- Home
- Documents
*Proof Theory for Lattice-Ordered Groups Keywords: Lattice-ordered groups, Proof theory, Hypersequent*

prev

next

out of 25

View

1Download

0

Embed Size (px)

Proof Theory for Lattice-Ordered Groups

Nikolaos Galatos1

Department of Mathematics, University of Denver, 2360 S. Gaylord St., Denver, CO 80208, USA

George Metcalfe2,∗

Mathematical Institute, University of Bern, Sidlerstrasse 5, 3012 Bern, Switzerland

Abstract

Proof-theoretic methods are developed and exploited to establish properties of the variety of lattice-ordered groups. In particular, a hypersequent calculus with a cut rule is used to provide an alternative syntactic proof of the generation of the variety by the lattice-ordered group of automorphisms of the real number chain. Completeness is also established for an analytic (cut-free) hypersequent calculus using cut elimination and it is proved that the equational theory of the variety is co-NP complete.

Keywords: Lattice-ordered groups, Proof theory, Hypersequent calculi, Cut elimination, Co-NP completeness

1. Introduction

A lattice-ordered group (`-group) is an algebraic structure

(L,∧,∨, ·, −1, 1)

such that (L,∧,∨) is a lattice, (L, ·, −1, 1) is a group, and · preserves the order in both arguments; i.e., a ≤ b implies a · c ≤ b · c and c · a ≤ c · b for all a, b, c ∈ L. It follows also from this definition that the lattice (L,∧,∨) is distributive and that

∗Corresponding author Email addresses: ngalatos@du.edu (Nikolaos Galatos),

george.metcalfe@math.unibe.ch (George Metcalfe) 1Supported by Simons Foundation grant 245805 and FWF project START Y544-N23. 2Supported by Swiss National Science Foundation (SNF) grant 200021 146748.

Preprint submitted to Annals of Pure and Applied Logic April 1, 2016

1 ≤ a ∨ a−1 for all a ∈ L. We refer to [1] for proofs and further standard facts about this class of structures.

Commutative `-groups include the real, rational, and integer numbers with the standard total order and addition. For non-commutative examples, consider a chain (totally-ordered set) Ω and denote by Aut(Ω) the set of all order-preserving bijections on Ω. Then Aut(Ω) constitutes an `-group Aut(Ω) under coordinate- wise lattice operations, functional composition, and functional inverse. It was proved by Holland in [10] that the variety of `-groupsLG is generated by Aut(R), where R is the real number chain, or indeed by any Aut(Ω), where Ω is an n- transitive chain for all n (i.e., for any two n-tuples of elements of Ω there is a bijection that maps the first tuple to the second). The standard proof relies on Holland’s embedding theorem, which states, analogously to Cayley’s theorem for groups, that every `-group embeds into an `-group Aut(Ω) for some chain Ω [9]. Although not every `-group embeds into Aut(R), each identity that fails in some `-group fails, by the embedding theorem, in some automorphism `-group, and a simple argument then shows that the identity must also fail in Aut(R). This generation result for LG was subsequently exploited by Holland and McCleary to provide an algorithm for checking if an identity is valid in all `-groups [11].

The first main contribution of this paper is a new syntactic (and first axiom of choice free) proof that Aut(R) generates the variety LG of `-groups. A proof system is defined in a one-sided hypersequent framework such that derivability of a hypersequent (interpreted as a disjunction of group terms) implies the validity of a corresponding identity in all `-groups. A rule is then added to the system and it is shown, following closely the Holland-McCleary algorithm of [11], that this augmented system derives all identities (rewritten in a certain form) that are valid in Aut(R). Finally, it is proved syntactically that applications of this rule can be eliminated from derivations. Hence an identity is valid in Aut(R) if and only if it is valid in all `-groups, and so, by Birkhoff’s variety theorem, Aut(R) generates LG. This proof illustrates the usefulness of proof-theoretic methods for tackling algebraic problems, and is similar to proofs of generation of varieties by dense chains via density elimination (see [4, 14]) or of properties such as interpolation and amalgamation via cut elimination (see, e.g., [7, 18]).

The second main contribution is the introduction of a first analytic (cut-free) proof calculus for `-groups. In contrast to the well-developed proof theory for well-behaved families of varieties of residuated lattices (which provide algebraic semantics for substructural logics, see [2,3,7,14,17,18]), there has been relatively little success in obtaining cut-free systems for algebraic structures related to `- groups. Hypersequent calculi have been defined for abelian `-groups and related

2

varieties in [13, 15–17], but a calculus for the general non-commutative case has until now been lacking. The virtue of such a calculus is illustrated by the fact that we obtain not only the known decidability result for the equational theory of `-groups, but also, via cut elimination, a (first) procedure for obtaining proofs of valid `-group identities in equational logic (i.e., using only defining identities of LG). More generally, the analytic hypersequent calculus presented here provides a crucial first step towards developing a uniform proof theory for the wide range of algebras and logics related in some way to `-groups: in particular, MV-algebras and GMV-algebras (which may be viewed as intervals in abelian `-groups [20] and `-groups [6, 8], respectively) and cancellative residuated lattices (which may be viewed as `-groups with a co-nucleus [19]).

The final contribution of the paper is a first proof that the equational theory of `-groups is co-NP complete, matching the complexity of the equational theories of both abelian `-groups [21] and distributive lattices [12].

2. Preliminaries

Let us call a variable x and its inverse x−1 literals. Using De Morgan identities valid in all `-groups, we consider only normalized `-group terms s, t built from literals and the operation symbols 1, ∧, ∨, and ·, with an inductively defined inverse:

1 = 1 (s · t) = t · s x = x−1 (s ∧ t) = s ∨ t

x−1 = x (s ∨ t) = s ∧ t. As usual we write st for s · t, omit brackets in group terms (built using literals, ·, and 1), and define t0 = 1 and tn+1 = t · tn for n ∈ N. We also write s ≤ t as an abbreviation for the identity s ∧ t ≈ s.

Given a class of `-groupsK, we letK |= s ≈ t denote that the `-group identity s ≈ t is valid in all members of K. Using standard distributivity laws in `-groups, every `-group term t is equivalent either to 1 or to a meet of joins of group terms (see [1] for basic facts about `-groups). That is, for some index sets I and Ji 6= ∅ (i ∈ I) and group terms tij , assuming also

∧ ∅ = 1,

LG |= t ≈ ∧ i∈I

∨ j∈Ji

tij.

Moreover, for any `-group terms s, t,

LG |= s ≈ t ⇔ LG |= 1 ≤ (st) ∧ (ts),

3

and for any finite family of `-group terms (ti)i∈I ,

LG |= 1 ≤ ∧ i∈I

ti ⇔ LG |= 1 ≤ ti for each i ∈ I.

Hence checking the validity of `-group identities amounts to checking the validity of inequations of the form 1 ≤ t where t is a join of group terms. Following proof- theoretic treatments of related classes of algebras (see [2, 3, 13–18]), we present such terms and identities here in the framework of sequents and hypersequents.

An `-sequent, denoted by Γ, ∆, Π, or Σ, is a finite (possibly empty) sequence of `-group terms, written

(t1, . . . , tn),

and interpreted as an `-group term by

I() = 1 and I(t1, . . . , tn) = t1 · . . . · tn.

The inverse of an `-sequent is defined as

(t1, . . . , tn) = (tn, . . . , t1).

For `-sequents Γ and ∆, we denote their concatenation by (Γ,∆) or simply by Γ,∆. An `-sequent will be called basic if it is a sequence of literals, and a basic `-sequent Γ will be called group valid if 1 ≈ I(Γ) is valid in all groups.

An `-hypersequent, denoted by G orH, is a finite (possibly empty) multiset of `-sequents, written

Γ1 | . . . | Γn,

and interpreted, when n ≥ 1, as an `-group term by

I(Γ1 | . . . | Γn) = I(Γ1) ∨ . . . ∨ I(Γn).

An `-hypersequent G will be called basic if it contains only basic `-sequents. It will be called valid in an `-group L if L |= 1 ≤ I(G), and `-valid if it is valid in all `-groups.

An `-hypersequent rule is the set of its instances, each instance consisting of a finite set of `-hypersequents called the premises and an `-hypersequent called the conclusion of the instance. Typically, such rules are written schematically using s, t to denote arbitrary `-terms and Γ,Π,Σ,∆ and G,H to denote arbitrary `-sequents and `-hypersequents, respectively. An `-hypersequent calculus GL is a set of `-hypersequent rules. A GL-derivation of an `-hypersequent G from a set

4

G | Γ (GV) G | ∆ | ∆ (EM)

Γ group valid

G | Γ,∆ G | ∆,Σ G | Γ,Σ (CUT)

G G | H (EW)

G | H | H G | H (EC)

Figure 1: The basic `-hypersequent calculus G`

of `-hypersequents S is a finite tree of `-hypersequents with root G such that each node is either in S or the node and its parents form an instance of a rule of GL. In this case, we write S `GL G and say that G is GL-derivable from S.

A rule is said to be GL-derivable if for each of its instances, the conclusion is GL-derivable from the premises, and GL-admissible if for each of its instances, whenever the premises are GL-derivable, the conclusion is GL-derivable. A rule will also be called `-sound if for each of its instances, whenever the pre