II Chapter

anclCombinatorics Fragments

Introdu,c'tion.In bhe present chapter we shall elaborate proofs of various combilatorial principLes in suitable fragments of arithmetic. In general, infinite principles deal .vith graphs, functions etc. on in.fi.nite sets, fi.nite principles relate similarly to finite sets. We prove both sonre infinite and some finite principles; furthermore, we show some infinite principles to be eqr-rivalentto certain collection principles and some finite principles bo be ecluivalent to certain consistency sbatemenl,s.Sections I and 2 deal with sl,rengthenings of the infinite and fi.nite Ramsey theorem (they will be formulated at the beginning of Sect. 1), in particular with various forms ancl instances of Paris-Harrington principle. This principle is very famous since it has been the fi.rsbexample of an arithmetical statement that has a clear combinatorial meaning, is true (in l/) and is unprovable in .P,4. Instances of Paris-Harrington principle will form a hierarchy of formulas, n-th of them will be proved in IEn (n > 1). As said above, in this chapter we deal with concrete proofs, not wibh unprovability; but unprovability resulLs immediately follow from the resurltsof this chapter ttsing Godel's incompleteness theorerns (elaborated in Chap. III). We shall mention this on corresponding places in this chapter: (n+l)-th instance is unprovable in ,[8". In Sect. 3 we shall deal with orclinals in {ragments, introcluce the nobion of cu-largesets (a an ordinal) and investigate another hierarchy of combinatorial stabements, related to the first one. Results of this section will be used in Clrap. IV for a characterization of functions provably recursive in IDn

(" 2 1).

Fragments cLnd Tlt,eorems l. Ramse?J)s

(a) Stateftrent of Itesnlts
1.-l-First i,veshall recail Ramsey's theorerrs in an inlorrnal formulation. If subsels X is a set of natural numbers,then [X]t' is the sebof all T.r-elemenL

I12

and Combinatorics II. Fla.srnent,s

of elemenl,s of X). al1 increasing u-element seclu.ences of .z{ (or, eqLrivalelil;ly, thaL ir is a mai:ping .Zr: [,'Y]u -t o (where c, is a natural number) mea,n.s '',vh.ose 'wlrosedom.ain is in {0, 1,..., ct - I}. It is range is iirclttclecl [X]" ancl nu,mber o1'colou'rs. Y Q ,V is to cali T.r{,he arity of. F ancl cl t,.[re cr-rstoma.ry to []']t' is cons'cant.Tire tn'firtite J].unse,y fot; .lr if. F rcstr:icbccl Ito'irtogc'n,eo'ut says thal, for each natural u and a ancl each F ' |X]" -> c'where X is tlt,eorem Y C; X i,vhichis homogeneoLls (i,e. infiniLe) there is an un.bouLncled. trnlrouLncled this by fb.r:.li'. It is customary to d.enobe

-' (co)"j,)or (Y'u,c)(to

(V"Xr, -, (,,t)\,,,);

of X ancl the seconcl r,iirbou.udectrr.r syml:olizing i;he r-in.bor-rncleclness the fi.rsl, ness of the hornogeneottsseL.

syrnbollr,yl -, (r/)) means that if 1.2 lbr natural nttmbers,tr)y)rL)z,c1,I)te numbers,tiren eachf ,lxl" -> z'has: X is the closedinterval lr,y]oI natr-rral q. The fi"ni'te Rnmsey'tlteorem'is sebof carclinaliLy a homogeneous

- kt)')) (Y*,q,u,, z)(1y)(l*,'a)

(cl)y rneans that each / as above has a hornogeneou.s The symbol [r, a] ] sef Y of cardinalit,y q wlrich is relutiuel'ylarge, i . e . m i n ( Y ) ! c a r c l ( Y ) . P a r i s io H arrington prin ci,ytle

( Y * , q , u ,z ) ( 2 y ) ( [ r , y ]

;lr

GDT)

Eviclently, Lhis is a strengthening of the finite Rarnsey theorem.l ParisFlarringl,on principle fbllo.,vs from the infinite Ramsey theorern using l{onig's lemma. (!Ve shal1 discuss the proof below.) 1.3 Clearly, Paris-Harrington principle as well as the finite Ramsey theorern is expressible by a lorrntLla of first order arithrnetic; let us wrile .PII(u,z) the ariby and z Ihe number of for (Vr,qxly)([r,U] : GDi) (recall that u is T'hus Paris-Harrington principle is (Vu, z)PI{(u,,2). Our: secluence color-Lrs). for n : of lbrmr-rlasthat are "h.arclerzrnd harcler to prove" is (Yz)PII(n,r) 'I^2"... Ramsey Lheorem in 1.4 On th.e other hand, 'we cannot Iorm.alize the infi"r-r.ibe quantify ovel arbitrary sgi,'; first-orcler z-lrithmeticas it stands since we cannot of natural numbels. llub lve carn cltrantify over se|:sof restlictecl comple,ril,y, e.g. over d,.,, sets (in J'X1) or lowzJ* (in EIE^).Thus we miry e,{press se'reral partil,ion relations saying th.at lor each l-1-defin.ableancl unboiLndecl is a f2-definzrltle unboLtndecl -{ and for: eacir /1-cle.fi.nableF' , l"]t' -r z {,heL:e

1. Ramsey's Theolerns ancl l\'agment,s

11J

lromogeneotrssel;(where f'1, .fz ale Ant, lo,tu,A^, or so). Denote such a fbrinulzr. by u : - - +( u ) ' ) ( h , 1 l z ) . .Recrtrsiue anolysis of Rarnsey's Lheorern consisl,s in. estahrlishing l,r'r-rl,h of assertions of this bype. (Pioneei:ing worlc 'was done by Joc.lcusch).OuLr ainr is sLill more ambiLioris: we want to esbablishprovability o.[ su.c]rasser:tions in suiLable fra.gm.e.tr.bs of ar:iLh.melic. trVear:e nor,vr.'eaclyLo presenl, Lhe m.ain resuits of this section.. proves 1"5 if'lreorern. .Foi:mrn ].'I, B Ern-1-r1 c , .-' r ( e u ) ! , r ( l o uA m + r , l o w A * a 1 ) , unbor-rn.ileclancl -F is a Iow /)n a1 i.e.: For each z, if ,V is Iow A*41 a:nc1. then. -F has a l.ow-l-\,rr1-1 mapping of [X]" into z (i.e. into {0,...,2 -I}) homogeneous unbounded set. proof (using lo'urbasis theorein) see belo',v.Note that; th.e asserbionis For t'r, 'meaningfa,l as a single formula using l.he cocling \n B E*11 and is e,xpressible of. Io'wAn"r.)l sets in B\m-ft (see Chap. I, Sect. 2). Due to some obvious I, BE*-l-n proves in.clusions,r,vehave e.g. the follo,,ving cor:ollary: for rn,n) u -> (co)\,u(Arn,A*+n) .

This assertion is wealcer bu.l,is meaningfr-rlalrea.clyin tD1 and is eqr-iivalent over -[X1 to B Em+p. 1.6 Theorern. Fbr n'r,n )-J., IDl B E^+n proves the tollowing:

: u) '-', (co)!,"(A. , A,n+I) .

(Here BEn +n is formulated as a single formula). By Tlreorerrr 1.5, BEm-fn proves an infinite Rarnsey type theol'em on ma.ppings of arity n ancl complexiLy low Am-rI. We shall see that this theory also proves a Ramsey type theorem on mappings of ariby (n + 1) ancl cornplexity on.ly a finite relati.zely large hornogeneous set bul, it gr-rarantees lou Av.,,..1--L; mulsbl:e stzr.nclard. and the number of color-rrs 1 . 7 T h e o l e r n . L e b f f i , D ) 1 . ( 1 ) I E n + n - 1 - p r o y e s t h e f o l l o r , v i n gI:f X i s L L , , , ancl if -F : [X]" -t z is LLrn, Lhen for each q l,hei'e is ar ancl unlroi-Lndecl relativel;, large homogeneous (fi.nite) set having al least g elemenl,s.This ctrn be exoresseclbv

(VzXVr/)(c..,
_J

GD7(L1",,))

II4

II. I\'agrnent's ancl Cornbinaborrcs

(2) Fbr each /c, I E*1-n*t

Proves (Vq)(r.r-,' (.1)[+1(L Ln) .

cortrplcxil,;' rr.eccl prc,i.es Lhat infinibe homoge.ucoussel,sof r.;oure II,,r-l-r, allst-r not exist. As an example we prove bhe fol.to,,ving. 1.8 Tlreorern. For m 2I, IEm-t-I pro'ves -cd --+ 1ro)?;.(Ar., Ern+r) (Thr_tsLhere is a arn F : lxlz -_r 2, x A"*, trnbounclecl, witlr no ),'r,r-;-1 homogeneous unbounded set). Note that a stronger result will be obtainecl in'I'28. Let us summarize the above resu-ltsfor arity 2,,2 colours and A1 mappings. We hav.e the following:

rrylF (vq),a e)7@) ,

'-+@)7@t, Ez) , If)2t- '-tLt) B Es l- w -, @)?r(A1,IowAs) . Now let us considerfinite Rarnseytype theorems. 1.9 Theorern.For n ) t,

(")
(b)

ID,I- (Yr,z,./)(:yxlt, d i @T)

z ) PH ( n ,r ) ) , ( i . e .I E " F ( V
for each k ,

l (rl)"+t) I E n l ( V x , q ) ( 1 y X l " , y7 k, IEnts PII(vTI,E). ( i . e .f o r e a c h

.Remark. Resttlbs of Chaps. III and IV enable us to add the fbllowing: First, all formulas in cluestion are -/72. Second,in Sect. 2 of the preseul,chapter we prove a bheorem implying thar.b .[)'1 proves (Vz)Pf] (r, "lrT, z) -, Con(I Ei) (consistency), thus we may appiy Godel's seconcl incompleLenesstl:eorem (provecl in I'1I.2.21) io cleituce l;,[re unprovability of (Vz)PII (n +1, z) in I En.T'hus .,vehave a strictly increasing hierarchy o'i II2 formulas. 1,10 Theorem. -I)-1 proves(Yx,z,q,u)(1y)(lr,Ul Q)T).

we shall elaborate prools' our lis1,of resLrlts.In what fbllo'',vs This comple{,es

1. Ramsey's'Iheolems

ancl fragtnenl,s

115

i..7,L.9) (b) Proofs (of 1".5,

-> a be /\'1, X z11 r-Lnbotrnclecl. An 1.1.1Ilefinitions (/)ir). Let -F ,lxl2 j < if fbr eacll i < of X is rtrehornogeneous increasing sequence s of elernenl,s cach i.c. lbl oniy on thc first argrtmcnt, /h(s) thc virluc F'((r);,(r)i) depencls i <. j < k < Ih(s) we haveF'((u);,(t)i) : F((u);'(u)r').If i < llt'(s)-I j < then the colour of (s); (in s) is bhe common value f((s);(s)1) for i < is prehomogeneolts,then the /h(s). If /h(s) ) 0, u > max(s) and s ^ (T.r) rnaximal elem.ent of s has a colour in s ,- (u), namely F(max(s),u). Let es x t e n s i o no f s i f s ^ ( t r ) i s I h ( s ) > 1 1s ^ ( u ) i s a . m i n i m a l p r e h , o r n o g e n e o u ^ (u) is prehomogeneolls ancl there is no u between max(s) ancltl such that s ^ (r). pr"ho*ogeneous ancl ma;c(s) has the same colour in s '- (u) as in s " (or h.:m.p'h.) if s is prehomogeneous, is hereditarily mini,maLprehomogenorLs - 1, s | (i + t ) ,s : 0 or (s)o : minX and for: each z bet',veen 1 and lh(s) is a minimal prehomogeneolls extension of s i i is the initial segment of s of l e n g b hf ) . 1.i.2 Definition. (IEt). A tree T is narrowly branchin.gif.there is a number c such that each s ?' has at most c immediate successors. 1 . 1 3 L e r n m a " ( . t E )L e t F : l X l 2 - > a b e a s i n l . . 1 1 a n d l e t T b e t h e s e t o f a l l '7' ts A1, and is an unbor,rndednarrowly bran"ching h.rn.p.h. sequences.Then tree. Proof. Evidently, 7' is a At tree. It is narrowly branching since each s 7 Tb prove thalb 7 is unbounded one easily has at most o immediate successors. - r. Indeed' slrows that lbr each r X there is an s e T such. that max(s) and (ss) is h.m.p.h. then (ro,r) istrivially prehomogeneous let s6:minX; Assume we have a h.m.p.h. sequences such that s -- (t) is prehomogeneous. If s ^ (r) is not h.m.p.h.,then there is a y ( r such that s,- (V) is h.m.p'h' and the colour of max(s) in s ,-. (y) it the same ersthe colour of ma,r(s) in n , ^ (r);then u ^ (E) ,--.(z) is prehomogeneous. 1.1-4 f,ernrna (1) (/X1). If ? is a A1 rarrowly branching unbounded tree, then f o r e a c h l e v e l r , b h es e t { s 7 ' l l h ( s ) : c } i s b o t t n d e d . 'I:f. (2) (Bt2) . T is as above then for each level z there is an s 7' sr-tch -l'hus bhat /h(s) : z and the z11 seb ?-' : {, e :f I t f ,} is unbouncled. brancb.ing bree. {t I s ^ t e f} is an r-rnboundedAt nzrr-r:owly r/(s) - s/of biresame lerrgth Proo!.(1) Associate with each s e T asecluence clefined as foliows,0' \r 0 and if s ^ (") e ?'then (, ^ (u))/ :5/ ^ (i) such of s in ?. Note thaL I/ is X1 that s ^ (1) is l,he i-bh immediate srlccssor ancl one-one on ?. Nor,v the set of all seclltencesf such that [/h(t) : r and

116

II. ftasments ancl Cornbinatorics

by some b;by SE1 each.member of I is ( a] is bor-rnded

: c ) - ' ( 1 , < d ) ( ue T k / { ( " ) : l ) l ' ( l c l ) ( V f< b ) [ ( ] se i I ' ) ( I I ( s )

.Lvidenl,ly, d is the clesireclbottn.d. ' [ ' 'o [ l e n g t h r . A s s u m e b h a ba,LI'.1.',, ( 2 ) t , l i v e n r , l e L c b e a b o u n c lt b r :a , l l. se ( s . T , l h ( s ) : z ) a r e b o u n d e d 'T h e n

(V, < cxlq)[s eTklh(s) : r + (Vt e Ry B L'2 rve obtain ( l q ) ( V s< thus ")[, e I ' k l , h ( s )- c

f s " - ' t :< c 1 \ ' ")(t

+ (Vl e ?')(t] s -> t <'/)l ;

( :q) ( Vt e r ) ( t < Q), a contradiclion.

TI

l e i . F : l , Y ] u - t & , u 2 3 , - , Yd 1 a n d u n b o u n d e d , (s /tr).Now L.L5 Oe"finit,ion if fol s of elemenLsof 'Y is preh,oftLogeneotls increasing secluence F'At.An eaclr i1 j7((r);r,...,(");*-',(o)i"l-r), i'"' tire value does not dependo ' n the last ar:i arrd s ,- (g) is prehornogereous, Lhen the coLo'Ltr If /h(s j )_ u g"-""t. of max(s) in s ^ (q) ir the finiLe mapping associating bo each increasing i1 < iz 1 seclr,rence q ) max(s), . The definition of a minimar,lprehomogeneous exLensiotl is as abovel s is h.m.p.lr. if. eitlt,erlh(s) 1 n -.1- and s consists of the first /h(s) elements o.f x or lh(s) ) u, s begins by the fi.rst (u - 1) elements of x and for ea,chi extension u-1 ancl lh(s)- 1, s | (i+t) is aminirrral prehomogeneoLrs bebween ofs li.

1.16 Lemrna (.IX1). LeL .F : lXl" --+a be as in 1.15 and Lel T be the set ot 'I'hen T is Al and is an unbouncled finitely branching all h.rn.p.h. sequences. tree.
ProoJ. Generalize the proof of 1.13 (bub clrop nar-row branching; finite brancl-,l-l ing is evidenb).

1 .1 ?L e mma (B E *1 1 ). Let u) 2 ancllet .F: [X]" - > a be low Am +1, X l ow 'Ihen the set of all h.m.p.h. sequeltces is a low Am1-I Am.-t7anclunbor-rndecl. tree. :finibelybranching unboundecl Hint. Relaiivize ihe above.
l . L 8 L e m m a . B E 2 | w - + ( c u ) f , , , ( l o t u L \ 2 , L o w A 2 )i,. e . i f X i s l o w A 2 U t n bouncled ancl f' : [.X] -r a then there is arr i < ct such thaL -l'-'-l(i) is utnbounclecl.

1. Ramsey's Theolems ancl Flargrnenls

It7

P,rooJ.A.ssume Lhe contrary, i.e. (V, < a)(:y)[(Vu)(F(u) - i -> rt" < A)]. t h . ef o r m u l a i n [ . . . ] i s A 2 a n c l b y B E 2 w e g e t ( - i y ) ( V t < Since.F is low.1-\2, -i-,LL a)[(t/u)r(u) tl contrzr,cliction. l . L 9 P r o o f o f T h e o r e m I . 5 . I l y i n c l u c t i o no n r z . F o r n : 7 . s e e 1 . 1 8 ( a n d L h e - , a b e a s a s s u m e c tB . y 1.17 take tlr.e o b v i o t L sr e l a t i v i z a t i o n ) . L e t , F , [ X ] " it is Iov'r Anr-,.1 and, by .t.3.10 (5) [r.asa tree T of a| h..m.p.h. seclrl.ences; Iow 1.\,n-s2unbounded branch. The branch clefinesa low Arn-F2 unbottnclecl seL -Z ancl J7 clefineson [f1n-t a function G : lYl"-r -, ct; G plehornogeneoLrtj is lo'w /Jm-t-2.By the induction hypoLhesis, G has an unbotrnded homogeneous set,Z wlrich is 1ow Arn*'tl-n-1, i.e. Iow A,n4n. .Z is homogeneous also at {; L.20 Proof oJ Theorem 7.'f. (1) The initial case for n : L is

tE ml

( Vq) ( Vz) ( a.' Q ) t ( I ' t ^ ) ) ;

clue bo relabivizertion, LLn may be replaced by A*. We also take nz : 1, i.e. we prove the following: if X e A1 is unboutndecland Iv e At maps X into (< z) then. there is a relatively large seb a of cardinality at least q such tlrab -F is constant on a. But this is easy: By 5X1 find a b e X such bhat for : i ) i m p l i e s ( l t < b ) ( r e X S zF ( r ) - i ) . e a c l rc o l o u r i < z , ( 1 r e X ) ( f ( r ) Then let Xs be the set of first (b -F 1) 'kz elements of X and let j be a colour such that o: {.* Xo lr has the colour:j} has more than b elements. Since min(c) 4 b, a, is the desired relativelv large homogeneous set. to th.e inducLion step in 1.19 but l'he induction step is now analogoLrs instead of low Arn+L and low Am-F2 one worlcs with LL* and LLm+t. (2) First let us prove IE1 F ru A G)?r(LLt). It is enough to replace ,ttr1 \ry A1 and then relativize. Assunre k standarcl to be givenl v/e proceed in IE1. Let X A1 be unbounded, F : lxl2 - k, F At, ,l arbitrary. Let ? be the bree of all h.m.p.h. secluences; it is unbouncled, k'-branching ancl 41. (See 1.13.) By 1.14 (1), fol each r there is an upper bound b fbr all elernents s 7 of length /h(s) < z. By LU1, there is a least such b; call it f/(r) and observe thaL /:[

is Xo(trr).

'Iai.e oul g anclpr-rtG(rc): H(") + (1.f 1) anctv - (!h*r(q). (Here we Llse tlre fact thal k is standard;we ma'yiberateG (k f 1)-manytimes.) Clearly, lengthl fix an s T such that lh(s) : r. For ? lras elementsof arbitrz-rry j : 0 , ...,1 , * 1 , l e t s; b e the initial segment of s havingthe tength Ci( q ) . : : tl(r';) * (/c-F 1)- Assign colours G(";) Tlrtts if r; : lh(s;) we get ri+7 to elements of s ir: the usal r,vayancl lel col(s;) be the seh of colours of anclIeI Z be the s eb e i e m e n ts o f s;. P i ck a n i su chthaL col( s;) : col( s;- ' - 1) for o[ elementsof si-F1.The cardinalibyof X is f/(r;) *, (h *.1), l,heretbre

J.. R.amsey's TLreolems a,nd F\'agmenl;s

119

NowasSL1me(*)formanc[pr:oveibfo'r:rn-|-.|..Th'LrsaSS.t|.IoeU_> * (r,r)l ,u(L\rn,1\rr), therefore by the in.clr-rcThen c,., (r)t,.,(A*1rtAm{r). tion hypol,h.esiswe have BEm-r1 V/e wanL Lo pro.le .R.A-,.Let 0 be I1,, ancl anci Lhe assume (C'r)(1y < o)0. Define 0/ a,sabove; by -BEml-7 0t i.s 1),rr-1-1 ll res[ is as above. This proves Lhe lemrna. now we have investigaLeclth.e co:m.binatorial relaLion (., -+ (ru)2( A ; , A j ) ( d e f i n e d i n 1 . 4 , c f . 1 . 5 ) . D e n o t e b h i sr e l a t i o n b r i e f l y b y A r r o u ( . r , i , j ) . l,eL us r:rolvconsider an appalenbly wealcerparti{,ion rela,Lion,denoLeclby e ^ - r 0( c o ) \ , " ( A ; , 1 ) ) o r b r i e f l y A r r o u j ( n , i , i ) (thus --, replaced by ---n0)*h.ose definition results from the definition of Arrow(n,i, j) by assuming X l,o be jr:st the whole universe V, thus: For each y'1 fl-rnction F , lvl -> z ('where z is any number) there is a Ai unbounded homogeneous seb. Firstconsiclerbhecasen:1.Evidenb1y,in.[D,,(m> Arrow(.I,m.,m) : Arrowl('I,*,,rn), since each A* unbounded set is isomorphic r,vith 7 by a A^ mapping (cf. I.2.65). W" prove even mol'e: o,f.Arrow(l,rn,rn) 1.25 Lenrma" IEl p.rovesthe eclr-rivalence Arrowo(I,m,m). and .1.24"'I'ill

on rn, sho'w lfh * Arrow0(L,,m,m) ts IUm. Assume Proof. l3y incigrction. this for rn ancl work in (/tr1 + Arrow\(I,*-F'1,'m + 1)). By the induction hypotlresis'we have I E*, thus ,4rrow(I,m, rn) and by I.22, B Em-rr. Given a non-empby Em-F7 set X such that r X : (1y)0(r, y) fbr some f/--formula 0, and a:n ct X, defi,nef'(s) '= rnin{r < a | (1y < s)0(r,,y)} if this set is non emptY,: d -F 1 otherwise. B y B D r n - r 7 , i : , F ' ( s ) i s A m * ' r ( a n d t o L a l ) a n c lb y A r r o u o ( 1 , - { I , r n { I ) , that I'-r(i) is unbor-rndecl. This i is rninX. This bhereis an i < a *2 sr-Lch tr proves LDn-t-fI ancl hence lDrnl-t. 1.26 tr,enirna" !-or n 2 2, m ) '1, I Erl pt-ovesthe fbllowing:

Arro.0 (rz)'r^rlrn t- nz) implies Arro"u0(, - 1,m 1- 1,n ),- nr) .

- , o b e 4 , , . 1 - 1 ;b y t h e l i m i t t h e o r e n rI . 3 . 2 t e t F ( r ) : Ifint: LeL fi ,lvl-t -) z. An lim" G(r, s) for a An" tunction G. We may assttme G , lvl unbounitecl b.ornogeneousset for G is homogeneous lor f' as .,ve11. (rz,rtz ) 1) 1.27 Ttreorern. Over IE1 the followir.rgare eclui.ralenL (i) B En+n,

118

II. F'\'agmentstr,nd Combina'bolics

has a cardinality some colour j, the set a: {r Z I z has the colour 7} is the maximum of elemenls on the lelrel than f/(r;). Now II(r) bigger -bh.ro )' s; ) max(s;) (maximttm of elements of s;)' And since Hft;\ - \' ,/ , z) "-*--* ",. get rnin(a) < max(st')< H(';) < card(a)''lihis pr:ovetj -

c'o'L(s;) rit1ioa1).,ve 'l'he inclr-icbion sbepis as above' or-rr'claim.

[l

no y ! . Z L P ro o fo f T h e o re mI.9. ( a) Assumethat x,z,Q ar e such ihat tbr r,ve have l"*,il -, (q)p. Thus for each y there is a counter-euample-mapping -; , uoithno homogeneorls relatively large sebl:aving ab leasl,q f , l*,yl;, of lr,oo]t: lJrtr[*,U]" Assumewe have fixed a C1 enumeraLion elements. of lengt]r by all [umbers such thab for each U, lx,y]t forms an initial segmenL s of length cl, such for is codedby a secluence ai. T'n"r. eachcounterexample determinesnaturally a f 1 dy, (r); < z. The set of all counterexamples "ich 41 tree ? which is Ar-estimatecl and unbounded;by the low basis theorem branch. This branch naturally deLermines 1i" rl1; it has a LL.l tinbor-rndecl with no a LLl mapping F : lVl --+ z (where v is the set of all ntr'mbers) set having ) g elements.But this conbraclicLs relativel.yf"rg" homogeneous 1 . 7( 1 ) . l-l (U)'neplacenby (n * 1), take k standardand apply 1'7'

1'10) tr.8, (c) Froofs(o[ ]""6'

neither on It remains to prove theorems 1.6, 1.8 and 1.10.The proofs depend bhe above proofs nor on each other' l-.22 Rernarh. observe that for: m,k )
.[ Etl.

l, the following are equivalent over'

-* (c,r)l u,' ,(A*, A*) o -- (ar)l,(A*,, A*+k)

jutst sets F-1(i), ( s i n c e f o r a F : X * > a maximal homogeneoussets are i<a).

l-"23Lernma.For m ) ( r.)

'1,

IEt -Fcu-' (c,,')1. ,(A*, A*) | BE'^-r1 '

proof,'By I.2.23, I3E7n1-.1 may be replacecl l:y RII.*-1. The proof is by (Cz)(ly < a),0(r'A)' incluctiorron rn. Lel rn: l-. l,et 0be tlo and asstLme s e l e c t o r ) ;t h e n 0 ' i s I I s ' i"t 0'(*,a) = 0(*,y)k(Ya' a y)-l(t,v/) (minimal

'tr6 n1"; : i: b'@,a)ky < a. F is lls, cLorn('F).is u.trrrr.b"i." and defi.n", g"'t a:ni < a suchthat -F-1(i) is ;;;;-,'i,i;]12r1,/li)'or" anclunboundecl; Rils' thus (Cx)0(r,d)' This pl'oves unJ:or-rnc1ecl,

2. Instances of Lhe Paris-Flarringion Plir-rciple and Consistency Sbateurents

L2'l

The fu.ncLion rrns is C1, l,oLal, ancl ib is easy to show by inclucbion on z -+ a there is a each f ,l"] thal, .[or each e of cardina)ity rms(u,,a,Q) an-cl secllrence .e (of elements of c) sr-Lch th.at lh(s) : q. tl h.omogeneolrs

of tlte Paris-tlcLrrington llrinciple 2. Instun,ces

and C onsiste?rcy Statements
(a) h"ltro'cluction avrdStaternen{ of trlesrih;s
2.1 Intloduction. In Sect. 1 lve introduceclthe noLion of a relabively large finite set (X is relatively large if rnin X < lXl) arid lhe "arrow" nol,ation (for each/ tlr,yl -> z bhere is a relativelylarge homogeneous l",Vl ; (_q)y sebhaving at least q elements).We put

PH (u,z) : (V x, ,ily)fu , yl ? ((q)!) ;

the Paris-Hamington principle was clefined as tlr.e sLatement (Yzt,z)PlI(u -F 1, z). Write (PII) for the lasL sbatement ancl (Pfl),,, for the lbrmula (Vz) PH (u * 7, z). Paris and Harrington showed that P.A proves

(PH ) : Con"(PA"+ 7r(/7i))

where in Con"(...), PA" stands for l,he natural /1 defrnition of PA and frQIil means the set of all true Zi-seni;ences. It follows by Godel's second incompleteness theorem that (PH ) is unprovable in P.A. As iL was sho'wn above (1.9) for each n ) 1, I t;n | (PH)"_t $.e. I En ts (Vz)PH (n, ")) i.e. fbr eadt lc, lEn I

a;ncl f En proi/es all numerical insLances of (PII),,

PII(n + 1,f). Tlre cluestionwhether the lbrmulas (PII). are related to staLerrentsassuring the consistency as follows of fragments of PA(+T'r(nr)) is answerecl by Paris's beautiful refinemenL of the result of Paris and Harringtoir: 2.2 Theorem. -If1 proves thab, for each zL) 1.,

(PH)" : con'(Ir:i,-l Tigli)) .

secLion. Th.e proof of Lhis result is the rnain conbenL of bhe preserrb

120

II. ft'agments and CombinaLorics

(ii) Art'otu(n,*,m l- n), i,e. + (a.,)p c,,' ,u(A*, An,+n) ( i i i ) t l r r o w o( r , ^ , m - l n ) , i . e . c,r--,u (u)),"(t\*, A*-Fn). Proof.T'l1eonly implicaLion l,o [:e proveclis (iii) -'(i); bub fbr n --I ib fo]lo-ws by L25 a1cl 1.23 ancl for n > 1it follo,,vstising 1.26: Indeecl assume (iii) foi hypobhesis)ancl let Arrow}("r-l l, rn,rn-l .o*" n ) 1 (ancl all nz - inch,rction n * 1). Then, in particr-lar, Arroro(i, n1)m * n * 1), thus Arrow\('I,m,m) -1L 1, rz f nz * 1) and BErr-p11 hence we may apply 1.26 and get'Arrowo(',* t-l hypothesis. hence B Z,"1,',1-1by th.e indr-Lction. Clearly, Theorem 1.6 follows. Proofs of 1.8 and 1.10 will be sketchy; the reader may elaborate cl.etailsas an exercise. 1.28 Theorern. Fbr rn F : lVl2 + 2 ( where v is the universe of all numbers) having no Err-p1 o. b.u. set. Hint: T.he proof in [Jockusch, 1972-JSL] (Theorem 3.1) formalizes easily and gives a A* mapping F : V2 + 2 with no o.t.u. Arn+I homogeneous set. B;i I.3.24 F has no o.L.u. Em*r homogeneousset. 1.29 Corollary" (1) For m ) 7, I Ern-r1 proves

-.[c,., --ro(u))72@*, E*+)) .

(2) Theorem 1.8 follows. I'Iint: I Em+t proves that a "Xp11 set unbounded, see I.3.23.

the following 1.80. We sketch a proof of the finite Ramsey theorem. IL r,Lses lemma: (*.) For each u, d,Q ) 1 there is a y such Lhat for each r of cardinaliby y *r a Lhere rs a prehonlogenoussecluenceof length g. ao.i f , l*l,t "rr.h (,r,) cloes not hold ancl consicler the tree of Suppose that for given rL)cL,q .orrntli"*amples like in 1.21. Ib is unboundecl and z11 estimated; an infinibe -i a wibh no LL1 branch determines an infinite LL1 mapping F , lvl prehomogeneoussequenceof length g. But this contradicts 1'15, 1'16' Now let Lls prove the finiLe Ramsey theorem' ('i); plt,ir; a1 ancl tobal. Define be the minimal y sz-utisfying L,et ph(n,cr,,c1) r m s ( ' I , c t , q )- a q I , a , q ) - - p l t ' ( n1 - \ , a , r m s ( r r ' , o ' , q ) )

rms(tt *

122

ancl Combinaborics II. Flagr-irenl,s

btib not in 2.3 Corollary"(i) Fbr each n > 1, ("PII), is provablein .IX,,-p1
.L u17.

r !-l

provesthaL(P.tr): Con(PA + :tr(II1)). (2) I\ theorem, (2) is innmeclial,e incomp]ete1-]ess (1) follows by Godel's secon-cl tcon 2.2 (anc[l,hecornirac {,nesst heor:en:-).
l.lrrrs 2"4 Disciission.Bobh efI)n anclCon(IEn-lTr(ru")) arc fI2-staternen.ts; (1) ibrming an increasing hicrar:chy(the n-th we have a hierarchy of senLencec of thern is prova[1ein IEr-,"1 br-rtnot in IEn), (2) being syntacticar,liysimplc (J7'2) anct (3) having a well understood doLrble meaning: (a) combinai;orial (mathem.ai,ical),an in.stanceof Lhe Paris-llarr:ington principle, ancl (b) logical. (meba.matheffratical)- claiming Lhe consistency of IEfl 1- T'r"(At), "lrhich is a certain reflecl,ion principle .tor .lEfl (as we shall see later). bu.t they follo,,v immediabely fro:n lrlon.-provabilities are negative resr,r.[ts; the positive result 2.1 via Goilei's second incomplel,eness theorem so il, is natural to men.Lionthem here. now plesent our general plan of the proof. In subsection ( b) we prove LeL uLs 'we fincl a some combinatorial facts relaLed to (Pfl)" and as a by-producl, simplifi.ecl[ormulation of (P.l/).,. In (c) we prove bhei.mpiication Con"(IDfl-lTr'(II1)) -' (PH)". We shall follow the correspondi.ngproof of Con(P.A 1Tr(.tI) -> (PI{) due to Paris and Halrington. Paris's proof of the forme.,: implication uses properties of a-large sets (a an ordinal) elabolatecl in the -> Con(I Di,-l ne,xt section. The subsections (d)-(e) contain a proof of (PH)"' elsewhere. things possibly r-Lseful fi.(17'1)), together with various au;*ciliary

(h) $iomeCornbinatorics
Recall that PH(u,z) meaus

,l ( V r ,q ) ( l y X l r y
hlote th.eobvious monotonicities:

-+

KDT)

r' 1 r, q' I cl, zt I z a:ndYt 2 Y then 2 .5L e mma (Jf1 ). If l r, y] -) kDi an.d, :13

l*'.,y') ,1kl)i,.
S/'e ar:e going Lo prove bwo lesulbs:

2 . 6 Tlreo-r'errt..[tr-1 proves that' for each zt ]

'I,

-Fl, z) : (Yz)(fvXl0, = (Yz)Ptt(u (PH'),, ul : Qt+ 2')i-rL) u 2 1, (Vz)(:v)(10, for eaclr vl ? (tt -r z7;t-t1 J'tr'1proves bhal, 2.? Ttreorene"
-> z f,herc impJiesthat for: eachz l,hereis a.y snclr tlrat fbl each / , [0, ,|tL-tt

Sl,at'ements Consisl,er-rcy 2. Ilstances of the Palis-I{;ulrington Principle ar-rcl

123

is an B' hom.ogeneous for f ancl satisfying l,ire follo,,ving:

< lHl . z ( min F/ < z*i.''(H)

prool 2.8 Rer-nark. We a::e going to prove 2.7; otrr proof fol.lows an anal'ogous sa-r-ne from [Paris-Flarrington.]. Then we inclicate how to prove 2'.6 hy Lhe rneLh.ocls. The following lemmas are pro. ecl in -I.X1: iff each Z.g L,ernma. Let .f , [0, b) u c. H g [0, b] is homogen.eottsfor / of f/ is. (e -F l)-element sr,rbseL bl -, ci, i : 1,..', & and let /(x) : (.fi('"), "', 2 . 1 0 L e r n r n a .l , e b " f ; , 1 0 , -, .IIc; and H q lO,b] is homogeneousfor / iff ilb is /r(")). Then / , [0, b\" homogeneous for each fi. -> c f- -1. sr-rch. 2.11 Lemma. Lel, / , [0, b] - c. Then there is an ,f' , [0, b]e+t -l1, is homogeneoLlsfor / iIf it is homogeneous that a set ff q [0, b], lfll > . f.or ft . for /, .f'(*) : proof. For x [0, b]e+I put //(x) : 0 iff x is hornoge:oeoLls for then clearly f/(x) : f ( * 0 , , , . . . , r " - 1 ) * 1 o t h e r w i s e .I f f / i s h o m o g e n e o l l s / '; *. prove * i' yftl-rr. Conversely, let H be homogeneotLs for f b'f", "u,.n tlrat the value of 7' on ll/]"+1 is 0, which implies lhat H is homogeneous for in H and //(x) : i:'I + f (t0,"',re-t)' J. Let x lre the least(e f 1)-tup1e Let y ) re+t be another element of tI; for each u [xle, f'(o,y): f ' ( r 0 , . . . , r u ) - 1 - l - , f ( " 0.,, . , r e - r ) : 1 - F f (rt)., n refining l,he

th.us x is homogeneous for J , contrary to our assumption. Z.lZ Remark. One can consrrucr an J' tl},,ble-Fl + construction. 1* 2/Zby

-.', B such thab, fbl. each 'I1 2,13 l,emlna. Fbr each b, Lhere is an -f , [0, b)2 relabively large and homogeneous hor f ,

2' <Y' r , ! l f / a n dr < A i m P l i e s

Let /s(*,y) - 0 if 2t < U,: 1o'w' Proof. <9' : 1 o'r'v' ft(*,y) : o if :':2 <U, : I o'w' J z ( r , y ): 0 i . f . 2 t

124

II. Flagments and Combinal,oncs

with Lemma 2.10 ancl let f/ be in accordance LeL f com.l)ineall bhese and relativelylarge.Leb a_ m.infI < lUl, e: max11, thrrs.; lromogeneous e o @ , y ) : 0 f o r e a c h( r , y ) [ H ] : 2 a 1 e , J o ( o , " ) : 0 a n c lt h e r e f o r J T h . * o 2 < - ) , h Q r , b ) : 0 a n c lL h e r e f o r/e 1 ( * , ' u ) : 0 f o r e a c h( r , y ) [ r r l 2 ' tr Similarly',2n ., e ancl 2' ( y fbr each (r, y) r= lhl)'L. For each l,,e there is an / , [0,b) -, e + 6 such bhat,for eacb 2"1-4'Lemxrrr, (*,y) l'H)t sric]rt,lLurl for / airclarrcleac:h large,,lromogeneous fJ'r'elal,ively r 1!J,,w e h a ve 2 * < A . By l ,e mma s2 .1 3 a n cl2.11. 2"15Lernma.For each b,e,c there is an / , [0,b]u* c +.1such that for each .torf and suchthaL lf/l ) e-l1 we havemin.Fl ) c. F.Ihomogeneous P r o o J . L eft( r 1 , . . . , t u ) - m i n ( r r , c ) . n

2.16 Lernrna.For each / : lct,bl'-> c there is an .f , [0,b]" * c(c-ll)(e -F 6) for i ancl relatively large then lhere such that if bhereis an I/ homogeneous is an f// hornogeneollsfor / such that c ( m in Ht <2r cinl.I,< l.H,l. 'U si n g L e mma 2 .1 5and Lem m a2.10,r eplacef by /ol0,b]" + c( c- F .L) Pro o J. for ,fO is homogeneousfor / and satisfies sr.rchthat each H homogeneous m in f/ ) c. L e t l o g r b e the m aximaiu such thal 2u ( r , let log( r 1,..., nn) b e ( l o g ( r 1 ) ,. . . , l o s ( r " ) ) . D e f i n e /r(" ) : /s( los( x) ) tbr x e [0,b]e. P : [0,b]' -, e + 6 from Lemrna 2'74 for f . ltl"t Let f (,r) combine.fu,p ancl\et H be relati,relylarge, homogeneous f*,fi, min H 2 cand we have2' 4 y'lor (",A) e lU)z. I1 is lromog"rr"olr. fb c /s , L e t I1 l - {1 o g :c I r H} ;we have llfll : lI/l , Ht is hom ogeneous Ll :log(minI/), thus 2minrI' <l{'l as desired. rnin-IJ'/ y] (Vz)(1y)(10, followsdirectly.To prove 2.6 assume Theorem2.'7 2.1?ltemar.lc" tnd let y be sr-lch and let r, q,z be given;Iet zt - z.max(x,c1) (n4-2)n'-Fr) 1 to an (tt+2 )i !1 . Assur ne t h a L[0 ,y] / ar bibr ar ilv f ,l*,r 1ui' I - - +z; exLend ? --> z andcombineit with.fi : [0, IJ]+'-n max(c,g) such Lhz'rl, /b ' f0, r1tt-'rr Let / be bheresulLing for /1 satisfiesmin(f1) ) u.,q.. eachfI homogeneous for .1, H'relatively large, lHl > u -l- l' fnnction ancllet H be homogeneous 'f'hen minil ) x,q Lrncl lbr /. f/ is hornogeneous

2. Instances of l,he Paris-Flarrington Principle ancl Consislency StaLemeni,s

125

(c) Proof of C on" (I EL * Tr (f{i)) -+ (P H)* (fortc 2 r")

Z.l8 Proo./. Recall 1.9: bherewe provectthai for each n ) 1 and lol each.r ) t, t). T]re p-r:oof I.Dn proves PtI(ni- 1, /c) (i.e. ploves (Vr, r/X:yX[", y] i (q)[' of tlris lacb tbrmalizes rn I8y as an easy inspection shows, ,so l,hat l,ve have the following:

Now work in.Itr1 -F Con"(IEn+ Tr(II1)). T|.e added axiom ca1 be evidenLly reformulated as saying that each Xl-sentence provable in 1X., is trr"re(otherinconsisbentwith /Xr)..N9* wise its negation.woriicl be a true -Il1-sentence -i GDy-Ft), take any n,z,Q and observe that, by ('), 'IXi proves (ly)([i,A] lvhich.is a .X!-sentence.Thus this sentenceis true (in the senseof satisfaction of xf -sentences).But then, by the "it's snowing"-it's snowing lemma, we get have proved (Vz)PH(u + I, z). (ly)(fr, al A Q)"));.,ve

r)) ('u-t-'L, I L'yl- (Y, 2 1)(Vua l) Pr'1 ",(P.tI

(')

(d) Strong Indiscernibles

Recall the Etn-formulasintroduced in Chap. I, Sect. 2 (").

2.19Definition. For eachXl-forrnula tp having Lheform (1yt)(Vaz)...p0(*,Y) let ,p I be the tr'l-formula (1w < ut)(Vyz< uz).. .,po(x, Y)
wlrere D!,...,,1)n are variables not occurring in p; they are called the desi gnated uariables of p f . Let (Et") l' denote the set of all p I lor tp I ut./ 17'

is meaningfulin.Itr1; thus in IE1 we have, 2.20 Observation.Defi.nition2.1-9 since (t!*l' )-fbrfor each u, the A1-set of all (X/, f")-formulas. Ivloreover, mulas are particular ffit-formulas and therefore we have a d1 satisfaction by F. u");we denoteit occasionally for a1i(Xl, f')-formulas (wibh ar'bibrary sebB : { b ; l t < ) } ( i n c r e a s i n ge n u i l r e r a 2.21 Definition (in IX1). A fi.niLe for (X,! fn)-formulasn if for each i < ,\ tion) is a set of strongindiscernibLes we have the followine: < i, such thaL t p ( x . , v ) For each (tl, f')-formula' g(x,v) (v clesignatect), ( b;, anil each' pair b, b/ each Lr-rplep of possible meanings of x, all (B b;])" of increasingu-tuples of elemenls of -B bigger tlian b i ,
\ 10,

( r')

F q r ( p , b ): q o ( p , b ' )

126

II. Flaemenls ancl Combinatorics

(Il.emember thaL in the last ecluivalence,tp must be strfficienbly srnall (< i), be sufficientl;' 5p,'11 (( b;) and irrcreasing tl-buples the paramebers p mr.t.si, elernents > b;)' la.rge(z11L b,'b/ of elements /l suFfi.ciently 2.22 Nxarnple" l-,eL'tl : 3, J.eh !a be

rc2',u)' (lyr 5 ut)(Yyz S rz)(1vz! q)tl''@1, p S i ; P t , p z < b ; ; i < j < k < q ; i < j ' < l c < q ' T h e n( ' r ' i)r n p l i e s / \ s s n t n, e F (llyi 3b)(Yyz 5 bft)(lvsS bq)'b@t,P2,u)
r11

l = (l vr 3 b i ') ( Yvz1' byXlvs 3 bq' ) ' h( pt,P2' ,u) Z.2BTheorem. Itr'1 provesthat, for each u,,(PH)r, irnplies bhefollowing: For' th.at for (X{, f u)-formulasstLch eachu thele is a set B of strong incliscernibles

lBl : ,.
'We

prove this theorem in the present section. The next secbion is cle-zotecl bo a proof of the fact that, for each u, the conclusion of 2.23 (existence of ar:bitrarily large sebs of strong indiscernibles for (X{, f")-formulas' irnpli.es

-r rr1n|). Con"(IEi

shall assulner, : 3. But bhe method is perfectly general. For simpliciby, r,ve Z.I4Conventions (only for this section). Define in IEt as follows:let (lr < u ) t p ( t , z ) b e a X f - f o r m u l a " s u c h t h a t r . rd o e s n o b o c c l l r i n g ; l e L p b e a t u p l e of possible meanings of z. Then the element def,ned in [0, 4 bV this fbrmr-rla with parameters p is the minimal a 1 cl such thab F g(o,p) (if there is such an a). Dnally, the element,definedin [0, 4 bV (V* < u)V(*,,2) is the minimal ct I d such that F -,p(o, p). If c/is a number and b g 10,d] then def q(d,b) denotes the set of all elements <c1,wit'I.r " of theabovetwoforrnssuchthat tf., of f0,dl clefineclbyformr-rlas ls paiameters from b. (In particular, You may use f.or ls any (EI" f")-formtila' or (tr{, f')-formulau (" > 1) w.r.t. its first clesignatedvariable.)

(n ac ne Anincreasingseque iet P <^t < 6 < dbegiven. lq < u)oI if, for eachq 1u -'),, (Ibr B,1,5,c1) sequ,ence lessthan B is a Pari,s elemenbs (1) lctqlt, Bl a defn(d, [0,oq]U {r, 6}) : 0, U {6}) : 0' n def,,(d,l},nql (2) lctq+r,71 (3) luint,6) o defo(d,[0, ctfl): $.
is a set of sbrong indiscernibles fbr Z.Zb Lemma (-ttr1)..A Paris seclLlence (Xi f")-formulasu.

2. InsLances of the Paris-I-Iarringlon Principle and Consistencv Slaternents

1z',l

Proot'.Let r/(w) * (lr)(Vy)(12)9@,,A,2,w) where <pis Et ; assLune (,ri f) < j ( i < < /c Q 1 u ,p 1 ctt. Then the followingis tr ue ( in the sense of F) :

(l'l)Ui,.y,6,p)( : 1 " 3 p ) N y 7 t ) G " < 6 ) t p ( r , l r , z ,= p)

: (l r : ai) ( Vy .,1) ( 1, .- i' Stp( r :,!t,z,p) (sinceotherwisethe smallesb z such that (Vy < l)(1t < 6)tp(r.,lJ,2,,tr) worlld b e i n l a i , 0 ) , i . e . i n d e f; ( d , [ 0 , a ;t]J { 1 , 6 } ) n l o t + t , B l , w h i c h c o n i r a c l i c t s ( 1 ) ) , < ap)(12 < 6)v@,y,z,p) "i)(Yy ( s i n cei f w e l e t rs b e su ch Lhat r s ( cziancl( dg,< ap) ( 12< 6) ?( r s,U,z,,p) --(!z < 6) l:ub not (Vy < l)Ft < 6X.. .), then Lhe minimal y such 1h211 g(xo,a,z,p).,voull d i e i n l a n , 1 ) ,h e n c e i n r l e if @ , [ 0 , o 7 J U {d}) )loj+t,l], which contradicts(2)) = (3n a < a6)(12 ! a)tp(r,a,z,p) "i)(Va (otherwiselet z6 1 cti be minirnal such that (Vy5 ctp)(12 S6)v@,?J,z,p) and take a go ( aA sLlch that -(12 1 oq),p(*0 l t,o ,z , p ) , then the minimal z such that tp(r6,u0,ztp) would lie in loq,6l, thus in defp(d, [0,or]) fl lor-rt,6], which contradicts(3)). Thus : (1" S

t r ( r bl ) ( p , " r , 6 , ,= p )( h l ' ) ( . , ;c ,ti,cth,p)

for aii i < j < k satisfying our condition, which sho.wsthat the Paris sequence of strong indiscernibles for (X! f )-formulas. n @q I q < z) is a sec{uence To cornplete th.e proof of Theorem2.23, it remains to prove the foilo,,ving 2.26 Lernina (J'X1). For each u,, (PIJ'),, implies the existence of a Paris sequenceof an arbitrary length. ProoJ (for u -- 3). Leb Lhe desired lengih u )_ 5 of a Paris sequence be given. 'We assurne(PiJ')3, ancl u.se2.'7. Take a. sufficiently large c (w.r.t. u;it turns out that c:2u is sufncient) ancl let d be such that for each .f : l},d)'t -n c tlrere is a horrogeneous f/ such that c ( min 17 < 2rnrnrI S lfll. -Fbl each a < B < 7 < 6 < d clefi,ne (cf. (1) above) o0:0

tLq*r: maxldef U h,6)) n [0,p]] + t ,,(cl,l},n',,)

- r). ( r: J o ,1 , . ' . , ,

I2B

IL f'\'agments ancl Combinatolics

thab if a,, 1B then p,l ) d e f Th r-ts [0,oq]U { t,6} ) : 0. Obser ve l a q -v1 n Ql1, bo'finc al< B < 7 < 6 < d s u c i r t h a t t h e c o r r e s p o n d i n g lVe.vanb c t ,1 { L q - t r . for eachq 1u -.I. o't crr'ssatisfiesthe f.olJ.owing secluence

(i') (2') (3')

glt\ d,e U t l , 6 j ) : V,) oq f( d , i O a q - f l . 'c >vi,. e .L * , ,, ,] ( 2 ) ,i . e .1 0 , 4 ) d e ,f , ( d , , [ 0 ,U { r]} ) : 0 , nq f ( d , 1a ( 3 ) ,i . e .l r , 6 ln d e n 0o , l) 6.

t t r n c t i o n F , [ 0 , . 1 ] 4 - - +c a s f o l l o i , v sf:o r e a c h ( o , 0 , ' Y , 5 ) [ 0 , t 4 u ' Defi.ne b h . e defi.nedabove and put Iet (ct, I c1( rz) be the secluence

an1 U { " y6 f Q I[,0 , q < u ) ( l * ,0 l ) d e n F(*, 9,t,5): (min , })+ 0)

if ther:e is stlch a c1,else

: c l 4 - F( m i n q<

")(B,fln

d eo l (d,l},altt{t}) I A)

if there is such a q, else

: c l 2 - F( m i nq < u ) ( l ^ / , 6 1d ) ef n ( c l , [ 0 , a r* )) 0))
if there is such a q, else :3c14 * I Eviclenlly, F : [0, r/]a --+ c; if we prove that there is a homogeneous FJ' strch that the comrnon value of .F on [fI]a is 3cl4* 1, then each cluadrttple of lengbh z. a Paris sequence \*,9,^/,5) -FIdetermin.es rI Now let // be hornogeneous for -F and such that c S min 71 < 2rnin a the l.F/l; let {h; I i < ei be its i.ncreasing enumeration. Firs't assume bhat common value of .F-on f1a is q < clL.Then for h,; < hi < hp < h* from I)' we have

a, nu J { h n , h , " }+ f (d,lo )0 l h , ; , h i ld ) en
-a n d si n cel h ;,h i l .td e fn -1 (1,[0,an- r 1U { hn,,h*} ) : 0 and aq m ax.ldef n- .( c l , f on t aq depend s that Note ht.. gel o"q we 1, n U [ 0 i,a 'h ,i tq -r1 {h t,,h ,n }) [0 ,hj]] : F( lzo,h' i,hk,h* ) : , h ^ b u i n o t o n h '; we have F( ho,l' ti,lt' p,h*) have t hs. Flence for all i g a n d we get o"n F ( h ;,h i ,h t,h *): "ve l h r h ; + t l ) d e f, , ( d ,[ 0 ,h o ]U { h " - z , l r " _ t } ) + 0 . r,vith( g free variablesand hgf 3 parameters, But we have ( g formulaseach. thus

< (h0+3)(c+t; l c t en f Q t , [h 0o , l| { h " - 2 , h " - r } ) l< ( c + r ) . ( l z o * 3 ) q

each of (.lfll - 3)12 (since q < u < c < h6). The larstseLmusl, inLersecl, clisjoinfinterva1sU,z;_t.,h2;)(:1,...,"-2)"}].rom2ho<

2. Ilstances of the Paris-I{arringLon Principle and Consisbency Statetnenl.s

'L29

2h-2<(2t,o-3)l2<(|fi|-3)lz;tlrtrsif1Vecan.I)roVe(/,o+3)(v+l;< ': 2". Thtrs 2hn-2" we have a coutradiction. Now remember that we boolc c o prove H e n c e i l , s r L f l i c ets > h0 ) c:2'. ( h 6 a 3 ; ( a + t ) a ( l r o+ 3 ) ' a n c l ( / 2 0 - F 3 ) u>_5 n : 2 ' i I b h i si s L r u ef o r r 1 2 " i m d | e s n ' < 2 r - 5 . B u . te v i c l e n t l y l,lrab '2 2," (if u 2 5). Thus we geb a. coni:racliction. :u zrrrdtlrerefole true f,rl ea.cJ.r ancl have exclucled th.e first possibility in Lhe definition of F. Similarly we eliminaLe the second trncl third possibility. Tzlhe the seconcl. ry(since il-refir:sLcase d.oesnol occtir:). that an ( hg for: eaclr. We alread.ylcno,,v h's' Thus c l 4 * r Jf o r a l l r e s p e c l ' i v e F'(h;,lt'i,hk,h*): Assr-rme

u Ihrrl]) :l'l, thus ,,(d, [0,on1 lhj,htrl ) clef u {.h"-t} ) :f) llti,lt ial n defo(d,[00]
for all / (as above), which ieads to a contradiction. The third case is analogotls. Tfur-rsthe common value of F' on [f114 is cla* 1 and the.r:eforefor ezrch of a's is a Paris secluence.This (n,,9,7,6) lf/]4 ih.e corresponciingsecluence n com.pletesthe proof of Lemma 2.26 and of the Theorem 2.23.

(e) Final Considerac;ions

2.27 Recall Theorem I.4.37 it will be used to complete the proof of (Vnz > 2) T r ( n l ' ) ) ) i " / X 1 . ( W e u s e l ? z- 1 i n s t e a i l o f ((PII),r,-1 -, Conn(Lnh-t'U to simplify our consiclerations.) Let S's be a u ancl Ln'r._, instead of IEI-y U Tr(IIl' )); assuming (P fI) n - 1 SkgTnl-1" finite set of closed instances o,f. we shall construct a A1-satisfaction F/ for ,9g such that F/ extencls F (the ustral 41 satisfactionfor X6-formulas) and F'So.By I.4.37, this implies the desireclconsistency. The satisfaction F/ is constructed using a suffi.cently long set of strong (Ek-t f')-indiscernibles, guaranteed by 2.26. The definition follows. 2.28 Definition (/X1). Let Q be (Qp1) . . . (Q x, ilrp(x, Y), tp L's- Recall the meaning of (@ l' ,)(*, y), namely

( Q " t < z ' r.). . ( Q # n < r i p ( x , y ) .

(d I oq -') Let A: {cts I q < ,} be a finite set in its increasingenr-rmerabion. clqlclql' r ...fbr zr t22... inLo ( d I o b v i or-rsl me y a n sth e re su l t of sr ibsLitr - r ting z),i.e. ( Q r " , . o o ). . . ( Q n t n 1 a n l k - r ) r ( " , Y ) ' Clivenc;,cl, assumethat g is the leasl,nurnber strch thal, ( l ) ( @f ) S q ' (2) - ci-r,rI4ctq,

130

II. Fragments and Combinatorics

(3)o)-k1u. -l pul, It Q;:

o ) ( o - . c ; , c l ) l. 1 - a q - r r ) l F ( s p ( ; )| a c t - t -2 f f ' ( * - c i , - L t c l ): ( r n i n c i

c;, d) : 9. pr-rt/.,@(+In all other cases 'Ihus we have interpyel,ecl all the function symbolsof Ske(O)by /-\1 functions on ,,1).Similarly .[ora finil,eset of formulas insteadjust one. This (clepenclent F/ for any fi.niteset of formulas in the usual way, a /1 saLislaction cletermines, of l.heforrn Sko@) (cf. I.'I.14). (PII)^-I. For eacirfinite set 56 of 2.29'.Iheorem(ID). Let rn ) 2 ar':d that if A is a, of sks(Lil'*-t" U Tr(IItt' )) there is a u such. instances closecl for (E'r"-t f')-formulas of the cardinality z ancl set o.f strong indiscernibles F/ is th.esatisfactionfor 5s given by definition2.2B,then F/ 50. 2.30 Corollary" Theorem 2.2 follows. ELabo'ration. 2 . 3 1 l .]e mma (IL ',). If O i s Et**1and A is a set of str ong( Ehti' ) - indis cernibles,then for eny cl saiisfying 2.28 (1)-(3) we have -'X* tf G- c,i-rtcl) - (min c;4 aqp)[F (O(;) | aq+2 (Obvious f'rom the definition of strong indiscernibles.) (/x1). Let (tr be El_1 or flfi_' and let g(s' u) be a closed 2.32 tr,emma F/ f)-indiscernibles, of 5ft6(O). Let A be a set of strong (th; insbance satisfactionand let g satisfy 2.28 (1)-(3) for c : v(s), the corresponcling cl: V(a). (I/ is the interpretationof Lheterm s)' 'lhen --+p(s, u) . tr' (O I an-p1)(n)) Proof. As in the proof of I.4.37 prove l = ' ( @f o s + r ( r ) - * ( O ( ; l ' ' c , q + l ) ( * -s ; , u ) . to show It su'Ffices (,) ,) F ' 1 O @f o q + r ) ( * r ; , u ) - , i O ( ; j - r ) | c t q + r ) ( es ; - p r u -; - p r , / ) . B L r t4 ( i )( - , t , y ) i , ( Q ; , + t r ; 1 1 ) O ( i + t ) ( , r ci,d)] '

2. Instances of lhe Paris-I{arrington Plinciple and Consistency SLa.temenbs

131.

Consicler Lhe following tvro cases: C a s e l . Q ; - n : V . S i n c e F / s ; - p 11 a , , , ( b y ( 2 ) ) , w e h a r v e

F t ( o U ) f a q + 1 ) ( * t ; , u ) - r ( V * ; - r r! c r , n 1 , t x - d ( t +l1 a) q+zX,-sirr;-p1,') -' (O(;+t)| os-rz)(*s;11,u)

- - ) ( d t ( i - l - 1|) a o 1 1 ) ( e - s ; - r r , u )

(by indiscernibility). but us s;,u) . Sir nilar lyas above, si+ 1 :41- t( C a se2 . QU -t: l . T h u rs of ff*r, we haveFt (O$') | .,q+t)('- si,u) -' (lr;-r-r S ing alsothe clefinibion a q , r )(Q1 'tr) | o q -rz)(,.- s;, { ;4.1,u) the proof. (d;(d-F1) | oq+t)(,-- r;+r,u).'Ihis cornpletes n

2.33Proof of 2.29.Let afinite set 5s e inst,(Sko(Ill--t"aTr(IIl" ))) b. given. r/(s1,...,s/c) 5b of the Slcolemizabion tirat for eachinstance LeL ug be sr-rch of an axiom V e Lnfi_ln U Tr(IIt' ) we have
( f | ) ( z s a n d s 1 , . .. , s 7 ,l us

and let A : {art I c11 us * 3m} be a set of strong (t'rr-t fn)-indiscernibles. with bounded form (As we shall see, if V e Lnt*_1 then its "prenex norrnal kernel" has ( 3m unbounded quantifiers.) OtLr aim is to show F' ,90 lbr the /l1-satisfaction given by A. (1) Le{. V e Tr(II'r),W - (Yn)$(n) where r/ is bounded/. l,et /(s) e S's; then F ,b(V(r)), i.e. F/ r/(t). ( 2 ) N o w L e tV , be L-Q,, whereA@t) is a Efi-1-formula (1"2) ...(Q.,rr")

g ( *t , . . ., n m). T h u s tr i s the following:

(Vr1)lo(r1v ) (ly1 < 11)(-@@) k ( V " r < y 1 ) i [ ( z ) ) ].

Hence an instance of Sks(tb) has the fbrm , p ( , eV ) (tt < si & -p(t) & ("r I't1 - - + , p ( r ) ) , i n s h o r t , t i 1 u ,t , t ; .

on the {brm of the terms s, t, r will be consiilered later.) (Precise condibion.s that (tr-@ f)) < s and V(u),y(t), VQ) 4 a,t. Let q be minimal strclr. We have ir,vo cases (o) F' r(r); then f (s, t, t) ancl we are clone. (b) F/ --9(s); then, by Lemma2.32, we have Ft (-Q I'oq-rrXrr). V(s) b: Let, e be bheleast number such that F (-'f lan11)(e); assurrle ( i . e .b 1 : V ( s t ) e t c . ) .

132

IL [\'agment's and Combina.torics

(3) Claim.V(tt) : .. Indeecl,ibr some h I c1,,

) , f( b ) V . c 1Z - b 1 : ( m i nc r < , a t , .F & (-<D I ot-rtXtr) JL*-rt(b) 8 z( Y z 1 < c 1 X @ | on-r^+t)Q) 1 trX@f ) r ( v z :< ( m i nq S o n ) F( ' - o f o q + r ) ( " r )8 "q-rr)Q) (rninq S nn)F-(-rF I crq-p1 Xyr )

Lhus 7(t1): ". i m p i i e sl = - V ( t ' t , t ' 2 , " ' , { * ) , (--O I'oq-rt)(/1) ( 4 ) N o w b y L e m r n a 2 . 3 2F is t; f.or i - 1.or i even and is given by -,O for i odd, i > t' Our l; wlrere tto are given by L=o. F (5) Claim, . o r i : I , . . . ) m ) F t ; : l f ; t h t i sF - p ( t ) . Fbrie.yerranc1i:1wetriviallyhaveFt;-t;;1briodd,i> t;- ) anclt; :- F*- r t( u,tl;- r ) ' tl - - Fi*( * b y i ::cl u cti o nNow . w e p ro ce e cl Ass1pe b : t/(s) and <- ci.-7:V(r- tr-t) - V(- t';-t) anclcornpute: V q ShStiOU) f *,-ro (b ,+- ci -r; - (mi n c;4 ao) ( p( b) & (Vr < ,t) Q | "p( "t) : (min q 3 ap)(--rP(;) ; ct'p-4?- c;) - (rn i nq 1a,) ( - O( ' t) | ct"q+r ( ?ci)
r \ : Jr-<b i (+- c;_r) .

f .,p- r tX* tl)

T h u s V (t;) : V (t';). (6) Now talce 11; if F 11 ) f 1 nothing need be proved. Thus assrlme tsV(?'t,..',r'*) where ,herefbre V(r) <V(t) - rh; then F Q I oq-4(t1)t prove r; : rt; for i even. trVe F ,,i : ,-i lbr z odd and similarly to iborr" we
have proved

Under the present assumption we have p(*). T'hus we have (7) Clur,rn. n the proof. proved,/(t,t,r), which completes

Sets Ilierarchyanda.-large 3. Schwich,tenberg-Wainer

studied insbancesof Paris-llarrington princiIn the preceding two sections \,ve ple ancl showecl(1) that the k-th instance (PII)k is provable in 1X641 and (2) (Evito Con'(IE"k-l Tr"(fr)). blrat, provably in IZy, (PH)t is ecg-tivalenb dently, (2) implies (1) since IEtr+tproves l,heabove consistency,cf. 1.21.33-3'1,

l{ierarchy and a-la,rgeSets 3. Schwichtenberg-Wainer

r33

is oI in.clepetl-cletrt btrb the explicit proof of (.Pfl)1, in I X'p-plthaL we pr:esentecl interesb.) combinaLoriaLpcinciple but rr-:lal,ec[ going to stLrcly a c1ilferenL Nolv r,ve.are ancl its instances. We shall call it (lV) or the p'rinciple of a-large inuerunls. we shall pi:esent Lhe priitcipJe rernar.t<s1 Iror baclcgrounclsee Jribliographici-r,l 'Ihis and and relate its insLancesLo instances of Paris-Ha,rringtoir p.rinciple. will be don.e in the following sLeps: (a) we introduce ordin.als in .I.11 ancl. clerive their important properties, (b) ,re show- r,vhichindnction is sufllLcier-ti 'rre introduce and stucly a-t;rrge seLs, to gel, enou.ghinduction for ordinals, (c) and (d) we define the principle (IrZ) ancl r'elaLeinstances of (tAr) bo insla,nr:es of Par:is-Harrington plinciple. Nole that results of bhis section ffIay be r-tse.-|. to geb the characterization of functions provably toLal in J'.Ii, (ancl P,4) usin.g moclel theoretic meansi this will be done in Chap. IV.

(a) Onclinalsin I E1
3.1. We are going to define in -tE1 a At class e linear:ly orclerecl by a 41 ordering { r,vith a least element 0 and with a ll1 operabion X assignirrg Lo each. each finite non-empty clecreasingsequencey4 . . .!,Lxof elements of e a:n.d sequenceof non-zero nttmbers rr1,...r&u of th.e same length an elemenLof e a; the oritering is related to E as follows: cot'i clenobedbV ILr

D'r'"; { I w';b;iff
i=! i:I

(1) there is an i 4 r,g such that prl f u; or ct; * bt, ancl, for: l,he lea,sl such i , l l i 1 u ; o r ( P " ; : u i a n c la ; 1 b ; ) o r (2) for each i I n, LLi: ui ancl ct; : bi. Furthermore, e is least 41-class containing 0 and closed under X. This is what we expect from orclinals ( e; we have to show that this can be achieved in /tr1. (!Ve alsoexpect well-orderl but, as we shallsee, this costs induction.) Thus let us rnake the following. of rrumbels sucir 3.2 Definition. A regular tree is a set t of fi.nite seqLrences that (i) / contains with each s ea.chinitial segmeirt of s, anil (ii) for each i, j,s i|s ^ U) I and i < i then s ^ (i) i. ( ' f h t r s u p p e l n e i g h b o r - r ro sf s i n t a r e s ^ ( 0 ) , s ' - ( 1 ) , . . . ) s ^ ( i ) f b r s o n r e zt. a rnapping e (e.ralLLai;iotr) A- pre-ordinalis each regular tree i together 1,vith. 'lbe 't lt'eiglr,i a non-zero number e(s). assigning to each non-empLy s of rl is the maximum of lengths of elemenl,s of l. We define art operntiot't of preI applicable to each pair (p,o) r,vhelep is a.non-empl,y seclllence -lh(u). of posil,ive num'bers such thab lh(p) orclinals ancl o is a secluence

1 C,l I d:t

lI. I\'agrnents and Combinatorics

1.el as follows: a l ):T ,l u ,r ' a; is definecl T h e p r:e -o r:d i n ('t,e 1,t;: ( t;,e;) er " nc l == ^ s I s t;} U { 0} , r ( ( i) ) a;, and for 0 /= u e t; 1e1; l e t f : l '_ Ji =r{(t) l,o joining the e ( ( z ) -- s) : e l (s).(It i s easilyseenthat bhis cor r esponcls ihe old .root of p; overa n.elvrool aucl evalLratiirg evalualeclbleesF;, . . . ) l.Lx: by o;.) O0,y: { 0} ,O*- st,y " L o L zr,l /\1 fu n c LionsO*,r-,.( r ,,as follor vs: N o w rl e fi rre {*,r, ai S Lt}i 4*,y using b]reobviotts [)',1truqct;I t-r;e 2r,a,pr; clesceniling m o d i fi ca L i o no sf 3 .1 (1 ), (2 ) above.
3"3 I'act. .Fbr each a, Or,y is a tota.l 41 :[nnction o.f ri Or,a g Or-F1', analogously tbr \(c,g. (Proofs in -/tr'1 eviilent). 0r,y e O*,!./,-l-7, 3"4 Ilerlnifion,
I l ru .t *'u

'

t l dy : U {',y
, l l

l)CI*,,, <; : U d',v

v
a
l

U 0*,r,
?J tU

$ra

3 . 5 F a c t , O y , 4 y , A * , 4 ' * , t , d a r eA 1 . (Eviclently, they are tr'1; but for pt: (.t,e) we have

p e O y - p e 9 r , . y ' f o r x : hei,sht(t) F e fr* -+ p e Or,, f.or y : m.ax(range(e))

3"6 Fact. (1) < linearly o.rders. (2) e is the smallest /1 class X containin.g a; (W X descending, a; positive). (3) Fo:: 0 ancl closeclunder sums |-r'' each:u,each non-empty d1 subset of.O* has a d-least (i.e. dr-least) elemenb. (To prove (2) show that each o*,y e x; to prove (3) observe bhat each O*,, is a finite set and Ou*r,y is an end-e,stensionof Ot:,y, i.e. each old elernenbprecedes each new one.) $"? Deflnition" 0 is isola'ted; It,rp'cr'; Limit,orrlinuL(or sirnply a limit). : 0; otherwise it is a if I,Lt: is isoLateiL

il.8 Fact. pt is a limii iff p, > 0 anil has no predecessor. 3,9 Sact" llach l-t. e has a successor.

d e f i . na e> p i f f u y d 1 , , , 3 . 1 0D e f i n i t i o n F . or cv: Ii up;ct;, P:L"u';b; (1r" is the least exponenl, in a, u1 the greatesti" 0). tr'urtherwe pub a ) 0

3. Sch',vichterrberg-\,Vainer {-l.ierar:c}ry a,ncla-Iar:ge Sel;s

1,.t(

it a =f 0. For cv > B define ot -l- p as follo,,vs: cv -l- 0 : cvi fuLrl;herrnore, i:f 1-t * c o e . t f i . c i e oa {1 ;s t . . . , a r , b 1 r . . . , b y ; i I [ t x : r z 1l , h e ni | ; i s g i v e n b y e , . i p o n e . n t s nc LL I r . . . , ( L x -7 , ( n , 1 - b t ) . br . ,. . . , b , , / . l L \ ,. . . , [ L :tcu 2t . . . , u , a n d c o e f i f i . c i cs : u3.3 -l-,r2 .B -l ,,,,0 ('I'hus e.g. (c,'3 .3 -y ,,.t2,4) .'L-F t^t0.7) .7.) * (co2

3.11 Lemima.Each limit oldin.al a e s can be trniqr-relyw.r:itten in bhe form .where(IID e (heacl)and I/D > r.,.'p'..[) or IID is ei:np1;y a,irclis; IID I c,.r/c.1. (trvideni.) clisregardecl. Theoi:ene, Th.ereis a d1 function -[a](r) clefined 3,.1-2 for each a e e ancleact:r r satisfying the following: if a is limit, a : HD * tur'.I (p 2 L) Lhen ( p * t is the pr edecessor ) ; if pris isolatecl r^ .r/^ .\_ | tl O *u tr-r 1aJ[cJ : \ nn -p 6,2{r}(') if pris limit. thus the result is t|D. For c:0 the memberuF-7r is cleletecl. ( i i ) co+ 0 i mp i i e s{a }(r) < a; a Q (9y,y and z < y implies {a}(z) . O*,yi B >> a > 0 i mp l i ." tt|-F a) ( r ) : P - t { *} ( ") an.d B > { *} ( ") . . For eachU ) 0, definethe function .Dr(a,r) : {a}r(r) on Oy x (< y) .Proof by the evident analogues of (i). Show that this is a d1 function with domain y) (< ancl th.at for z ) U, Du is a restriction o'f. Ov >< D". (T" this end, show by induction on r that D, | (0*,a x (l y)) i* a well-defined function with (9r,r. r-ange includedin is /1; thus D, is A1.) D, is X1 and its dom.ain 'Ihe last claim followsfrorn bheevident fact thaL, unclergiven assumptions, iff a is isolateclancl if they are limib then for a : .H.D * coq 0 -f * is isolal,ecl tr we have 0 -f * : (0 -f I)D) 4- utu. m e a n s t h a tt h e r e i sa f i n i t e s e q u e n c e s : ( 4 0 , . . . , d , . ) ; B s is thab cvg: e, &r. - B and, fbr i < r, o.i-F!: {al}(r). The secltrence suLclr. called the uitnes.e of a ? P, 3.lS.Definition *. 3 . 1 4 L e mu ra .(i ) * : P i s 41. ( 2) If X e e is X1 or IIl then so is Y: {a I

( i ) I f a - 0 - r t t h e n{ a } (r ) - 0 ; { o } ( r ; : s '

(VpX* : B.-,.p x)).

of a -+ B show: if a 9r,,y, Proof. By inctr-Lction on bhelength of the .,vitness a:ncl C,!( ct,z-)- z < : r)a n clcv ? ,6 th e n 0 0*,y.Let, F( a,z) : Oh"tghr ( a) ,2
z

of elemeirts of -F(a,z). F,G are 41 a,nd bhe set of all dec:reasing secluences cluzrntifi.er (3s) in the a'bo're clefined lor all rl e ancl all z. The e,risbenbial cleJirrition o f a " - , g ^ n y b e . r e p l a c e db y ( l s G ( a , m a r c ( " , r " ' ) ) v v h e r ez " ' i s

tJll

ancl Combinatot'ics II. f,,\'agrnenLs

Llre rna,rimal nrrmber from all the eva,luabionsin a and z. p:r:oof of (2) is simiiar. is 41. T'.[re * : B (Propei:l;iesof -r) Tllir:oi:e;rn 13"1.5

'I'his

shows ihab L]

o "y (1) (\ -,\ 0 ;, implies 7 (2) 7i B'>: a > 0 ancla ) 7 t'hen0 -f a ;, B -Fj
l) i f a > , , r 0 ' b a , n cz more generally, ;0; t,)o.n Ln".y (4,) r '- lJ imp.Lies ; -, w6 for each z'-r 0 t,t6-l-L fb) \ / : 4 (:i) * (C) a - -, 0 iroplies too -n w0
. z Z

lc then a

-t

,ro .lt

( 7 ) :u1 l J i rn p l i e s {*}(Y ) (B) r < y and a -

; { *} ( r ) B imPlies s -+ I a -+^0 l-t Bimplies a'-L I

( 9 ) s ) 1 ' , d - 1 B a n da > o,

( 1 0)Ib r a ) 0 a n d r )'1 , wo - + co{ o} ( ' - 7) .r -' -+ P implies ( 1 1 )a \ / r , ' {a}( ' z) $ {pi(t);

ila'> B and*:

P t h e n{ a } ( " ) ;

0.

i.e. for the corresponding onestands for c,.r*.1, .pe.ntark.Needless to say, c,,,o elemenb sum. ProoJs.(2) fbllows from the lasb clairn in 3.12 (ii). (Inspect, by induction, each uumber ol: the witnessing sequence.) (3) Let, k I z ( r. Let o be the smallest elemenb o'fo* str.chthat * >- .0 .k * > ,0 .k, {a)(z) < o, {*}(t) (?' and one ani not o -n ,0.p, b5.en
z

propertiesin 3.12 that {a}( z) 2 roj.k; thus shows by checkingthe clefrning

t*je) 7 ,0.k.
,*.*.

(+) ny (3), cu*(y - x)

wo.n ) 0; fr-rrtherlrlore'

,o.(Lr- z)' Thus, by (2),

Loo.y ::

-, ,u6.2 -> t,,'6 (tiy (+)). (5) cuo-F1 \ / z z

,,. First assumecvLr ,: -lIirnit: the' {c,ro}(r) : rrt"}Q) : ufl - Now let a ='^l 1, bhel 0 :'Y ancl,by (6) It is enoughLoshow:rl p : {*x")
,P.

then roa

(5), c,;o

(7)Tri.lia1fbraisolatec1;aSSumea1imit.Le|'r,,y< \n Or. Lei a - IID * top; ihen wt' e 0t- If' IID uu"'pr.o.""clby incLuction and gives{ ""} ( g) th e n th e i nductionassum ption i , in i { "} ( r ) " mp b y, 'Ihus assull-Ie fID empt4,& : cot'.If. p is isolzi,tecl fbl.lowsby (z). ihe resrtl.b

3. Schwichienbelg-Wainer [Iielarchy ancl e-large Sel,s

IJT

t l r e n se e(a ); i f i b i s a l i mi t then { c,.r r c} (:y),{ u} fu) 7 r .,r { r ) ( ,':) { tot } ( r ) by t h e in d trcti o na ssru n p ti o n (since lt,ltj( y) ,{ pXr ) e 0,) . (9,, (B) z\.sstLme a z -" -:, a by (7), thrrs {c,-}( y) ,: {.,}(.c) : p a;L.cL e -- i3. y {ct}(a) l [cv](:r) l J y (9) It strffices to assu:.ne pro'ire a:ndto * To this enct, B {o}(r) *lrg+.r. ib str.ffices to prove the following for each fixed z and r + l. (. z, u ) I: F br ca ch a e Oz, i f {a }(r ) : *#, 0 - t- t ,t 'hi s 0 bhen( p+ i) e O, ancL is pr:ovedby inducbionon a. 'Ihe caseof a being isolated is Lrivial; assume a: HD -l-co,. First assumeHD non-empty ancJ. put {o:6](") : Bs. Then : -F-1 a*cl HD * Fo and, by the induction assumpLion, ,u6 {*X") .;rg, 'f'hus a: HD -F cud*' FID+ l3o-f I: {a}(r) -F 1 : ,6-F 1 ancl 0o -F I e 0r. one easilyseesbhat g +! e 0" (io+I e O, anclthe last exponenL in IID is strictly greaberthan the exponenb of 0o). Assume HD empty, thtLs o : (u6.If d is p -F 1 then a -rr{Lr6}(" -F 1) =, cop(n +1): wP.:t-Frr 1 ( s i n c eb , y ( 3 ) ,, r t h u s* ; ; . r 0 + ' t . ;:irr,t\.r-f dr1), If 6 is a limjt then c..,6,;1,,r{0}1'+t) e o, (sincexi -F 1 < ,); by ('/) ancl (6)

,{a}(c+1) -r cu{o}(*)ancl

may be replacedby (bv (a) since 1 < r). ? i By the induction hypothesiswe get r[0](c-r1) -+.cr{6}(ii)-r t which gives *
c-F.1

&.: e6 -u_,r{a}(t) + 1 : g + tas clesirecl.

c-F1

( 1 0 ) A ssu .mi n g p{ "} ( *- 1) .c. Clea r l y , a ) 0 a nd r ) L we pr ove," ; - 1) - F 1. ,Ihr _r s, a - ' {a }(" - i ) (a n d r.- 1 > 1) ; by ( 9) , o bv (6) , ; { a} ( r
coo --+ ap{"}(t-1)-F1 -* p['](t-1).r x: x: ancl we are clone.

(11) This is a triviality: a : B implies {cz}(c) : Ul}(") ancl a > p and a --+ B implies that the witnessingsequence is a, {a}( n), . .. , B, bhr-rs

{*Xr) ;

p rhis gives {'}(")

; {0}@).

3.16 Definition. For each p e we put coff: p",ul_r_1 : 1,1Lop . 3..1T Renrar.h, (1) Evictently,if y >'I and 1,r, Oy then a,lf Oy for ez-r.ch c. : toj.n we shall wrii;e cuf insl,eacl (2) II 1-r of c..'11.(3) EvidenLly, L , t y ; :r a * - r furl,h.errnore, for r ) 0, and z > 0,
,.0*
, . 2 LUx-7

fbly)1ancl z]0;

138

II. Fla.gmenls aucl Combiuatorics

.- ,o],_t for r > 0). The fir'st relation is by 3.15 (5,6), the (note that c,:"9 by (6). ,7 -->c,,'0.2 trom c,.,1 seconcl floreach a e, ancl each.r, a e ,Qn iff cv ( r,:]'. (Ptoue (a) A tso eviclenbty, on r.) of lu bY iri.lr-rcbion i,hisII1-i,rro;.,er:tY

and I'ragrnerd;s trincluction (h) Transiimii;e

ii".f-$Theoretn" Foi: eacJ:nz, l;,n )'.L, II)ynylc-1 proves |;he following: each h.as a least elemenL in bhe non-empt y En set of ordinals less than c,",fr ordering d. Proof. By inclucl,ion on k. First, for each n, IDnT proves thab each non-empby X,,, subset of (3 c,,'f ) has a least elemenb:this can be proveclby indurction on n. i, ile., since .IX1 proves that (a ,rl) is 41 isornorphic witir The case n:7 the universe of all nLrmberswith (. It also proves ihat cuf is /J1 isomorphic bo the caltesian power of n copies of ihe universe ordered lexicographically. Assnme'we ha.veprovecl on IErn the claim for n and let 'Y 9 w. x '." x (-o.
(n{1) tirnes

X. Such Let a be the least number such that, for some n-tuple u, ((") ^ ") secluences all such since the condibion is En . Let Y be the set of an a e,xisbs s; by the induction asstrmpbion,we can prove that Y has a least elemenb sg Thus (o) ^ sg is leasbin'Y; this completes the in tlrelexicographic orclerir-}g. proof in I E^. Now assume we can plove the theorem for k, n and (* * 1); we show that it hoicls for (ft * 1), n ancl rn. This will complete the whole proof. We proceed E.^,x q (s tut+). Defineafunctionll asfollows: in IElr-p*. Let X *0b.

r(0)ist1reminima1(pg,c6)suc1rt1ratpg(<wft),ao>
contains an element wpoag-l-... . (Exisi;enceclear';minimality is understood le;ricographically,using { and <.) Let rC("): (pr,or) be given; we define

r(r + 1).
F (r-l 1 )

- lF " ' i f b h e r e is . ..*L)tt'clx;{cotL**taa,+ a n d X c o n t a i n sa n e l e m e n t c o t ' o a*O -l1) : F(t). snclr an elementl othertvise F@ and lhe set Y : clearly, such F is well-defi"nedin ID6-p* (it is a*+t) C (5 ,,[). By the induction {pr* | r} is Xr.,r.'1and non-empty; furbhermore,Y f.or some y. Butt ttris rnearnsthal, lrypothesis Y has a ieast elemenb pt;pr,is pr,o n as is the least element of X' blre element 11 Lop" Remark, Lef L(cof,,D*) be the s|atemenb "each non-empl,'YD* subseb of 1, (< r'f) has a least element". Thus we pfovect,fbr eachrn,k,n)

i s th e mi n i rn a l( l,r + r tdt*r ) suchthat p*- p1e ( < o' f) , csal ) 0

I E nt-i-r,:L I L(afr, E*)

I{ieralchy a'nd a-large Sets 3. Schi,vichienber'6-lVaiuer

139

(c) cu-largeSets i.n tr IJ1

very r:seful ancl is aiso appealing ilhe nobion.of an cr-large seb is Lech.nically in its o'',vnrighL. In tire n.exLsubsection (d) we shall shor,vtirat ib is naLLrraliy of tr-rncLions. hierzr,rchy Io the SchwicJ.rtenbelg-VVi'r,irter' reliLi,ecl L , e f ,t 4 b e a f i n i t e s e t a n d . l e t ( o s , . . . , c r , u ) b e i t s i n c r e a s i n g 3.19 Deft,trition" A : 0 then {*}A - c\; en.nmeration; r,ve clefi.ne{*}A lo:: each. a e. l.:fo t h e r w i s ew e p u t . [ a ] ( c r s , .. . , a , t ) - { [ a ] ( a o ) ] ( a r ,- . . , c l q ) ( C l e a r l y ,t h i s c l e f i n e s { a } A a s a t o b a ld 1 f u n c t i o n . ) A i s a - l a r g e i f { a } A : 0 . if ancl oniy if card(A)> x. 3.20 Fl.emark.(1) A is r-large (i.e. r,.,0.r-large) (2) A is ar-large (i.e. arl.1-large) iff. card(A) > min A. i t r t L - ( o o , . . . , c l i )i s { a } ( c s , . - - , ( l i ) - I a r g e ' ( 3 ) A - ( c s ,. . . , a q ) i s a - l a r g e (Evident.) 3.21 Theorem (1) If A is a-large, z ( rninA and cv -+ p then;l is B-large. ( 2 ) I t A , : ( c t o , . . . , a q ) ,B - ( b 0 , . . . , b , ' ) ,Q S r , ' I o t i I q w e h a v e b ; S c t ; ancl A is a-large then B is B-large; (in particular) i:t A e B and A is a-large then B is a-large. 'Ihen that A is (a 1B)-large iff ther-e are B,C suLch (3) Let a ) 0 ) 0. A: B U C, max -B < min C, B is p-large and C is a-large. ( ) Leba 2 I, A: ( a 0 , . . . , , a , t )c , t s) 2 , a o - r o then Lhere is an loO,oq) is decornposed into as intervais). If A is c,.,o-large' i < a o s t r c ht h a t ( * ; , * ; - r t ) o A i s 6 p { o } ( o o - t ) - l o t g " . (Here

(r;*;+l : {, I r; 1 z ! r;+r} .)
(5) If Ai s c . r f l - l a r ga en ,d a0,U) l thenboth (or"',no) and (cts,...,rrg) ( o 0 , . . . , a c r - - r )a r e . l - t - I a r g . .

e : m i 1 . A .P t L tc u ,:i { " } ( o 0 , . . . , (t) B V 3 .1 5 (B ), w e may asstlme Pr o ofs. : a r g o : 0 , a q 1 l: 0 , a 0 a ; - t ) , 0 t : { 0 } @ 0 , . . . , c 1 i - 7 )t;h u s c v o e0o get ai we similarly thus a1 0;., l,hus By 3.15 (11), ar i ;t 0t; ;: Pt a ( t + . l. > g q * L i . e .0 : c \ q 1 - t Z \. qat:0.
q n

(2) Let A : a; for 1 ui-rr : {*;}(";)

(.,0,..,aq), B

( b 0 , . . , b , ' ) ,r

and similar:ly for B. First, a1 :

fu (for cLo

140

Il. Itragmeni;s and Combinatolics

a;+t a;+1

p ;-rt, d -l -.t 'i 1 1 I:f . l3;+t :0

,[,f, : { c t ,r - r } ( r , ; + r ) ; r , p ; - v z . l l h u s in any ca,se b y 3 . f 5 ( 1t ) a ; - p 2 {a}B:0. n - ( c t s , . . . , a q )t o b e ( a 1 - 0 ) ( 3 ) l , e t a , 0 ) 0 , a ) B . F i r s La s s u m e F l a r g e .P u l , . , \ ; : { * [p]krot'.'tc;i-r). We have l 3 } ( " 0 , ' . . , & i - r ) ,0 i : f. Th.en\r n: Q, sr - Lch - 0 ; L h .u .s, r nbe r ninim.al i , 13;:0. .[,eL L o r so m.e t r o - pt a;nd(ct*,'..,aq):L'isa-large' ( r , 0 , . .. . , & n z - L ) -B i s B - I a r g e let A. : B l.JC', m.ax.B < min C, Let,.B be p-large, C' a-1.atge; Conversej.y, 0, {a -l P}(B) : Q, . hen {p}(B): a s s u . r rtlh ea t . B i s t h e l e a s t p o s s i b l eT

then { t} ( n) :0; \f l3u- t ) 0 l.hen : 0;+ z and a;- p1 > 0+ z ; g ;+t,;-',{0 ;-rt}(" ;+r ) ;,{ 0;- r r } ( b;- r r )

{a -t 0}@ u c) : {a}(c) : 6.
( < t ) B y 3 . J . 5( 1 0 ) , u "

t h u s b y 3 . 1 5 ( 1 - 1 ) ,{ c , . ' * } ( o o ) il ;:1-s{"}(oo-t).ao; (no) is {c,.r*}(oo)LhenA p { " } ( o o - r ) . a o. I f A . : ( o 0 , . . . , a , r ) i s a . , o - l a l g e large a,nd bherel6ts,r'[o](cs-1)as-large.By (S) this means that .A-(as) mav be clecomposedinbo 81 ,,...,,8o that are mutuirlly disjoint and such {,l-r,a'r,t B; is p{o}(oo-t)-1otg". Thus if we have the maxB;.( min B+I u,ncl "u,.h ro decomposil,ionag (r;,*;-rtl must contain B; ancl theretbre is gp{o}(t'o-1)-large. inber.i/al L o a * - 'w e h a v e ( 5 ) t r v i d e n t l y , ( a 1, . . . , a , t ) i s { o f } ( a s ) - l a r g e ; s i n c e , H ;: (r)__ ...',aq) { c , r f l } ( a 6 )d , 0 , - t ( 3 . r 7 ) ; t h e r e s u l t f o l l o w sb v ( t ) . F o r ( a s , "t"

(d) Schvrichtenberg-Wainer Hierarcfryi

S.22 \Me strall investigate the hierarchy of nurnber theoretic f.unctions cLefinecl informally as fbliows:

/b(') : r -l.L / * + r ( " ) : / i ( z - F1 ) + 1) for ) limit ' /.r(") : "f1.r11';(r ( e; f*o)means r-thiterationoff ,i.".f0(A): llerea,,\ varyo,rerorclinals bv investigated : f(f'(y)). This is a variantof the hierarchy ,, 7:v't'1(y)

from bwohierarchiesinvestigatecl and'Wainer and the second Scfir,vichtenberg this hierarchy is 1\1 defrnable that by Solovay and Ketonen. We shall sho,,v conclitionssufficientto prove in -fX1 as a hierarchyof parbial functions,sho.,v sets (cl. thal a givel function /o is botal and relal,ethe hielarchy to cv-large Theorern3.30). (1) Ib is easy boshow irL IE1ihab if -F is a total one-algumeuf 3.ZB.[i.ernarh" [uncLionClsuchthat -41 tunction then there is a uniquetol,altwo-ai:gumenb r,y. Ci(r,y): Fr(y) fbr each

3. Schwichtenberg-Wainel l{ierarchy and a-lalge Set's

141

(2) SimilarIy, IDl proves that if g is t-r, finiLe one-argum.entfunction then there is a 6.nibel,wo-argumenLtuncLion q/ which is bhe maxirnaL fnnction such o r e a c h ( r , y ) e d o m ( q ' ) , q ' ( r , y ) - q ' ( y ) . E a s y p r o o f ' sa r e l e f l , ' r o t h e t h z - ufL reacler. (:3) We sha.ll worlc with. fi.nite two-argtrrnenl, ltt.ncLions clefineci for so'rne pairs (*, *) where a e and r is a nurriber. If q is such a lirnction then qo will be the unique function such that for all rc, c1o(r) is defineclift q(a,z) is definecland then Ar@) : g.(a,r). (lieedlesc to say, "r1o(r) is definecl" means r e rlornqo.) 3.211 llefinition. Define a predicabe WD(q,d,r,A) (read "g is a derivation of A" or, pedantically, "g is a clerivation of the fact Lhabthe v'altte oL r in f ("): the a-th function in the Schwichtenberg-Wainer hierarchy is y") as follo'',vs: cu; (1) q is a finite function, dom(r1)C e ><

(2) q"(x) -- v

(3) whereverqp(z) is definedthen (i) i:t z > 0 then qy! - 1) is defrned, and equal to qBQ), (ii) i:f B: t +'I then q](z * 1) is defi.ned Io qBQ). (iii) if B is a limit then q1p11,7Q + 1) is definedand eclual (1 ) WD i s A t.Q) If 3 . 2 5Le rn mu " W D( q,a,r ,Lt) , qp( z)is definedancl B -> 7 (3 ) WD( q,d,r ,!J) and W D( Q|,e,r ,yt) im pliesy : lJ t. t h e n q r(z) i s d e fi n e .l .

can Proof (in /I1) is easy.(1) Note that e is 41; all "is defined"-c1uan.tifrers that "u be bounded by q. It remainsto olrserve f ,zt,x,y. f'(y)" iu /1 i.n. : use (iii) and (i); if (2) r\ssume1 13is 0o + I {0}@. If B is limit then n tlren {B}(z) : 0o and use (ii) and (i). 3.26 Lernrna.Assume WD(q,d,x,y) and.qp(z) defined.Then ( 1 ) q B Q )) z - F1 qBOu)<(lp/) (2) w<z implies i m p l i eq seQ)>q^,(z). (3) z)0,0 7 Tand A*t (1) & (2) U (3) bV induction on B running over' ProoJ.Prove sirnulbaneously tl Ihat q, 10. all 7 surch
3"2T Cottollary" LeE IVD(q)&.)rjU), a Q.O,,u and g ( u. Then the resl,riction c1t o f q t o O u , > <( S V ) s a L i s f i e sW D ( q ' , a , r , U ) .

Proo!. Verify the conditions3.23 (i), (ii), (iii) for t1tby induciion on 0 e 0,,,, tr u s i n g3 .2 6 (1 ),(2 ). 3 " 2 8D e f i n i t i o i tA . : f .(r) ift (:r/)WD(r,a,:c,y). fr,y)l -: {zlr 1 .z < y } ; l ( " , y ) l: { . r l r < z < y } ; s i m i l a r ' l[y (t'yl .

1,,L2

II. Fragments ancl CombinaLorics

a parbial l,.wo-argrlmentfunction. ObS.29 It,emark. (:1.)C.l.earlyl,his de.fi.nes serve that 3.2,'l implies that th.is funcbion is ,1-,-\1. clefi.ned: Nol,e also Lhal,by 3.26, Lheluncbions satisfy l,helollor,vingrnr.lrenever

(ii) r.o( z implies Jo(r) .: l'"Q), j T=B impliesfe(r) > f^,(r). (iii) z >, 0, 0 : L(wft,Ez). Using {,hiswe (2) Recail l,hat,fbr:,k ) l. ancl any n1 lDufll : supr { ' y} ( r ) . ( ' Ihis is a m a y p ro vei n /,1 6 1 r th a L for eachlim it 7 - .cofr .,"l the leasL consicler -[2 condition on 7; for its proof, .L(':ft,82)is sufficienb: 7 -<tuT, each a proves that:tbr < o'bserve that IDtr+t Lhis.) Also satisfying not 0 *. (This again is II2in 0.) Llrere is a z.such that 0 : 3.30 Theorern. For k > 0 anclany n, IL'61l provesthe foilowing: (1) For each a \ wft, /o is tobal. ufr, th.ereis a z such that (2) For each a -< B -'1

( i )f " ( , ) Z z - r ' 1 ,

. (V. > z)(f < f B@)) "(u)

(3) For each a -1 cofrand each x, fo(r) is the least gt such that the interval l*, y)) is c.ro-Iarge. by applying n times Remark 3.23. Fbr /t > 0 Proof. (1) For k : 0 this [ol].ows obserwethat the sLabementin quesbionis II2 so that L(rT, X2) suffices to prove it for all a -( cuf (using Remark 3.23). *; by 3.2,9(1) (iii), this irnplies a z such that B (2) Bv 3.29 (2), ter,ke 7 * and bherefbre r.u ) z then, by 3.15 (B), ,,vehave B < f '(r).If f ;l "(t)

/ "( . r ) < f p (*). (3) lVe sha1lproceed by induction on a < ,T; observe that assuming in questionis LI1. The casecv:0 is totality of all f ,, a <ru\, the assertion clear. Clai,m.Assttme the assertionof (3) for a; then fbr each y > I and each interval beginning wibh r. Proof r,, lr,/H(t))] is bhe minimal (cu".y)-large is 41 in y). For y : 1 this is our on y (ihe present asserbion by indr-rcbion for y-1. The"ly, r)l is oo.g-large assumption the present assume assumption; ( b y 3 . 2 1( 3 ) ) i t r ' > i f Ff / * ( r ) , 2 ) l i s , u o . ( y- 1 ) - l a r g e fH-'(/"(")) : fH(*). This proves bheclairn. 1. For r : 0 we easily see LhaL Continuing the proof of (3), considercv-l-. = 0, thus the one-elemenL set (0) is c,.,"-Fl-large. and {cud-l-l}(0) /-+r(0):1, iff l" + 1,2)] is c ) 0 anil use the claim: fu,2)) is cuoi-1-12'urge Thtrs assu.rfre -F 1) : l"+ r ( c) . i ff r . , , ,o .z-i a l g e ,z > f|(r iff l:"--lI,z)) IL rernainsbo considera.being limib. Then l*,") is cu''-large ' iff z ) ft*y1*;(r + 1) : /.,("). his completesthe pi'oof. n 1sse{o}(')-1arge

3. Schwichtenberg-War,iner l'lierarchy ancl e-lzr.rgeSets

143

3.31 flemalh.

The reacler may verify the follor,ving as an e,xercise:

- 2Y (r -F1) - 1 fYt@) fz(*):2*(x+2)-1'

a, rjz) z: J*@) ifl z is 3.32'Iheorem. -II'1 provesthe following:fbr eacl.r minirnal such that l*,r)l is c,.,o-large. (Observe l,hat we do not claim that ./,(r) exisbs,but we claim. thai z/ is c,:o-large and .z is mi.nimal wiLh that and ecluals.z then lr, exisis it ")] ancl z is minimal property; and if there is a z such that lr,")l is c,,o-large z. OtLrproof is an inspecl,ion with this property then /r(r) existsand ecluals o f t h e p ro o f o f 3 .3 0(2 ).) ProoJ. Let a e Oq and r, z < q ) 1. We prove by incluction on a e Oq t'he following 41 property of a: . f "(*) iff z is minimal suchthat l*,r)) is c,.'*-lerrge) This is clear for a - 0. Assume (i.) for cvand let a -l- 1 Oa. C l a i m. F o r a l l L < y 1 c!, z 1q, z: fH( *) iffz is m inimal suchthat l",t) l .y)-Iarge.(Seethe proof of 3.30.) is (c,"'o r ) I. By the claim, 1r,,")l ir.o-F1-1trge iff l" + 1,2)] is We may assr-lrne wo.r-Iarge iff z > f|(r -F 1) : f o-rt(x) - u. in 3.30. Similarly {br a being tl limit. 3.33 Definition" Lel (ItV)",be the formula (Yr, z)(1y)(l",yl is c,.,fl-large) (the principle of ordinal-largeintervals). 3.34 Facts.(I) (W), is a J72-formula. ( 2 ) ID !ts (V u )((V V ).: (V") ( V' - < Ll' i- ) ( /, i' total) ) (3) For eachk ) 0, -[X61.rI QV)k and, for eacltn, I En-t F (Vr)(]y)([", y] is of,-ur-large). ((2) follows by Theorem3.32.) 3.35 Corollary. ID1 provesthe fol1o'',ving: (V"X Con(I D" t Tr(tI1)) -' (t,Y)") . statement This follo''vs from 3.30 (formalizedin /tr1) exactly as the analogous i n 2 .I'7 . 3.36 Theorern.For eachk, i'2,1 proves(Ptr)7, = QV)U thlrs it proves(W)p = C o n (.t D p + T r(II1 )). (,r) (Yz < q)(V* < cl(z:

'144

IL l\'agrnenbs ancl Com.binatoeics

Lhanhs1,ot.[re Commen't One implication (easy) is 3.35. For the converse? to show in If1 the follo'rzing: main resulb of Seci;.2 is enor-Lgh

('r)

-, (Pil),') , (Vu)((W),,

or,, at least, Lo prove each instance of this. Here we have two possibilil,ies: (a) Sol.o.ra.y pnoved th.at, fbr each li > 1, c ) 2, b t a )1 ?:, ar:.cl.K.etor:]er:t thab th.eir-proof (k I \!+1. If or." checlcs then [o, b] if 1), a1i, ,r,_1-i5-lorge ? worlrs in IE1 (which we expecL but ha-re not cheched.)bhen the irnplicabion (,r,)is provecl. (b) Paris has a moclel-theorel,icproof of (lV)p -, (J'.tI)p (for any stanilarcl /c). lVe shall elaborate il, in Chap. IV (see 1V'3'37). I D1 | (W)" - (PII), or, at 3.37 Froblern, Fincl a reasonably simple proof o'f. l- (Yu')(W)" -' least, I Dt F (Yn)(W)" -' (Vu)(P.H)") or, albernativelv,I \ paper dispensable? Cort\.PA { Tr(II) ). Are itetails of Soio.ray-Kebonen's