Decorated Linear Order Types and

the Theory of Concatenation

cic
Vedran Ca

Pavel Pudlak

Department of Mathematics

Mathematical Institute

University of Zagreb
Bijeni
cka 30, 10000 Zagreb

Academy of Sciences of the Czech Republic

a 25, 115 67 Praha 1
Zitn

Croatia

Czech Republic

veky@math.hr

pudlak@math.cas.cz

Greg Restall

Alasdair Urquhart

Department of Philosophy

Department of Philosophy

The University of Melbourne

University of Toronto

Parkville 3010

215 Huron Street, Toronto

Australia

Canada

restall@unimelb.edu.au

urquhart@cs.toronto.edu

Albert Visser
Department of Philosophy
Utrecht University
Heidelberglaan 8, 3584 CS Utrecht
The Netherlands
albert.visser@phil.uu.nl

August 27, 2007

Abstract
We study the interpretation of Grzegorczyks Theory of Concatenation
TC in structures of decorated linear order types satisfying Grzegorczyks
axioms. We show that TC is incomplete for this interpretation. What
is more, the first order theory validated by this interpretation interprets
arithmetical truth. We also show that every extension of TC has a model
that is not isomorphic to a structure of decorated order types.

Introduction

In his paper [Grz05], Andrzej Grzegorczyk introduces a theory of concatenation TC. The theory has a binary function symbol for concatenation and two
constants a and b. The theory is axiomatized as follows.
TC1. ` (x y) z = x (y z)
TC2. ` x y = u v ((x = u y = v)
w ((x w = u y = w v) (x = u w y w = v)))
TC3. ` x y 6= a
TC4. ` x y 6= b
TC5. ` a 6= b
Axioms TC1 and TC2 are due to Tarski. Grzegorczyk calls axiom TC2: the
editor axiom. We will consider two weaker theories. The theory TC0 has the
signature with just concatenation, and is axiomatized by TC1,2. The theory
TC1 is axiomatized by TC1,2,3. We will also use TC2 for TC.1
Andrzej Grzegorczyk and Konrad Zdanowski have shown that TC is essentially
undecidable. This result can be strengthened by showing that Robinsons Arithmetic Q is mutually interpretable with TC. Note that TC0 is undecidable since
it has an extension that parametrically interprets TC but that TC0 is not essentially undecidable: it is satisfied by a one-point model. Similarly TC1 is
undecidable, but it has as an extension the theory of finite strings of as, which
is a notational variant of Presburger Arithmetic and, hence decidable.
We will call models of TC0 : concatenation structures. We will call models of
TCi : concatenation i-structures.
We will be interested in a special class of concatenation structures: those
whose elements are decorated linear order types with as operation addition, or
concatenation of decorated order types. Let a non-empty class A be given. An
A-decorated linear ordering is a structure hD, , f i, where D is a non-empty
domain, is a linear ordering on D, and f is a function from D to A. A
mapping is an isomorphism between A-decorated linear order types hD, , f i
and hD0 , 0 , f 0 i iff it is a bijection between D and D0 such that, for all d, e in
D, we have d e d 0 e, and f d = f 0 d. Our notion of isomorphism
gives us a notion of A-decorated linear order type. We have an obvious notion
of sum or concatenation between A-decorated linear orderings which induces a
corresponding notion of sum or concatenation for A-decorated linear order types.
1 The theories TC are theories for concatenation without the empty string, i.o.w. without
i
the unit element. One can show that TC is bi-interpretable with a corresponding theory TC
via one dimensional interpretations without parameters.2 The theory TC1 is bi-interpretable
via two-dimensional interpretations with parameters with a corresponding theory TC1 . The
situation for TC0 seems to be more subtle.

We use , , . . . to range over such linear order types. We write DLOT(A) for
the universe of A-decorated linear order types with concatenation. Since, linear
order types are classes we have to follow one of two strategies: either to employ
Scotts trick to associate a set object to any decorated linear order type or
to simply refrain from dividing out isomorphism but to think about decorated
linear orderings modulo isomorphism. We will employ the second strategy.
We will call a concatenation structure whose domain consists of (representatives of) A-decorated order types, for some A, and whose concatenation is
concatenation of decorated order types: a concrete concatenation structure. It
seems entirely reasonable to stipulate that e.g. the interpretation of a in a concrete concatenation structure is an decorated linear order type of a one element
order. However, for the sake of generality we will refrain from making this
stipilation.
Grzegorczyk conjectured that every concatenation 2-structure is isomorphic
to a concrete concatenation structure. We prove that this conjecture is false. (i)
Every extension of TC1 has a model that is not isomorphic to a concatenation
1-structure and (ii) The set of principles of valid in all concrete concatenation
2-structures interprets arithmetical truth.
The plan of the paper is as follows. We show, in Section 2, that we have,
for all decorated order types , and , the following principle: ()
= = = .3 It is easy to see that every group is is a
concatenation structure and that () does not hold in the two element group.
We show, in Section 5, that every concatenation structure can be extended to
a concatenation structure with any number of atoms. It follows that there is a
concatenation structure with at least two atoms in which () fails. Hence, TC
is incomplete for concrete concatenation structures. In Section 3, we provide a
counterargument of a different flavour. We provide a tally interpretation that
defines the natural numbers (with concatenation in the role of addition) in every
concrete concatenation 2-structure. It follows that for every extension of TC1 is
satisfied by a concatenation 1-structure that is not isomorphic to any concrete
concatenation 1-structure, to wit any model of that extension that contains a
non-standard element. In Section 4, we strengthen the result of Section 3, by
showing that in concrete concatenation 2-structures we can add multiplication
to the natural numbers. It follows that the set of arithmetically true sentences
is interpretable in the concretely valid consequences of TC2 .

Acknowledgments
Some of the results of this note were obtained during the Excursion to moun e`za of the inspiring Logic Colloquium 2007 in Wroclaw and, in part, in
tain Sl
the evening after the Excursion. We thank the organizers for providing this
wonderful opportunity.
3 This

fact was already known. See: [KT06], problem 6.13. Our proof, however, is different.

We thank Dana Scott for his comments, insights and questions.

A Principle for Decorated Order Types

In this section we prove a universal principe that holds in all concatenation

structures, which is not provable in TC. There is an earlier proof of this principle.
See: [KT06], problem 6.13. Our proof, however, is different.
Theorem 2.1 Let 0 , 1 , 2 be decorated order types. Suppose 1 = 0 1 2 .
Then 1 = 0 1 = 1 2 .

Proof
Suppose 1 = 0 1 2 . Consider a decorated linear ordering A := hA, , f i
of type 1 , By our assumption, we may partition A into A0 , A1 , A2 , such that:
hA, , f i = hA0 ,  A0 , f  A0 i hA1 ,  A1 , f  A1 i h,  A2 , f  A2 i,
where Ai := hAi ,  Ai , f  Ai i is an instance of i , Let : A A1 be an
isomorphism.
Let n A(i) := hn [A(i) ],  n [A(i) ], f  n [A(i) ]i. We have: n Ai is of order
type i and n A is of order type 1 .
Clearly, A0 is an initial substructure of A = A1 . So, A0 and A0 are
disjoint and S
A0 adjacent to the right of A0 . Similarly, for n A0 and n+1 A0 .

Take A0 := i i A0 . We find that A

0 := hA0 ,  A0 , f  A0 i is initial in A

and of decorated linear order type 0 . So 1 = 0 , for some . It follows

that 0 1 = 0 0 = 0 = 1 . The other identity is similar.
2
So, every concrete concatenation structure validates that 1 = 0 1 2
implies 1 = 0 1 = 1 2 . We postpone the proof that this principle is not
provable in TC to Section 5.

Definability of the Natural Numbers

In this section, we show that the natural numbers can be defined in every
concrete concatenation 1-structure. We define:
x y : x = y u (u x = y) v (x v = y) u, v (u x v = y).
x ini y : x = y v (x v = y).
x end y : x = y u (u x = y).
e a : mini n (m = a k ini m (k 6= m m = k a)).
n:N
e a for: m : N
ea n : N
e a . Etc.
We write m, n : N
4

We prove the main theorem of this section.

Theorem 3.1 In any concrete concatenation structure, we have:
e a = {an+1 | n }.
N
e a is precisely the class of natural numbers in tally representation (startI.o.w, N
ing with 1). Note that on this set is addition.

Proof
Consider any concrete concatenation 1-structure A. It is easy to see that every
e a.
an+1 is in N
e a is either a or it has a predecessor. The
Clearly, every element of of N
axioms of TC1 guarantee that this predecessor is unique. This justifies the
e a . Let be the order
introduction of the partial predecessor function pd on N
e a . If, for some n, pdn 0 is
type corresponding to a. Let 0 be any element of N
undefined, then 0 is clearly of the form k+1 , for k in .
We show that the other possibility cannot obtain. Suppose n := pdn 0 is
always defined. Let A be a decorated linear ordering of type and let Bi be
a decorated linear ordering of type i . We assume that the domain A of A is
disjoint from the domains Bi of the Bi . Thus, we may implement Bi+1 A just
by taking the union of the domains.
Let i be isomorphisms from Bi+1 A to Bi . Let Ai := (0 i )(A).
Then, the Ai are all of type and, for some C, we have B0 = C A1 A0 .
Similarly B1 = C A2 A1 . Let
be the opposite ordering of . It follows
e a .4 A contradiction. 2
that 0 = = 1 = pd(0 ). Hence, 0 is not in N
e a defines the tally natural numWe call a concatenation structure standard if N
bers. Since, by the usual argument, any any extension of TC1 has a model with
non-standard numbers, we have the following corollary.
Corollary 3.2 Every extension of TC1 has a model that is not isomorphic
to a concrete concatenation 1-structure. In a different formulation: for every concatenation 1-structure there is an elementarily equivalent concatenation
1-structure that is not isomorphic to a concrete concatenation 1-structure.
Note that the non-negative tally numbers with addition form a concrete concatee a ),
nation 1-structure. Thus, the concretely valid consequences of TC1 +x (x : N
i.e., the principles valid in every concrete concatenation 1-structure satisfying
e a ) are decidable.
x (x : N
4 Note

that we are not assuming that is in A.

Definability of Multiplication

If we have two atoms to work with, we can add multiplication to our tally
numbers. This makes the set of concretely valid consequences of TC nonarithmetical. The main ingredient of the definition of multiplication is the
theory of relations on tally numbers. In TC, we can develop such a theory. We
represent the relation {hx0 , y0 i, . . . , hxn1 , yn1 i}, by:
bb x0 b y0 bb x1 . . . bb xn1 b yn1 bb.
We define:
r : REL : bb end r,
:= bb,
e a bb x b y bb r.
x[r]y : x, y : N
adj(r, x, y) := r x b y bb.
Clearly, we have: TC ` u, v u[]v. To verify that this coding works we need
the adjunction principle.
Theorem 4.1 We have:
e a ) (u[adj(r, x, y)]v (u[r]v (u = x v = y))).
TC ` (r : REL x, y, u, v : N
We can prove this result by laborious and unperspicuous case splitting. However,
it is more to do the job with the help of a lemma. Consider any model of TC0 .
Fix an element w. We call a sequence (w0 , . . . , wk ) a partition of w if we
have that w0 wk = w. The partitions of w form a category with the
following morphisms. f : (u0 , . . . , un ) (w0 , . . . , wk ) iff f is a surjective and
weakly monotonic function from n + 1 to k + 1, such that, for any i k, wi =
us u` , where f (j) = i iff s j `. We write (u0 , . . . , un ) (w0 , . . . , wk )
for: f f : (u0 , . . . , un ) (w0 , . . . , wk ). In this case we say that (u0 , . . . , un ) is
a refinement of (w0 , . . . , wk ).
Lemma 4.1 Consider any concatenation structure. Consider a w in the structure. Then, any two partitions of w have a common refinement.

Proof
Fix any concatenation structure. We first prove that, for all w, all pairs of
partitions (u0 , . . . , un ) and (w0 , . . . , wk ) of w have a common refinement, by
induction of n + k.
If either n or k is 0, this is trivial. Suppose (u0 , . . . , un+1 ) and (w0 , . . . , wn+1 )
are partitions of w. By the editor axiom, either (a) u0 un = w0 wk
and un+1 = wk+1 , or there is a v such that (b) u0 un v = w0 wk
and un+1 = v wk+1 , or (c) u0 un = w0 wk v and v un+1 =
6

wk+1 . We only treat case (b), the other cases being easier or similar. By the
induction hypothesis, there is a common refinement (x0 , . . . xm ) of (u0 , . . . , un , v)
and (w0 , . . . , wn ). Let this be witnessed by f , resp. g. It is easily seen that
(x0 , . . . xm , wk+1 ) is the desired refinement with witnessing functions f 0 and g 0 ,
where f 0 := f [m + 1 : n + 1], g 0 := g[m + 1 : k + 1].
2
We turn to the proof of Theorem 4.1. The verification proceeds more or less as
one would do it for finite strings.

Proof
Consider any concatenation 2-structure. Suppose REL(r). The right-to-left
direction is easy, so we treat left-to-right. Suppose x, y, u and v are tally
numbers. and u[adj(r, x, y)]v. There are two possibilities. Either r = bb or r =
r0 bb. We will treat the second case. Let s := adj(r, x, y). One the following four
partitions is a partition of s: (i) (b, b, u, b, v, b, b), or (ii) (w, b, b, u, b, v, b, b), or
(iii) (b, b, u, b, v, b, b, z), or (iv) (w, b, b, u, b, v, b, b, z). We will treat cases (ii)
and (iv).
Suppose := (w, b, b, u, b, v, b, b) is a partition of s. We also have that :=
(r0 , b, b, x, b, y, b, b) is a partition of s. Let (t0 , . . . , tk ) be a common refinement
of and , with witnessing functions f and g. The displayed bs in these
partitions must have unique places among the ti . We define m to be the
unique i such that f (i) = m, provided that m = b. Similarly, for m . (To
make this unambiguous, we assume that if = , we take as the common
refinement with f and g both the identity function.)
We evidently have 7 = 7 = k and 6 = 6 = k 1. Suppose 4 < 4 .
It follows that b v. So, v would have an initial subsequence that ends in b,
which is impossible. So, 4 6< 4 . Similarly, 4 6< 4 . So 4 = 4 . It follows
that v = y. Reasoning as in the case of 4 and 4 , we can show that 2 = 2
and, hence u = x.
Suppose := (w, b, b, u, b, v, b, b, z) is a partition of s. We also have that :=
(r0 , b, b, x, b, y, b, b) is a partition of s. Let (t0 , . . . , tk ) be a common refinement
of and , with witnessing functions f and g. We consider all cases, where
1 < 6 . Suppose 6 = 1 + 1 = 2 . Note that 7 = 6 + 1, so we find: b x,
e a . Suppose 2 < 6 < 4 . In this case we have a b
quod non, since x is in N
as substring of x. Quod non. Suppose 6 = 4 . Since 7 = 6 + 1, we get a b
in y. Quod non. Suppose 4 < 6 < 6 . In this case, we get a b in y. Quod
impossibile. Suppose 6 6 = k 1. In this place there is no place left for z
among the ti . So, in all cases, we obtain a contradiction. So the only possibility
is 6 1 . Thus, it follows that u[r]v.
2
We can now use our relations to define multiplication of tally numbers in the
usual way. See e.g. Section 2.2 of [Bur05]. In any concrete concatenation 2structure, we can use induction to verify the defining properties of multiplication
7

as defined. It follows that we can interpret all arithmetical truths in the set of
concretely valid consequences of TC.
Corollary 4.2 We can interpret true arithmetic in the set of all principles valid
in concrete concatenation 2-structures.

The Sum of Concatenation Structures

In this section we show that concatenation structures are closed under sums.
This result will make it possible to verify the claim that the universal principle
of Section 2 is not provable in TC. The result has some independent interest,
since it provides a good closure property of concatenation structures.
Consider two concatenation structures A0 and A1 . We write ? for concatenation in the Ai . We may assume, without loss of generality, that the domains
of A0 and A1 are disjoint. We define the sum B := A0 A1 as follows.
The domain of B consists of non-empty sequences w0 wn1 , where the
wj are alternating between elements of the domains of A0 and A1 . In
other words, if wj is in the domain of Ai , then wj+1 , if it exists, is in the
domain of A1i .
The concatenation of := w0 wn1 and := v0 vk1 is
w0 wn1 v0 wk1 , in case wn1 and v0 are in the domains of different structures Ai . The concatenation is w0 (wn1 ? v0 ) wk1 ,
in case wn1 and v0 are in in the same domain.
In case is obtained via the first case, we say that and are glued together.
If the second case obtains, we say that and are clicked together.
Theorem 5.1 The structure B = A0 A1 is a concatenation algebra.

Proof
Associativity is easy. We check the editor property TC2. Suppose 0 1 =
z0 zm1 = 0 1 . We distinguish a number of cases.
Case 1. Suppose both of the pairs 0 , 1 and 0 , 1 are glued together. Then,
for some k, n > 0, we have 0 = z0 zk1 , 1 = zk zm1 , 0 = z0 zn1 ,
and 1 = zn zm1 .
So, if k = n, we have 0 = 0 and 1 = 1 .
If k < n, we have 0 = 0 (zk zn1 ) and 1 = (zk zn1 ) 1 . The case
that n < k is similar.
Case 2. Suppose 0 , 1 is glued together and that 0 , 1 is clicked together. So,
there are k, n > 0, u0 , and u1 such that 0 = z0 zk2 u0 , 1 = u1 zk zm1 ,
u0 ? u1 = zk1 , 0 = z0 zn1 , and 1 = zn zm1 .

Suppose k n. Then, 0 = 0 (u1 zk zn1 ) and 1 = (u1 zk zn1 ) 1 .

Note that, in case k = n, the sequence zk zn1 is empty. The case that k n
is similar.
Case 3. This case, where 0 , 1 is clicked together and 0 , 1 is glued together,
is similar to case 2.
Case 4. Suppose that 0 , 1 and 0 , 1 are both clicked together. So, there
are k, n > 0, u0 , u1 , v0 , v1 such that 0 = z0 zk2 u0 , 1 = u1 zk zm1 ,
u0 ? u1 = zk1 , 0 = z0 zn2 v0 , 1 = v1 zn zm1 and v0 ? v1 = zn1 .
Suppose k = n. We have u0 ? u1 = zk1 = v0 ? v1 . So, we have either (a)
u0 = v0 and u1 = v1 , or, for some w, either (b) u0 ? w = v0 and u1 = w ? v1 ,
or (c) u0 = v0 ? w and w ? u1 = v1 . In case (b), we have: 0 w = 0 and
1 = w 1 . We leave (a) and (c) to the reader.
Suppose k < n. We have:
0 (u1 zk zn2 v0 ) = 0 and 1 = (u1 zk zn2 v0 ) 1 .
2

The case that k > n is similar.

It is easy to see that is a sum or coproduct in the sense of category theory.

The following theorem is immediate.
Theorem 5.2 If a is an atom of Ai , then a is an atom of A0 A1 .
Finally, we have the following theorem.
Theorem 5.3 Let A be any set and let B := hB, i be any concatenation structure. We assume that A and B are disjoint. Then, there is an extension of B
with at least A as atoms.

Proof
Let A be the free semi-group on A. We can take as the desired extension of
B, the structure A B.
2
Remark 5.4 The whole development extends with only minor adaptations,
when we replace axiom TC2 by:
` x y = u v ((x = u y = v) (!w (x w = u y = w v)
!w (x = u w y w = v))
Here is exclusive or.

References
[Bur05] John Burgess. Fixing Frege. Princeton Monographs in Philosophy.
Princeton University Press, Princeton, 2005.
[Grz05] Andrzej Grzegorczyk. Undecidability without arithmetization. Studia
Logica, 79:163230, 2005.
[KT06] P. Komj
ath and V. Totik. Problems and Theorems in Classical Set
Theory. Springer-Verlag, New York, 2006.

10