AN INFINITE NATURAL SUM 5 PAOLO LIPPARINI 1 0 2 Abstract. Asfarasalgebraicpropertiesareconcerned,theusual b additionontheclassofordinalnumbersisnotreallywellbehaved; e for example, it is not commutative, nor left cancellative etc. In F a few cases, the natural Hessenberg sum is a better alternative, 4 since it shares most of the usual properties of the addition on the 2 naturals. A countably infinite iteration of the natural sum has been used ] O in a recent paper by Va¨a¨n¨anen and Wang, with applications to infinitary logics. We present a detailed study of this infinitary L operation, showing that there are many similarities with the ordi- . h nary infinitary sum, and providing connections with certain kinds t a of infinite mixed sums. m [ 3 v 1. Introduction 3 2 There are different ways to extend the addition operation from the 1 set ω of natural numbers to the class of ordinals. The standard way 5 0 is to take α ` β as the ordinal which represents the order type of α 1. with a copy of β added at the top. This operationcan be introduced by 0 the customary inductive definition and satisfies only few of the familiar 5 properties shared by the addition on the naturals. 1 : On the other hand, again on the class of the ordinals, one can define v i the (Hessenberg) natural sum α#β of α and β by expressing α and β X in Cantor normal form and “summing linearly”. See below for further r a details. The resulting operation # is commutative, associative and cancellative. It can be given an inductive definition as follows. 0#0 “ 0 (1) α#β “ suptSpα#β1q,Spα1#βqu α1ăα β1ăβ where S denotes successor. 2010 Mathematics Subject Classification. Primary 03E10;Secondary 06A05. Key words and phrases. Ordinal number, sum, infinite natural sum, left-finite, piecewise convex, infinite mixed sum. Work performed under the auspices of G.N.S.A.G.A. 1 2 PAOLO LIPPARINI It is relevant that the natural sum, too, admits an order theoretical definition. If α, β and γ are ordinals, γ is said to be a mixed sum of α andβ iftherearedisjointsubsets AandB ofγ suchthatγ “ AYB and A, B have order type, respectively, α and β, under the order induced by γ. P. W. Carruth [C] showed that α #β is the largest mixed sum of α and β. He also found many applications. In V¨a¨an¨anen and Wang [VW] the authors define a countably infinite extension of # by taking supremum at the limit stage. They provide applications to infinitary logics. Subsequently, we have found appli- cations to compactness of topological spaces in the spirit of [Li], in particular, with respect to Frol´ık sums. Carruth Theorem, as it stands, cannot be generalized to such an infinite natural sum. Indeed, every countably infinite ordinal is an “infinite mixed sum” of countably many 1’s, hence in the infinite case the maximum is not necessarily attained. See Definition 4.1 and the comment after Theorem 4.2. However, we show that Carruth Theorem can indeed be generalized, provided we restrict ourselves to certain well behaved infinite mixed sums. In order to provide this generalization, we need a finer descrip- tion of the countably infinite natural sum. We show that any infinite natural sum can be computed in two steps: in the first step one takes the natural sum of some sufficiently large finite set of summands. In the second step one adds the infinite ordinary sum of the remaining summands. In other words, the infinite natural sum and the more usual infinite sum differ only for a finite “head” and they agree on the remaining “tail”. This is used in order to show that the infinite nat- ural sum of a sequence is the maximum of all possible infinite mixed sums made of elements from the sequence, provided one restricts only to mixed sums satisfying an appropriate finiteness condition. In the end, we show that the infinite natural sum can be actually computed as some finite natural sum, and can be expressed in terms of the Cantor normal forms of the summands. We show that a sequence has only a finite number of mixed sums satisfying an additional con- vexity property. This extends a classical theorem by Sierpinski [S1], asserting that one gets only a finite number of values for the sum of some fixed countable sequence of ordinals, by changing their order. 2. Natural sums We now give more details about the definitions hinted above and list some simple facts about the natural sums. Here and below sums, products and exponentiations will be always intended in the ordinal AN INFINITE NATURAL SUM 3 sense. See, e. g., the books Bachmann [B] and Sierpinski [S2] for a detailed introduction to ordinal operations. Recall that every ordinal α ą 0 can be expressed in a unique way in Cantor normal form as follows (2) α “ ωξkr `ωξk´1r `¨¨¨`ωξ1r `ωξ0r k k´1 1 0 for integers k ě 0, r ,...,r ą 0 and ordinals ξ ą ξ ą ¨¨¨ ą ξ ą k 0 k k´1 1 ξ . 0 Definition 2.1. The natural sum α#β of two ordinals α and β is the only operation satisfying α#β “ ωξkpr `s q`¨¨¨`ωξ1pr `s q`ωξ0pr `s q k k 1 1 0 0 whenever α “ ωξkr `¨¨¨`ωξ1r `ωξ0r k 1 0 β “ ωξks `¨¨¨`ωξ1s `ωξ0s k 1 0 and k,r ,...,r ,s ,...,s ă ω, ξ ą ¨¨¨ ą ξ ą ξ . k 0 k 0 k 1 0 The definition is justified by the fact that we can represent every nonzero α and β in Cantor normal form and then insert some more nullcoefficients forconvenience just inordertomaketheindicesmatch. The null coefficients do not affect the ordinals, hence the definition is well-posed. See, e. g., [B, S2] for further details. An elegant way to introduce the natural sum is obtained by express- ing equation (2) in a conventional way as α “ ωξ, where F is the ξPF finite multiset which contains each ξℓ exactly rřℓ times. This is justified by the fact that, say, ωξ ` ωξ “ ωξ2. In this way, α “ 0 is expressed by summing over the empty multiset. Of course, when expanding the above summation, one should be careful to consider the terms with larger exponents first, that is, write them on the left. If α “ ωξ ξPF and β “ ωξ, then α # β is defined as ωξ, whereřin the ξPT ξPFYT union F YřT we take into account multiplicitieřs. In this note, however, we shall follow the more conventional notations. It can beshown by induction onpmaxtα,βu,mintα,βuq, ordered lex- icographically, that Definition 2.1 is equivalent to the definition given by means of equations (1). This shall not be needed in what follows. Notice that the assumption ξ ą ¨¨¨ ą ξ ą ξ in Definition 2.1 is k 1 0 necessary, since, for example, p1 ` ωq # p1 ` ω0q is ω # 1 “ ω ` 1, while summing “linearly” we would obtain 2 ` ω “ ω. However, the assumption that ξ ą ¨¨¨ ą ξ ą ξ can be relaxed to ξ ě ¨¨¨ ě ξ ě k 1 0 k 1 ξ . 0 Proposition 2.2. Let α, β and η be ordinals. 4 PAOLO LIPPARINI (1) The operation # is commutative, associative, both left and right cancellative and strictly monotone in both arguments. (2) suptα,βu ď α`β ď α#β. (3) If α,β ă ωη, then α#β ă ωη. (4) If β ă ωη, then α#β ă α`ωη. (5) If β ă ωη, then pα#βq`ωη “ α`ωη. Proof. Everything is almost immediate from Definition 2.1. For example, to prove (4), let α “ ωξkr ` ¨¨¨` ωηr `... with, as k usual, the exponents of ω in decreasing order, and where we can allow r to be 0. Then α`ωη “ ωξkr `¨¨¨`ωηpr`1q, while, if β ă ωη, then k α#β “ ωξkr `¨¨¨`ωηr`..., since β doesnotcontribute tosummands k where the exponent of ω is ě η. Thus surely α`ωη ą α#β, with no need to compute explicitly those summands which are ă ωη. (cid:3) Parentheses are usually necessary in expressions involving both ` and #; for example, p1 # 0q ` ω “ ω ­“ ω ` 1 “ 1 # p0 ` ωq, or p1`0q#ω “ ω `1 ­“ ω “ 1`p0#ωq. Definition 2.3. Suppose that pα q is a countable sequence of ordi- i iăω nals, and set S “ α #...#α , for every n ă ω. The natural sum n 0 n´1 of pα q is i iăω # α “ supS i n iăω năω The above natural sum is denoted by # α in [VW]. iăω i ř In the above notation α #...#α we conventionally allow n “ 0, 0 n´1 and assume that 0 is the outcome of such an “empty” sum. Notice that the notation is not ambiguous, in view of Proposition 2.2(1). Proposition 2.4. Let α , β be ordinals and n,m ă ω. i i (1) α ď # α iăω i iăω i (2) řIf βi ď αi, for every i ă ω, then #iăωβi ď #iăωαi (3) If n ă m, then S ď S ; equality holds if and only if α “ ¨¨¨ “ n m n α “ 0. m´1 (4) S ď # α ; equality holds if and only if α “ 0, for every i ě n. n iăω i i (5) If π is a permutation of ω, then # α “ # α iăω i iăω πpiq (6) More generally, suppose that pF q is a partition of ω into finite h hăω subsets, say, F “ tj ,...,j u, for every h P ω. Then h 1 rphq # # # α “ # α “ pα #α #...α q i j j1 j1 jrphq iăω hăω jPFh hăω Proof. (1)-(4) are immediate from the definitions and Proposition 2.2 (1)-(2). AN INFINITE NATURAL SUM 5 Clause(5)isaremarkintheproofof[VW,Proposition4.4]. Anyway, (5) is the particular case of (6) when all the F ’s are singletons. h To prove (6), define, for h ă ω T “ # α #...# # α h j j ˆjPF0 ˙ ˆjPFh´1 ˙ Thus the right-hand of the equation in (6) is sup T . For h ă ω, hăω h let m “ max F . The maximum exists since each F is finite, 0ďℓăh ℓ ℓ and we are considering only a finite number of F ’s at a time. Then ℓ each summand in the expansion of T appears in S (taking into h m`1 account multiplicities), hence, by (4) and monotonicity of the natural sum, # α ě S ě T . Hence # α ě sup T . The reverse iăω i m`1 h iăω i hăω h (cid:3) inequality is similar and easier. The assumption that each F is finite in condition (6) above is neces- h sary. Forexample, takeα “ 1, foreveryi ă ω,thus# α “ ω. Sup- i iăω i pose that there is some infinite Fh¯. Then #jPFh¯ αj “ ω. If ωzFh¯ ­“ H, then # # α ě ω #1 ą ω. hăω jPFh j Not every`thing fro˘m Proposition 2.2 generalizes to infinite sums. For example, the operation # α , though monotone, as stated in iăω i (2) above, is not strictly monotone. E. g., # 2 “ # 1 “ ω. iăω iăω Actually, # α “ ω, for every choice of the α ’s such that α ă ω, iăω i i i for every i ă ω, and such that there are infinitely many nonzero α ’s. i Condition(5)abovecanbeinterpretedasaversionofcommutativity, and (6) as a version of the generalized commutative-associative law. However, not all forms of associativity hold. We have seen that we cannot associate infinitely many summands inside some natural sum. Similarly, we are not allowed to “associate inside out”. Indeed, ω`1 “ 1## 1 ­“ # 1 “ ω. This is a general and well-known fact. For iăω iăω infinitary operations, some very weak form of generalized associativity implies some form of absorption. Example 2.5. Suppose that ‘ is a binary operation on some set X, and a P X is such that a‘x ­“ x, for every x P X. There is no infinitary operation on X such that À x ‘ x “ x 0 i`1 i àiPω àiPω for every sequence px q of elements of X. Indeed, taking x “ a, for i iPω i every i P ω, and letting x “ x , we get a‘x “ x, a contradiction. iPω i À 6 PAOLO LIPPARINI 3. Computing the infinite natural sum Theorem 3.1. If pα q is a sequence of ordinals, then there is m ă ω i iăω such that the following hold, for every n ě m. # (3) α “ α i i nďiăω nďÿiăω # # α “ pα #...#α q` α i 0 n´1 i iăω nďiăω (4) “ pα #...#α q` α 0 n´1 i nďÿiăω Proof. Let ξ be the smallest ordinal such that the set ti P ω | α ě ωξu i is finite. Let m be the smallest index such that α ă ωξ, for every i i ě m. The definition of ξ assures the existence of such an m. If ξ “ 0, then all but finitely many α ’s are 0 and the proposition is trivial. i Suppose that ξ is a successor ordinal, say ξ “ ε` 1. By the mini- mality of ξ, the set ti P ω | α ě ωεu is infinite, hence unbounded in ω. i Then # α ě α ě ωεω “ ωε`1 “ ωξ. Suppose that ξ is nďiăω i nďiăω i limit. By the definiřtion of ξ, we have that, for every ε ă ξ, there are infinitely many i ă ω such that α ě ωε. In particular, we can choose i such an i with i ě n. Then α ě α ě ωε. Since this holds for nďiăω i i every ε ă ξ, we get # řα ě α ě sup ωε “ ωξ. The nďiăω i nďiăω i εăξ inequality # α ě α ěřωξ is proved, no matter whether nďiăω i nďiăω i ξ is successor or limit. ř On the other hand, because of the definition of m, if i ě n ě m, then α ă ωξ. By Proposition 2.2(3), α # ...# α ă ωξ, for every i n ℓ´1 ℓ ě n. Hence α ď # α “ sup pα #...#α q ď ωξ. nďiăω i nďiăω i ℓăω n ℓ´1 In conclusion,ř (5) # α “ α “ ωξ i i nďiăω nďÿiăω thus we have proved (3). Let us now prove (4). The inequality # α ě pα #...#α q` iăω i 0 n´1 # α is trivial, since every “partial sum” on the right is bounded nďiăω i by the partial sum on the left having the same length, by Proposi- tion 2.2(2). For the other direction, and recalling that S denotes ℓ α # ... # α , observe that, by associativity, for every ℓ ě n, we 0 ℓ´1 have S “ S # α # ... # α ă S ` ωξ “ S ` # α , where ℓ n n ℓ´1 n n nďiăω i the strict inequality follows from repeated applications of Proposition 2.2(4), since α ,...,α ă ωξ. The last identity is from equation n ℓ´1 (5). Since # α “ sup S and since S is increasing, we get iăω i ℓăω ℓ ℓ # α ď S `# α . iăω i n nďiăω i AN INFINITE NATURAL SUM 7 The identity # α “ pα #...#α q` α is now imme- iăω i 0 n´1 nďiăω i (cid:3) diate from (3). It can be also proved in a wayřsimilar to above. Notice that the sum ` in equation (4) cannot be replaced by a nat- ural sum #, that is, we do not have, in general, # α “ S # iăω i n # α , nor we have # α “ S # α . This is similar to nďiăω i iăω i n nďiăω i the argument in Example 2.5: just take řαi “ 1, for every i P I; then # α “ ω but S ## α “ S # α “ n#ω “ ω `n. iăω i n nďiăω i n nďiăω i However, in Corollary 5.1 we shall showřthat the computation of a countable natural sum can be actually reduced to the computation of some finite natural sum. Remark 3.2. Notice that equation (3) in Theorem 3.1, together with Proposition 2.4(5), imply that if pα q is a sequence of ordinals, m i iăω is given by Theorem 3.1, and n ě m, then α “ # α “ nďiăω i nďiăω i nďiăωαπpiq, for every permutation π of rn,řωq. Actually, equation (5) iřn the proof of Theorem 3.1 shows that it is enough to assume that π is a bijection from rn,ωq to rn1,ωq, for some n1 ě m (equation (5) does not hold if ξ “ 0, but this case is trivial). The result in the present remark can be obtained also as a conse- quence of a theorem by Sierpinski [S1], asserting that a countable sum of nondecreasing ordinals is invariant under permutations. Just notice that every sequence of ordinals is nondecreasing from some point on. On the other hand, Sierpinski’s result is immediate from equation (5). Thus parts of the present note can be seen as an extension of results from [S1] to natural sums. 4. Some kinds of mixed sums The definition of a mixed sum of two ordinals can be obviously ex- tended to deal with infinitely many ordinals. Definition 4.1. Let pα q be any sequence of ordinals (with no re- i iPI striction on the cardinality of I). An ordinal γ is a mixed sum of pα q i iPI if there are pairwise disjoint subsets pA q of γ such that A “ γ i iPI iPI i and, for every i P I, Ai has order type αi, with respect tŤo the order induced on A by γ. i Intheabovesituation, wesaythatγ isamixedsumofpα q realized i iPI by pA q , or simply that pA q is a realization of γ. Notice that α i iPI i iPI i can be recovered by A , as embedded in γ. i Notice that we could have given the above definition just under the assumption that γ andtheα ’s arelinearly orderedsets, not necessarily i well ordered. In this respect, notice that any finite mixed sum of well ordered sets is itself necessarily well ordered; however, in case I is 8 PAOLO LIPPARINI infinite, the α ’s could “mix themselves” to a non well ordered set. i For example, starting with countably many 1’s, we could obtain every countably infinite linear order as a mixed sum. Throughout this note, however, andnomatter howinteresting thegeneralcaseoflinear orders is, we shall always assume that γ is an ordinal, that is, well ordered. Theorem 4.2. (Carruth[C], Neumer [N]) For every n ă ω and ordinal numbers α ,...,α , the largest mixed sum of pα q exists and is α # 0 n i iďn 0 α #...#α . 1 n As we hinted in the introduction, and contrary to the finite case, the set of all the mixed sums of an infinite sequence of ordinals need not have a maximum. If we take α “ 1 for every i ă ω, then every infinite i countable ordinal is a mixed sum of pα q , thus the supremum of all i iPω the mixed sums of pα q is ω , which is not a mixed sum of pα q . i iPω 1 i iPω Hence there is some interest in restricting ourselves to well-behaved mixed sums Definition 4.3. We say that γ is a left-finite mixed sum of pα q if γ i iPI can be realized as a mixed sum by pA q in such a way that, for every i iPI δ ă γ, the set ti P I | A Xδ ­“ Hu is finite; in words, for every δ ă γ, i the predecessors of δ are all taken from finitely many A ’s. i Given a realization pA q of γ and i P I, we say that A is convex i iPI i (in γ) if ra,a1s “ tδ P γ | a ď δ ď a1u Ď A , whenever a ă a1 P A . γ i i We say that γ is a piecewise convex (resp., an almost piecewise con- vex) mixed sum of pα q if γ can be realized in such a way that all i iPI the A ’s (resp., all but a finite number of the A ’s) are convex in γ. For i i brevity, we shall write pw-convex in place of piecewise convex. If γ is a pw-convex mixed sum of pα q , as realized by pA q , then, i iPI i iPI for every i ­“ j P I and δ,ε P A , δ1,ε1 P A , we have that δ ă δ1 if i j and only if ε ă ε1. In this way, if each A is nonempty, the order on i γ induces an order (in fact, a well order) on I. Hence we can reindex pA q as pA q for some ordinal θ and some bijection π : θ Ñ I i iPI πpιq ιăθ in such a way that δ ă δ1, whenever δ P A , δ1 P A and ι ă ι1. πpιq πpι1q Then an easy induction shows that γ “ α . If in addition γ is ιăθ πpιq left finite, then necessarily θ ď ω. ř Conversely, if γ “ α , for some reindexing of the α ’s, then ιăθ πpιq i trivially γ is a pw-conřvex mixed sum of pαiqiPI, and if θ ď ω, then γ is also left finite. We have proved the next proposition. Proposition 4.4. Suppose that pα q is a sequence of ordinals, and i iPI α ą 0, for every i P I. Then γ is a pw-convex (pw-convex and left- i finite) mixed sum of pα q if and only if there are some ordinal θ (with i iPI θ ď ω) and a bijection π : θ Ñ I such that γ “ α . ιăθ πpιq ř AN INFINITE NATURAL SUM 9 Remark 4.5. There might be infinitely many left-finite mixed sums of the same sequence. Indeed, take α “ ω, for every i ă ω. Since ω is the i union of countably many disjoint countably infinite sets, we see that ω is a (necessarily left-finite) mixed sum of pα q . By moving just one i iăω copy of ω “to the bottom” we get that also ω`ω is a left-finite mixed sum of pα q . Iterating, for every n ă ω we get ωn as a left-finite i iăω mixed sum of pα q . Also ω2 is a left-finite mixed sum of pα q ; by i iăω i iăω Proposition 4.4 it is the only one which is left-finite and pw-convex; actually, it is the only one which is left-finite and almost pw-convex. Theorem 4.6. If pα q is a sequence of ordinals, then # α is a i iăω iăω i mixed sum of pα q . In fact, # α is the largest left-finite mixed i iăω iăω i sum of pα q , and also the largest left-finite and almost pw-convex i iăω mixed sum of pα q . i iăω Proof. By equation (4) in Theorem 3.1, we have # α “ pα #...# iăω i 0 α q` α , for some n ă ω. By Theorem 4.2, γ “ α #...# n´1 nďiăω i 1 0 αn´1 is ařmixed sum of α0,...,αn´1. By the easy part of Proposition 4.4, γ “ α is a left-finite pw-convex mixed sum of pα q . 2 nďiăω i i nďiăω Putting thře members of the realization of γ1 at the bottom, and the members of therealization ofγ at thetop, we realize # α “ γ `γ 2 iăω i 1 2 as a left-finite and almost pw-convex mixed sum of pα q . i iăω To finish the proof of the theorem it is enough to show that if γ is any left-finite mixed sum of pα q , then γ ď # α . Let γ be a i iăω iăω i left-finite mixed sum of pα q as realized by pA q . If all but a finite i iăω i iăω number of the α ’s are 0, then the result is immediate from Theorem i 4.2. Otherwise, leftfiniteness impliesthatγ isalimitordinal. Ifγ1 ă γ, then pγ1XA q witnesses that γ1 is a mixed sum of pβ q , where, for i iăω i iPI every i ă ω, β is the order type of γXA ; thus β ď α . Left finiteness i i i i implies that only a finite number of the β ’s are nonzero, thus, again by i Theorem 4.2, γ1 ď β #...#β , for certain distinct indices i ,...,i . i1 iℓ 1 ℓ Taking n “ supti ,...,i u, we get γ1 ď β #...#β ď α #...#α ď 1 ℓ i1 iℓ i1 iℓ α #...#α ă # α . Since γ is limit and γ1 ď # α , for every 0 n iăω i iăω i γ1 ă γ, we get γ ď # α . (cid:3) iăω i 5. Expressing sums in terms of the normal form The proof of Theorem 3.1 gives slightly more. Let α and ξ be or- dinals, and express α in Cantor normal form as ωηkr ` ¨¨¨ ` ωη0r . k 0 The ordinal αæξ, in words, α truncated at the ξth exponent of ω, is ωηkr `¨¨¨`ωηℓr , where ℓ is the smallest index such that ℓ ě ξ. The k ℓ above definition should be intended in the sense that αæξ “ 0 in case that α ă ωξ. 10 PAOLO LIPPARINI Corollary 5.1. Suppose that pα q is a sequence of ordinals, and let ξ i iăω be the smallest ordinal such that ti ă ω | α ě ωξu is finite. Enumerate i those α ’s such that α ě ωξ as α ,...,α , with i ă ¨¨¨ ă i (the i i i0 ih 0 h sequence might be empty). If ξ ą 0, then (6) #α “ pα #...#α q`ωξ “ αæξ #...#αæξ #ωξ and i i0 ih i0 ih iăω (7) α “ α `¨¨¨`α `ωξ “ αæξ `¨¨¨`αæξ `ωξ; i i0 ih i0 ih iÿăω moreover, for every ε ă ξ, we have #α “ #αæε and α “ αæε i i i i iăω iăω iÿăω iÿăω Proof. The ξ defined in the statement of the present corollary is the same as the ξ defined in the proof of Theorem 3.1; and the α defined ih here is the same as a in that proof (if the sequence of the α ’s is not m´1 iℓ empty). Equation (5) in the proof of Theorem 3.1 gives # α “ mďiăω i ωξ. By commutativity and associativity of #, and using Proposition 2.2(5), equation (4) in Theorem 3.1 becomes exactly the first identity inequation(6). Thesecondidentity iseasyordinalarithmetic, noticing that α`ωξ “ αæξ `ωξ and pα#βqæξ “ αæξ #βæξ, for every α and β. The proof of (7) is similar, using the fact that α “ α `¨¨¨` iăω i 0 αm´1` mďiăωαi. Thenoneshouldusetheidentitřy β`γ`ωξ “ γ`ωξ, holdingřwhenever β ă ωξ. Indeed, if γ ă ωξ, then all sides are equal to ωξ; otherwise, if γ ě ωξ, then β is absorbed by γ, since it is already absorbed by the leading term in the Cantor normal expression of γ. See [S1]. To prove the last two identities, notice that if ε ă ξ, then ξ is also the least ordinal such that ti ă ω | αæε ě ωξu is finite. Hence we i can apply (6) twice to get # αæε “ pαæεqæξ # ... # pαæεqæξ # ωξ “ iăω i i0 ih αæξ#...#αæξ#ωξ “ # α . The last identity is proved in the same i0 ih iăω i (cid:3) way, using equation (7). Notice that Corollary 5.1 furnishes a method to compute # α iăω i and α in terms of the Cantor normal forms of the α ’s, in fact, iăω i i of juřst finitely many αi’s, once ξ has been determined. One cannot expect that, for every sequence pα q of ordinals, there i iPω is some permutation of ω such that # α “ α . The coun- iăω i iăω πpiq terexample has little to do with infinity: just takře two ordinals α0 and α such that α # α ­“ α ` α and α # α ­“ α ` α , for exam- 1 0 1 0 1 0 1 1 0 ple, α “ α “ ω ` 1. Then, setting α “ 0, for i ą 1, we have 0 1 i α #α “ # α ­“ α , for every permutation π. Of course, 0 1 iăω i iăω πpiq ř