Functorial Equations for Lexicographic Products

Franz-Viktor Kuhlmann, Salma Kuhlmann and Saharon Shelah 17. 4. 2002

Abstract We generalize the main result of [KKS] concerning the convex embeddings of a chain in a lexicographic power . For a xed non-empty chain , we derive necessary and sucient conditions for the existence of non-empty solutions to each of the lexicographic functorial equations ( )0 , ( ) and ( )<0 .

1
modified:2002-07-22

Introduction

revision:2002-07-22

Let us recall the denition of lexicographic products of ordered sets. Let and , be non-empty totally ordered sets. For every , we x a distinguished element 0 . The support of a family a = ( ) is the set of all for which = 0 . We denote it by supp(a). As a set, we dene H ( , 0 ) to be the set of all families ( ) with well-ordered support (with respect to xed distinguished elements 0 ). To relax the notation, we shall write H instead of H ( , 0 ) once the distinguished elements 0 have been xed. Then the lexicographic order on H is dened as follows. Given a = ( ) and b = ( ) H , observe that supp(a) supp(b) is wellordered. Let 0 be the least of all elements supp(a) supp(b) for which = . We set a < b : 0 < 0 . Then (H , <) is a totally ordered set, the lexicographic product (or Hahn product) of the ordered sets . We shall always denote by 0 the sequence with empty support in H . Note that if all are totally ordered abelian groups, then we can take the distinguished elements 0 to be the neutral elements of the groups . Dening addition on H componentwise, we obtain a totally ordered abelian group (H , +, 0 <).
2000Mathematics Subject Classication: Primary 06A05, Secondary 03C60. First and second authors partially supported by an NSERC research grant, third author partially supported by the Edmund Landau Center for research in Mathematical Analysis, supported by the Minerva Foundation (Germany). Publication number 615

615

Functorial Equations for Lexicographic Products

Lexicographic exponentiation of chains: If = for every , we x a distinguished element in (the same distinguished element for every ), and denote it by 0 . In this case we denote H by , and call it the lexicographic power (with respect to 0 ). In other words, is the set {s; s : such that supp(s) is well-ordered in }, ordered lexicographically. This exponentiation of chains has its own arithmetic. In this paper we study some of its aspects (cf. also [K] and [HKM]). Note that if and are innite ordinals, then lexicographic exponentiation does not coincide with ordinal exponentiation (cf. [H]). Lexicographic powers appear naturally in many contexts. For example, NN is the order type of the nonnegative reals, and ZN that of the irrationals (cf. [R]). Also, 2 is (isomorphic to) the chain of all well-ordered subsets of , ordered by inclusion. The chain 2N has been studied in [H]. However, the main motivating example for us was that of generalized power series elds. If k is a real closed eld and G a totally ordered divisible abelian group, then the eld k((G)) of power series with exponents in G and coecients in k is again real closed. The unique order of k((G)) is precisely the chain k G . It was while studying such elds that our interest in the present problems arose. In [K-KS], we considered the problem of dening an exponential function on K = k((G)), that is, an isomorphism f of ordered groups f : (K, +, 0, <) (K >0 , , 1, <). We showed that the existence of f would imply that of a convex embedding (that <0 is, an embedding with convex image) of the chain G<0 into the chain k G . On the other hand, we proved: Theorem 1 Let and be non-empty totally ordered sets without greatest element, and x an element 0 . Suppose that is a conal subset of and that : is an order preserving embedding. Then the image is not convex in . Now for any ordered eld k, the chain k has no last element. Similarly, G<0 has no last element if G is nontrivial and divisible. So, using Theorem 1 one establishes that no exponentiation is possible on generalized power series elds. If we omit the conditions on and in Theorem 1, the situation changes drastically. In this paper, we study conditions on the chains and under which a convex embedding of in exists. In particular, we seek for non-empty solutions to the functorial equations: ( )0 , ( ) , and ( )<0 .

revision:2002-07-22

modified:2002-07-22

(if T is any totally ordered set and 0 T is any element, we denote by T 0 the initial segment (including 0), and by T <0 the strict initial segment (excluding 0) determined by 0 in T ). None of the three equations hold if both and have no

615

Functorial Equations for Lexicographic Products

last element (for the rst, this is trivial, and for the second and third it follows from Theorem 1). In Section 2 we start by proving a strong generalization of Theorem 1 (cf. Theorem 2). In Section 3, for each of the three functorial equations, we give simple characterizations of those chains for which non-empty solutions exist. In Section 4 we study simultaneous solutions to all three equations.

Nonexistence of convex embeddings

In this section, we shall prove that Theorem 1 remains true in the case where is arbitrary, but 0 is not the last element of . This will follow from the following more general result: Theorem 2 Let and , , be non-empty totally ordered sets. For every , x an element 0 which is not the last element in . Suppose that has no last element and that is a conal subset of . Then there is no convex embedding : H .

Proof: For every , we choose an element 1 such that 1 > 0 . Take d = (d ) . If S is a well-ordered subset of such that d = 0 for all S, then we set
modified:2002-07-22

d S := (d )

with d =

d for S / 1 for S .

Observe that the support of d S is contained in supp(d) S and thus, it is again well-ordered. Note also that S S dS < dS . = (1)

615

revision:2002-07-22

Indeed, let 0 be the least element in S \ S . Then (d S )0 = 0 < 1 = (d S)0 . On the other hand, if and < 0 then (d S ) = (d S) : if S then S (by minimality of 0 ) and (d S ) = 1 = (d S) ; if S then S / / and (d S ) = d = (d S) . Now suppose that : H is an order preserving embedding such that the image is convex in H . We wish to deduce a contradiction. The idea of the proof is the following. Let ON denote the class of ordinal numbers. We shall dene an innite ONN matrix with coecients in , such that each column (n) ( )ON is a strictly increasing sequence in . Since is a set, every column of this matrix will provide a contradiction at the end of the construction (cf. gure).

Functorial Equations for Lexicographic Products

0 . . . . . . . . . (1) . . . . . .

(1)

0 . . . . . . (n) . . .

(n)

(n+1)

. . . . . .

... ... ... ...

(n+1) . . .

(1) (n) . . . = ? . . .

... ...

...

...

...

To get started, we have to dene the rst row of the matrix. We construct sequences (n) (n) , n N {0}, and 0 , n N, in . We take an arbitrary (0) . Having (n+1) constructed (n) , we choose 0 and (n+1) as follows. Since has no last element, we can choose (n) , (n) such that (n) < (n) < (n) . Hence, (n) < (n) < (n) . Let (n) be the least element in supp (n) supp (n) for which ( (n) )(n) < ((n) )(n) , and (n) the least element in supp (n) supp (n) for which
modified:2002-07-22

(2)

((n) ) (n) < ( (n) ) (n) . Since is conal in , we can choose (n+1) such that (n+1) max{ (n) , (n) } . Further, we set d(n+1) := (d(n+1) ) with d(n+1) = ((n) ) for (n+1) 0 for > (n+1) .

(3)

revision:2002-07-22

Then by (2) and (3), (n) < d(n+1) < (n) . Thus, d(n+1) by convexity, and we can set 0
(n+1)

:= 1 d(n+1) .
(n+1)

Now for every n N we have that (n) < 0 , hence every well-ordered set (n+1) (n) S with smallest element 0 has the property that (0 ) = d(n) = 0 for all S; and moreover, 0
(n)

< 0 S < (n1) .

(n)

615

Functorial Equations for Lexicographic Products

(n)

Thus, 0 S by convexity. Suppose now that for some ordinal number (n) 1 we have chosen elements , < , n N, such that for every xed (n) n, the sequence ( )< is strictly increasing. Then we set
(n) (n+1) := 1 (0 { | < }) (n+1) (n+1) for every n N. If < , then { | < } { | < } and thus, = (n) (n) < by (1). So for every ordinal number , the sequences ( )< can be extended. We obtain strictly increasing sequences of arbitrary length, contradicting the fact that their length is bounded by the cardinality of . 2 (n) (n)

Corollary 3 Assume that 0 is not the last element of . If there is an embedding of in with convex image, then has a last element.

Solutions to the Functorial equations

We start with a few easy remarks and lemmas. Throughout, x a chain with distinguished element 0 . Remark 4 1) If 0 is last in (respectively, least), then 0 is last in (respectively, least), for any non-empty chain . 2) Let I be any chain, and C a non-empty convex subset of I. Let c C. Then the initial segment determined by c in C is a nal segment of the initial segment determined by c in I. Remark 5 If <0 has no last element, then also ( )<0 has no last element, for any chain : If not, let s be last in ( )<0 and set = min supp(s). Then s() = < 0 . Take < < 0 . Consider s dened by s () = and s ( ) = 0 if = . Then s ( )<0 , but s > s, contradiction. Lemma 6 Let and be chains, and suppose that : is a chain embedding. Then lifts to a chain embedding

modified:2002-07-22

revision:2002-07-22

: . Proof: For s and x , set (s)(x) = 0 if x Im / 1 s( (x)) if x Im .

(here, Im denotes the image of ). Now, it is straightforward to check the assertion of the lemma. 2 In view of this lemma, if F is a subchain of a chain , then there is a natural identication of F as a subchain of .

615

Functorial Equations for Lexicographic Products

Lemma 7 Let be a chain and F a non-empty nal segment of . Then F is convex in (and 0 F ). Proof: Let si F , and set i = min supp(si ) F , for i = 1, 2. Let s be such that s1 < s < s2 . If s = 0, then s F . So assume s = 0 and set = min supp(s). Suppose that F . If s > 0, then s() > 0 . On the other / hand, < 2 (otherwise, F ). Thus, s > s2 , a contradiction. Similarly, we argue that if s < 0, then s < s1 , a contradiction. Hence, min supp(s). Since F is a nal segment of , this implies that s F , which proves our assertion. 2

Corollary 8 Assume that has a last element. Then embeds convexly in , such that 0 is mapped to 0 . If moreover 0 is last in , then F embeds as a nal segment in , for any non-empty nal segment F of . Consequently, if has a last element, and 0 is last in , then embeds as a nal segment in . Proof: The rst assertion follows from Lemma 7, applied to the nal segment consisting of the single last element of . For the second assertion use Remark 4, parts 1) and 2). 2 We now give a complete solution to the rst functorial equation, and a sucient condition for the existence of solutions to the third functorial equation: Theorem 9 There is always a non-empty solution for the functorial equation ( )0 . If <0 has a last element, then there is also a non-empty solution for ( )<0 . Proof: Set 0 := 0 . Since 0 has a last element, embeds convexly in 0 . Consequently, 0 embeds as a nal segment in 1 := (0 )0 . By induction on n N we dene n := (n1 )0 , and obtain an embedding of n1 as a nal segment in n . We set := nN n . Since every n is a nal segment of , every well-ordered subset S of is already contained in some n (just take n such that the rst element of S lies in n ). Hence, an element of ( )0 with support S is actually an element of n+1 = (n )0 , for some n. This fact gives rise to an order isomorphism of ( )0 onto . To prove the second assertion, we set 0 := <0 . Since 0 has a last element by assumption, embeds convexly in 0 , and the same arguments as above work 2 if we dene n := (n1 )<0 . Remark 10 Note that 0 has a last element and embeds as a nal segment in the constructed solution (in both cases considered in the proof). Thus, has a last element, and there is no contradiction to Theorem 2.

615

revision:2002-07-22

modified:2002-07-22

Functorial Equations for Lexicographic Products

Note that if 0 is least in , then the rst equation has the trivial solution = {0 }. We next turn to the second functorial equation. Remark 11 Suppose that 0 is last in . Then the solution to the rst equation given in Theorem 9 also solves the second equation. Indeed, in this case, 0 is last in , so ( )0 = . We also have the converse: Corollary 12 Assume is a chain such that the functorial equation has a non-empty solution . Then 0 is last in . Thus, the functorial equation has a non-empty solution if and only if 0 is last in . Proof: Assume 0 is not last, and choose some element 1 > 0 . This provides us with characteristic functions. If S is well-ordered, then let S denote the characteristic function on S dened by: S () = 1 if S 0 if S . /

modified:2002-07-22

Note that these characteristic functions reect inclusion: if S is a proper wellordered subset of S , then S < S . Now assume for a contradiction that i : , and let = card(). We shall construct a strictly increasing sequence { ; < + } in . Set 0 = i1 (0), and assume by induction that { ; < } is dened, and strictly increasing in . Then dene = i1 ({ ;<} ). It follows that { ;<} < { ;<} , whenever < . Since i1 is order preserving, it follows that < as required. 2 We now turn to the third functorial equation. We deduce a simple criterion for the existence of solutions: Corollary 13 Assume that 0 is not the last element of . Then the functorial equation ( )<0 has a non-empty solution if and only if <0 has a last element. Proof: The if direction is just the second assertion of Theorem 9. So assume now that is a non-empty solution. Assume for a contradiction that <0 has no last element. Then by Remark 5 ( )<0 has no last element as well. Thus, the same holds for the solution . This contradicts Theorem 2. 2

615

revision:2002-07-22

Functorial Equations for Lexicographic Products

Simultaneous Solutions

Recall that by Remark 11, the chain given in Theorem 9 solves the rst and the second functorial equations, if 0 is last in . By we denote the ordinal with the reverse ordering. Theorem 14 Assume that 0 is last in and that embeds as a nal segment in . Then the solution given in Theorem 9 to the rst and second functorial equations solves ( )<0 as well. Proof: Recall that embeds as a nal segment in the given solution . Thus, embeds as a nal segment in as well. In particular, has a last element 0. Since = ( )<0 {0} and , we nd that ( )<0 \ {0}. But \ {0}, since is a nal segment of . 2 We now turn to the question of whether the sucient conditions given in this last theorem is also necessary. We need to introduce a denition: Say that a solution (to any of the three equations) is special if embeds as a nal segment in . Note that special solutions are necessarily non-empty. Proposition 15 Every non-empty solution to is special.

modified:2002-07-22

Proof: Necessarily, 0 is last in (by Corollary 12). Thus, has a last element, so by Corollary 8, embeds as a nal segment in , and thus in . 2

Corollary 16 Assume that is innite and is any non-empty chain which solves simultaneously ( )<0 . Then 0 is last in and embeds as a nal segment in . Proof: Since , 0 is last in (Corollary 12). Therefore, 0 is last in by Remark 4, and so also has a last element 0. The assumptions imply that \ {0} . This is equivalent to the assertion that embeds as a nal segment in . Now note that is a special solution by Proposition 15, i.e., embeds as a nal segment of . Since is innite this implies that embeds as a nal segment in , as required. 2

revision:2002-07-22

Corollary 17 Assume that is innite. Then the following are equivalent: (a) 0 is last in and embeds as a nal segment in . (b) There exists a (special) simultaneous solution to all three equations. (c) There exists a (special) simultaneous solution to the second and third equations.

615

Functorial Equations for Lexicographic Products

Proof: (a) implies (b) by Theorem 14. (b) implies (c) trivially. Finally, (c) implies (a) by Corollary 16. 2 We conclude with the following question: Are special solutions unique up to isomorphism? We can give a partial answer to this last question: Proposition 18 Assume that 0 is last in . Let = n be the solution to the second equation given in Theorem 9. Then embeds as a nal segment in any other solution. Proof: Let be another solution. Then it is a special solution, by Proposition 15. So = 0 embeds as a nal segment in . Since 0 is last in , 1 = 0 embeds as a nal segment in . By induction, n is a nal segment of for every n N. Thus. embeds as a nal segment in as well. 2 References
[H] [HKM] [K] [KKS]
modified:2002-07-22

[R]

Hausdor, F. : Grundz ge der Mengenlehre, Verlag von Veit, Leipzig (1914) u Holland, W. C. Kuhlmann, S. McCleary, S. : The Arithmetic of Lexicographic Exponentiation, preprint Kuhlmann, S. : Isomorphisms of Lexicographic Powers of the Reals, Proc. Amer. Math. Soc. 123 (1995), 2657-2662 Kuhlmann, F.-V. Kuhlmann, S. Shelah, S. : Exponentiation in power series elds, Proc. Amer. Math. Soc. 125 (1997), 3177-3183 Rosenstein, J. G. : Linear orderings, Academic Press, New York - London (1982)

Department of Mathematics and Statistics University of Saskatchewan 106 Wiggins Road Saskatoon, SK S7N 5E6

Department of Mathematics The Hebrew University of Jerusalem Jerusalem, Israel

615

revision:2002-07-22