc involutions, taking the subset {(A, A ): A ∈Mn(Aˆ)}:

• We get the symmetrized version of the classical Lie algebra Bn when (∗) is the transpose and n is odd. • We get the symmetrized version of the classical Lie algebra Cn when (∗) is the symplectic involution and n is even. • We get the symmetrized version of the classical Lie algebra Dn when (∗) is the transpose and n is even.

10.2.1. Lie -super-semialgebras. Let us superize the Lie theory by means of Deﬁnition 10.1.

Deﬁnition 10.10. A Lie -super-semialgebra with a negation map is a module L with a negation ′ ′ map (−), endowed with super-(−)-anticommutative multiplication L × L → L, written as (b,b ) 7→ [bb ]s, ′ ′ ′ ′ called a -super Lie bracket, satisfying [[bb ]s v]s [b[b v]s ]s (−)[b [bv]s ]s for all homogeneous b,b , v ∈ L.

′ ′ ′ ′ Thus, for all b,b , v ∈ L we have [[bb ]s v]s [b[b v]s]s(−)[b [bv]s]s. (The negations all appear in the same degree in the super-version, so cancel out.)

Proposition 10.11. Any associative semialgebra R with negation map becomes a Lie -super-semialgebra under the -super-Lie bracket

ij [bibj ]s = bibj (−) bjbi, bi ∈ Ri, bj ∈ Rj . (10.3)

Proof. Reread the Leibniz identities (Lemma 2.35) in terms of (10.3).

10.3. Poisson semialgebras and their module congruences. The Leibniz -identities of Lemma 2.35 motivate the next notion.

Deﬁnition 10.12. A Poisson -semialgebra is an associative semialgebra A with a negation map, together with a bilinear operation { , } : A×A→A, called a Poisson bracket, satisfying

{ab,c} a{b,c} + {a,c}b, {a,bc}{a,b}c + b{a,c}, ∀a,b,c ∈ A.

(This takes into account Deﬁnition 10.4, as well as Proposition 10.7.) Then { , } yields a Lie - structure as in Proposition 10.7.

Example 10.13. The following are commutative Poisson -semialgebras. In each case we get a triple, where T is taken to be the set of monomials.

(i) If L is a f.d. Lie semialgebra with negation map, having base a1,...,an, then, viewing the ai as commuting indeterminates in the commutative polynomial semialgebra R = F [a1,...,an], introduce a Poisson bracket on R by deﬁning {ai,aj } to be the Lie product in L and extending the Poisson bracket via the Leibniz identities, i.e.,

{ab,c} = a{b,c} + {a,c}b, {a,bc} = {a,b}c + b{a,c}, ∀a,b,c ∈ A.

(ii) Suppose V is a f.d. vector space with an alternating bilinear form (in the sense that hv, vi 0). Take a base {x1,...,xn} of V . The polynomial semialgebra F [x1,...,xn] becomes a Poisson -semialgebra, where one deﬁnes {xi, xj } to be hxi, xj i.

The super-version is obtained by taking instead Deﬁnition 10.10 and Proposition 10.11. ALGEBRAS WITH A NEGATION MAP 49

11. Areas for further research We have concentrated on the system as an algebraic structure. This leads to the following questions: (i) What is the complete classification of strictly negated systems for which e′ = 1? (See Proposi- tion 4.10 and Theorem 6.50 for motivation.) (ii) What systems are hypersystems with respect to some hyperfield? This is answered in [4]. (iii) What systems other than hypersystems are of hypergroup type? Which satisfy the properties given in §4.5, namely T -strictly negated, T -reversible systems? One might want to throw in idempotence, which implies that (−) is of the second kind and e′ = e. (iv) What can be proved in linear algebra? (See [5] for some results.) (v) What is the theory of affine and projective geometry over systems, starting with Remark 5.7)? (vi) What is the theory of matroids and valuated matroids over systems? This has been started in [4] and [23]. (vii) How does the representation theory of systems fit in with [17]? This is begun in [49]. (viii) How far can one develop Lie structure theory along systemic lines, starting with Lie’s theorem and Engel’s theorem? (The negation map (−) could be either the identity or the switch map on the symmetrized algebra.) (ix) How does one develop Hopf systems? (Hopf semialgebras are in the literature.) (x) What is the geometry of systems?

12. *Appendix A: Major examples of hypergroups and hyperfields and their power sets

We bring the major examples of hyperﬁelds, cf. [7] into the systemic setting. Example 12.1. The ﬁrst four of these examples correspond to (−)-bipotent systems (each of height 2), but the last two do not.

• The supertropical hyperfield. Define R∞ = R ∪ {−∞} and define the product a b := a + b and max(a,b) if a 6= b, J a ⊞ b = ({c : c ≤ a} if a = b. Thus 0 is the multiplicative identity, −∞ is the additive identity, and we have a hyperfield called the tropical hyperfield, a special case of Proposition 6.52. This is easily seen to be isomorphic (as hyperfields) to the supertropical semiring of Definition 3.4, which goes back to Izhakian’s extended tropical arithmetic, identifying (−∞,a] := {c : c ≤ a} with aν . We have a natural

semiring system isomorphism with the sub-semiring R∞ of P(R∞), sending the tangible elements to R∞, because g b if b>a; (−∞,a]+ b = (−∞,a] if b = a; (−∞,b] ∪ (b,a] = (−∞,a] if b < a. • The Krasner hyperfield. Let K = {0;1} with the usual operations of Boolean algebra, except  that now 1 ⊞ 1 = {0;1}. Again, this generates a sub-semiring of P(K), having three elements, which is isomorphic to the supertropical semiring system of the monoid K with tangible elements 0 and 1, where we identify {0;1} with 1ν. • The hyperfield of signs. Let T := {0, 1, −1} with the usual multiplication law and hyperaddi- tion defined by 1 ⊞ 1 = {1}, −1 ⊞ −1 = {−1}, x ⊞ 0=0 ⊞ x = {x}, and 1 ⊞ −1 = −1 ⊞ 1 = {0, 1, −1} = T . Then T is a hyperfield called the hyperfield of signs. • Valuative hyperfields ([7, Example 2.12]) also are systemically isomorphic to the extended semiring in the sense of [42], in the same way. As noted in [27, Example 6.9], the four elements {{0}, {−1}, {1}, T} constitute the sub-semiring† T of P(T ), comprising a metatangible system, as seen in Example 6.42. • The phase hyperfield. Let S1 denote the complex unit circle, together with the center {0}, and takee T = S1. Points a and b are antipodes if a = −b. Multiplication is defined as usual (so 50 LOUIS ROWEN

corresponds on S1 to addition of angles). We call an arc from a to b of less than 180 degrees short, and denote it as (a,b). The hypersum is given by

(a,b) if a 6= b; ⊞ a b = {−a, 0,a} if a = −b 6= 0; {a} if b =0.

Then T0 is a hyperfield called the phase hyperfield. 1 At the power set level, given T1,T2 ⊆ S , one of which containing at least two points, we define T1 ⊞ T2 to be the union of all (short) arcs from a point of T1 to a non-antipodal point in T2 (which together make a connected arc), together with {0} if T2 contains an antipode of T1. Thus the system A spanned by T is not metatangible; the sum of two distinct points of T is never in T , so this is as far from metatangible as one can get. Its negation map is of the second kind. Its elements can be described as follows: (i) {0}, which has height 0, (ii) T , the points on S1, each of which has height 1, (iii) Short arcs (the sum of non-antipodal distinct points), which have height 2, (iv) The sets {a, 0, −a} = a − a, which we write as a◦, which have height 2, (v) Semicircles with 0 adjoined, having the form a◦ +b where b 6= ±a, which have height 3 (which go clockwise or counter-clockwise depending on the relation from b to a), (vi) S1 ∪{0} = a◦ + b◦ where b 6= ±a. This also can be written as the sum of three points of an 1 2π equilateral triangle on S , i.e., at angles of 3 , so has height 3. The additive structure is described as follows in terms of the hyperfield T : (i) A short arc T plus a point c with −c∈ / T is a short arc. (ii) A short arc T = (a,b) plus a point c with −c ∈{a,b} is a semicircle with {0} adjoined. (iii) A short arc T = (a,b) plus a point c with −c ∈ T \{a,b} is S1 ∪{0}. (iv) a◦ + b + c is either a◦ (for b,c ∈ {±a}), a semicircle with {0} adjoined (if b,c are on the same side of ±a), or S1 ∪{0} if b,c are on different sides of ±a). It follows that any finite sum of elements of T0 is one of the sets given above, so these comprise a system (A, T , (−), ⊆). On the other hand, any proper arc of S1 can be obtained as the product of two short arcs and a point. Hence, A is not closed under multiplication! S1 ∪{0} itself is obtained as the sum of three points (say with each 120 degrees apart). Thus the system A has height 3; the elements of height 3 are precisely the semicircles with the origin, and S1 ∪{0}. Distributivity fails since certain arcs cannot be obtained as unions of arcs. For example, take a1 and a2 almost to be antipodes, b1 = a2, and the arc connecting b1 and b2 just passes the antipode of a1; then (a1 ⊞ a2)(b1 ⊞ b2) is the arc from a1 to b2, a little more than a semicircle, 1 whereas a1b1 ⊞ a1b2 ⊞ a2b2 is already all of S . But this can be remedied by defining multiplication instead to be the convex union of hulls of the points on the arcs. (This can be viewed as a special case of Theorem 2.45.) It is easy to check that Lemma 2.30 and the proof of Proposition 2.40 are applicable, since set inclusion is antisymmetric, so (A, T , (−), ⊆) remains a hypersystem for either choice of multiplication. Viro [68] also presents another version. • The “triangle” hyperfield T defined over R+ by the formula a ⊞ b = {c ∈ R+ : |a − b|≤ c ≤ a + b}. Again distributivity fails for A. Here

T + T = {[a1,a2]: a1 ≤ a2}, so A obviously is not metatangible. The negation map is of the ﬁrst kind, satisfying

a◦ = a ⊞ a = {c ∈ R+ :0 ≤ c ≤ 2a}.

a1+a2 a2−a1 ˆ ′ ′ It A has height 2 over T . Indeed, [a1,a2] = 2 + 2 ∈ A, whereas [a1,a2] + [a1,a2] is ′ some interval going up to a2 + a2. ALGEBRAS WITH A NEGATION MAP 51

13. **Appendix B: Fuzzy rings as systems

Another concept which turns out to provide systems was introduced in 1986 and refined in 2011 by Dress [20], and Dress and Wenzel [21]. This treatment also is inspired by [27]. Let A := (A, +, ·, 0, 1) be a cbimagma, for which (A, ·, 1) is a commutative monoid. A× denote the set of invertible elements of A. Definition 13.1 ([20, Definitions 2.1, 2.8], [27, Definition 2.14]). A is a fuzzy ring if it is an A×-module and has a distinguished element ε and a proper ideal A0 satisfying the following axioms for a,ai ∈ A: (i) ε2 = 1; × (ii) a = ε, iff a ∈ A with 1 + a ∈ A0; (iii) If a1 + a2, a3 + a4 ∈ A0, then a1a3 + εa2a4 ∈ A0; (iv) If a1 + a2(a3 + a4) ∈ A0, then a1 + a2a3 + a2a4 ∈ A0. The fuzzy ring is coherent if A× spans (A, +). Note that (iii) is the fuzzy property of Definition 6.53. In line with the systemic approach, it is natural to generalize the definition slightly, replacing A× by a monoid T . On the other hand, conditions (iii) and (iv) do not initially enter into our proofs (and also did not enter into the proof of [27, Theorem 3.3]). This motivates us to suppress them at the outset, to get a more straightforward structure theory using triples. Definition 13.2. A pre-fuzzy triple is a cancelative cbimagma triple (A, T , (−)) where T is a multiplicative submonoid of A, together with a distinguished element ε ∈ T and a proper ideal A0 satisfying the following axioms: (i) ε2 = 1; (ii) For any ai ∈ T , a1 + a2 ∈ A0 iff a1 = εa2; (iii) (−)b := εb.

The pre-fuzzy triple is T -coherent if (A, +) = { ﬁnite ai : ai ∈T}. ◦ × Remark 13.3. Condition (ii) implies A ⊆ A0. TheP fuzzy ring deﬁnition takes T = A . Remark 13.4.

(i) In a pre-fuzzy triple A, the sub-T -cbimagma generated by T and A0 is also pre-fuzzy, so we assume from now on that it equals A. × (ii) Condition (iii) of Definition 13.2 matches Definition 13.1(ii) for a1 ∈ A . (iii) Adjusting multiplication according to in Theorem 2.45 always enables us to dispose of Condition (iv) of Definition 13.1. Definition 13.5. A pre-fuzzy triple is T -strictly negated if

a + b ∈ A0, a ∈ T , implies b = εa + c for some c ∈ A0. (13.1) The next result reconciles fuzzy rings with Deﬁnition 13.2.

Lemma 13.6. Assume that (A, T , (−)) is a pre-fuzzy triple with A = T + A0, and put ε = (−)1.

(i) Condition (iii) of Deﬁnition 13.1 holds whenever a1,a2 ∈ T . (ii) Condition (iii) of Deﬁnition 13.1 holds whenever A is T -strictly negated.

Proof. (i) a1 + a2 ∈ A0, so a1 = εa2 by Deﬁnition 13.2(ii). Hence a1a3 + εa2a4 = a1(a3 + a4) ∈ A0. (ii) The assertion is obvious if a1,a2 ∈ A0, so we may assume that a1 ∈/ A0, i.e., a1 ∈ T . But we are given a1 + a2 ∈ A0, so, by Deﬁnition 13.5, a1 + c = εa2 for some c ∈ A0. Hence a1a3 + εa2a4 = a1(a3 + a4)+ ca4 ∈ A0.

Proposition 13.7. Any strictly negated system with respect to ◦ satisﬁes the fuzzy property of Deﬁni- tion 6.53. ′ ◦ ′ ′ Proof. This is clear if bi,bi ∈ A for either i, so by (4.1) we have either bi bi or bi bi, so we may ′ ′ ′ assume that bi bi for i = 1, 2. (We can replace bi,bi by (−)bi, (−)bi if necessary.) But now writing ′ ◦ bi = bi + ci we have ′ ′ ◦ ◦ ◦ b1b2(−)b1b2 = b1b2(−)b1b2(−)c1b2(−)c2b1 ∈ A . 52 LOUIS ROWEN

In a metatangible pre-fuzzy triple A, we can also replace A by { finite ai : ai ∈T} , in which case A becomes T -coherent. Condition (iv) of Definition 13.1 then becomes superfluous, in view of [61, Theorem 7.34]. In short, pre-fuzzy triples often are fuzzy rings. P 13.1. Fuzzy rings versus pre-fuzzy triples and systems.

Theorem 13.8. A T -coherent pre-fuzzy triple A gives rise to a system (A, T , (−), A0 ), cf. Proposi- tion 2.40(iii), where (−)a = εa. Proof. The map a 7→ εa obviously is a negation map. Furthermore, if a(−)b ∈ A◦ for a,b ∈ T , then 1 + εba−1 ∈ A◦, implying εba−1 = ε, and thus b = a. a(−)a = a + εa ∈ A0. For to be a surpassing relation, we need to verify the conditions of Deﬁni- tion 2.28. Conditions (i)–(iv) are clear; for (v), suppose a0 a1 for ai ∈ T . Then a1 = a0 + c for c ∈ A0 ◦ implies a1(−)a0 = a0 + c ∈ A0, and thus a1 = a0. Since T ∩ A◦ = ∅, this is a system, by Lemma 2.30 and Proposition 2.40(iii).

MAJOR NOTE 13.9. We are back to the deﬁnition of system, where ANull = A0. Thus systems provide a straightforward way of viewing fuzzy rings, and their theory includes that of fuzzy rings. Conversely to Theorem 13.8, the notion of pre-fuzzy triple also encompasses cancelative triples. Proposition 13.10. Suppose that A := (A, T , (−)) is a cancelative cbimagma triple with unique quasi- ′ ◦ negatives. Then A gives rise to a pre-fuzzy triple A with the same operations, where A0 = A and ′ (A , +) is generated by T and A0, and ε = (−)1. Proof. Note that (A′, T , (−), ) is a triple, so we may assume that A′ = A. The conditions of Deﬁni- tion 13.2 are clear.

References

[1] M. Akian, R. Bapat, and S. Gaubert. Max-plus algebra, In: Hogben, L., Brualdi, R., Greenbaum, A., Mathias, R. (eds.) Handbook of Linear Algebra. Chapman and Hall, London, 2006. [2] M. Akian, S. Gaubert, and A. Guterman. Linear independence over tropical semirings and beyond. In Tropical and Idempotent Mathematics, G.L. Litvinov and S.N. Sergeev, (eds.), Contemp. Math. 495 (2009), 1–38. [3] M. Akian, S. Gaubert, and A. Guterman. Tropical Cramer Determinants revisited, Contemp. Math. 616, Amer. Math. Soc. (2014), 1–45. [4] M. Akian, S. Gaubert, and L. Rowen, Systems of hyperfields and related examples, preprint (2020). [5] M. Akian, S. Gaubert, and L. Rowen, Linear algebra over systems, preprint (2020). [6] F. Baccelli, G. Cohen, G.J. Olsder, and J.P. Quadrat, Synchronization and linearity. Wiley Series in Probability and Mathematical Statistics: Probability and Mathematical Statistics. John Wiley and Sons, Ltd., Chichester, 1992. [7] M. Baker and N. Bowler. Matroids over hyperfields, arXiv:1601.01204v4, (2016). [8] M. Baker and N. Bowler. Matroids over partial hyperstructures, arXiv:1709.09707v1 (2017). [9] A. Bertram and R. Easton, The tropical nullstellensatz for congruences, Advances in Mathematics 308 (2017), 36–82. [10] G. Bergman, An Invitation to General Algebra and Universal Constructions 1998. (pub. Henry Helson, 15 the Crescent, Berkeley CA, 94708) 398 pp. ISBN 0-9655211-4-1. [11] G. Blachar, Exploded Layered Tropical Lie Algebras, Masters Thesis, Bar-Ilan University (2016). [12] G. Blachar and E. Sheiner, ELT Linear Algebra, arXiv:1603.02204 (2016). [13] N. Bourbaki. Commutative Algebra. Paris and Reading, 1972. [14] A. Chapman and L. Rowen, Quadratic forms over systems, preprint (2020). [15] P.M. Cohn, Universal Algebra. Harper and Row, 1965. [16] A. Connes and C. Consani, From monoids to hyperstructures: in search of an absolute arithmetic. In Casimir Force, Casimir operators, and the Riemann hypothesis, Walter de Gruyter, Berlin (2010), 147–198. [17] A. Connes and C. Consani, Homological algebra in characteristic one, arXiv:1703.02325v1 [math.AG] 7 Mar 2017. [18] A.A. Costa. Sur la thêorie générale des demi-anneaux, Publ. Math. Decebren 10 (1963), 14–29. [19] D. Dolzan, P. Oblak, Invertible and nilpotent matrices over antirings, Linear Algebra and its Applications 430 (2009), 271–278. [20] A. Dress, Duality theory for finite and infinite matroids with coefficients, Advances in Mathematics 93(2) (1986), 214–250. [21] A. Dress and W. Wenzel, Algebraic, tropical, and fuzzyq geometry, Beitrage zur Algebra und Geometrie/ Contributions to Algebra und Geometry 52 (2), 431–461, 2011. [22] R. Eilhauer, Zur Theorie der Halbkörper I, Acta Math. Sci. Acad. Hung 19 (1968), 23–45. ALGEBRAS WITH A NEGATION MAP 53

[23] L. Gatto and L. Rowen, Grassman semialgebras and the Cayley-Hamilton theorem, Proc. Amer. Math. Soc., to appear, arXiv:1803.08093. [24] S. Gaubert. Théorie des systèmes linéaires dans les diodes. Thèse, Ecole´ des Mines de Paris, 1992. [25] S. Gaubert and M. Plus. Methods and applications of (max,+) linear algebra. In R. Reischuk and M. Morvan, editors, STACS97, number 1200 in LNCS, Lubeck, March 1997. Springer. [26] J. Giansirancusa and N. Giansirancusa, A Grassmann algebra for matroids, arXiv:1510.04584v1 (2015). [27] J. Giransirussa, J. Jun, and O. Lorscheid, On the relation between hyperrings and fuzzy rings. [28] J. Golan, Semirings and their Applications, Springer-Science + Business, Dordrecht, 1999. (Previously published by Kluwer Acad. Publ., 1999.) [29] M. Gondran and M. Minoux. Graphs, dioids and semirings, volume 41 of Operations Research/Computer Science Interfaces Series. Springer, New York, 2008. [30] A. Grothendieck, Produits tensoriels topologiques et espaces nucleaires, Memoirs of the Amer. Math. Soc., no. 16, 1955. [31] D.K. Harrison and M.A. Vitulli, V -valuations of a commutative ring I, J. Algebra 126 (1989), 264–292. [32] S. Henry, Symmetrization of monoids as hypergroups, arXiv:1309.1963v1 (2013). [33] I. Itenberg, G. Mikhalkin and E. Shustin. Tropical Algebraic Geometry, Oberwolfach Seminars, 35, Birkhäuser Verlag, Basel, 2007. [34] Z. Izhakian. Tropical arithmetic and matrix algebra. Commun. in Algebra, 37(4) 2009, 1445–1468. [35] Z. Izhakian, M. Knebusch, and L. Rowen, Supertropical semirings and supervaluations. J. Pure and Appl. Alg. 215(10) (2011), 2431–2463. [36] Z. Izhakian, M. Knebusch, and L. Rowen, Categorical notions of layered tropical algebra and geometry, Algebraic and combinatorial aspects of tropical geometry, Contemp. Math., 589, Amer. Math. Soc. (2013), 191–234. [37] Z. Izhakian, M. Knebusch, and L. Rowen, Supertropical linear algebra, Pacific Journal of Mathematics, 266(1) (2013), 43–75. [38] Z. Izhakian, M. Knebusch, and L. Rowen, Categories of layered semirings, Commun. in Algebra 43(5) (2015), 1807– 1836. [39] Z. Izhakian, M. Knebusch, and L. Rowen, Supertropical quadratic forms I, J. Pure Appl. Algebra 220 no. 1 (2016), 61–93. [40] Z. Izhak ian, M. Knebusch, and L. Rowen, Supertropical Quadratic and Symmetric Bilinear Forms on Modules with Unique Base over a semiring, arXiv:1509.01039 (2015), to appear in Documenta Matematika. [41] Z. Izhakian, A. Niv, and L. Rowen, Supertropical SL, arXiv:1508.04483 (2015). [42] Z. Izhakian and L. Rowen, Supertropical algebra, Advances in Mathematics, 225(4) (2010), 2222–2286; arXiv:0806.1175 (2007). [43] Z. Izhakian and L. Rowen, The tropical rank of a tropical matrix. Commun. in Algebra, 37:11 (2009), 3912–3927. [44] Z. Izhakian and L. Rowen, Supertropical matrix algebra, Israel J. Math., 182(1) (2011), 383–424. [45] N. Jacobson, Basic algebra II, Freeman, 1980. [46] J. Jun, Algebraic geometry over hyperrings, arxiv:math.AG/1512.04837 (2015). [47] D. Joo and K. Mincheva, Prime congruences of idempotent semirings and a Nullstellensatz for tropical polynomials, arXiv:1408.3817v1 (2014). [48] J. Jun, K. Mincheva, and Rowen,L., Projective module systems, arXiv:1809.01996, J. Pure and applied algebra, to appear (2019). [49] J. Jun, K. Mincheva, and Rowen,L., Homology of systemic modules, arXiv:2005.12760 (2000). [50] Y. Katsov, Tensor products of functors, Siberian J. Math. 19 (1978), 222-229, trans. from Sirbiskii Mathematischekii Zhurnal 19 no. 2 (1978), 318–327. [51] Y. Katsov, Tensor products and injective envelopes of semimodules over additively regular semirings, Algebra Collo- quium 4 no. 2, (1997), 121–131. [52] J. Kuntzman, Théorie des Réseaux (1972), Dunot, Paris. [53] O. Lorscheid, A blueprinted view on F1-geometry, in Absolute Arithmetic and F1-geometry (edited by Koen Thas), European Mathematical Society Publishing House, 2016. [54] A. Niv. Factorization of tropical matrices, J. Algebra Appl., 13(1) (2014), 1–26. [55] B. Parker, Exploded manifolds, Adv. Math. 229 (2012), 3256–3319. [56] M. Plus (M. Akian, G. Cohen, S. Gaubert, R. Nikoukhah, and J.P. Quadrat), Linear systems in (max,+)-algebra. In Proceedings of the 29th Conference on Decision and Control, Honolulu, Dec. 1990. [57] C. Reutenauer and H. Straubing, Inversion of matrices over a commutative semiring, J. Algebra, 88 (1984), 350–360. [58] L.H. Rowen, Graduate algebra: Commutative view, AMS Graduate Studies in Mathematics 73, 2006. [59] L.H. Rowen, Graduate algebra: Noncommutative view, AMS Graduate Studies in Mathematics 91, 2008. [60] L. Rowen, Algebras with a negation map, first version, arXiv:1602.00353v1. [61] L. Rowen, Algebras with a negation map, fifth version, arXiv:1602.00353v5. [62] L. Rowen, An informal overview of triples and systems, arXiv:1709.03174. [63] C. Semple and G. Whittle, Partial fields and matroid representation, Adv. in Appl. Math., 17(2) (1996), 184–208. [64] E. Sheiner, Doctoral dissertation, Bar-Ilan University, 2015. [65] H. Straubing, A combinatorial proof of the Cayley-Hamilton theorem. Discrete Math., 43(2-3)(1983), 273–279. [66] M. Takahashi, On the bordism categories III, Math. Seminar Notes Kobe University 10 (1982), 211–236. [67] Y. Tan, On invertible matrices over antirings, Linear Algebra and its Applications 423 (2007), 428–444. [68] O.Y. Viro, Hyperfields for tropical geometry I. Hyperfields and dequantization (2010). ArXiv AG/1006.3034. 54 LOUIS ROWEN

[69] D. Zelinsky, Nonassociative valuations, Bull. Amer. Math. Soc. 54 (1948), 175–183.

Department of Mathematics, Bar-Ilan University, Ramat-Gan 52900, Israel Email address: [email protected]