This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
paarweise
mit
q _> o
Folgen
Peirce-Datum fur v versteht man ein Tripel
~ =
und den folgenden
verschieden
141
n,
~ = <l
Eigensehaften:
und alle Terme erster Art;
die
jedes
n,
Glieder
P
~ =
a n v o n _a s i n d
ist eine Peiree-Form, n
und t n ist
ein
v o n Xp
verschiedener
Peirce-Sobterm
yon tPn
; jedes
gn ist
ein
Endo-
n morphismus
von ~ mit
]* = v ,
gn[Xp
g n [ t n]
= an
falls
[Xp
term von tpn] oder
[Xp
n
L-uniform
und t n R-Subterm
von tp ) , gn[tn ]* = a n n
n Z-uniform
und t n Z-Subterm
u n d t n L-Sub-
R-uniform
n
van
tp
, Z = L
oder
Z = R . Ist
c ein
falls Xp n
weiterer
Term aus
R,
n
so versteht
man unter der
, wobei
~'
Transformation ven
=
n
dureh e
und g~ sich van gn
das Peirce-Oatum
nur for die Variable
Xp O
unterscheidet
und g~[Xp
]* = c
gilt,
und welter
~ =
nund
g~[t n) = b n
n
respeKtive
g ~ [ t n]
Nunmehr
erkl~rt
und vorgegebener
[vS] [v~]
und
[vs)
gemOss den
angegebenen
man Mengen 2 yon Beweisen
Menge P van Peirce-Formen
EOtlen
gilt.
bei vorgegebenem
w~rtlich
wie zuvor,
Axiomensystem verschOrft
Beweise
aus S
der FunKtionen , so liegen
F 5 , FSo
, F51
die S , <sil
auch die Funktionswerte
von S , S ~ = S l e ~ , S I = S l e I . Sei
So[e o] = Sole I) = I , s3[eo]*
= s3[e I] , s3[e o) = v , yeA .
Oann existiert
zu jedem w aus s4[e o) ein e W in N[e I] und ein Beweis
dass S W das SKelett
[s1[e W] - { s 4 [ e l ] [ e w ] ) ]
v
Sei
, sei
nun
[i,e, und
ausserdem
vEL
q > o ] Peirce-Oatum sei
die
von Sle W
{w*}
S W in
und das Ergebnis
hat. D e = s l [ e o)
for
v
Transformation
gebe es for das zuvor bestimmte
oder
i
in S
Sei Se2 , s e i e n e o, e I Knoten
derart,
aber
wie folgt:
Sind von den Argumenten und S ~
[vsP)
= bn
Ist
-
{v}
, Sei
D nR o
nicht
leer,
von
durch
c
e w in N[e 1] einen Beweis
[O o
(c}] W [sl[e w]
im[~]]
[D O
{c)] ~ [sl[e w)
[{v*) V
V
sofern
ein
nicht
so s e i
, Zu j e d e m
leeres
c daraus, w~s4[e o]
S~ in ~ mit dem Ergebnis
im[~] im[~]]]
V
[{c) ~ im[~]]
v ~ S l [ e w]
,
,
142
W.
Felscher
oder, falls es kein c gibt,
oder
O~
[ s l [ e w)
O
(sl(e
0
W
im[~))
]
-
({v*}Vim[a)]J
sofern
--
Oann gibt es in S auch zu jedem
wes4[e
--
von S~ hat
seheidet,
)
v e s l [ e w)
einen Bewels S %W , der das Skelett
0
und dessen Ergebnis sich yon demjenigen van S~
d a s s d a s E l e m e n t s q ( e l ] ( e w)
entfernt
dadureh unter-
und d u t c h w
ersetzt
worden i s t .
Oamit kann man auch noch eine Funktion F6o erKli~ren, indem man for S ~ und S%w aus [vs P) als F6o(S~
~)
einen der Beweise S %w
w~hlt.
Ein Beweis S sell sohlank heissen, wenn for keinen seiner Knoten e met
s [e) = 1 o
ein Nebenterm zugleich Parameter ist, es sei denn, dass der Typ I vorliegt, for den P ein Nebenterm aus R auch Parameter sein daz-f. Es folgt aus [Vs), dass jeder Beweis S aus einer Menge ~ von Beweisen mit Hilfe von F 5, F 4, F 3 transformiert werden kann. o Ivon
S
: ist e m i t
in einen sehlanken Beweis in
Oas sieht man sogleich dutch Induktion Ober die Rangzahl
So(e) = 1
der Kleinste Knoten von S , so kann man (v~) derart
auf die oberen Nachbarn von e anwenden,
dass Keine Nebenterme mehr Parameter bleiben,
und danach kann man eventuell verloren gegangene Parameter dutch AbschwBchungen wieder hinzufOgen.
Oamit formuliert man das
Eliminationslemma mit Peirce-Daten. Typs;
fixierten
und s e i
~ abgeschlossen
Knoten yon s l [ e e)
S
,
S~
v o n S~ '
Datum f o r
v
positiver
Lange
Olm R Gibt
sofern
j
veO
,
S1
gilt
0
Menge v o n s c h l a n k e n
des
Typs I
gelte
die
= Dl ~ { v * }
nicht alle
so se•
, auch S ~
o
Glieder
die
c in mit
, so s e i a
~ aus
v eD 1
dieses
von a seien
n
--
Transformation
Oln R
,
Schnittr~ngen
so g i b t
eines
Sei
, Sei
d ~1 < j
,
Beweisen
L•
Gj
sei
, nicht
Element
,
do ~ < j veO
das
Funktion
, s 1 = sle~
gelte
q j
es e i n
Seine
gegen
sl%
=
= DO , s l [ e 1)
o , d~ 1 o~
aus
im Falle
Sei
Se~ , s e i e n e
R ,
Ivl
= J
. FOr die , Sei
D
0
yon
, Ist
o, .
e1
Peirce-
leer,
i,e,
c ein
yon j
@
Sei
ein
, die
yon
Element
dutch
es i n ~ B e w e i s e
unterhalb
<
SchnittrSnge
Datum n i c h t in
j
c
,
S ~ und, in
der
W.
A b s c h l i e s s u n @ von S ~ , S 1
143
Felscher
unter den F u n k t i o n e n F 1 ..... F6o, Gj
liegen
und die die Ergebnisse
(~]
[o
im[~)) v im[b] v (01
o o
haben.
Ist
v
[{c}
-
{c})
u im[~]]
respeKtive
C{v} v
im[~]]
u D1
01mR
leer, so gibt es in S Beweise S ~ und,
sofern
veO
--
auch S ~} S~ , S 1 (~]
[0
C~]
o
mit S c h n i t t r ~ n g e n unterhalb von j , die in der A b s c h l i e s s u n g von unter den FunKtionen
F 1 ..... F6o, Gj
llegen und die die Ergebnisse
im[~}} v 01
o
Co e
[{v}
respektive
v im(~]]
u 01
haben.
Die Beweise
S~ , S1
und deren Rangzahlen noch allgemein-reKursiv.
W~hlt man hier im Falle
veO
Ergebnis wie Gj+I[v,sO,s1]
Sw , S~
und ihre Rangzahlen berechnen sich aus den
das leere P e i r c e - O a t u m <~,~,~>
o
, so hat S ~
dasselbe
.
Die Beweise S ~ und S ~
werden reKursiv Konstruiert.
Man nimmt zun~chst an
dass die Konstruktion for alle Beweise S 11 mit einer Rangzahl Beweise S ~ und for alle P e i r c e - O a t e n
o~ I < ~
for alle
ausgefOhrt sei; alsdann nimmt man in einer
zweiten ReKursion an, dass die Konstruktion mit S 1 for alle Beweise S ~176 mit einer Rangzahl
o~ ~ < o F
und for alle Peirce-Oaten ausgefOhrt sel.
e Sei zun~chst o I = o .
O~nn ist O ~ ein Axiom,
und da die Terme a n elnes Peirce-
Oatums positiven Grad haben mOssen,
Kann nut das leere O a t u m vorliegen.
I oI = o
schliesst man wie im Falle
[a) des @ e w S h n l i e h e n Eliminationslemmas.
o1 I > o
wendet man die Konstruktion
fern
s 3 [ e o)
gew6hnlichen werden s o l l e n .
oF > o Oberhaupt
. Man b e t r a c h t e definiert
Eliminationslemmas, Im F a l l e
(ba)
ist.
daraus reKonstruiert man S ~
zun~chst
den F a l l ,
Diese Situation
dessen Unterf~lle
schliesst
Im Falle
au~ S ~ und die P a r t i a l b e w e i s e von S I an,
(w
die in den oberen Nachbarn von e I enden;
S e i nun
Im Falle
dass
entspricht
vermBge
(v~).
v ~ s 3 I e o) g i l t , dem F a l l
(b)
sodes
und B e z e i c h n u n g e n
hier
Obernommen
man genau w i e im g e w B h n l i c h e n
Fall,
w o b e i man
in der InduKtion das v o r g e g e b e n e P e i r c e - D a t u m verwendet.
Ebenso verf~hrt man im
144
Falle
W.
[bba]
, sofern
a n = s3[e o] gibt.
[bbaa]
Felscher
, n~mlich nieht vss4{e o]
, gilt und es
auch kein a n mit
Tritt aber ein an mit dieser E i g e n s c h a f t auf, so betrachte man
die v e r s c h i e d e n e n anm , die in s4[e o]
liegen und v o n d e r
Gestalt
anm = gn(tnm]
for
direKte P e i r c e - S u b t e r m e von t n sind; da S ~ schlank ist9 sind sie von allen a n , versohieden.
N u n m e h r vergr~ssert man das gegebene Peirce-Oatum,
anm der Folge _a hinzufOgt, noch g u m
lauter
gleiche
ebenso _P um alle
gnm'
gnm=
diesem neuen P e i r c e - O a t u m w •
gn "
Ist
indem man alle diese
verl~ngert
c vorhanden,
wie im frOheren Fall
und schliesslich
so s o h l i e s s t
[bba]. Ist e nicht vorhanden,
so wendet man die I n d u k t i o n s e n n a h m e ebenfalls wie frOher auf Sleoo warden im Ergebnis von S' alle a t EA n
und alle a
n
nm
entfernt.
9 folgt aus der Oefinition der Peirce-Formen,
Wegen
nm
Man betrachte nun den Fall
Beweis mit dem H a u p t t e r m s3[e o) erh~lt man S w respektive S ~
[bbab), n~mlioh ves4[eo], a n = s3[e o)
9 so wendet
[v~)
liefert dann einen
9 wobei v als N e b e n t e r m verschwunden ist. Aus •
, indem man v oder c gegebenen Falles dutch A b s c h w ~ c h u n g
Tritt aber ein a n mit a n = s 3 [ e ~ ] auf, so v e r g r ~ s s e r t man das Peirce-
Oatum auf dieselbe Art wie zuvor. die Konstruktion [v~]
Oaher hat man somit
[~) w i e d e r auf Sieoo und Sle I an; im Ergebnis van S' liegt dann
v nebst aller weiteren Elemente van sq[e o) . Anwendung von
hinzufOgt.
a eA , also auch n
auftreten.
sodass der Typ I nicht vorliegen Kann. Gibt es Kein a n mit man die KonstruKtion
, Sle I an; dabe•
zusammen mit der Termsehluss-
Bedingung 9 dass ella E l e m e n t e aus s4[e o) unter den a bereits S w oder S ~w erhalten.
man m i t
Ist c vorhanden,
so wendet man bei der Induktion
[w167 mit dem neusn P e i r c e - O a t u m an; alsdann rekapituliert man vermSge
den ursprOngliohen Schluss, wobei der N e b e n t e r m c verschwindet;
ung mit v oder c erh~lt man daraus S w und S ~ bei A n w e n d u n g der Konstruktion
[~] s~mtliche a
sodass man damit bereits S w e r h g l t j
dutch Abschw~ch-
. Ist c nicht vorhanden 9 so entfellen o
und a
nm
, die von v v e r s c h i e d e n sind,
b e i Anwendung d e r K o n s t r u k t i o n
[~w
entf~llt
Oberdies v , und man erh&It S ~w
Im Falle v = s4[eoJ[eo•
(bbb] betrachte man als Unterfall
(bbbaJ
, dass es Kein i mit
gibt; dann ist die frOher in [bbb] erkl~rte Menge 12 leer 9 also 11
nicht leer. Gibt es kein e n mit
an
s3Ce o] , so wendet man die I n d u k t i o n s a n n a h m e
mit dem gegebenen P e i r e e - O a t u m for alle i , i
W. Felscher
145
verwendet man zur Bildung von S ~ Oberall die KonstruKtion S~
0berall die KonstruKtion
verm6ge
[v~]
.
(~]
, und zur Bildung von
[~j] ; aus den entstehenden S~l erh@it man S w und S w
Tritt ein a n mit
a n = s3[e o)
auf, und 1st a u s s e r d e m e vorhanden,
so vergr~ssert man das P e i r c e - O e t u m und schliesst wie seeben. so w~hle man eines der anm , und dazu findet man ein • mit
Ist c nicht vorhanden,
anm = Sq(eo)(eoi)
Bildet man unter V e r w e n d u n g der I n d u K t i o n s a n n a h m e nun des zugehBrlge Si , so ist dort a
nm
entfallen.
Oaher erh~it man S w und S w
von e ~ , die nicht in s2(eo](eoi) ist zu bedenKen,
auftreten,
, i n d e m man allf~llige P a r a m e t e r
dutch A b s c h w ~ c h u n g hinzuf0gt.
Hier
dass der Typ I wegen des nicht v o r h a n d e n e n c a u s g e s c h l o s s e n ist;
liegt der Typ II vet, so muss v der einzige in R gelegene P a r a m e t e r gewesen sein, sodass auch dann alle jene A b s e h w ~ e h u n g e n mit Elementen aus L geschehen. fall
(bbbb)
bleibt die MBglichkeit,
nehme z u n ~ c h s t
an,
dass
es e n t w e d e r
dass Kein
v = s4(eo)~e an mit
.) el
for ein i gilt. Man
an = sl(e ~ ) gibt,
an = s3[e o] gilt, jedoch XPn nicht d i r e k t e r S u b t e r m yon tn ist. so verf~hrt man wie im Falle vor,
[bbab].
Als Unter-
oder
dass
Ist c n i c h t
zwar
vorhaqden,
Ist e vorhanden und liegen die Typen I I o d e r
so wendet man bei der Bildung der S~ l
die Konstruktion
und zwar for
a n ~ s3[e o] mit dem gegebenen,
Peirce-Oatum.
Oamit bleibt v e i n
Nebenterm,
for
[5] for alle i , •
a n = s3[e e)
und verm~ge
III , an,
mit dem v e r g r B s s e r t e n
[v~) f0hrt man dann s3[e o]
wieder als H a u p t t e r m ein, wobei v verschwindet;
man erh~lt S ~ und S w
v oder c w i e d e r durch A b s c h w ~ c h u n g e n einfOhrt.
Liegt der Typ I vor, so wendet man
bei der B•
der Ss
for i mit
v = s4(eo)(eoi)
die Konstruktion
, i n d e m man nun
(5} anj for
die restlichen i , bei denen v noch als P a r a m e t e r zu e . auftreten meg, verwendet man 01 die Konstrukt•
(5) falls S w gesucht ist, und die KonstruKtion
sucht ist; aus den entstehenden S i die MBgliehKeit,
dass
c nicht vorhanden,
a n = sl[e o)
[w167 falls S w
erhalt man S w und S 55 vermSge gilt
(v~} 9
Es bleibt
und dass v als eines der a nm auftritt.
so verf~hrt man wie im Falle
III vet, so wendet man bei der Bildung der S~ for alle i die i
KonstruKtion
an; dann l~sst man bei der folgenden A n w e n d u n g von
term c verschwinden und erh~lt S w und S 55 dutch a n s c h l • der Typ I vet, KonstruKtion
so wendet man be• der Bildung der Ss
(~]
Ist
(bbab). Ist c vorhanden und liegen
die Typen I I o d e r (w
ge-
for i mit
anj for die restliehen i verf~hrt man wie •
(v~} den Neben-
Abschw~chungen. v = s4Ieo)(eoi)
Liegt die
ersten Tell dieses
14@
W. Felscher
Unterfalles
{bbbb].
Es bleibt ein weiterer im gew~hnlichen obere Nachbar
Eliminationslemma
Peirce-Schluss
Peirce-Oatum
vermOge
so sei
Endomorphismus
hlxo]*
oder
Z = R . Oaher ist
und Fs1
Z-Subterm
von tPn
einem neuen Endomorphismus = h(t o)
So(e oj bestimmte
und Xp
Man vergrOssere
weshalb
falls
von s4(e el entfernen
S 5 ebenfalls
= <3,1>
Peirce-Form
und beh~It
Z , Z = L
f{Xp
) = v , f{t n = n n Peirea-Oatum dureh s4{eo), auf $1e' und Sle I an. Bei
for
f'(t n)
ausgefghrter
InduKtion
Peirce-Datum
sowie
der x O in
h'(t O] = f'(to(tn]).
also b n
evident.
Gilt
um die durch
auf SIe' (v~)
wendet
,Sle I
Sei nun
So(eo]
Zur Konstruktion
g j alsdann
verm6ge
ist s3[e o)
h(t el = s4(e o) ,
so folgt
v ~ ss(e o} ersch6pft.
Induktion.
um s4(eo)
Peirce-Schluss
f', der sich aus h
for 0 und h' das Transformationsresultat
nut noch
P und den Endomorphismus
mit diesem vergr6sserten
Tell von gn zu
= sq[e o) . Ist nun h' der Endomorphismus,
, so ist die InduKtion
Peirce-Datum
nochmaligen
Wegen
mit demselben
von f ein Endomorphismus
und sonst mit h Obereinstimmt,
dutch eine evidente
das vorgelegte
for den dann
f'It n) = g~{t n) ; n a c h
sind alle M6glichkeiten
So(eo)
= s4(e el j seien
n
nun das gegebene
Kann man dutch einen Peiree-Schluss
Oamit
in tp
und s4[e o) durch f'[tO{tn ]] falls
h(t O)
f'(t n] , also b n * , sendet Mithin
Z-uniform
se• h der
Peirce-Oatums.
und wende dann die InduKtion
an die Stelle
durch f'[tn )*
dutch f'[tO(tn))
h{to]*
Teil yon h mit dem auf t n wirKenden
und f
tritt
zusammensetzt,
daher ersetzt
oder
{sodass der Typ III
Peirce-Fermund
Teile des gegebenen
% zusammenfassen,
.
<,tO[tn)>
und gnJ
s3(e el = a n
eine Peirce-Form, deren Auftreten in Beweisen n Oa die Variablen von t o und t n voneinander verschieden sind,
Kann man den auf t o wirKenden
der Transformation
man die InduKtionsannahme
an und fOgt dann den gegebenen
an. Gilt aber
h(t O) = s4(e o)
der
So[e o) = 4 ; se• e' der
n
ist.
f(tO{tn]]
n&mlioh
und SIe I
die durch
= an .
v ~ s3(e ej zu batrachten,
der a n , so wendet
n
aus ~ zul~ssig
8ioeo
auftrat,
Sle'
die zu a n geh6renden
ist dann t
Gilt
nicht
auf
O =
mit
und gn
(v~]
a cR n
und
des Falles
van e ~ . Ist s3(e el haines
mit dam gegebenen
vorliegt),
Spezlalfall
v = ss{e o}
= 4 , so erhalt
von S ~w
s4(e o) = 4
vergr6ssert
. Es ist klar,
man
bestimmte
man die KonstruKtion an und erh~lt
man
daraus
dass es genau
(~) S w167durch dieser Fall
W.
Felscher
147
ist, der die EinfOhrung yon Peirce-Oaten Oberhaupt notwendig macht. Es bleibt nun der Fall
sole o) = I
einzugehen.
zu behandeln.
Oabei wird es nGtig, auf das Verhalten yon sole 1)
Bezeiehnet man die oberen Nachbarn von e I dutch elj , so wird man ver-
suchen, die InduKtionsannahme auf Sle ~ erhaltenen Beweisen dann S w und S ~
und die S]elj
vermSge
(v~)
anzuwenden und aus den so
zu rehonstruieren.
Ergebnissen yon S I heine Peirce-Oaten zu transformieren solange unproblematisch,
Da in den
sind, bleiben diese I n d u K t i o n ~
wie das Element c aus
D I~ R
auch noch in allen Sl[elj)
auStritt, respehtive solange, wie zusammen mit
01 n R
auch alle
sind.
Man nehme daher zun~chst an, dass diese Verh~ltnisse vorliegen.
struktion von S w und S ~
leer
Oie Rehon-
aus den durch die Induktionsannahme gelieferten Beweisen ist
dann evident, sofern nicht ein,
s1[elj)n R
So[e I) = I , s3[e I) = v*
gilt. Tritt dieser Fall aber
so entspricht die Situation dem Fall [db) des gewBhnlichen Eliminationslemmas,
Im Falle v~A
Kann man dle dort verwendeten Bezeiehnungen Obernehmen.
Kann man das Ausfege-Verfahren
nicht sogleich auf Sleoo anwenden,
erh&it.
Im Falle
(dbab] ist S[eoo ebenso zu ersetzen;
induKtive KonstruKtion der Sw%, S w% sogleieh so einrlehten, dabei transformiert werden. zu transformieren, F~lle zurOck.
Im Falle
und den Fall
[dbba)
dass v mit v*
[dbaa]
ersetzt man die
dass aueh die Pe•
sind in S' ebenfalls die Peirce-Oaten
(dbbb] fOhrt man wie frOher auf die vorangehenden
und der Index o m i t S W = S]ew
frOheren
S %w
(dbba)
durch Beweise S w %
wieder die S % w
und fegt dann aus. Im Falle , indem man S' aus
und S %W
betrachtet.
entspreohend vor; zur Bildung der S B w je nach dem, ob bestimmten S %w
ves2(eoB(e w]
(db) so zu
dem Index 1 vertauscht wird. Im Falle , welche die InduKtion aus
Sw und Sle I liefert; alsdann betraehtet man start der alten S w gelieferten
{5) aus Sleoo und
ausserdem muss man die
Gilt vsB , so sind die frOheren Bezeichnungen im Fall
~ndern,
[dbaa)
sondern muss an
dessen Stelle einen Beweis setzen, den man mittels der Konstruktion Sle I
Im Falle
(dbab)
Sle ~ , Sleqo
die dutch (vs P]
verf~hrt man gem~ss dem
honstruiert und statt der S w und S w
Im Falle [dbba]
geht man dem frOheren
verwendet man die Konstruktionen
gilt oder nieht. Auf diese S~
wendet man dann das AusTege-Verfahren
kombiniert man jetzt S' aus (dbab] mit den S %w
oder (w
und die vermSge
mit Slelo an.
aus (dbba).
(~)
[dbab]
Im Falle
(vs P] [dbbb]
,
148
W.
Es bleiben das Auftreten formation und
nun die Ausnahmef~lle
anderer
der Peiree-Oaten
Sole I] = I
liefert
Elemente
oder
Vorhanden
fOhrt.
sffnd. Vermittelst
Iv P]
vermSge
sind v und v ~ entfernt (v~)
vorliegenden
sl[eljJ~R
der InduKtion
wieder
weiter
betrachte
ist, w~hrend
dass nicht
CSSl[elj)
an, dass
Peiree-Oatum
man den Fall,
es aber wenigstens respeKtive
s3[e I] ~ v*
gilt.
Sleoi
, Sle I
FOr die eljo. mit
nicht
v*ss~[e~., ~Jo)
Sle o
9
S Jl &
das Resultat
SleljI
Induktionsannahme [~) Oaten
jedoch noch Schluss,
Nachbarn
ejl mit
geblieben.
man nun den bei e 1
liefert.
Oie geschach-
sich dann vermeiden, Peiree-Oatum
Ergebnis
sowohl
wenn
liefert
v wie die Ele-
man S!j , indem man im[b] --
ein zweites
nicht
dutch Abschw~ch-
setzt man
S& Jl
elj von e I gibt so,
leer ist; man nehme zun~chst [w
liefert
oberen
SlelJo
"
liegt respektive [w167 mit gegebenem
Nachbarn
S w , Slelj 2
S~ J2
Naehbarn
S~3o
Mal an und definiert aus
gilt oder c gar nicht
mit dem gegebenen
S~ , aus denen man wie zuvor S ~ erh~lt.
v und v ~ sind verschwunden,
Aus den S~ Jo
c # sB[e 1]
einen oberen
in s1(eqJl)
FOr die restlichen
transformiert 9
dass
der KonstruKtion
mit leerem Peirce-Oatum
vorhanden.
Beweise
. In ihren Ergebnissen
repetiert
Oie Konstruktion
8eweise
v ss1[elJl ) , for die aber e n o c h sei
Subterme
von c abet erhalten
in dessen
sl[eljJ m R
aus
ist,
S!j
der S! l~sst J
aus ihm erh~lt
, Sle I
vorliegenden
o
mit dem gegebenen
einen Beweis,
sind;
Beweise
und damit S ~
(w
Sleoi
Mal an und konstruiert
durch Abschw@chung, einfOhrt
s3(e 1) = c
eoi von e ~
Mit den oberen
ein zweites
Nebenterme
dass
einfOhrt.
Als n~chsten vorhanden
allf@llige
Oie ~onstruktion
entfernt
so S w .
zur KonstruKtion
aus Sle ~ , Slelj
mente von im(aJ -ungen
worden,
der c wieder
leer ist.
dann n&mlich
und erh<
aus S ~ , Sleqj
, oder auch direkt
Schluss,
telte Anwendung
[w
aus
von c o d e r
bei der Trans-
Nachbarn
Peirce-Oatum
man den bei e
man nun die InduKtionsannahme
leerem Peirce-Oatum
den Fall,
die v aufbauenden
repetiert
wird,
zun~chst
Mit den oberen
transformiert,
durch den v wieder eingefOhrt
Vermittelst
gilt.
in denen das Fehlen zu SchwierigKeiten
Man betrachte
(~) mit dem gegebenen
die Peirce-Oaten
von e I wendet
zu behandeln, sl[elj)~R
sole I) = <3,1>
die KonstruKtion
S', i 9 in denen
in den
Felscher
Konstruiert
elj 2
S~ 32
s1(elJl)n R Peirce-Oatum
von e I wendet
als Resultat
. Dann bleiben jedoch
FOr die eli1 mit
ist
aus
man die
der RonstruKtion
in S~32
die Peirce-
st[eli 2) - (v*)
man S w167verm~ge
leer
[v~)
wieder
W.
Felscher
149
Nun blelbt noch die MBgliehKeit zu diskutleren, setzungen des letzten Falles gilt
ss{e I) = v*
dass unter den sonstigen Voraus-
. Oies b e e i n t r O e h t i g t die Bildung von
S ~ nieht, die unverOndert Obernommen werden Kann. Man betrachte als ersten Unterfall den, dass c in s1(e I) vorhanden ist. Oann ist e P a r a m e t e r und kann nut dann in einem sl{elj) fehlen, wenn
v*eB
OisKussion der FOlle
gilt und mehrere Pr~missen auftreten.
{dbaa} - {dbbb)
for
vEA,
Betraehtet man die
so findet man, dass eine InduKtion
mit diesen Pr~missen nur be• der Bildung der Sw%, S w% in {dbab), und damit in (dbbb} auftritt. Is~ nun S
w
oder S w v o n d e r
Art, dass c n i c h t
in seinem Ergebnis auftritt,
,
so
findet man die neuen Sw%, S w% , indem man die I n d u K t i o n s a n n a h m e ein zweites Mal, und zwar mit der Konstruktion < ~ )
und leerem Peirce-Oatum,
Als letzter Unterfall bleibt, dass s 1 { e l ) n R ganz wie soeben sehliessen. [dbbb) for veB
Im Falle
v*EA
leer ist.
auf S w
S
w
oder S ~, S w anwendet.
Im F a l l e v*eB
Kann man dann
lehrt die OisKussion der F~lle
[dbaa) -
, dass dort eine InduKtion mit der PrOmisse von S I nur be• der Bildung
von S' aus Sle ~ , Slelo die I n d u K t i o n s a n n a h m e Oatum anwenden,
in (dbab]
, und damit in [dbbb)
, vorKommt.
ein zweites Mal, n&mlich auf S ~ , Slelo
Hier muss man
und mit leerem Peiroe-
um dos neue S' zu erhalten.
Dos beendet den Beweis des E l i m i n a t i o n s l e m m a s mit Peirce-Oaten.
Oie Rangzahlen der hier Konstruierten Beweise S ~ und S ~ ihre AbhOngigKeit
von den Rangzahlen der A u s g a n g s b e w e i s e S ~ und S I wird durch Funk-
tionen beschrieben,
die w e n i g s t e n s
so stark wie die A c K e r m a n n ' s c h e FunKtion wachsen
und die deshalb nicht p r i m i t i v - r e k u r s i v sein kBnnen. sich auf die Rangzahlen o 3 zu beschr~nKen,
kSnnen. Eine F u n k t i o n f3 liefert,
muss
wachsen sehr schnell;
da die
' welche Absch~tzungen
Um dies zu sehen, genOgt es,
01
, 02
nur noch g r 6 s s e r werden
o 1 o~ ~ f 3 [ o 3 , o 3 ]
%~ o 1 0 3 ~ fS{OS,a3]
nun neben den im @ew6hnlichen E l i m i n a t i o n s l e m m a angegebenen Un-
g l e i c h u n g e n ouch noch d e r U n g l e i c h u n g fS(f3(x',y)§ genOgen.
dann
~
fs(X,y)
x' < x , y' < y
Betrachtet man nOmlich den so genannten A u s n a h m e f a l l mit zweifach anzuwenden-
der InduKtionsannahme, s•
§ 1
etwa in seinem ersten Fall
x, x i, y, yj
die o 3 der S~
durch
die
f(xi,Y)
o3
von
Sle ~ , Sleoi
s3{e I} = e , So(e I} = I , und , Sle I , Slelj
, dos o 3 von S ~ also durch
f{x.,y}§ l
, so werden abgesch~tztl
150
W.
for die a 3
Felscher
der S!j erhOlt man Absch~tzungen dutch f[f(xi,Y)+1,y j) , for das a 3 von S ~
also eine Abschatzung dutch
f[f(xi,Y)+1,yj)
+ I
Falle liegt hier also eine Ungleichung der Art
Bereits im einfachsten, f[f[x-l,y),y-1)
sie for die Ackermann'sche FunKtion charakteristisch ist. Anfangsbedingungen
y < f3[o,y)
Anfangsbedingungen tionsbeweis,
und
x _< f3(x,o)
63
Es ist zu vermuten,
Inspiziert man dan Elimina-
(s) , p. 211
oben , ebens
eine
die dasselbe Wachstum dar BeweislOngen nach sich zieht.
dass die UnmBglichKeit
einer primitiv-rekursiven
hier nicht in der zufOllig gewahlten KonstruKtion
Des Eliminationslemma R1imT'nationstheorem.
Oa f3 ausserdem den
for Sequenzenkalkule mit Klassischer Peirce'scher
Regel gegeben hat, so bemerkt man dort im Fall geschachtelte InduKtion,
vor, wie
$enOgen muss, sind auch die
der AcKermann'schen Funktion $egeben.
welchen CURRY
< fix,y)
finit~ren
AbschBtzung
sondern prinzipiell begrOndet ist.
zieht nun auch nach sich das
Sei S e i n e
Menge von schlanKen Beweisen eines fixierten Typs;
im Falle des Typs I s
das Linksprinzip.
Oann 8ibt es zu jedem S aus S
ein S j in S mit demselben Ergebnis wie S , des schnittfrei ist und in der Abschliessung yon S unter den Funktionen
F I ..... F6o
liegt.
Der Beweis ist klar.
Als Beispiele von Regeln,
deren HinzufOgung z u den Oblichen logischen Sequenzen-
kalKulen $mmer noch Schnitteliminationss&tze
in GOltigkeit belBsst, findet man daher
neben der bekannten Peirce'schen Regal etwa mit einfachen Peirce-Formen den Obergang yon
mit
nicht
M,
(x + y)vz
M,
[ x + y) A (x + z)
einfachen M,
~
x
Peirce-Formen 1 X
~
den X
~
x
zu
M
zu
M
0bergang
oder
X
VOIq
yon
ZU
M
M, ~ ~ ~ x
~
x
zu
M
M, ~ [ x ^
y]
-~
x
zu
M
M
~ ~ x , x
zu
M
~
X
~'
-~
X
9
VOIq
X
9
VOFI
x
yon
W. F e l s c h e r
151
Referenzen BACHMANN
55
: Baohmann,
H.: T r a n s f i n i t e Zahlen
(Ergebnisse der Math. N.F.
Berlin-G6ttingen-Heidelberg CURRY
63
: Curry, H.B.:
I)
1955 .
F o u n d a t i o n s of Mathematical Logic. New York 1963 .
FEFERMAN
68
: Feferman,
S.: Lectures on Proof Theory. S u m m e r School in Logic, 7o , I-Io7
SMULLYAN
66
: Smullyan,
R.M.
Lecture Notes in Math.
(1968)
: First O r d e r Logic
(Ergebnisse der Math. N.F.4@~
B e r l i n - H e i d e l b e r g - N e w York TAIT
68
: Tait, W.W.:
P r o c e e d i n g s of the
1968 .
Normal Derivability in Classical Logic. and Semantics of Infinitary Languages, Notes in Math.
72
, 2o4-236
[1968]
The Syntax Lecture
C A L C U L A B I L I T Y OF THE P R I M I T I V E RECURSIVE FUNCTIONALS OF FINITE TYPE OVER THE NATURAL NUMBERS ( a
revised
version
Yoshito
We
consider
arithmetic T-Z
in
has,
notion
of Z(t)
the
:
'
T-z
(ii)
For
(iii)
By proof
meral
n
a
notion
equal
is
to
a
w ,
neutral 73
of
)
intuitionistic
.
types,
natural
restricted then
a
This
Theory
more
restricted
by
number.
'
Z.
:
Vx~
~ each
closed
term
inspection without
of
cut
a
nor
t
of
proof
type
of
0,
Z(t)
induction,
T-Z
in
of
t
T
= ~
F- Z ( t ) Z,
,
.
we
for
can
ob-
some
nu-
.
As than
usual
is
remark
(i)
a
t
N-HA
( Troelstra
:
induction We
tain
0
of
type
the
type
Hanatani
variant
finite
besides Z
And
a
all
)
such
a
proof
calculation
calculability
of
without
the
cut
term
t
nor
,
induction
we
have
is
thus
the
none
other
effective
.
Acknowledgements This The
original
University
'
and
published
67,
71
cal
results
and
tude
to
The
author
a
version
result
was
presented
Toky8
Ky8iku
in
1966
Hinata
author
67
The by
ShSji
sincere
positive form
to
thanks
the '
in
version,
for also advice
and
of
here
) his
his
those
result.
Dr.
to
at of
the 1965,
of
Tait
metamathemati. sincere
inestimable
which
equally
beginning from
etc.
to
former
master-thesis the
differs
express
go and
author's
applications
Maehara
criticism this
as
method the
the
theorems
wishes
Professor
to
.
of
Daigaku
( cut-elimination
author's his
revised
of
The
for
is
Jean-Yves encouraged Professor
gratiguidance. Girard the Justus
Y.
Diller this
for
Benedikt
I.I.
of
the
type
for
from
theory
T-Z
type
induction
I.Z.
sion
of
zen
the
) with
the type
the
bound
of
the
following
(
a 1 ~
{
Tp
type
~.
g,
T c Tp
Tin0) note
p, (
[
~
them
advice
by
we
that
on
our
Z
V-elimination
the
is
T-Z
has
notion
~~ is. befor
Z
and
that
the
be
by
the
ver-
ZJ
terms
the
ZJ
defined
many-sorted
calculus
and
should
N-HA
T-
on
also
as
of
.
take
structure
theory
constant
sequential
Tp
type
sub-system
term
~
g ~
~,
T
of
Tp
}
classes
by
x~
Tml)
0
~
Tm
Tin2)
N
~
Tm
0 0 ~0
)
of
Tm~y
(Gent-
.
That
is,
3-introduction same
this
as
that
of
version.
inductively
by
the
. any
as
are
.
We
(~J 1, ~ 2 , - - - ,
terms the
type
is
classes
the
abbreviate a n ~
~ ),
and
union Tma
defined
of
the
terms
inductively
by
of
the
: of
,y~r
c Tp
denote
the
clauses
variables
r)
)---)
Tm e
seven The
its
denote
~rn--
cIass
The
following
and
.
,
c~ 2 . . . .
Tma
Mr
clauses
0c
The U
the
structure
following, (
type
type
) the
of
to
restricted
predicate
framework
. We
TpI)
In
and
finite
finally
.
finite
~~ is
more
unary
concerning
two
a
restricted
variable The
Tp2
the
find
first-order
of
N-HA
intuitionistic
application
the
an
T-Z
logical
Gentzen's
35
from
is
of
version
many-sorted
structure by
and
to
revision
this
functionals
schema
Description As
a
difference
usual
valuable
~
is
recursive
were
careful of
N-HA
0 , represented
the
his
T-Z
essential
sides
version,
revised
theories
primitive
The
the
formulation
The the
which
first
Difference
153
suggestions
Peppinghaus the
Formal
on
kind
formulation
writing
i.
his
Hanat ani
each
,z~
type , w~Z
u ).
belong
to
Tm~
,
( we
de-
154
Y.
Tin3)
K {~, T
Tin4)
S p , ~ , ~
Tin5)
Rcr
Tin6)
t
In
the
( ---
t n )
languages
(E)
as
,
our Z
follows
or
for
Vx~
Vy ~
Vx~ axioms Vx ~
Vy'r
V x (' V y l r , , 0 - - , , )
(DRZ),~
Vx ~ V r l ~
Axioms
axioms
are
have
an
L
[=,
of
.
~
T)
r
We
abbreviate
equality
Peano
' = '
and
( Nx=O
for
Z
(zP1)
z(o)
(zpz)
vxO(Z(x)
(ZE)
Vx0Vy ~
the
sub-
any
formula employ .
(E)
,
(D)
respectively
an
,
(P)
,
(Z)
. and
So
~ y=x)
V z c~ V w rr Ka,
T
,
(x=y
i z=w
Sp,C,,T
'
n xz=yw)
R~
:
r,,Txy=x
)
-(r
( S p.,,,rXyZ=xz(yz
VzP
~ x:y
~ Z(~x)
)
)
)
e~ xrz)
z )
)
)
^ Z(y) of
in the
A y--z
)
A D A [Nx]
sequence
L[=],L[Z]
and
:
Sequent-schema Vx0(
,
:
Vx 0 ~
0 ],
by
Z]
0 ( R~, x y ( N z ) = r (
(P4) Axioms
,
( R a xy0:x
0 ( Nx:Ny
manner
term
p
,
:
Vx 0 Vy
finite
Tm
by =
(P3)
We
we
without
(K
(DR1)o_
A
0 .... ) (tu) e
t n
for
VxIP'"--T)VyP
A[
,
t 1 t 2 ---
V z ~' ( x = z
(DS)p,~,r
(IndZ)
~, )
any
VY ~ ....
Defining
(Z)
~ ) ~
( p
,
denote
aquality
(EZ)~
(P)
.-
)
:
(x=x)
(DK)a,r
"
r
language
Vx~
r
)
~-
u
as
(E~)~
(D)__
a
t
non-logical
Axioms
(SAP)or,
p,
a
'
without
)
( (
--
constants
denote
The ( IndZ
,
predicate I Z
T
( ~, ( ~, 0T ' u e Tm(5
T m -
( t 1 t 2 ) - --
one
a ,
Tm
s
We
usual
c
following
unary
with
Tin(
Tm
e
e
As
is
e
Hgnat ani
the
notation
T - Z Ff0
of
V--A
)
induction
),
L [=,
~ Z(x)
Z(t) Z]
--
3x0
and 'F-' for
on
t of
:
(t=x
any the
example
axiom-formulae
Z
^A term
) e Tm
0
deducibility means of
T-Z,
. in
the
that
there
such
that
Y. Hanat a n i
the
sequent
(IndZ)
F 0 , F--A
supplementally
.
We
denote
sequence
.
sistent
,
T-
is
of
N -HA
denote
i.e.
by
without
T-
the
the
axioms
Consistency
of
case
the
schema
of
beginning
where
means
that
If-T
F
is
empty
T-Z
is
incon-
to
Proof.
Evident
.
If-T
+NJ
Gentzen
35,
Corollary
and
the
the
(Z)
nor
predicate
induction-free
logic-free
If-T-F- t=u
example,
apply
Prawitz
the
71 T-
1. 3. 3.
For
In
the
' t and
u
fact,
Z
subsystem
equational
calculus
, for
consider
any the
normalization
t , u c Tin. intermediate
theorem
for
sysNJ
(see
) . is
by
also
consistent
Remark.
measure
~=~
If-T-
~
also
result
Closed A CTm
by
and
closed
of
is
term
and
class
T
Nt=0
for
any
CTm
CTm
=
terms
with
t=u
is
. ' The
property (see
show
axiom
73)
Lemma
theorem
for
derivations
.
reduction
interpreted
of t h e
Troelstra
this
the
(E2)
by
proof
.
is
system.
then There
.
directly,
lf-T
as
in
proving
induction
on
some
consistent sketch
of
the
.
term
those
~- INt=0
term
models
if-T-
1. 3. 4.
Z. 1.
a common
can
If-T-
, when
Church-Rosser
the
Main
0
of t h e
If-T
normalization on
Lemma
for
other
We
t c Tm
system
to
the
many
of
any
a model reduce
satisfied
then
the
~
For
i. 3. 2.
give
2.
containing
without
0
Lemma
sort
T-Z
T-
T-F- t = u
rules
of
.
denote
I. 3. i.
t ~ Tm
subsystem
with
T-
Lemma
is
the
' T-ZF--'
identical
corresponding
Let
with
schema
w
Let
the
T-Z ~-A
that
essentially
i. 3.
a
LJ
.
Z)
are
in
accepted
by
Remark
We
tem
deducible
the
sequent
(Ind
is
as
155
is
a
a
denote
of
type
U{
CTmal
term the a
containing class The
a ~ Tp}
no
of
the
inductive and
by
variable closed
terms
definition the
six
.
of
clauses
and CTm
156
Y.
CTml)GTrn
CTmd)
,
each
place
of
Tm
in
2. 2.
Main
Theorem that
T
For
F-
being
exactly
as
Tml)
- Tin6)
hut
whith
o
result
.
Hanat ani
.
any
t e
CTm
0
,
there
exists
a
numeral
fi s u c h
t=5
The
following
follows
from
this
by
Lemmata
i. 3. I.
,
1.3.3. Corollary.
For
numeral
fi
such
Remark. tive,
proof
actual
2. 3.
t e CTm
that
Every our
each
any
If-T-F-
actual
will
Our
of
proof
(a)
t
(b)
T - Z F- Z ( t )
(c)
T-+
(d)
T
an
of
the
exists
cut-eliminition
give
example
Steps
, there 0 t=6 .
t c
and
process
effective CTm
one
only
being
calculation
one
effec-
process
for
0
proof
follows
the
steps
: (a)~
(b)~
(c)~
(d)
;
E CTm 0
In
the
(ZPl)
+ (ZP2)
~-t=fi
above
for
Remarks
(i)
(c)
is
(if)
(d)
implies
(iii)
(b)
does
(iv)
Our
proof
The
(i)
step In
say
that
'
type
0
each
type
a
0 ),
(a)
give
:
.
any
term
T- + (Z)~-
points
E Tm
0
Z(t)
of
of
fact,
terms
for of
any
the
u
form
such
that
K0,
UXT
. with
the
the
(a)~(c)
In
the
(if),
following
[ T-ZF-Z(t)}=
{ t
(iii)
above
relation
and
the
for
some
consis-
:
E Tm 0 [T-~
t=n,
f i } c T m 0.
proof_.
. the
T-ZF- Z(t)
'
fi for
(c')
(a)
together
will
view
to
imply
not
0 c {t e Tm0
2. 5.
stands
)
(b)
example
but
CTm
t
not
(b)
T-
A Z(x)
numeral
equivalent
( for
of
(t=x
.
satisfy
tency
some
schemata,
2. 4.
T-Z~- Z(u)
~-- 3 x 0
The
point
a
predicate
'
equivalence
between
represents
' t
of
the Fa
proof in
of L [Z]
is this such
(b)
and
a calculable step that
is
to
'T-ZF-
(d),
we
term assign F a (t)
can of to '
Y.
represents The
' t
is
assignment
form (ii)
to
Step As
the
by
an
apply
we
see
of
proof
by
is
a
to
restate
induction
on
the
interpretation
(iii)
Step
2. 4
,
term our
the
strong
tool,
this
a proof
of t y p e
step
in
cr .'
a general
definition
step
in
of
of
CTm.
we
of
not
Z
only
a
that
a very
method
, followed
of
mean
the
. Our
theorem
does
need
essentially Z(t)
interpretation
. This
for
is
T-Z
cut-elimination
calculus
much
second-
we
weak
use
a
system
for
.
(c)~(d) have
with
of
from
of t h e
too
We
(i)
second-order
predicate
proof,
calculable
us
the
(IndZ)
application
order
enable
157
(b)~(c)
elimination of
a hereditarily
will
suitable
Hanat ani
.
only
to
respect
to
consider each
the
purifications
predicate
of
symbol,
the
given
remarking
that
m
the
axioms
of
the
axioms
(ZPI),
the
cut-elimination
should and
3.
remark
the
3. 1.
T
awe
any
For
every
a e Tp
in
L [Z] (t)
Lemma the
of
e Tma_
3. I. 2.
- CTm5)
,
whith
evidently
properties
of
we
T- +(ZPI)+(ZP2)
.
main
reduction
F
(t)
.
with
t e Tm
inductive
a denotes
rules
a
:
,
(F~(x)
CTm,
we
T-Z ~ F o-
of
common
it is
main
definability
let
following
any
(F)
in
for
. The
following
Vxa
T and
LJ
tool
~(b)
and
the
Z(t)
:
the
1. 1.
F a
by
=
(t)
collection
CTml)
,
for
(a)
of
3. t
the
main
symbol
T
Definition
definition
Lemma
. Our
explicit
step
of
through
tive
the
_Proof
F
predicate
theorem
of
F 0
no
(ZP2)
consistency
formuIa
Then,
contain
steps
corresponding
establish
our
T (t) u e
are
all
formulae
Tm
as
:
F 0 (0)
(F2)
:
F0~
and a
T-Z ~
0 (N)
to
step
F a (u)
the
(a)~(b) ~T-ZF-
induc.
FT(tu) ,
for
.
provable
.
(FI)
D F T (tx))
follows
in
T-Z
, where
corresponding
(F)denotes to
158
Y.
(FB)o
.
(r4)p,,.
~
(FS)~ Lemma
3. 1. 3. As
by
3 . 1. 1
:
F(~,~_o)(Ko,~
:
F ( (p
:
F(o
and
is
~
,
it
3. 2.
So_ m e
Lemma
3. 2. 1.
T- +
formula
A[x"]
Lemma
3. 2. 2.
Let
(IndZ)'
:
A[0]
,
(IndZ)"
:
A[0I
, Vx0(Z(x)
any
for
where
A
stands Then
Remark.
The
These pect 3. 3,
axioms
T-+
in
T-Z,
(FI)
T-Z
and
to
formula
for
Z(t)
in
and
induction
3 . 1. 3
prove
any
(IndZ)"
] ),
t e CTm is
. evident
3. I. 2.
--A[t]
and
L[--,Z]
(IndZ)"
be
t,
u ~ Tm
as
follows
t7
"
:
--A[t] )),
Z(t)--A[t]
L [=, Z]
,
and
t,
for
any
term
T-Z~(IndZ)" is
evidently
are
therefore
the
strongest
form.
equivalent
with
res-
of
T - Z F- ( F 2 )
T - Z F- ( F 3 )
Fa
and
axioms we
3 . 1. 2 .
and
.
by
(DK)
apply
are
and
3. 2. 1 and
(IndZ)" .
evident,
T-Z~- (F4) applied
(DS)
rest
are
to
is
by
A
by using
To
to
are
= Fa(x),
3. 2. 2
similar
they
immediate
respectively
admitted
The
as
,
to
the
prove the
case
the the
(F5)
ease of
(F3)
and
The
only
pri-
.
4.
Proof
4. 1.
Description The
mitive
of
the
quantifier
the
VZ
formulae
are
1)
t , Tm 0 ~
2)
A
is
(b)~(c)
of f o r m a l
is
added
on
The
step
language
symbols
predicate
of
any
only
A[u]
any
and
o (A oA[Nxl
Lemma
A [ z ] = F a (RaxYz) (F4)
t=u,
o AINx
for
)
(Z)
T-Zt-
defining
F-
(IndZ)'
of
of
definition
T-Z
~ )
0 -- ~ ) ( R o
definition
of
in
schema
Proof
of
by
(IndZ)'
for T-Z~
schemata to
(Z)
Vx0(A
-- ~ ) ,
remains
properties
e Tm 0
, (p_.),p_T)(Sp,~
T - Z F- F a ( t ) ,
evident
3 . 1. 2 .
)
~,_~) (~,0
t ~ GTma
3 . 1. 1
H a n a t ani
are
terms The
theory extension
the
e Tm 0
T
,
2
of
variables
additional
and clauses
L[=,
Z]
(denoted the
by
@ ) of
corresponding of
unary
universal
inductive
definition
:
r
is
a formula logical
an
.
not rules
a formula, containing are
all
the
V2~ rules
V2r
A
of
LJ
is
a
formula.
extended
to
159
Y. Hanat ani
this
language
V2 @ A
sequence and
and
of
additional
occurrence
B
'
rules
V2.E
stand
for
:
but
@
denotes
the
result
~(t)
in A
for
F
containing
We
V2
:
A[~tx0B],r--@ V2~A ,F--O
a formula,
formulae
A[Ax0B]
L J2
the
FAA(~F) F --V2@
V2-1 : where
and
of
and
at
|
most
substituting
denote
this
, for
one
a finite
formula
B[t ] for
logical
, each
framework
by
. The
4. 2.
non-logical
Remarks
on
axioms
the
of
weakness
T
2 are
and
the
exactly
those
of T
cut-eliminability
of
L J2 The LJ
. The
from
the
that
a
logical
weakness
as
definition
of t h e
successive We
for we of
firstly
is
difference not
Our cial Lemma
of
4. 2. LJ2
4. 3.
And
: let
Let
the
our
LJ~-
Proof. fact
provable
of
(~(0)
^ Vx0(~(x)
4. 3. 1.
that
the
in
T 2 .
comes
example
excluded
.
in
use
the
and
theorem
of t h i s
classical
55),
and
tool,
version
limit
of t h e
secondly
intuitionistic
that
versions
.
is
to
obtain
some
spe-
@
on
transformed
Then,
@ ~ in
T-2
~ ~(Nx) of
the
o 3x0(t=x
substituting
as
follows:
^ ~(x)
) ).
Z "(t)
for
each
.
T-2~-F"-
@*
complexity axioms
formula
)
respectively ~
.
second-order
results
T - Z F - F -- @
]By i n d u c t i o n
F--
the
the F
the
L [=, Z]
T-Z
V2~
in
is
strength
provable
theorem
in
denote
Z(t)
for
of
:
Z "(t)
0 ~ denote
V2-I
(Takeuti
the
@ be
~
be
calculus
cut-elimination
for
restricted
of
F ,
the
classical
Let
occurrence Lemma
e0
following
F- F - - @
F',
to
extension
remark
rule
theorem
to
Interpretability
Z "(t)
the
the
of t h e
following
up
for
. We
proof-theoretic
application
cases
formulae
the
weak
predicate
the
between
essential
a very
second-order
ensured
induction
is
the that
explicitly
transfinite
is
in
Concerning
remark
the
a
L J2
application
suppose
L J2
L J2
framework
of t h e
(__Z)" a n d
deduction,using (IndZ)"
are
all
160
Y.
Corollary Proof
4 . 3. Z .
.
By
4. 4.
T-Z
4. 3. 1
(b) ~(c)
axioms
are
in
all
of
T-2
T- + (ZP1) Proof
5 . 1.
Explicit
Lemma
of
5. 1.
:
where
all
As
its
Remark were
step
by
the
.
The
~
of
based
is
But
we
5 . 3~
a term
:
(Z)
implies
(2')
:
u
satisfying
proof
5. 4.
in
Proof We
cy
of
T-
shall
.
Given
(Z')
(i) by
such any
that type
expli-
U ~- A [ t ] , .
calculus form, for
even a
(ZP2)
LJ
when
more
L J,
this
is
and im-
. the
logical
detailed
(ii)
F- 3 x 0 ( t = x
satisfying
t=u
base
reasoning
.
Z(u)
both
(1)
) and
by
Then
(2)
:
. when
T-F-
t=u
(ZP1) can
^ Z(x)
,
we
prove
the
following
:
,
+ (ZPZ)~
be
none
considering
Z(u),
other
than
a
numeral
.
s=s
the for
transformation each
of
occurrence
the
Z(s)
in
it.
.
show
Evidently a
the
.
substituting of
a ,
true
accomplished
Immediate given
.
satisfies
V-prenex
then
+
~
(l')
of
system
intuitionistic of
u e Tm 0
implies
Proof
sequent
.
+ (ZP2)
(ii)
(Z)
t c Tm a and
also
T- + (ZP1)
because
T-+
theorem
T- + (ZPI)
be
above
non-
have
concerning
of
a
an
need
(2)
(1)
is
are
+ (ZP2)F-
(i)
U
on
lemma
will
the
we
the
.
formula
T- + (ZPl)
proof
But
,
^ Z(x))
the
cut-elimination
.
T - Z F- Z * ( t ) .
in
4. 2,
(c) ~(d)
(1)
The
by
have
A Z(x))
formulae
F- 3 x 0 ( t : x
axioms
Suppose is
the
there any
is
we
.
reduction
there
4. 3. I, 3x0(t=x
subsystem
for
U
Main
5. 1,
the
A
classical
5. 2.
By
and
(ZP2)
non-logical
mediate
,
(ED)
stands
Proof.
.
I. 3. 4.
Therefore
+
Any
U ~-3xa A
.
definability
definability
(ED)
and
(ZPZ)-
L [ = , Z]
5.
4. 2
T - Z F- Z ( t )
T-ZF-(ZP1),
logical
cit
,
consistent
.
Suppose therefore
is
Hanat ani
cut-free
that the
(2) latter proof
implies is of
(2')
impossible Z(u)
in
or
the by
inconsisten1. 3 . 4 .
T- + (ZP1)
+ (ZP2)
.
Y.
By
the
in
it is
lae
subformula
in
either
the
obtain through This
in
least
the
L[=]
or
6.
Remarks
(i)
The
follows
:
sion
T-Z
on
original Let
denote
concerning
Z
,
I)
are
all
provable
with
the
type-free
3
but TC-Z
~- Z ( t )
(ii)
The
proof
the
sical
formal
via
the
If-TC-
above
, the
(iv)
Still
through
the In
any
formu-
evidently in
of
was
type-free
L[Z]
the
,
proof
.
essentially
as
(combinatory)
without
is
an
the
ver-
axioms
e CS by
to
a the
numeral the
same by
say
stratifiable of
t e CS 0 ,
sort
the of
is
of
be
of
such
all
to
one
the
clas-
reduce up
to
CS~
type ,
the E0
, and any
If-TC
in
~-t={~
passing
CSa
and
only
.
theorem
theorem,
denote
~
for
that
the
normalization
the
of
: Let
calculable
union
.
.
Then n
Church-Rosser
preserves
to
.
present
order
55)
TC-
numeral
our
induction
following
ours
TC-F-t=s
choice
in
(Takeuti the
as
the
was
as
0,N,K,S,R ~,
transfinite
result
(__F) a r e
constants
choice
the
unique
terms
where
caused
version
: Any
fact,
TC-Z,
unique
This
terms
reduction
following
can
purely
result
atomic
an
to
more
corresponding that
.
logic-free
, there
we
the
.
sub-system
in
results
closed
t e CS 0
this
the
difference
Takeuti's
of
are
parts
version
essentially
explicitly
The class
was
theories
Professor
(iii)
, its
for
detailed
calculability
so
formula
Then,
2)
except
or
redundant
of
and
in
proof
L[=]
every
.
version
TC-
a
in
original
TC-Z
proof,
, because
such
the
,
(F)
L[Z]
wanted
the
161
cut-free
purely of
we
a
From
proof
what
in in
.
one
elimination
implies
of
principle
end-sequent
at
Hanat ani
for by
stratifiability
, we
the
of
If-TC-,
the
fact
have
the
:
Lemma.
Let
restricted
to
If-TC-
If-TCCS
.
~ t=u
/CS
Then
~
denote for
any
If-TC-/CS
t,
subsystem
u e CS
5- t = u
~ ,
If-TC-
,
.
162
Y. Hanatani
References
G.
Gentzen
1935
Untersuchungen
fiber d a s
Mathematische K.
logische
Zeitschrift
39,
Schliessen.
pp.
176-210,
I. II. 405-431.
G~del
1958
I~ber
eine
bisher
noch
nicht
benfltzte
Erweiterung
des
finiten S t a n d p u n k t e s . Dialectica Y,
pp.
280-287.
I-lanatani
1966
Calculabilit4 de
type
des
fini s u r
Annals
of the
Science, S.
12,
fonctionnelles les
nombres
Japan
Vol.
3
r~cursives
naturels.
Association
N o . 1,
primitives
pp.
for
Philosophy
of
19-30.
Hinata
1967
Calculability
of p r i m i t i v e
recursive
functionals
of
finite type.
D.
Science
reports
Vol.
No. 226,
9,
of the pp,
Tokyo
Kyoiku
Daigaku,
S e c t i o n A,
42-59.
Prawitz
1971
Ideas
and
results
Proceedings sium.
Ed.
of the by
J.E.
(North-Holland
Wo W ,
ait
1967
Intensional
1971
in p r o o f
theory.
Second
Scandinavian
Fenstad
interpretations
of s y m b o l i c
Normal
form
for
235-307.
of f u n c t i o n a l s
Logic,
theorem
Sympo-
Amsterdam-London
P u b l . C o . ), pp.
Journal
Logic
32, bar
pp.
of finite t y p e I.
198-212.
recursive
functions
of
finite t y p e Proceedings sium.
Ed.
of the by
J.E.
(North-Holland Go
Second
Scandinavian
Fenstad.
Publ.
C o . ),
Logic
Sympo-
Amsterdam-London pp.
353-367.
Takeuti
1955
On
the
Journal pp.
fundamental of
the
249-275.
conjecture
Mathematical
of Society
GLC
I. of
Japan,
7,
Y. Hanat ani
A. S.
1973
163
Troelstra Metamathematical Arithmetic Springer New
York
and Lecture
Investigation
of
intuitionistic
Analysis. Notes
Vol.
344.
Berlin-Heidelberg-
.
5,1, r u e 78240 France
de
Montaigu
Chambourcy
OBSERVATIONS ON A RECENT GENERALIZATION OF COMPLETENESS THEOREMS DUE TO SCHUTTE
For K. Sch~tte on the occasion of his 6~
th
birthday
G. Kreisel
Summary.
The generalization in question is contained in [i~] which will be
referred to as [KMS]. function symbols or
It concerns the language of ordinary predicate logic without = 9 (i) two kinds of valuations
p~ namely
'total' ones (defined
for all formulas of the language considered)~ and 'semi-valuations' in the sense of Sch~tte's article [22]~ (ii) the complexity
~
validity (of a formula) for valuations
is considered provided
C~
of the valuations used~ that is~ ~
satisfies
some simple closure conditions depending on the exact choice of data determining the valuations of the kind in question -- and not only 'logical' validity~ that is~ validity for the class of arbitrary valuations 9 and (iii) instead of finite (derivation) trees 3 socalled ~ - f o u n d e d
trees are used~ built up according to suitable rules
Rp~ in particular~ rules with cut correspond to the total valuations mentioned in (i) and rules without cut correspond to semi-valuations.
As in [KMS] the familiar
completeness theorems~ for finite (well-founded)
trees built up by use of the finitary
rules
and logical validity (for valuations
p)~ are generalized to ~-founded
p-valuations of complexity
The generalization also applies to
Rp
trees and e-logic.
~
resp.
These generalizations suggest a reappraisal of some work in the fifties
on 'constructive' models by Mostowski [18] and Vaught [24]~ in which the complexity of a model is measured by its valuation on the atomic formulas. -- The significance of the difference between rules with and without cut is analyzed in 'extensional' terms by showing that9 for
~
of suitably low complexity~ for example~ for recursive
9 there are formulas which are true for all total valuations
C~
~ but not for
all such semi-valuations. The observations fall into 3 groups.
The most controversial ones concern
defects of traditional views~ for example9 (i) of the restriction to finite~ and~ in particular~ formal derivation trees (which would make the use of infinite founded trees teratologieal~
~-
if not 'illegitimate') and (ii) of so-called operational
semantics which provided -- in effect~ if not in intention -- the only permanent~ not merely heuristic~ significance for cut free rules:
this semantics can now be
compared to the usual model theoretical interpretation refined by 'definability' requirements
~.
observations~
'related' because earlier work was presumably influenced by~ and
Only slightly less controversial are the related historical
certainly formulated in terms proper t~ traditional views.
The least controversial
observations are about open problems stated by essential use of the concepts involved in the generalization.
Some of these problems are new~ some are more precise versions
G.
Kreisel
168
of questions scattered in [KMS]. It is certainly possible to read this article with very little previous knowledge of the subject provided one simply ignores the issues arising from the literature specifically cited.
But the principal aim is to serve the needs of those readers
who have a detailed knowledge of
our (venerable)
subject which goes back
> iOO
years; readers who wish to test to what extent this detailed work agrees with the expectations they -- or~ for that matter~ its founders -- have had of our subject. In short~ the article is intended to have pedagogic use for the so to speak logically over privileged (with genuine problems of their own); a class which is created by progress~ and therefore liable to be neglected by those who follow uncritically (once)
i.
'reasonable' pedagogic traditions.
Backsround.
Completeness proofs for (the usual rules of) predicate calculus
were given in the first half of this century by G~del and Henkin.
As far as clarity
of the arguments is concerned~ neither proof leaves anything to be desired. only room for improvement was in the discovery of significant refinements.
The A first
step was made in Hilbert-Bernays where (a version of) GSdel's proof of the completeness theorem was formalized in first order arithmeti~ but without special stress on the complexity of the predicates involved -- except that~ trivially~ they are definable in the language of arithmetic.
Around the middle of the century the
complexity was stressed explicitly in publications by Kleene [8] and myself [iO]~ in two different forms (which later turned out to be equivalent): both Kleene and O I ([iO] pp. 275-276) used the arithmetic hierarchy~ ~2; I also used a peculiar circumlocution of what we should now call 'recursive in the jump' ([iO]~ p. 7 8 and footnote i on p. 39).
Both were concerned with the complexity of the satisfac-
tion relation on the atomic formulas~ my main stress in [I0] being on the 'negative' results~ establishing a conjecture in Hilbert-Bernays II~ p. 191 ~ that there are consistent formulas of predicate logic which have no recursive model.
Evidently~
such a negative result is optimal for the measure of complexity on atomic formulas. Soon afterwards Hasenjager [6] improved the 'positive' result by formalizing Henkin's proof; he showed that any consistent formula has a model for which the satisfaction relation on all formulas (that is~ in Henkin's terminology~ a complete and consistent O extension) is A 2. Both the positive and negative results were steadily refined; v 0 perhaps most satisfactorily in terms of Ersov's hierarchy for ~2 sets; cf. [7]. Having been skeptical from the very start about the value of the piddling business of 'formalizing' convincing proofs~ I was very much taken by the possibility of formulating the interest of then-current formalizations in terms of definability theory, and coined the term 'basis theorem' [ii] for the type of result I was after. (It is too late now to change this mediocre terminology.)
It remains open whether
significant further improvements are possible by closer attention to the (metamathematical) principles of proof needed for establishing the 'definability' results:
166
G.
Kreisel
if so~ it seems certain that an imaginative classification needed~ and that current favourites 0 to ~2-predicates) are not adequate.
In the middle fifties a bunch of new completeness Hintikka~
Sch~tt%
and probably others.
as 'simplifications'
[i])
papers.
proofs appeared;
by Beth~
in view of the pellucid character
But already then (cf.~ for e x a m p l %
my review of
it seemed that -- in effect if not in intention -- a detailed model theoretic
foundation of the choice of lo$ical rules was being attempted; one starts with the (model theoretic) precisely~
of counter model (to the formula
A
considered)~
finite if and only if there is no such model. obstructions a complete
Precise details~
counter models~
It turns out that the possible
Of c o u r s %
'natural'
'sophisticated'
conditions).
'sophisticated'
data
or only on the atomic
of formulas were noticed by almost everyone
But the distinction
insignificant
for a given domain of individuals~
determines
'sophisticated'
value in the context of infinitary
between the 'extreme' and the
unless (something
after all~ the satisfaction
The use of the
are of interest
only the terminology differed (model sets~ semi valua-
data remains
the data is involved;
procedure provides
predicate on all f o r m u ~ s
the relevant classes
who looked at the procedure; tions~ consistency
and this then yields
which are given below~
somewhere between the two extremes discussed a b o v %
that is~ between the satisfaction
formulas.
and builds up a tree
involved and which is
to getting an infinite tree are easily enumerated~
set of rules.
in as much as they show that this determining
in terms of [KMS]:
notion of infinite countable model or~ more
of formulas which codes up to isomorphism all the models
ones.
is
or even its restriction
It seems fair to say that they were presented
This was quite unconvincing
of Godel's and Henkin's
of such principles
(first order arithmetic
like) complexity
of
predicate on the atomic formulas~
the satisfaction
predicate on all
data turned out to have some expository
languages
tions which are so to speak non-archimedean
[9]~ where simplifications
or distinc-
in ordinary predicate calculus have a
chance of becoming noticeable. Early in the sixties Schutte [22] choice of data discussed
proposed a much more imaginative
in the last paragraph;
ordinary predicate c a l c u l u %
but analogues
to be p r e c i s %
of those data for two particular
tions of 'applied' many sorted predicate calculus~ theories of types with and without cut. jectured
Takeuti~
is -- still -- not well known~
Sch~tte then observed~
of semivaluation
to those
'applied' calculi~
If a formula~
sense)~
(Since the significance in Appendix
of the notions of total and
that they are complete for validity
So Takeuti's conjecture
of the language considered~
(in the modified
formal
the latter~ con-
it is briefly discussed
for his natural extensions
all total~ resp. all semi valuations.
formula-
known as the impredicative who formulated
that the two theories have the same set of theorems.
of the conjecture I.)
use of the
not for the case of
is equivalent
is true in some semivaluation
it is also true in some total valuation.
to:
in
G.
Kreisel
167
Evidently~ this does not require that every semivaluation can actually be extended to a total valuation (though~ of c o u r s %
in general not of the same complexity).
But it raised genuine model theoretic questions concerning such extensions;
'model
theoretic' because structural properties of models are involved~ not merely the existence of some model satisfying conditions formulated in the language considered (and thus 3 by completeness~ reducible to a 'proof theoretic' question).
It also
raised the possibility that 3 once we have learnt to make use of our model theoretic knowledg%
Takeuti's conjecture would be almost as obvious as -- the truth of --
Gentzen's
Hauptsatz~ and thereby refute the particular consequences which Takeuti
(and others) obviously expected from a proof of the conjecture. reasonj Sch~tte's formulation~
But~ for this very
in terms of total and semivaluations~
provided a
(possible) new and perfectly legitimate use of these notions~ as a tool for correcting misconceptions about the nature of the conjecture. However 3 if both these notions are to be principal objects o f study, one will look for contexts in which they are no_._~tequivalent (not even w.r.t, their 'logic'); not~ of c o u r s %
contexts manufactured for this purpose~ but as venerable as possible.
Such a context is the theory of models of low (recursion theoretic) complexity~ and the generalization discussed in this article belongs to this subject.
2.
Minimal semi-valuations and cut-free rules.
The exposition below~ of the
'simplified' completeness proofs that appeared in the fifties~ should be regarded as an instance of the expository principle which dominates (the exposition of) mathematics in this century.
One starts with the proofs that the pioneers stumbled
on~ either by experimentation in the subject concerned or in attempts to solve problems outside it (for e x a m p l %
algebraic manipulation or algebraic theorems
inspired by geometric problems); and then one tries to find concepts to reformulate those proofs appropriately; known as: utiles.
Trouver le bog cadre~ d~gager les hypotheses
To the inexperienced the principle seems delicate because it suggest an
infinite regress:
one sees no reason why this choice of appropriate concepts should
not be subject to an analogous analysis (why for any explanation one should not want a further explanation).
As Bourbaki stress~ at least implicitly~ the discovery that~
occasionallyj we have no need for this may provide genuine information about the notions considered and about the structure of our thinking; the parallel in the natural sciences is obvious. It is in the nature of the expository principle above that changes i__nnterminology may be essential. We consider two sets
F+
and
P-
of formulas of predicate calculus without
and without function symbols (except so-called and put
P = F + U P-.
0 -- ary ones~ that i %
The logical operations used are:
~ ~ v~ 3
(and
=
constants) A3 ~
V may
be considered as 'defined'). -- The 'basic' question in the fifties was to give simple criteria for the existence of some realization (of the language of
F) which
108
G.
Kreisel
is a model of
{F : F C P +) [J {~ F : F C P-] ,
in terms of the formal underivability of -- what is usually written as -- F + ~ A more
'structural'
F-.
question is to code up all models of (*) of some suitable
kind, where, as with all structural questions, the proper choice of kind of model and of the data determinin$ those models is an essential part of the problem. We consider term models (also called F
'canonical' in [14]) in the language of
enriched by familiar Henkin constants~ eF~ associated with existential formulas
3xF
in
the enriched language.
Furthermore~
the models are required to satisfy
(according to the meaning of 'Henkin constant'): known, if (~) has any
that of the language generated by card
F.
F[x/eF] v ~
3xF.
As is well-
model at all, it has such a 'Henkin' model; its cardinal is F
or, equivalently~
the first infinite cardinal
For our principal aim (explained in para. i), namely definability refine-
ments, a more important fact is this:
The logical complexity of (Tarski's) adequacy conditions for truth (or satisfaction) and
3xF
in Henkin models is reduced:
if
D
is the domain
is, say, closed, we have
T(e3xF ") -> T(rF[x/eF] l)
instead of:
T(r3xF ") -~ (3a C D)
NB.
Sat(a : rF1) .
Readers who are interested in such matters will easily find the relation between
the term models above, introduced by Henkin, and older term models in languages obtained from
F
by adding suitable function symbols~ either by themselves or by
looking up the literature, for example, [14], Ex. 2 on p. ii0. The data used to determine those term models of (~), that is, the classes of formulas of the language generated from made to depend on
(F +, F').
F
which are assigned truth values, are
They are semi-valuations in the sense of Sehutte [22]
in which (*) holds, but -- in addition -- required to be minimal.
Roughly speaking,
the formulas in question arise directly in the lo$ical analysis of the formulas in f
or, more precisely, of subformula occurrences.
The difference from Schutte's
notion of positive and nesative part is dictated by the minimality requirement: a given subformula occurrence FI, F 2
F I v F2
(with the value:
is assigned a truth value (true), and for an occurrence of
with value:
true) only
F[x/eF]
is assigned a value.
for
true), only on____eeof the 3xF
(again~
Since formula occurrences
G.
Kreisel
are involved 3 the obvious I representation
Exercises.
169
is in tree form.
Before turning to the representations
of (minimal)
semi-valuations,
some easy facts should be verified -- or looked up in the literature,
for example,
in [KMS].
(a) Any semi-valuation 3 say SV (tacitly,
for term models of the kind
considered
here), can be extended to a total one; in fact, the class of total ones
compatible with SV are just those which agree with SV on the atomic formulas which are given a value in SV, and are
otherwise
tion SV can be of much lower (recursion
arbitrary.
theoretic)
complexity
compatible with SV; and almost equally trivially some recursive
semi-valuation~
contrast: increase
but in no recursive
the restriction this complexity,
(of arbitrary
much less precise: of a minimal
semi-valuations)
in
SV
a semi-valua-
than any total valuation
(F +, F-)
are true in some
total valuation.
at least for finite
minimal one which is contained
(b) Trivially,
(c) Perhaps,
in
to minimal ones does not
F, since for any
and primitive recursive
SV in
there is a SV.
(d) This is
there seems to be a quite sharp sense in which no proper subset
semi-valuation
has the (obviously)
essential
properties
of semi-valua-
tions.
(i)
Brutal representations
branchings
of minimal
of the trees will occur even if
formula gets the value:
false);
semi-valuations; F
'brutal' because
infinilte
is finite (unless no existential
equivalently,
infinite sets of formulas are
'put'
at the nodes. NB.
If
F
is finite,
is also finite; For each ~(F)
and
F C F, stage
On(F)~
F
~
of the representation
their elements are pairs,
true and false resp., leads from
the height of the trees used in brutal representations
it is determined by the lo$ical complexity
to
and sequences
F'.
Specifically,
+
involved
~-), the second is
o+
or
sequence of
O~ i.
whether
or
F0
3x ~ (F 0 v FI)
FI
or
Here
and
mines whether
~ , 3
vI
or
indicate,
gets the value false~ hence each
Fo[X/eGi]
or
of
by two sets F
with values
the sequences consist again of pairs:
s
Fl[X/eGi]
the
the mth stage of the analysis VO' Vl~ Vs if
is given a value (but not both),
value true for all Henkin constants
F'
E ?.
n, coding the logical analysis which
(to indicate whether
v0
is determined
of subformulas
of length
first element is
of the formulas
where
F0 v F2 and
vs
s
is an infinit~
is given the value true, is used when
Fo[X/eG i] v FI[X/6G i]
assumed to be in order
~, and
gets the s
deter-
is given a value.
i Opinions differ; but it would seem that the criterion of obviousness or elegance is useless here since the business is so simple that, realistically speaking, we understand any representation that comes to mind. Presumably a good choice can be made if one thinks about concrete implementation by a computer program.
170
G.
dO(+F)
= ~F}
if
F C l~+
The intended definition readerj
and
-
for
Oo(F ) = IF}
the well-foundedness LK, %
genesis
do(F ) = ~ .
Representation
vs
to the familiar
% > ~ -- and like the less familiar
languages
infinitary
LK~%~
in
are well ordered outwards).
of minimal
semi-valuations
of cut-free rules (of proof).
from the use of
are well founded is obviously related to
of the subformula relation (in contrast
for card
which strings of quantifiers (ii)
F C F-; otherwise
were envisaged).
The fact that the trees considered
languages
if
d+ and J- (n > O) is clear (enough for the intended n n -- subject to modification if~ for example, some computer
necessarily
implementation
and
Kreisel
in (i)~
We consider
in general
by means of finitary trees: finite sets
F.
As is evident
the set of distinct minimal
semi-valuations
which make
(~)
each
F : F C F+
true and each
has the power of the continuum.
F: F C F-
In any c a s %
false ,
at some nodes of the
sentation there are infinite sets of formulas~
'brutal' repre-
provided some existential
formula
is given the value false.
Exercise.
Given
semi-valuations
F+
satisfying
Nevertheless
and
F-~ show how to determine whether any or all minimal
(~) are infinite.
all minimal
semi-valuations
mined by
(F +, F-); specifically
formulas,
say
FN+
Warninss. fifties,
and
FN, at each node
familiar~
all minimal semi-valuations the property:
those trees, cultivated
the only point that needs verification
are coded (since completeness
if (~) has a model at a l l
sets seems pointless
or unordered
the sets
ability)
refinement
footnote
1 concerning much more delicate refinements
recursive
operations
when operations 2For example, F + U (B,B}
is that
only a tree with
iN, FN
theorem)
satisfying
are to be regarded
(but permitting repetitions)
in the present context,
(of the completeness
requires
since the
the tree codes some semi-valuation
(b) The familiar question whether
as ordered (as finite sequences) 'ordinary'
'coded' by binary trees, deter-
N.
(a) The rules for constructing
are perfectly
(~) above).
can be
by their infinite paths, with only finite sets of
or as
that is, for the on_._~e(defin-
discussed
in this paper; cf.
than such things as primitive
or -- as in [KMS], Part II -- the effect of those distinctions 2
on trees are involved.
the use of 'sets' with repetition ~- F" 0 (A~A}
derive
F + U (B}
and a contraction
~- ?- U CA}
as the rule Rep, discussed at length loc. cit.
rule:
from
has much the same effect
G.
Kreisel
171
Perhaps the most familiar construction is this (which differs from the literature of the fifties only in the use of constants of constants
c13 c2~ ...
N
O~ there is one node
is at level i
2n~ N
'successors ', if
if
FN
where 3xA
is
FN+
is
~ G~ A v B
cG i
often.
N'
or
3xA
FN+
is
(NB:
A[x/eG i]
~ G
A v B, namely
FN
and
F+ ~
F-.
If the node
or
~xA
namely
FN
--and < ~ ,
B} ~
FN .
is placed at the end to ensure that all formulas
2n+l~ N
has only i 'successor' at level
2n+2;
we have reSpo
is the first Henkin constant s.t.
A[x/cG i]
to ensure that
3xA
does not occur in
~N;
will be analyzed infinitely
the word 'successor' or 'descendant' conflicts with the familiar tree
orderin$ of nodes to
is at level
is placed before
FN+
(FN~ ~N>.
<~, A[x/eA]} ~-- FN ;
(Note that -- the analysis of -- F + N are eventually analyzed.) N
resp.
< }~ 'equipped' with
2n+l) if
~+F N [-
If the node
(F~ ~},
has
'successor' (at level
2
for our term models in place
indexed by numbers instead of formulas).
are regarded as sequences of the form At level
eG i
N
where the path from < }
to
N
'precedes' the path from < }
if and only if it is an extension of the latter.)
For any
N~ at odd or even levels~ if
+ resp. FN~
FN
is atomic~ then
N
has
just i 'successor'~ namely ~+ + <%, Fp
resp
Finally 3 the construction terminates at
+
<% N
if
i~+ N
and
PN
have an atomic
formula in cormnon.
Exercise.
Verify that there is a level
nO
beyond which the tree has only
atomic formulas at its nodes if no existential formula appears in any Adapting the terminology of tree of
F+ ~
F-
[KMS]~ we call the tree:
w.r.t, minimal semi-valuations~
canonical
FN. 'refutation'
in symbols
TCF
(r+r -) The seneralized completeness theorem says that~ for a class
C
(specifically:
of
172
G.
infinite paths of suitable formulas)~
Kreisel
closed under~
say~ primitive recursive
opera-
tions
T CF
is C-founded
(r+,F)
makes
if and only if no minimal semi-valuation
(F : F C F +} U ('~ F : F E F-}
In view of Ex. (c) on p. 6~ is a special case if predicates C
theorem i is large enough to be a 'basis' for socalled strict ~i
C
~
classes
C-founded binary recursive
are well-founded
and therefore
finite.
we have axioms (in the usual cut-free derivations
F N+ ~
of
FN
of inferenc%
except when
one requires
A[x/t]
For in this c a s % formulations
N~
logic)~ and the
of
N
is
3xA
has an even level and
for
trees)
at the terminal nodes
of predicate
are the usual rules (since ordinarily
_t that does not occur at all in
~A~ dictated by our interest
Readers
device.
First~
say
which are
all variables 3xA
~; in other words~
~+F N
in the particular
nor in
FN
in
term models
by Henkin).
Exercise.
is
of subsets of
trees (like our canonical refutation
F N+
N
should verify (by themselves
ture) that usual derivations
aN
The usual completeness
from the immediate descendants
for some
place of the specific considered
true.
'minimal' can be dropped.
or 3 equivalently 3 for
such that
~C
can be brought
one standardizes
or by inspection of the litera-
to the present form by the following
the given derivation by use of distinct variables 3
'quantified'
(3~)
at the node
that are never quantified.
N
where the
Then one replaces
aN
aN by
differ from cA
if
F~
and the other variables
arbitrarily by cG,. -- NB: This standardization l operation (for the usual topology on binary trees) and~ in particula~j
is a continuous
does not assume the trees to be finite. Readers will have noticed the following seneralization (for classes
C
of the kind above):
tree built up locally according F+ ~
F-
If
true~ that is~ F + ~
F-
This is so because an infinite path in recursively
from an (hypothetical)
Warnin$
C
'pure' 0~ i
Sur%
theorem
C-founded
which makes
IF : F C F +} U
is valid for all semi-valuations
~C.
can obviously be defined primitive
It would be quite simple minded to assume operations
theoretic
our proof~ resp. refutation
trees (just as formulas
letter alphabet).
T
SV C C
under primitive recursive
operations defined in familiar recursion sent context.
recursive)
SV.
(cf. also para. 3 below).
that closure of
of the soundness
is a (primitive
to the usual cut free rules with the end formula
then there is no semi-valuation
U [~ F : F ~ F-}
even
T
or~ for that matter~
terms is appropriate
in the pre-
trees can be recoded by suitable
and sequences
of formulas can be coded in a
But the trees and paths so obtained are
'sparse' (for familiar all purpose measures
under
'special' and~ p e r h a p %
of density).
Nothing is easier
2
G.
173
Kreisel
here than to concoct precise 'sensible' problems on refinements in familiar terms~ and thereby create a so to speak vested interest in the subject of those refinements. When it comes to choice of problems~ delay is usually preferable to error~ as (apparently:
3.
only) experience shows.
Open problems.
Granted the interest of models of relatively low complexity
mentioned in para. 13~ there are at least 3 areas of (potential) application of the observations in para. 2. First of all~ for infinite sets the familiar finiteness theorem plexity).
F+
we should expect some generalization of
(for total and semi-valuations of restricted com-
The paradigm for the generalization is provided by
founded but infinitary 'proof' trees.
Curiously~
~-logic~ with well-
this type of problem was not
mentioned in [KMS]. Secondly~ there is the grey area of those classes (i)
C
for which:
C-founded binary trees (built up according to the usual rules with and
without cut) are in general not well-founded or~ equivalently~ (total or semi) valuations
eC
formulas valid for
are not logically valid~ but
(ii) exactly the same formulas have
C-founded proof trees built up by means
of rules with and without cut. This grey area should be of interest~ in analogy with the heuristic value of cutfree formulations of ordinary predicate logic which were used in the discovery of Herbrand's theorem~ Beth's definability theorem or Craig's interpolation lermna. -NB:
This problemj mentioned and discussed in some detail in [KMS]~ may be deceptive
inasmuch as an improper choice of the classes
C
is liable to trivialize the probl~m
for reasons given at the end of para. 2~ for example -- but only as an example7 -if
C
is required to be closed under all recursive operations. Thirdly~ there is the extension (of 'logic') to mathematical theories which is
mentioned in [KMS] in the course of some 'reappraisals'~ but without an adequate bibliography.
9
V
.V
For example~ the apparently relevant papers by Pljuskevlcus [19] and
Rogava [21] were overlooked. 3Certainly~ there is a massive literature on recursive(ly presentable) models~ with satisfactory positive results when a model is given by its satisfaction relation on all formulas~ for example~ generally for the case of recursively decidable theories~ and 3 less easily~ for specific theories with prime models~ perhaps most recently by Harrington [5]~ there are satisfactory negative results of the kind mentioned in para. i~ for formulas which have no model with a recursive satisfactian relation on its atomic formulas. As pointed out in [KMS]~ the (negative) results of Mostowski [18] and Vaught [24] are less convincing~ by [KMS] and para. 2 above the validity predicates for recursive total and for recursive semi-valuations are certainly arithmetical since the property of~being a C-founded tree is arithmetical if C is the class of recursive (or~ say~ ~ sets.
174
G.
Kreisel
It is perhaps worth mentioning that the problems above~ involving infinite 'proof' figures~ are not intended to bear on the idea that our reasoning is best represented by infinite data, an idea with which I have flirted occasionally and which some of my betters~ for example, Zermelo [ ~ ] and fury.
have embraced with much sound
The simple idea behind the problems is the observation~ coming from
experience with ordinary predicate logic and its finite proof figures~ that the latter have generally been of marginal and at best heuristic use -- certainly
not
often for any (realistic) analysis of reasoning except for its most external features. A subsidiary idea is this:
traditional proof theory is full of 'transfinite inductions'
tacitly restricted to predicates of low logical complexity;
there is just a chance
that this restriction may have a sensible use in some context~
like the C-founded
trees above, where one has to do with figures which are not well-founded~ but Cfounded for
~.
C
of low complexity.
Adequacy of 'semantics' for predicate lo$ic:
an attempt at speaking for
the 'silent majority', SM for short~ of logicians (and mathematicians);
especially
in connection with the wide-spread view that this SM is apathetic or even hostile when it comes to foundations and~ more generally, towards logic. unconvincing;
The view is
such logical problems as the independence of the parallel axiom excite
the SM~ and there is little doubt that a proof (of the asymptotic version) of Artin's conjecture on zeros of polynomials in p-adic fields~ by Ax-Kochen and Er~ov, would have caused less exci~ment if it had not used logic.
Familiar
'explanations' of
this view which are current in the logical literature, range from the great depth and difficulty of (logical and~ in particular)
foundational problems to their lack
of meaning or precision; in short 3 however much the explanations may differ~ they certainly have a dramatic character.
The depth of foundational problems is commonly
supposed to be such that it is even difficult to find any coherent
'account' (of
the area of experience for which foundations are to be given), let alone a correct analysis or theory. Generally there will be some restrictions on the type of 'account', at least tacitly; in the case of predicate logic two types are familiar, namely model theoretic and operational semantics; the latter was touched on in passing by Gentzen (on the meaning of logical operations being determined by rules for their 'use')~ developed in an imaginative way by Lorenzen in
his operative Lo$ik and his dialosues (where
the 'use' consists in playing certain two-person games with the odds in favor of the defense)~ and somewhat more polished by Prawitz. 4
The area of experience invol~ed
gFor details, see~ e.g.~ App. i of [13] ~ where in particular various different ways are pointed out how this kind of thing can be done almost mechanically, starting with various versions of familiar realizability.
G.
Kreisel
is here supposed to consist of the linsuistic logical reasoning~ too often~
in familiar
from a genuinely empirical
restrictive arguments
and codified
178
evidence
point of view this
because we use constantly mathematical~
to prove logical
actual acquisition
theorems.
of knowledse
provided by our ordinary
formal systems.
Consequently~
is concerned
As cannot be repeated
'evidence'
-- there is something
proofs is prima facie implausible
'need' of further evidence.
(NB:
Of course~
proofs although~
in fact~ they are often
tion of set theoretic
axioms and auxiliary
proofs is the selec-
-- explicit -- definitions
to which those
are applied.)
Both model theoretic and operational to the 'precision'
blems orj in effect~ considered w.r.t,
are unimpressed:
-- minded
accounts have the -- unquestionable
logician in starting with
adequacy conditions:
the relevant
adequacy conditions.
completeness
semantics).
Each
a reason which would certainly
is sound and c o m p l e t e
(semantics)
'account' verifies
but~ in the case of operational
pro-
(of the rules
its -- own --
logicians
as Brouwer
appeal to both is that there are
for which the $iven
that is~ this evidence
--
'fundamental'
and soundness
Not only the SM but also such vociferous
not only many different meanings
principle~
in
these purely logical inferences may
but the obvious heart of the mathematical
attraction
'special' about
and certainly
omitted in practice:
inferences
set theoretic
the view that -- as far as
purely logical (first order)
be said to occur also in the mathematical
is itself highly
in particular~
'linguistic
evidence'
leaves the semantics undetermined
semantics~
in
it is wildly undetermined~
and simply not even remotely connected with the meaning we have in mind in ordinary reasoning when we use logically complex expressions However~
at all.
there is a subtler adequacy condition which forces itself on those of
us who have enlarsed our experience beyond ordinary or mathematical studying Gentzen's work:
There simply is no doubt about the
practice~
free rules (which of course still leaves doubts about an adequate analysis interest).
Certainly~
there is no evidence
that I know for supposing
found those rules by use of any kind of operational to it as a passing afterthought
semantics;
of this
that Gentzen
he may have referred
at a time -- the thirties -- when the slogan
meaning of a word is its use' was at least not as hackneyed as later. and I believe I am speaking here for a SM (albeit hypothetical) with cut free rules -- operational
by
'raw' interest of cut-
'the
Nevertheless
--
which is familiar
semantics was best in at least one respect~
for
Tin such circumstances~ at least the SM would suppose that what these differenL meanings (compatible with the $iven evidence) have in common would be 'superficial'~ and significant progress would require selection among these meanings; naturally in terms not included in the evidence considered. (Of course~ it is not disputed that -- the analogue of -- an operational semantics is perfectly natural for such things as lo$ic free numerical computation. It is the 'grand' scheme of regarding all mathematics as essentially computational that is in doubt -- and even the more 'modest' idea that computational features are senerally significant for mathematica~ reasoning.)
176
G.
Kreisel
the simple reason that it was the only semantics which gave a suitably distinguished place to cut-free rules.
In contrast, for model theoretic semantics these rules
had initially heuristic value, but no obviously permanent place. theorists), using clumsy jargon familiar to mathematicians~
The SM (of model
spoke of not understanding
cut-free rules though, of course, they knew as well as anybody what these rules were (was sie sind); they did not know any convincing use for those rules within the model-theoretic scheme (was sie sollen). The generalized completeness theorem discussed in the present article provides a candidate for such a use.
Of course it is coherent and capable of development;
that is not the issue at all.
What is needed to establish the proposal, is indicated
in the open problems of para. 3.
Or, put slightly differently, what has to be
established is that
the expressive power of first order language and its formal laws are not only (occasionally) useful in areas of mathematics which consider arbitrary structures, but also in those that fall, for example, into the'grey area' of para. 3.
It is probably superfluous to add that there is a quite separate area of study, of structural properties of proofs, which at least the SM does not expect to be analyzed adequately in model theoretic terms.
Remark.
The reader will have noticed that intuitionistic rules and their
semantics are not considered here at all.
Some reasons for this omission are given
in Appendix 2.
APPENDIX i:
Takeuti's Conjecture
The account below is autobiographical~ easy to write but b e c a u s %
not only because such expositions are
it would seem, the average (informed) reader has probably
had much the same experiences with this subject. Originallyj
the interest of the conjecture (for type theory without the axiom
of infinity) was said to consist in its connection with the consistency of type theory with the axiom of infini~
; this was later refined to an equivalence, by
primitive recursive methods~ with socalled i -- consistency [12].
Like most people
I felt very skeptical about the value of this connection for a consistency proof~ I although it may fairly be said that it was of value to Takeuti in the case of ~lanalysis. Sch~tte's discovery [22] of a relation between the conjecture and 'model theoretic' problems concerning semi and total valuations certainly
did not remove that skep-
ticism; once one considers this sort of problem in model theoretic terms at allj one
G.
Kreisel
177
is almost bound to speak of~ say~ e-models of the theory of types, and once one does this one has a trivial consistency proof anyway; cf. the obvious model theoretic proof of Gentzen's Hauptsatz (by use of the completeness of cut-free rules and soundness of rules with eut).
And certainly~ if one is interested in model theoretic
problems at all~ the particular one corresponding to Takeuti's conjecture would be low on anybody's list.
If anythingj the relation with model theory reduced my
interest (wrongly, inasmuch as I am interested in the present paper). When various proofs of Takeuti's conjecture were found~ consisting essentially of a closer inspection of Henkin's exposition of 'general' models for type theory (by Tait~ Takahashi~ also Prawitz)~ I was interested -- necessarily more so than by the originally proposed u s %
for a consistency proof:
surely there was something
that could be done in this direction~ and I mentioned possible directions to various people (obviously~ the circumstances behind the particular choices are of no general interest~ and they were ephemeral anyway). A first attempt was to find something of interest in the socalled theory of species~ the formally intuitionistic analogue of impredicative analysis.
Here one
had the old stand-by of 'explicit definability results' established by Prawitz [20]. One of my own favori~s was closure under Markov's r u l %
which I sketched very briefly
[12]~ and Girard [3], then a student at Paris, and others later worked out properly. A -- to me -- much more interesting direction,
suggested by Takeuti's conjecture,
was a so to speak radically nonmodel theoretic problem~ nowadays called normalization procedure (in contrast to the original normal form theorem) namely whether certain specific (cut elimination) procedures terminate.
In
1969/70 I
learnt a bit, from
Prawitz, about natural deduction systems which I had previously ignored on being assured that they were 'equivalent' to the ones I knew; cf. [26] to see how far this is from the truth.
Whatever doubts remained about any special significance of
natural deduction formulation%
it was at least clear which normalization procedures
were to be considered.
1969/70 Girard,
Also in
who had learnt GSdel's functional
interpretation from Shonfield's book Mathematical Lo$ic, worked out such an interpretation for the theory of species by use of functionals with a Erima facie somewhat peculiar type structure.
Again~ whatever doubts remained it was clear that (i) the
work should be brought to the attention of proof theorists~ in particular at the Oslo Symposium 1970 ~ and
(ii) if the work was basically correct it should be easy
to adapt it to a proof of normalization of the theory of species formulated style of natural deduction. normalization,
6
in the
I myself was particularly taken by one application of
to finding the realizations of existential theorems provided by a
6For people with practice in such systems; (ii) had a striking proof since I mentioned (ii) to Martin-L~f (and Prawitz) in the train to Oslo and gave Girard's text to Martin-Lof~ who already during the Oslo conference convinced himself that (ii) could be carried out.
178
G.
Kreisel
given proof [13] J later extended by Mints normalization
problem had displaced
This impression questions
[17].
turned out to be wrong in the following sense:
about self-referential
sentences
for cut-free analysis
know for which classes (of generally undecided be proved in analysis
At this stage it seemed that the
the normal form problem.
itself -- whereas
sentences)
the termination
dures considered cannot be so proved (for the classes answered
for certain
it was necessary
Takeuti's conjecture
of the normalization
in question).
accordin$
in contrast
to lo$ical form).
to procedures
by Girard in [4] where~ Finally~
however~
to classical
[23] for analyzing
kind of structural
properties
analysis)
referred
is left implicit.
APPENDIX 2:
procedure according
to
was made in Chapter III of Statman's
properties
of proofs of
~l-theorems;
the
to at the end of the main text above.
Intuitionistic
To what extent are the considerations intuitionistic
The work was greatly extended
use of the normalization
structural
proce-
to the content (of the end formula~
the basic procedure
a really convincing
logical form (adapted dissertation
according
proce-
This was
in [16] for certain useful classes by means of primitive recursive
dures which we called normalization
to
can
Rules
of para. 4 inadequate
because they neglect
rules altogether?
Put differently: Should we do better if we looked for a view that applies equally to classical as to intuitionistic
rules? (though obviously such a view would not be close to
actual reasoning at the present time where
the relative
importance
of the two kinds
of rules differs enormously). On (i)
present knowledge
I am doubtful
For Kripke models
in finding analogues
results
that there is great difficulty
to para. 2 even if some slightly subtler notions are needed
than minimal semi-valuations. in the models
for two reasons:
it seems hard to believe
themselves~
(There may be a difficulty
if one is not interested
but merely wants to use them as a tool for certain formal
such as closure properties
since such results
tend to require the formaliza-
tion of model constructions in the theory considered.) (ii) Heyting~
For the originally which involves
intended
logic seems simply uninteresting. look at proofs~
intuitionistic
interpretation 3 by Brouwer and
proofs -- and not expanding universes
in particular~
If one wants to know about proofs~
at their structural
properties
only at such statements 3 as those of first order logic~ quite i__nndirectly~ namely only via the interpretation NB:
as in (i) -- predicate
This view is of course absolutely
one should
and relations;
not
into which proofs enter
of the logical operations.
orthodox (for Brouwer's
--
original intentions).
G. K r e i s e l
179
REFERENCES [I]
E. W. Beth, La crise de la raison et de la logique, Collection de logique math. A, No. 12, 1957; rev. JSL 23 (1958) 35-37.
[2]
J. Y. Girard, Une extension de l'interpr~tation de Godel ~ l'analyse et son application ~ 1 .i . . / ellminatlon des coupures dans l'analyse et la theorie des types, pp. 63-92 in: Proc. Sec. Scand. Logic Symposium, ed. Fenstad, Amsterdam 1971.
[3]
, Quelques resultats sur les interpretations fonctionnelles, pp. 232-252 in: Springer Lecture Notes 337 (1973).
[4]
Three valued logic and cut elimination: the actual meaning of Takeuti's conjecture, Dissertationes math. (to appear).
[5]
L. Harrington, Reeursively presentable prime models, JSL 59 (1974) 305-509.
[6]
G. Hasenjager, Eine Bemerkung zu Henkins Beweis fur die Vollstandigkeit des Pr~dikatenkalkuls der ersten Stufe, JSL 18 (1953) 42-48.
/
.
1
[7]
0 G. C. Jockusch, Hi-classes and boolean combinations of recursively enumerable sets, JSL 39 (197~) 95-96. See also recent numbers of: Algebra and Logic.
[8]
S. C. Kleene, Introduction to metamathematics, Amsterdam, 1952.
[9]
H. J. Keisler, Model theory for infinitary logic; logic with countable conjunctions and finite quantifiers, Amsterdam, 1971.
[I0]
G. Kreisel, Note on arithmetic models for consistent formulae of predicate calculus. FM 37 (1950) 265-285 and Part II, Proc. XI Int. Congress of Philosophy, 14 (19531 39-39.
[ii]
, A variant to Hilbert's theory of the foundations of arithmetic, Brit. J. Phil. Sc. 4 (1953) 107-127.
[12]
Church's thesis: mathematics, pp. 121-150 in: Amsterdam, 1970.
a kind of reducibility axiom for constructive Intuitionism and Proof Theory, ed. Myhill et al.,
[15]
, A survey of proof theory II, pp. 109-170 in: Logic Symposium, ed. Fenstad, Amsterdam 1971.
[14]
G. Kreisel and J. L. Krivine, Modelltheorie, Spinger Hochschultext 1972.
[15 ]
G. Kreisel, G. E. Mints and S. G. Simpson The use of abstract language in elementary metamathematics; some pedagogic examples, pp. 38-131 in: Proc. 1973 Boston Logic Colloquium ed. Parikh, Springer Lecture Notes 453; vgl. PPS.
[16]
G. Kreisel and G. Takeuti, Formally self-referential propositions for cut-free classical analysis and related systems, Dissertationes math. i18 (1974) 4-50 .
[17 ]
G. E. Mints, On E-theorems, Zapiski 40 (1974) 101-118; siehe auch PPS.
[18]
A. Mostowski, On recursive models of formalized arithmetic, Bull Ac. Pol. Sc., cl. III, 5 (1957), 705-710.
[19]
A. J. Pljuskevlcene, A sequential variant of R. M. Robinson's arithmetic system not containing cut rules, Proc. Steklov Inst. Math. 121 (1972) 121-150.
9
v
.V
Proc. Sec. Scand.
180
G. K r e i s e l
[20]
D. Prawitz~ Some results for intuitionistic logic with second order quantification rules~ pp. 259-269 in: Intuitionism and Proof Theory, ed. Myhill et al.~ Amsterdam~ 1970.
[21]
M. G. Rogava~ Sequential variants of applied predicate calculus without structural deductive rules 3 Proc. Steklov. Inst. Math. 121 (1972).
[22]
K. Schutte Syntactical and semantical properties of simple type theory~ JSL 25 (1960) 305-326.
[23]
R. Statman, Structural complexity of proofs~ Dissertation~ Stanford~ 1974.
[24]
R. L. Vaught~ Sentences true in all constructive models~ JSL 25 (1960) 39-53.
[25]
E. Zermelo~ Grundlagen einer allgemeinen Theorie der mathematischen Satzsysteme~ FM 25 (1935) 136-146.
[26]
J. Zucker, Cut-elimination and normalization, Annals of Math. Logic 7 (1974) 1-112.
[27]
E. G. K. L o p e z - E s c o b a r and W. Veldman, I n t u i t i o n i s t i c completeness of a r e s t r i c t e d s e c o n d - o r d e r logic, this volume.
[28]
D. Prawitz, Comments on G e n t z e n - t y p e procedures notion of truth, this volume.
and the classical
P.S. At least two papers in this volume - [27] and [28] which were added to the list of references above at the last minute - are d i r e c t l y r e l e v a n t to the present paper, for the f o l l o w i n g reasons: [27]: The p a r e n t h e t i c a l w a r n i n g in (i) of A p p e n d i x 2 (of the present paper) does not apply if, instead of Kripke models, some such variant as that of [27] is used which can be proved to be complete by formally i n t u i t i o n i s t i e methods. - Remarks on [27] (to avoid any misunderstanding): The use of [27] just m e - ~ o n e d is not affected by the w e a k n e s s e s of its general i n t r o d u c t i o n concerning c o m p l e t e n e s s of formal rules, say, R w.r.t. (intuitionistic) validity or provability. Contrary to p. 2 of [27] this, obviously, does not require each ~ (method of) proof, #, to be actually formalized by some R - d e r i v a t i o n , or - in Heyting's language cited lee. cit. - to be 'embraced' by R; it is enough that some R - d e r i v a t i o n , p o s s i b l y quite d i f f e r e n t from ~, has as endformula the p r o p o s i t i o n proved by ~. More generally, it can not be assumed that possible uses o f [27] will depend on any (close) r e l a t i o n between its semantic and the o r i g i n a l l y intended m e a n i n g of i n t u i t i o n i s t i e validity - quite apart from the fact thatthe latter suggested itself before anybody had much experience of the subject (and so its r e l e v a n c e has to be tested in the light of later experience). [28]: Many readers may find the present paper more useful after first reading [28], whe#e the completeness of out-free rules w.r.t, validity for s e m i v a l u a t i o n s is proved in more detail than here (and also reading Part Z of [KMS] for its - c o r r e s p o n d i n g - details on the completeness of rules with out w.r.t total valuations, and on the incompleteness w.r.t, the latter of cut-free rules, for example in the case of r e c u r s i v e valuations). - W a r n i n g s on some implicit assumptions in [28] which are capable of being tested, but have not been checked: (i) The particular, usual choice of cut-free rules may be uniquely determined in a much less s u p e r f i c i a l sense than that of [28], p. 2, i. 3; namely - as e x p l a i n e d in [KMS] - up to c o n v e r t i b i l i t y by means of suitable rules p r o v i d e d completeness for a s u f f i c i e n t l y wide class of c is required (in analogy to 'glamor' theorems of modern m a t h e m a t i c s where global analytic conditions, here taken to c o r r e s p o n d to a suitable
G. Kreisel
181
choice of r ensure an algebraic structure). (ii) Para. 4 of [281 (pp. 14-15) assumes that the role of socalled side-formulas should be analyzed in terms of constructivity in its modern sense, which requires restrictions on the (metamathematical) methods of proof; in particular, the strategy of the present paper, which makes dist~ctions only in terms of definability or 'complexity', is inadequate. But this assumption has not been tested. (Indeed, before accepting the assumption, one would first test the strategy of [KMS], Part If, where not the complexity of the derivation trees themselves is used as principal tool of analysis, but definability and other properties of the transformations of proof figures into proof figures). - These warnings evidently do not affect the following more limited aim of [28 ] and of the present paper w.r.t. cut-free rules (and of [KNS], Part I, w.r.t, the usual rules with cut): to show how 6iven encodings, by means of finitely branching trees, of all minimal semivaluations (resp., of all countable total valuations) indeed d e ~ r m i n e the rules in question. As a by-product, the relevant completeness properties of those rules are evident, once one has convinced oneself that all valuations of the type considered are included in the given encodings. P.P.S. (added at the very last minute). The paper by G.E. Mints Finite investigation of infinite derivations, Zapiski 49 (1975) 67-122, contains the ~etails ~of tr~usformuations sketched in Part II of [15] and extensions of the E-theorems of [17], for example, to the functional interpretation introduced by GSdel in Dialectica 1958 (and much besides).
STRONG NORMALIZATION FOR ARITHMETIC (VARIATONS ON A THEME OF PRAWITZ) Daniel Leivant Mathematisch Centrum, Amsterdam The Netherlands
We refer to GENTZEN's natural deduction system for intuitiouistic arithmetic (GENTZEN [36] w
PRAWITZ [71] III.I), for which we give a variant of PRAWlTZ's ([71]
app.A) proof of strong normalization. The main point of departure of this note from PRAWITZ's treatment is this: we define (an analogue to) "strong validity" (called below "stability") explicitly in arithmetic, instead of giving an inductive definition. The raisons d'etre of this variant seem to be: I. It is an alternative, which some people might like (e.g., I do). 2. The formalization of restricted versions of the results within arithmetic is direct (compared to TROELSTRA [73] IV.5,I.4). 3. A more economical measure of complexity on formulae, for the definition of "strong validity'~ is obtained (2.1). As a consequence the role of implicational complexity in normalization proofs is stressed (7.3.). 4. This kind of treatment is applicable to infinitary derivations, where inductive definitions seem to fail altogether.
(This matter will be treated in detail else-
where). In section 6 we indicate an even simpler variant of the proof, but a variant which works only for the disjunction free fragment (and for which 4. fails). A number of metamathematical corollaries (such as consistency of Peano's
Arith-
metic) require normalization only for the negative fragment (i.e.- where B and v are not used). The reader wishing to look only at this fragment may obtain a simple proof of strong normalization for it by reading only the following: l.l(i)-(iii);
1.2; 1.5; 1.6.
2.1; 2.2(i)-(iii); 2.3-2.9. 3.
except 3.12(ii).
In 4.2 consider case (i) only, and take i(H) := w(H0) + w(Hl). Cases [a], [b] in the proof become one trivial case, and case [c] concludes the proof. 4.3. is also trivial. 5.1; 5.2. I am indebted to J. Diller and to J. Zucker for a number of helpful remarks.
D.
183
Leivant
I. REDUCTION STEPS; STRONG NORMALIZABILITY
The reductions assumed
of 1 . 1 - 1 . 3
are defined by PRAWITZ
to satisfy the convention
on parameters
to be replaced by others in course of reductions 1.1.
Detour reductions
z0
z 1
(i)
A0
AI A0&A 1
Z. l A.l
>
([65],[71]).
of PR[71]
All derivations
1.2.4, and parameters
are
are
so as to conform to this convention.
(i=0,1)
A. 1
[A]
A
Z
[A]
(ii) A§
A
B
Here [A] is a set of some open assumptions charged"
of Z of the form A, which are "dis-
(or "closed") by the indicated § Also, no open assumption of A A in [A] (compare LE[73] p.14 and ZU[74] p.36). Z
is discharged
Z(a) Aa
Z(t)
(iii)
> VxAx
At
At
(iv)
Z
[A 0 ]
[A 1]
Z
A.i
A0
A1
C
C
[A.] i A. i c
AoVA
1
>
c
(v)
Z
[Aa]
Z
At
A(a)
[At]
3xAx
B
A(t)
B
B
>
(i=0,1)
184
D.
Leivant
1.2. induction reductions
if the normal form ~ of t is ~, and where
Am l
a(a)
Z
Ao
A(Sa)
Ao
rt,o
(i)
F t's is a logic free derivation for t=s. (Cf. e.g. TR[74] 2.2.33, 2.3.]I for the
t=o
existence and uniqueness of the normal form
At
At
of t and for an eventual formalization of Am Z
A(a)
AS
A(Sa)
(ii)
this reduction).
Am A(a) A(Sa)
Ao
if ~ is Ss.
As A(s)
At
F t'Ss
A(Ss)
t=Ss At
].3. Permutative reductions A
A
A
(i)
- -
A
(Fi) i
BE
A
w
-
-
p ~
[Fi)i
>
p
E
B BE
B
B
Al
A2
Al
A2
A
A
A (Fi) i - p B
A
(ii)
vE A (Fi) i - - p
> ~
(Fi) i p B vE
B
B
where P is an elimination rule. Note that the premise A of p in (i), (ii) above is always the left, i.e.- the major one. 1.4. Semi-proper reductions [A 0]
[A I]
A0
g1
[A.] (1)
AoVA |
B
B
>
(2) B
(i=0,1)
[Aa] >
3xAx
z
B
B
[Aa]
1 A.
A B
(These reductions preserve the derived formula, but may alter the set of open assump-
D.
Leivant
tions, except if the indicated set of discharged they are identical with PRAWITZ's
185
assumptions
([71] 3.3.2) immediate
is empty,
in which case
simplifications).
1.5. Inner reductions If A is a proper subderivation
of E , A > A' by one of the above,
and E' comes
from E by replacing A with A', then E 9 E'. We say then that ~ ~ E' by an inner re-
duction; A > A' by I.I-1.4 we call a main reduction. 1.6. A is strongly normalizable (s.n.), > A2 >
... > An is impossible.
minimal n satisfying 1.7. Remark.
if there is natural number n such that A ~ A I
If A is strongly normalizable
The treatment below may be modified
of permutative
reductions,
(for the case
of §
generalization
to apply to a more general definition
where 0 is allowed to be any inference rule except induction
such a reduction may, however,
of the derivation).
we write ~(A) for the
the above condition.
For applications
is superfluous.
alter the set of open assumptions
of the strong normalization
theorem, however,
this
We therefore prefer to treat the restricted definition,
allowing a greater clarity of the proofs.
2. IMPROPER REDUCTIONS,
STABILITY
2.1. A measure of complexity The measure ~ on formulae is defined by recursion on their length: ~(A)
:= 0 for A atomic
~(A&B)
:~ ~(AvB)
~(VxAx)
:~ ~(3xAx)
~(A§
:= max[~(A)+],
:= max[~(A),~(B)] := ~(A~) ~(B)]
For a derivation A with a derived 2.2. Improper reductions
formula A we also write U(A)
:= ~(A).
*)
Assume that the notion "stability"
and the reduction-step
>o are defined
for deri-
vations A such that ~(A) < n. For A s.t. ~(A) = n we then define
(i)
EO
El
AO
AI
>~
E. A.l l
(i=0, I)
A0&A I
*) Similar notions have been used byH.R. Jervell,
by P. Martin-Lof
and by R. de Vrijer.
186
D.
EA] (ii)
B
A EA] E
-to
A§
whenever
J
,
is stable
A
B
Z(a)
E(t) At
Aa
(iii)
Leivant
for every term t
VxAx E
(iv)
A.
E
l
(i=O, ! )
A, 1
AoVA1 E (v)
At
>o
At
3xAx
Note that these reductions have a combinatorial
do not preserve
the meaning of derivations.
They only
role in the proof of strong normalization.
2.39 We write A >'r A' if for some n _> 0
A -- A0>r A l'zr ...>~ An - A', where>~ is either
or>-. 2.4. Stability It is seen outright A >>
that if A >~ A' then U(A') <- ~(A). Hence the definition
A' uses the notion of stability only for derivations
may define: A is stable if A > ~ A '
from A by substituting
stable derivations
2 96. A is stable under substitution 2.7. Lemma. A is stable i f f A ~
F s.t. ~(F) < ~(A). So we
~ A' is s.n..
2.5. We write A ~-+ A* if A* is obtained free in A and then substituting
of
terms for parameters
for some open assumptions.
(s.s.) if A ~-+ A* ~ A* is stable9
A' implies that A' is stable.
2.8. Lemma. EVery s.s. derivation is stable, and every stable derivation is s.n. . 2.9. Lemma. I f A' is a subderivation of A, and A is s.n. then A' is s.n. and
~(~') _< ~(A). 2.7-2. 9 2.10. Remark.
are immediate
from the definitions.
If A is stable,
then every >~-chain
starting with A is finite (this is
easily proven by induction on the usual logical complexity of the derived formula of A).
D.
Leivant
To prove the converse one needs, prima facie, the fan theorem, uniform bound required
in the definition
servative over Heyting's Arithmetic alternative
characterization
so as to obtain the
of s.n.; the fan theorem is, however,
(TROELSTRA [74]). Note that, in any case,
of stability
3. TREATMENT OF INTRODUCTION
187
conthis
is H 1 I"
INFERENCES AND INDUCTION
ao(a I ) p where p is an introduction-rule an atomic (Post)
If A =- ~
3.1. Proposition.
rule or the replacement rule, and A 0 (and A I) are s.s., then A is s.s.. Proof.
By 3.3, 3.7, 3.11 and 3.12 below.
3.2. Lemma. If A0,AI are stable then so is A B
A0 A
A1
-- A
B A&B
Proof. By induction on ~(A0) + W(Al).
If A > A' then this reduction
is necessarily
an
inner one,
i
v
A0 A' z A
A1 B A&B
where v(A~)+~(Ai)
A O A'
say, then
< ~(A0)+~(AI) , hence A' is stable by induction hypothesis.
A'
is stable by assumption.
By 2.7 A is stable.
If A ~o A'
D
3.3. Lemma. I f A0,A l of 3.2 are s.s., then so is A. Proof9 Immediate from 3.2. 3.4
Definition.
D
Let EA] be a derivation,
"
where [A] is a set of open assumptions F
A
of A
F
of the form A. We say that A is s.s. at [A] if for every stable derivation A' [A] is A stable. 3.5 9 Lemma. Let EA] A be s.s. at EA], EA] A ~ [A]' A' where [A ] ' is the set of copies of ele-
ments of [A]. Then A' is s 8. at EA]' Proof.
Immediate by induction on ~(A). (Note that the same P is substituted for every F occurrence A ~ [A] in 3.2, and that no assumption of F maybe discharged inA in[A].) D --
3.6. Lemma. I f
A
E~] i s s . s . B
a t EA], then
188
D.
Leivant
[A] A
Z -
B
A§
is stable. a v
Proof.
By induction
on v(A).
If Z > E' ~ ~
then v(A')
< v(A), A' satisfies
dition of the lemma by 3.5, and we are done by ind. hyp..
the con-
If
F Z ~o 2' = EA] A
(F
is stable)
B
then Z' is stable,
since A is s.s. at [A] by assumption.
3.7. Lemma. If E A]A is s.s.,
Hence by 2.7 Z is stable.
then
B
[A] A = l B
A§ is
S.8.
Proof.
.
Let
EA**] Z ~-+ 2" -
A, B
A* § B* A is s.s.,
hence A* is s.s. at [A*],
3.8. Lemma. If a is free in Z(a),
so by 3.6 Z* is stable.
So 2 is s.s..
Z > Z', then a is free in Z' (if it occurs there) and
Z(t) 9 Z'(t) for every term t. 3.9. Lemma. If A ~+ A* and a does not occur in any open assumption of A then A ~-+ A*[t/a]
for every term t.
The proofs
of 3.8 and 3.9 are immediate.
3. I0. Lemma. If a is free in A(a) and A(t) is stable for every t then Aa A(a) -
Aa VxAx
D.
Leivant
189
(if at all a correct derivation) is stable. Proof.
By induction
on v(A)
(as in 3.2).
If
A' (a) E >E'
-
Aa VxAx
then ~(A')
< v(A) and by 3.8 a is free in A' and A'(t)
the induction
hypothesis
by assumption. 3.11.
A' is stable.
By 2.7 2 is stable.
is stable
for every t. Hence by
If E ~o l' - A(t) then 2' is stable outright
D
If A(a) is s.s. then so is Aa
Lemma.
A(a) E
Aa
-
VxAx
Proof.
Let A*(a) I ~--+ E* -
A a
VxA*x By 3.9 A~-+ A*(t) required. 3.]2.
for every t, so A*(t)
is stable.
By 3.]0 then E* is also stable,
D
Lemma.
A (i) I f A t i s
A A F t's A A At ond At t=s . (ii) I f AA is s.s., then so are A , A AvB BvA BxAx As A (iii) If A is s.s. then so is ~ p where 0 is an atomic (Post) rule. Proof.
s.s., t h e n s o a r e
Similar
3.]3. Lemma.
to 3.2-3.3. (Note thatF s't is logic free, hence stable outright).
E is stable, and for every term t If A~
[At] is s.s. at [At], A(t)
then
[Aa] E
A(a)
H ~ A5
A(Sa)
IND
At
is stable for every term t. Proof.
By induction
on ~(I) + ~(A) + T(H), where T(H) is defined
T(t)
:= O,
if t is a term and for no term s
T(t)
:= T(S) + l, if ~ is SS,
~ = Ss,
as follows:
as
D.
190
T(H)
Leivant
:= T(t), if the main inference-rule
of the derivation H is IND,
with t as a proper term.
NOW if H > H' by an inner reduction in the proof of 3.10 H' satisfies hypothesis
then T(E') = T(H), 9(Z')+9(A')
the assumptions
and as
of the lemma. Hence by the induction
H' is stable.
If t = 0, H > H'
as in 1.2(i),
then H' is stable by 3.12(i)
and the assumed
stability of Z. If [Aa] A(a) H > H' = Ao
A(Sa) [As] Ft,Ss
A(s)
t=Ss
A(Ss) At
then T(H0) < T(H), so by the induction hypothesis s.s. at [At], so K' is stable by 3.]2(i). 3.14. Proposition. Proof.
By assumption A(t) is
The lemma follows by 2.7.
If %, A(a) of 3.13 are s.s. then so is ~.
3.14 follows 3.13 like 3.11 follows
4. TREATMENT
H 0 is stable.
3.10.
OF ELIMINATION-INFERENCES
4.0. Notations
and definitions.
For the sake of brevity we shall skip cases for disjunction-rules, be treated in complete analogy to the 3-rules. subderivation
of H, and h(H)
denote
which are to
Let H0(H I ) denote the left (right) main
the height of H (as a tree).
If H0,H 1 are s.n., we define for H the measure i(H) byi(H) := <~(H0),%(H0),w(HI)> (where <,,> is the lexicographic 2 .~(~0) + ~.X(n0) + ~(~I)).
ordering.
One may also think of the ordinal-measure
Let Z z 3xAx" Z [At] . stable under z at [At] if A is stable, A IS 0 At
Z ~ ... ~
then
BxAx
[ 2t ]
and whenever
is stable.
A
4. I . Lemma. If [A]is stable under Z at [A] then (i)
if [A]), [A~' then A' is stable under Z at [A]'/ A
A'
(ii) if Z ~ Z' then A is stable under Z' at [A]. Proof.
(i) is analogous
to 3.5. (ii) is immediate
from the definition.
D
D.
4.2. Main lemma. Let H ~ H0~HI s A
)
191
Leivant
be given s.t. either
p
(i)
p
is an elimination-rule other than 3E, and HO,HI are stable; or
(ii)
p
is 3E, [A(a)]
Hl(a )
H0
~
3xAx
B
say,
B
Hl(t) is stable under HO at EAt].
~0 is s.n., and for every t Then H is stable.
Proof. By induction on i(H). I.e., we assume that every @ satisfying the lemma and i(@) < i(H) is stable,
the conditions
and we prove that H > A ~ A is stable
of
(which
implies that H is stable by 2.7). Case [a]: ~ > A by an inner reduction in ~0. Then w(A 0) < w(H 0) so i(A) < i(H). If (i) applies to H (and to A) then A satisfies (ii) applies - by 4.](ii).
Case [b]: H > A by an inner reduction < i(H). A satisfies if (ii) applies. Case [c]: argument
the conditions
of the lemma by 2.7, and if
So by the induction hypothesis
A is stable.
in H I. Then A 0 = H O, w(A I) < v(~ ]) so i(A) <
the lemma's conditions by 2.7 - if (i) applies,
By the induction hypothesis
(i) applies,
A is stable.
and H > & by a main detour reduction.
is similar for &E and VE).
H!
I EA]
H ~
H0 ] [
L
F
B
HI
A§
>
A
H 0 and HI are assumed stable,
[A] F
~
A
B
so
ii I [A] H 0 >o
hence A is stable Case [d]:
A
-
F
(2.7).
(ii) applies,
and H > A by a main detour reduction.
H0
[Aa]
At
H|(a)
H ~ BxAx
B B
H0 [At] ~
HI(t) B
and by 4.|(i) -
~
A .
Take the case p = §
(the
192
D. Leivant
By condition Case [el:
(ii) A is stable outright.
(i) applies and H ~ & by a permutative
II O
H ---
!xO
reduction.
r 1 (a)
3
B B
3E (II 1 )
[Aa] Fl(a) ro
B
(E 1 )
3xAx
C
H 0 is stable by assumption,
p
_-- &
3E
H 0 > F! (by a semi-proper
reduction, 1
cf.
1.4), so F| is 0
stable, and v(Fl) < ~(H0).-- H ] is stable by assumption, hence A is stable. A = F 0 is 0 a subderivation of H , hence it is s.n. by 2.9, and v(A 0) s v(H 0) while %(A0) <%(~0). So i(A) < i(H). To show that A satisfies condition
(ii) of the lemma, it remains to see that
whenever F0 3xAx > "'" >
(*)
@ At 3xAx
then @ [At]
r1(t)
-: E
B
(~1)
P
is stable. But if (*), then @ [At] H0 >
... Y
0
Fl(t)
=
;
B
so ~(E 0) < ~(H 0) and i(E) < i(H). H0 is assumed stable, hence H 0 is stable (2.7), while E I ~ H I is assumed outright.
Hence E satisfies case (i) of the conditions
duction hypothesis
E is stable. Hence A satisfies
and by the induction hypothesis
A is stable.
stable
of the lemma, and by the in-
case (ii) of the lemma's condition,
D.
193
Leivant
Case Eli: (ii) applies, and H ~ A by a permutative reduction.
(i) [Aa]
r0
rl(a)
3xAx
3yBy
(2) EBb]
(1) 3E
HI(b)
3yBy
C
[Aa]
(1)
(2) [Bb]
rl(a )
~[l(b )
3yBy
C
F0
C
3xAx
H 1 is stable at [Bb] under H0,
l
(2) 3E
(2) ~E
! A (a)
-- A .
I
(1) 3E
H 0 y FI, Hence (by 4.1(i)) H 1 is stable at [Bb] under
F I. We conclude that A| is stable, and that i(A) < i(H) like in case [el. It remains to show that for every t
Al(t) is stable at [At] under F0; i.e.,
that if
(*)
FO> ... >
0 At 3xAx
then @ [At]
[Bb]
rl_t.( )
(b) C
3~By
z: Z 3E
C is stable. But, like in [e], (*) implies that H 0 >
... > _0,a so z-0 is s.n., v(E 0) <
< ~(H 0) and i(Z) < i(H). (~(E |) is well defined, because Z | ~ HI which is stable by assumption, E satisfies case (ii) of the lemma's conditions by 4.1(ii),
so s.n.).
so by the induction
hypothesis E is stable. Hence A satisfies case (ii) of the lemma's conditions, the induction hypothesis A is stable. Case [$]: (ii) applies and H ~ A by a semi-proper reduction:
Hl
H0
~ 3 E
A
! ~ H
~ A
then A is stable by assumption.
This concludes the proof.
0
and by
194
D.
Leivant
~0 (~i) p where p is an elimination-inference,
4.3. Corollary.
If H = - A then H is s.s. .
s.s., Proof.
If p is &E, §
or VE this follows case (i) of 4.2 outright.
and HO,H j are
If O is 3E (or anal-
ogously - rE) then, if
H ~.+ H* -
N0*
[Ba] Hl*(a)
3xBx
A
3E
then a does not occur in any open assumption
HI~
of H, so we have that
[Bt] H1*(t )
for any t and any stable ~t" In particular,
H O* >
... >
0 Bt
>o
if
@ Bt
3xBx then Bt@ is stable, conditions
since H 0* is stable
(by assumption).
of case (ii) of 4.2 are satisfied,
5. THE STRONG NORMALIZATION
@ [Bt] So ~l,(t)
hence H* is stable,
is stable,
THEOREM
5.1. Theorem. Every derivatfon
H is s.s..
Proof. By induction on %(H). For h(H) = I, i.e., H is a singleton-derivation, theorem is immediate
from the definition
rule,
subderivation(s) elimination
rule, an atomic
(Post) rule or the
then H is s.s. by 3.1 and by the ind. hyp. applied to the immediate of H; if p is IND then H is s.s. likewise by 3.|3, and if p is an
rule - by 4.3.
5.2. Corollary.
the
of s.s..
If the main rule p of H is an introduction replacement
and the
as required.
(strong-normalization
Proof. By 5.1 and 2.8.
theorem). Every derivation
is s.n..
D.
195
Leivant
6. ANOTHER VARIANT OF THE NORMALIZATION-PROOF
6.0. For the disjunction-free
fragment we may give an even simpler alternative
as below. Note that disjunction
is eliminable
in intuitionistic arithmetic
proof,
(TR[73]
1.3.7, LE[73] IV.l). 6.1. Define the reduction relation > as follows. (I) Detour reductions
- like 1.1.
(2) Detour reductions
through BE:
Z0 A 80
ZI B
Z0
A&B
&0
A
BE 81
BE
A&B
~
gl
A
BE
BE
A&B Ak
A
Ak
9
BE
BE
A&B
A
A and similar clauses corresponding correspond
to the q-reduction
to the other detour-reductions.
(These reductions
of LEL73] IV.2.3 for a system with existential
instant-
ation). (3) Induction reductions
- like 1.2.
(4) Inner reductions - like 1.4. The definition of strong-normaiizability
and the measure v(A) follows as in I.
6.2. To the clauses of 2.2 add to the definition
80
z0
zl
A
B
improper reductions
through BE:
~0
A&B
A0 - -
BE A&B &k
of ~
A BE A
>o &k
BE
BE
A&B
A etc.
The definitions
of stability,
and s.s. follow.
6.3. The treatment of the introduction-rules that of the elimination-rules and permutative obtained.
reductions
is considerably
simplified
the same as in 3, while
(cases are treated separately
do not occur). Thus the strong normalization
Here we get all the corollaries
mutative reductions,
is now essentially
of normalization
because of the presence of reductions
theorem is
without referring through BE.
to per-
196
D.
Leivant
6.4. Permutative reductions of the most general kind (|.7) may be reinserted into the treatment without destroying its simplicity. Let p|,...,O k enumerate the instances of BE in a given derivation 4, and ~|'''"~k their respective major premises. Say that A. is s~7~ordinated to A~ iff A. occurs in -i -j -l the minor premise of pj. Define
oA(~i)
:= __max[o4(A j)_ I A._l is subordinated
h4(Ai)
:= the height of ~i in 4
~3(4)
:=
nr
:= Z{hA(~i ) I ~
Then, if 4 ~ 4' by a permutative
where i) = r}
reduction,
(usual summation)
relative reduction sequences
by induction on <~(A),~3(4)>,
A > A' by a non-permutative
.
then ~3(4') < D3(4) and ~(4') = ~(4).
Now, every derivation 4 is strongly normalizable allowing permutative reductions,
to A.]_j + 1
since
reduction ~ ~(&') < ~(A);
4 > 4' by a permutative reduction ~ ~(4') = ~(A) and ~3(4') < ~3(4).
7. FORMALIZATION OF THE PROOF IN ARITHMETIC
7.1. The formalization within arithmetic of our proof of strong normalization tine, except for one point: the arithmetization Let Stk(F) be a tentative abreviation is stable" (where derivations
is rou-
of the stability predicate.
for the formalization of "~(F) ~ k and F
are identified with their ~odel-numbers).
Strong norma-
lizability is seen outright to be formalizable as a E20 predicate (Sn say). Since
Sto(4) +-+ Sn(A) if ~(4) = 0, St 0 is also a E 02 predicate. If ~(4) = n+1 then 4 >o 4' is in general formalizable as a predicate of the form (3r < A') [Stn(F ) & F(&,4',F)] where F is a p.r. relation. 4 > > A '
is 2|0 in >~
is E 0I in St n.
hence A ~ > A '
Stn+l(r) ~ vA'EA>>A' § Sn(A')J, so Stl(A) is of the form V[E
§ E J which is classically a H3-predlcate;
we can see by induction that St
n
is classically equivalent to a H~+2-predicate.
7.2. Consequently, we may formalize within H~+k-arithmetic n) the normalization-proof
(where k is fixed for every
for all derivations A, satisfying:
~(A) ~ n". Some consequences
and for n e 2
"if A occurs in A then
of this are given by TROELSTRA ([73] IV.4).
D. Leivant
197
7.3. Our proof of normalization illustrates the essential role of implication in formulae complexity, since implication is the only logical symhol counted for the measure ~. By 7.2 normalization of derivations with a bound on the (negative) nesting of implications in the formulae (but with no bound on the alternations of quantifiers) is formalized within arithmetic. Thus, for example, Heyting's Arithmetic (HA) is not conservative over 0 Griss' positive arithmetic (NA. Cf. LOPEZ-ESCOBAR [74]) even for Nl~sentences, because the consistency of NA is provable in HA (in fact even in a simple fragment of HA).
REFERENCES G. GENTZEN [36], Die Widerspruchsfreiheit der einen Zahlentheorie, Math. Ann. I12 (1936) 493-565. D. LEIVANT [73], Existential instantiation in a system of natural deduction for intuitiouistic arithmetic, Report ZW 23/73, Mathematisch Centrum, Amsterdam, 1973. E.G.K. LOPEZ-ESCOBAR [74], Elementary interpretations of negationless arithmetic,
Fund. Math. 82 (1974) 25-38. D. PRAWITZ [65], Natural Deduction, Stockholm,
1965.
D. PRAWITZ [71], Ideas and results of proof-theory,
in: FENSTAD (ed.), Proceeding8
of the 2nd Scandinavian logic symposium, Amsterdam, 197|, pp. 235-307. A.S. TROELSTRA [73], Metamathematical
investigation of intuitiouistic arithmetic and analysis, Berlin etc., 1973.
A.S. TROELSTRA [74], Note on the fan theorem, Report 74-14, University of Amsterdam, Sept. 1974. J. ZUCKER [74], Cut-elimination and normalization, Annals of Math. Logic ~ (1974) |-||2.
INTUITIONISTIC COMPLETENESS OF A RESTRICTED SECOND-ORDER LOGIC Dedicated
to Kurt
of his
E.G.K.
w
INTRODUCTION,
icate calculus
Sch~tte
65 th
on o c c a s i o n
birthday
L O P E Z - E S C O B A R and W. V E L D M A N
The completeness of the c l a s s i c a l f i r s t - o r d e r pred-
is over
40
years old, n e v e r t h e l e s s most of the proofs
given for it are,
if not wrong,
at least misleading.
From the prelim-
inary discussions
one is often led to believe that what will be proven
is that every i n t u i t i v e l y valid formula of the ~ l a s s i c a l 2 r e d i c a t e ~alculus~
CPC, is derivable using the axioms and rules of CPC.
ever, what is shown,
is often no more than:
ValcPc(A) ~ > where
"ValcPc(A)"
valid
DercPc(A),
stands for "the formula
(i.e., true in all s e t - t h e o r e t i c
is an a b b r e v i a t i o n for "the formula Of course,
How-
A
A
of the CPC is f o r m a l l y
structures)", is derivable
and
"DercPc(A)"
in the CPC".
it doesn't take much to remedy the situation.
All that
remains to be shown is that every i n t u i t i v e l y valid formula of CPC is f o r m a l l y valid. ity of
A
tations,
The latter is justified on the grounds that the valid-
entails that
A
is true in all p o s s i b l e kinds of interpre-
including the s e t - t h e o r e t i c
structures.
The i_ntuitionistic predicate ~alculus, better.
IPC, has not fared much
To start with there is the t r a d i t i o n a l view that i n t u i t i o n i s m
is solely concerned with questions p e r t a i n i n g to specific m a t h e m a t i c a l constructions. terest
Thus,
from a t r a d i t i o n a l viewpoint,
there is little in-
in a t t e m p t i n g to c l a s s i f y those sentences which are intuition-
istically true i n d e p e n d e n t l y of the interpretation.
In addition,
Heyting has e x p r e s s e d the opinion that his system for the predicate calculus
(and the systems that have since been introduced)
are not
E.G.K. demonstrably
complete;
Lopez-Escobar,
for example
W. Veldman
in H e y t i n g
1966,
199 page
i02,
he
states:
It must be remembered that no formal system can be proved to represent adequately an intuitionistic theory. There always remains a residue of ambiguity in the interpretation of signs, and it can never be proved with mathematical rigour that the system of axioms really embraces every valid method of proof. In view of the fact that m o d e l - t h e o r y classical kind
of
mathematics,
there have
(intuitionistic)
such a m o d e l - t h e o r y (w.r.t.
if it is to be part
model-theory
is one of the
(w.r.t.
methods.
heuristic
use then any valid
istically
true
with respect E.W.
must
then be by
if the m o d e l l i n g (in the
sense
interpretations)
for any
probably
the
first
Predicate
sentence
A
person
Calculus.
In Beth
(0.2)
B - ValIPC(A)
~>
DerlPC(A) ,
(0.3)
ValidlPc(A)
~>
B - ValidlPc(A),
"B - ValIpc(A)"
that
and that
is an a b b r e v i a t i o n
"ValidiPc(A)"
combining
(0.4)
(0.2)
stands and
(0.4) no m e n t i o n
for
(0.3)
ValidlPc(A) in
Furthermore of the
intuitionisti-
of being
intuition-
also be valid
a modelling
1956 he
set out to
of IPC
B - ValIpc(A),
Note
theorem
the proof
who p r o p o s e d
~>
and
if
is to be of any
should
DerlPC(A)
models"
some
Obviously
requirements.
(0.i)
where
to obtain
a completeness
mathematics
use to
to the modelling.
Beth was
that
first
sentence
in all p o s s i b l e
for the I n t u i t i o n i s t i c prove
attempts
the m o d e l l i n g )
Finally
of great
for intuitionism.
of i n t u i t i o n i s t i c
theorem
cally a c c e p t a b l e
been many
is to be of any use then
that m o d e l l i n g )
completeness
has been
"A
for
"A
is true
is i n t u i t i o n i s t i c a l l y
we obtain
that:
--> DerlPC(A),
is made
in all
of the
Beth-models.
Beth
valid".
2~
E.G.K. Just
(0.3)
Lopez-Escobar,
as in the c l a s s i c a l
are of a quite
explicitly
defined
ValidiPc(A),
The
same
case,
as-constructions
counterparts, nature.
mathematical
although
definition. itionistie
different
and
obvious,
is true of
the proofs
(0.2)
constructs;
probably
ValidlPc,
W. Veldman
so it is not
closer
in
solely with
however
mathematical in the intu-
to the concept
immediately
and
(0.3),
no explicit
ValidcPc(A);
is much
(0.2)
is c o n c e r n e d
while
has
of
obvious
of proofs-
that
(0.3)
should
hold. As a m a t t e r cause
there
yields
(see Kreisel
principle
for p r i m i t i v e
Kreisel's
proof was not with
respect
to species
"S - ValIPc(A)".
is that the
to be more
easily
be-
of
of Kreisel
recursive
is yet another
of IPC.
It was
that
predicates
by K r i p k e - m o d e l s ;
(classical)
~>
interpretation, of
which
we shall
S - Vallp C
assumptions
in Beth ab-
over
can be shown to
S - ValIPC(A)
than of
introduced
is played
to v a l i d i t y
implication:
notion
of semantics
respect
An a d v a n t a g e
accepted
treatment
correct
point
(see the report
was a result
(which under a few r e a s o n a b l e
There
was a moot
1982).
ValidlPc(A)
formulae
(0.2)
Markov's
be equivalent)
tends
of
(0.3)
damaging
by
B - Valip C
in the proof
that
Even more
but with
breviate
out
1961).
Actually models,
it turned
was an error
Dyson/Kreisel (0.2)
of fact
(0.3).
(mathematical) in Kripke
for i n t u i t i o n i s t i c probably
completeness
1965.
formal
because
proof was
validity
for the
In the c l a s s i c a l
systems,
of the
a main part
fact that
given with respect
the first to this
type of models. Kripke
models
intuitionistic vative
theories.
extension
itionistically
are often
In addition,
properties,
acceptable
used to i n v e s t i g a t e through
some of those
(see T r o e l s t r a
the
of some
the use of some conser-
results
1973).
strength
can be made
intu-
E.G.K.
In the "valid
following
in e v e r y
The
abbreviate by
the
201
intuitionistic
notion
"K - V a l i P c ( ' ) " .
ValidiPc(A)
~>
in K r i p k e
1965,
justified
acceptance,
However,
even
intuitionistic
and
W. Veldman
implication:
is p a r t l y
an
we w i l l
Kripke-model"
(0.5)
versal
Lopez-Escobar,
S - Valip C
it has
its
if one
accepts
shown
and
although
it does
not h a v e
uni-
charms.
completeness can be
K - VallPc(A).
of
(0.5), IPC
Kripke
(i.e.
models
of
to be e q u i v a l e n t
are no use
(0.4))
since
(under
suitable
for
K - Valip C assump-
tions). In v i e w
of the
tuitionistie istie
proof
proof
with
to c o n c l u d e
notion
of
COMPLETENESS ity for
that
intuitionistic
state
the
FOR
of the
of
that
is not does
(logical)
unless
yield one
obviously
not
validity.
quite As
Is t h e r e
that an
an in-
intuition-
identifies
the
case.
correspond
a matter
intuitionistic
a mathematical predicate
such
that
(A)
the
notion
(B)
the
implication
[ValidiPc(A)
~>
(C)
the
implication
[ViPc(A)
DeriPC(A)]
intuitionistic
state
of
We to the
fact we
problem:
IPC:
VIPC(A)
to
IPC w o u l d
However,
S - Valip C
following
PROBLEM
formulae
it is c u s t o m a r y
principle.
S - ValiPc,
prefer
to
remarks
of the c o m p l e t e n e s s
of M a r k o v ' s
ValidiPc(A)
like
above
has
a semantical
~>
mathematics
without
notion
calculus,
of v a l i d -
say
VIPC(A)
character,
VIPC(A)]
is p l a u s i b l e , is p r o v a b l e
making
use
in c u r r e n t
of M a r k o v ' s
principle?
In this can be
solved
restricted
p a p e r we
will
try to
for a r e s t r i c t e d
second-order
show
that
second-order
language
in the
the
completeness
minimal
sense
that
logic the
~.
problem It is a
second-order
202
E.G.K.
variables
are
intended
definable
species.
are p r i m i t i v e towards order
to range
concepts.
is a
The format
over a subclass
It is minimal
problem
(conservative) of the paper
language
A calculus
w
~
w
A formal
w
Soundness
w
Explicit
theories.
w
A spread
which
w
Construction
of a u n i v e r s a l
w
Construction
of the
of a r e s t r i c t e d
as an extension
of the
semantics theorem
for for
generates
second-order
of the
spread
~.
w
"V"
spread
~.
w
"DerR"
w
The c o m p l e t e n e s s
w
Realizations
and the
explicit
IPC.
logic.
of
predicate
theories.
of explicit
theories.
Z.
~.
~, and
Kripke
models.
THE LANGUAGE OF A RESTRICTED SECOND ORDER LOGIC, can be b r i e f l y
of
described
The lanzuage
as follows:
~.
A denumerable
set
Var
A denumerable
set
Par I
For each
a denumerable
n,
calculus.
realization.
spread
spread
of
intuitionistic
spread
,,n,, and the
Symbols
second-
~.
w
I.i
it is a contribution
the r e s t r i c t e d
of the
falsity
~.
Some p r o p e r t i e s
~
nor
~.
w
of
negation
is as follows:
w
i.
neither
we believe,
extension
The
for
of the f i r s t - o r d e r
for IPC because
w
and the
because
Nevertheless,
the c o m p l e t e n e s s
logic
Lopez-Escobar, W. V e l d m a n
of i n d i v i d u a l of i n d i v i d u a l
variables:
v0, Vl,...
parameters:
set
p(n)
of
set
~ (n) of
a0,al, . . . .
n-ary predicate
variables:
n-ary
parameters:
p(n) _(n) (n) 0 ' ~I ' P2 ' .... For each
n,
a denumerable
predicate
E.G.K. Q(n) 0
^(n) ' UI
First-order
connectives:
universal
symbols:
Symbols
1.3
Pseudo-formulae,
1.4
not of
1965.
V
I~
of v a r i a b l e s
,
but u s e d
formulae
terms.
and s e n t e n c e s
occurrences.
Given
Ix0...Xm_iF
term.
only used
in the o p e r a t i o n
will
The a b s t r a c t i o n
Some n o t a t i o n a l
: U nin(n),
are d e f i n e d
in w h i c h
A sentence
as done
all o c c u r -
is a f o r m u l a
be omitted.
will abbreviate w i l l be u s e d
instead
occurring
ters o c c u r r i n g IXl...XnF,
variables
be c a l l e d terms
are
an
F
occurring m-ary
such that
in
F,
then
elementary
in the m e t a l a n g u a g e
ab-
and are
in
in
F.
F
duction having
FOR
of
variables
(parameters)
'(Aoi)'
then and
Parl(F) Par2(F)
Furthermore
then we d e f i n e
A CALCULUS
and p r e d i c a t e
'VPoP 0'.
is a p s e u d o - f o r m u l a ,
parameters
pseudo-formula
Par 2 : UnQ(n).
will usually
w
IR
conventions:
in the q u a n t i f i e r s
F
of
~, I, ~.
of s u b s t i t u t i o n .
Superscripts
If
a atomic
individual
straction
'~A'
in a b b r e v i a t i o n s :
is a p s e u d o - f o r m u l a
are b o u n d
are all the
the e x p r e s s i o n
'I'
V (2)
any p a r a m e t e r s .
x0,...,Xm_ I
P
n. (i)
quantifier
A formula
Abstraction
1.5
(i)
( , ).
1.2
in P r a w i t z
A, v,
quantifiers:
Second-order
without
20S
' ....
Propositional
rences
W. Veldman
~(n) ' U2
Auxiliary
Lopez-Escobar,
R,
the u s u a l
if
Pari(T)
With
~
T
is the
the set of p r e d i c a t e is the a b s t r a c t i o n
= Pari(F),
we a s s o c i a t e
introduction
set of i n d i v i d u a l parame-
term
i : i, 2.
a s y s t e m of n a t u r a l
and e l i m i n a t i o n
rules
for
de-
A, v,
204
E.G.K.
n, V (I)
and
3 (1).
Lopez-Escobar,
W. Veldman
For the s e c o n d - o r d e r q u a n t i f i e r
t r o d u c t i o n rule is standard
V (2)
the in-
(e.g. as in Prawitz 1965), h o w e v e r the
e l i m i n a t i o n rule is w e a k e n e d to: v p ( n ) A ( p (n))
(V(2)E)
A(T) where
T If
is any F
is a set of formulae of
d e r i v a t i o n in 'mere(A) '
w
R
n - a r y e l e m e n t a r y a b s t r a c t i o n term.
~
of
A
instead of
from ~_~A
F.
R
then
F~A
iff there is a
O c c a s i o n a l l y we shall write
@
AS AN EXTENSION OF THE INTUITIONISTIC PREDICATE CALCULUS,
Let us assume that the i n t u i t i o n i s t i c predicate caloulus,
IPC, has
been f o r m a l i z e d as a system of natural d e d u c t i o n w i t h the falsum symbol
's
as a p r i m i t i v e symbol and negation as a defined concept.
Then given a formula
A
of IPC let
A*
be the formula of
tained by r e p l a c i n g all occurrences of the atomic formula by the sentence
I ;
if
&
~ ~
obin
A
is a set of formulae of IPC then we let
&* = {A* : AEA}. is an e x t e n s i o n of IPC in the following sense:
3.I
THEOREM.
then
AU{A}
is a set of formulae of IPC and
A ~ iPC A
A* ~ A * .
PROOF. of
If
~
The only rule of inference of IPC not included in the rules is the rule for
plexity of
A
Moreover that if
&*~A*
I.
shows that,
~
However a simple i n d u c t i o n on the comfor any
A,
I~A
.
is a c o n s e r v a t i v e e x t e n s i o n of IPC in the sense then
n o r m a l i z a t i o n theorems make the following:
& ~ i P C A. for
~.
The latter is a c o n s e q u e n c e of To be a little more
specific let us
E.G.K. 3.2
DEFINITION.
iff
A
Lopez-Escobar,
A formula
is built up from
V, n, V (I)
and
A
of
~
is e s s e n t i a l l y f i r s t - o r d e r
and the atomic formulae by means of
A,
3 (1)
It should be clear that formula
I
A
205
W. Veldman
of IPC,
A
is e s s e n t i a l l y f i r s t - o r d e r iff for some
A = A*.
The d e f i n i t i o n of a normal d e r i v a t i o n in long to write down, however,
the d e f i n i t i o n
s e c o n d - o r d e r logic have been given
~
w o u l d take too
for the case of full-
(explicitly or implicitly)
in
Girard 1971, Prawitz 1971 and T r o e l s t r a 1973 so that it is a relatively simple m a t t e r for the reader to make the a p p r o p r i a t e changes r e q u i r e d for
~.
E i t h e r using the
(strong) n o r m a l i z a t i o n
for full s e c o n d - o r d e r
m i n i m a l logic or m o d i f y i n g the proof for f i r s t - o r d e r i n t u i t i o n i s t i c logic
(V (2)
causes no p r o b l e m because
formula than
VPA(P)),
ization for
A(T)
is always a simpler
it is possible to obtain a (strong) normal-
~.
Because of the r e s t r i c t i o n we have placed on derivation
3.2
~
in
PROPOSITION.
tially first-order order formulae
~
If
V(2)E
a normal
has the following kind of s u b f o r m u l a property.
~
formula
is a normal A
derivation
from a set
then every formula
occurring
F
in
~
of an essen-
of essentially in
~
first-
is essentially
first-order. An immediate c o n s e q u e n c e of 3.2 is the f o l l o w i n g c o n s e r v a t i v e e x t e n s i o n result.
3.3
then 3.4
THEOREM.
If
AU{A}
is a set of formulae
of IPC and
A*~A
A~IPcA. REMARK.
From (the proof of)
3.1 we obtain that n e g a t i o n in
~
206
E.G.K.
Lopez-Escobar,
W. Veldmsn
behaves in the same way as does i n t u i t i o n i s t i c n e g a t i o n in IPC. example the following are theorems of
An
For
~:
(~AnB)
(A n B) n ~(A
(~B ~ ~A) ^ ~A)
~ 3 x A n Vx~A
An~A w
A FORMAL SEMANTICS
FOR
R,
Our formal m o d e l l i n g for
~
will
be in the style of Kripke 1965.
4.1
DEFINITION.
such that
K
r e l a t i o n on for all
A model-structure
is an inhabited set, K
and
DI, D 2
is a quadruple ~
a reflexive and t r a n s i t i v e
are unary functions on
Dl(e)
is an inhabited
subset of
Parl,
(.2)
D2(a)
is an inhabited subset of
Par 2,
(.31
if
~ s ~
then
DEFINITION.
unary function
M
M(e)
(.2)
if
A (M(e)
(.3)
if
e S 8
4.4 Dz,M>
4.5
such that
on
K
and
D2(~) ! D2(8)"
on a m o d e l - s t r u c t u r e such that for all
is a
~,8 E K:
is a set of atomic formulae,
REMARK.
that "A
DI(~) ! DI(8)
A model
(.i)
4.3
K
e,B ~ K:
(.i)
4.2
then then
Parl(A) ~ DI(~)
A (M(a)
has been v e r i f i e d by stage
such that
DEFINITION.
A realization M
Par2(A) i D2(~),
M(e) c M(8).
If the atomic formula
DEFINITION.
and
of
then we shall say
e".
~
is a structure
is a model on the m o d e l - s t r u c t u r e
Given that
~ =
~
=.
is a r e a l i z a t i o n of
E.G.K. ~,
aEK,
A
Lopez-Escobar,
is a formula
Par2(A)
~ D2(a) ,
symbols
~I=eA
either
A
then
"A
is an atomic and
or
A = (BvC)
and either
or
A = 3xB(x)
or
A : (BnC)
or
A : VxB(x)
or
A = vP(n)B(P n)
4.6
(or evident)
and
~ Dl(a)
at
~
and
in
~",
in
A (M(a)
~ ]= B or
C,
a ( D l ( a ) , ~ l=aB(a),
8 ~ ~,
and for all
n-ary elementary
~[=
B ~ a, ~
and for all
I=~C
~ I=6B(a)
B ~ e,
abstraction
whenever
~ I=BB,
whenever
~I=BB(T)
a (DI(B) ,
whenever
term such that
T
Parl(T)
is
~ DI(B),
Par2(T) ~ D2(B). A sentence in symbols:
DEFINITION. Va~(A)
formula
Parl(A)
B, ~ l=aC,
and ~or all
DEFINITION.
bols:
~I=
and for some
=,
4.7
such that
is true
formula
A = (B^C)
and
~
207
iff:
or
an
of
W. V e l d m a n
A ~I=A,
A sentence iff
~ I= A
is formally
is true in a realization
A
iff for all
~(K,
~ l=aA"
of
~
is formally
whenever
~
is a realization
valid iff its universal
closure
valid,
in sym-
of
~.
A
is formally
valid. w
SOUNDNESS
THEOREM
FOR
length of the derivation 5.1
THEOREM.
and
Der~(A)
5.2
COROLLARY.
PROOF.
Let
If
A
R,
The usual proof by induction
on the
gives us the following:
is a sentence
of
~,
~
a realization
of
then ~ I=A.
~0
I
is not a theorem of
be the realization
such that: K (01 D~0)(~)
~.
=
{~}
=
{a 0 }
(0)>
208
E.G.K.
Then
•
5.3
is not true
REMARK.
ample
let
Then
I
There
~I
w
=
{Q~O),Q(O)}
M(O)(~)
=
{Q(O)}.
in
except Thus
THEORIES,
is a set of f o r m u l a e
Par.(F), 1
i : i, 2.
If in a d d i t i o n (i)
(AvB)
(ii)
3xB(x)
then
F
6.1
formulae
~
AEF
Also,
or
B(d)
A set
A(F
I
then
Parl(F)
has the
F
way;
for
for e x a m p l e
= UA( F Parl(F).
is c l o s e d u n d e r deri-
FI--~ A
and
following
Pari(A)
properties:
B(F
( F
F
whenever
although
nevertheless
mention
the r Sle of
i.
6.2
If
LEMMA.
to
conventions
in the o b v i o u s
for some
of f o r m u l a e Par.(A) l
individual
constant
F
is overcomplete
c Par.(F), -1
Note that we a l l o w the p o s s i b i l i t y plete.
For ex-
= {Q~0),Q~0)}.
the n o t a t i o n a l
whenever
F
I~
d,
an explicit theory.
is c a l l e d
A,
that
A(F
E F ~>
DEFINITION.
of
a theory
( F ~>
is true.
M(1)(a)
is a theory iff
of f o r m u l a e
in the sense
•
of
"the U n p r o v a b l e " .
to sets of f o r m u l a e
vations
be a t h e o r e m
s h o u l d not be c o n s i d e r e d
We e x t e n d
if
F
cannot
in w h i c h
that
as r e p r e s e n t i n g
formulae
A set
•
•
for h o w c o u l d the False e v e r be true?
single F
and h e n c e
@0
~i"
"the False";
EXPLICIT
~0
be like
c o u l d be c o n s i d e r e d
W. Veldman
D~O)(ct)
are r e a l i z a t i o n s
is true in
represent
Lopez-Escobar,
iff for all
i = i, 2.
that a t h e o r y
be o v e r c o m -
we do not m a k e use of the f o l l o w i n g
it b e c a u s e
it gives
some m o r e
is a set of formulae of
conditions are equivalent:
~
lemma,
information
on
then the following
we
E.G.K.
Lopez-Es6obar,
209
W. V e l d m a n
(.I)
F
is overcomplete,
(.2)
F
is a theory and
(.3)
F
is a theory and for some formula
(.4)
r
is a theory and for every sentence
IEF,
(AA~A)
A,
( F,
S(F.
S,
A SPREAD WHICH GENERATES EXPLICII THEORIES, Spreads are basic
w
constructions
in i n t u i t i o n i s t i c
consists of a spread
law
Z,
set theory.
Z =
w h i c h is a function from finite se-
quences of natural numbers to
{0,1},
which is defined on the nodes admitted
and a c o m p l e m e n t a r y
by
{<no , .... nk_l > : Z ( < n 0 , . . . , n k _ l >) = 0), (previously)
A spread
~,
i.e., on
c o n s t r u c t e d m a t h e m a t i c a l entities. (i -i)
function from the
set of finite sequences of natural numbers onto the set ural numbers such that
0
~
of nat-
is the code for the empty sequence.
will be used for the c o n c a t e n a t i o n function,
e.g.
> = < n 0 , . . . , n i _ l , m 0 , . . . , m r _ l >.
Greek letters: functions and [(i)
If
Ec,
and whose range c o n s i s t s of
To simplify matters we use a standard
i.e.,
law
e
~
e,8,..,
is the c o u r s e - o f - v a l u e s
function d e t e r m i n e d by
e,
= <~(0),...,~(i-l)>. is such that
a member of
will be used for number t h e o r e t i c
Z,
Vi(Z(~(i))
and write:
It will be later shown, a spread
~ =
mappings
F(n), Dl(n) , D2(n)
: 0)
then we say that
~
is
a(Z. in Sections
9 through 13 that there is
whose c o m p l e m e n t a r y law
Zc
such that if we set
F
=
Um(~ F(am)
Die
=
Um(l~ DI (~m)
consists of three
210
E.G.K.
Lopez-Escobar,
D2e then the following Condition
7.1.
UmEi~ D 2 ([m)
conditions
If
set of formulae,
:
n
W. V e l d m a n
are satisfied:
is admitted
Dl(n) ! Par I
by
and
Z,
then
7.2.
If
eEZ,
then
Die 9 = Pari(F e) ,
Condition
7.3.
If
aEZ,
then
Fe
Condition
7.4 9
If
eEZ
and
following
are equivalent:
(ii)
(A~B)
7.5.
the following ({) (s163
VxA(x)
and
If
(s
i = i , 2,
then the
and
BEF8]. Pari(VxA(x))
! Die,
i = i, 2,
then
E Fe ~
7.6.
If
vp(n)A(p (n))
and eEZ
If
and
A(a)
6 FS].
Pari(vp(n)A(p(n)))
! Die,
i : i, 2,
( F e, and
term such that 7.7.
aEDIs ~ >
are equivalent:
VBSEzVT[Fe!F 8
Condition
theory.
are equivalent:
then the following (s
is an explicit
Par I.(AnB) ~ Die,
AEF B ~ >
eEL
V88E~Va[F e i F B
Condition
i = I, 2 .
E F ,
VBBEz[F ~F B
Condition
is a (finite)
D2(n) i Par 2.
Condition
({)
F(n)
T
Pari(T) A
is an
n-ary elementary
! Pari(F$),
is a sentence
of
abstraction
i : i, 2 = > ~
A(T)
e F8].
then the following
are
equivalent: ({) (ii)
w
Der~(A), VaaE~(AEFe).
CONSTRUCTION OF A UNIVERSAL REALIZATION, The spread
previous
section
can be used to define
a realization
~
~ of
of the ~
such
E.G.K.
Lopez-Eseobar,
that for all sentences
S
(8.1)
~ I=S = >
The d e f i n i t i o n of
~ B
~
iff
e,6 ( ~
=
DIs,
D2(e)
=
D2e,
]M(a)
=
{B : B
211
of Der~(S).
is as follows:
DI(~)
W. V e l d m a n
and
Let
F ~ ! F B,
is an atomic formula and
B s F }
and then set
= We prove
< ~ , S , D I , D 2 , ~ >.
(8.1) a s s u m i n g that the spread
~
satisfies conditions
7.1 -7.7. 8.2
LEMMA.
If
Par.(A) c D. 1
PROOF,
--
~
and
A
(i = 1,2),
is a formula of
~
such that
then
i~
By induction on the logical c o m p l e x i t y of
Basis ste~.
A
is an atomic formula.
quence of the d e f i n i t i o n of Induction step.
that
and
Then it is an immediate conseI= .
Let us c o n s i d e r the case when
Assume thus that ~I=~ A . V6BE~VT[8 ~ ~
~
and
T
A.
A = vP(n)B(p(n)).
Then
is an
n-ary e l e m e n t a r y a b s t r a c t i o n term such
Pari(T) ! D i ( 8 ) ,
Then u s i n g the d e f i n i t i o n of
i = 1,2 = > ~
~I=8B(T)].
we obtain that:
212
E.G.K.
V 8~EzVT[F B ~ F
and
such that
T
Lopez-Escobar,
is an
Pari(T)
From the induction
n-ary elementary
~ Pari(Fs),
hypothesis ~I=$B[T]
Using condition
that
<=>
( F s.
B(T)
7.6 we then conclude
is similarly
The proofs
abstraction
i = 1,2 ~ >
we obtain
term
~6~(T)].
that ~ F
vp(n)B(P (n))
The converse
W. V e l d m a n
proven.
for the other
compound
formulae
are analogous
(and
well-known). 8.3
If
COROLLARY.
A
i8 a sentence of
(i)
~I=A
iff
V~aEZ(AEF ) ,
(ii)
~ I=A
iff
VaaE%3m(A E F(~m)),
(iii)
~ I=A
iff
Der~(A).
PROOFS.
Of (i),
(ii), immediate.
For
w
CONSTRUCTION OF THE SPREAD ~
next
5 sections
we shall adhere
~,
then
(iii) use condition
OF EXPLICIT THEORIES, For the
to the following
FO,FI,...
is an enumeration
of the formulae
~0,~i~...
is an enumeration
of the derivations
~0 U 9 1 U
... is a partition
into a denumerable
of the set
sequence
7.7.
Par 2
conventions:
of in
of predicate
of pairwise
disjoint
parameters
deDumerable
sets. ~i = {Qio~Qil ''''} ~ o u ~ l U ...
is an enumeration
is a partition
of
of the set
~i Par I
of individual
param-
E.G.K. Lopez-Escobar, eters into a denumerable
sequence
Veldman
W,,
of pairwise
213
disjoint
denumerable
sets
Mi = {ciU'Cil''''} ~k,s
is an enumeration
is a (i-i) mapping ~k,Z,3~
= ek,i,2~
The functions recursion
from
+ 2,
F(m), Dl(m) ,
on the length of (the finite
Basis step.
Xi
~2 x {1,2,3}
+ 1 = ek,s
Z(m),
of
onto
and
and
~
such that
~0,0,1~
D2(m)
sequence
= 0.
are defined by
coded by)
m.
Z(< >) = 0
r(< >)
r
=
DI(< >) = K0 D2(< >) = Q0 Reeursion
step.
F(m), Dl(m)
Suppose that
and
ural numbers
k, s
D2(m) r
m = <m0,...,mp_l >
have been defined.
=
We then determine
nat-
~k,Z,r~
and proceed by cases depending
on the value of
that if for some
its value is to be
E(m),
such that p
the convention
and that
i,
s,
E(m*~)
r.
We shall follow
is not specified then
and thus the finite
sequence
(coded by)
.,o^
m"s = <mo,...,mp_l,S> Case i
~.
r = i.
SHb~__!a. (i)
is not admitted by
If
if for some
Pari(F k) ! Di(m), q ~ p,
q
i E i, 2
is a derivation
then we set: Z(m*l)
=
0
r(m*l)
=
F(m)
Dl(m*l)
=
Dl(m)
U {F k}
then of
Fk
from
F(m)
214
E.G.K~
D2(m*~)
=
hand
for all
(2) and if on the other tion
of
Fk
from
Lopez-Escobar,
F(m)
W. V e l d m a n
D2(m) , q ~ p,
~q
is not a deriva-
then we set:
Z(m*~)
=
0
Z(m*2)
=
0
r(m*~)
=
r(m) U {F k}
r(m*9)
=
F(m)
Dl(m*l)
=
Dl(m)
Dl(m*2)
=
Dl(m)
D2(m*~)
= D2(m)
D2(m*2)
= D2(m).
S~b~e_ib.
If e i t h e r
Parl(F k) ~ Dl(m)
or
Par2(F k) ~ D2(m)
then
we define Z(m*O)
: 0
r(m*~)
: r(m)
Dl(m*0 ) = Dl(m) D2(m*0) Case
= D2(m).
2
r = 2.
In this s > 0
Case
3
B(x),
functions.
That
is,
for each
~(m*s)
=
0
r(m*s)
=
r(m)
Dl(m*s)
=
Dl(m)
U {Cpj : j < s}
D2(m*s)
=
D2(m)
U {Qpj : j < s}.
r = 3. case we c o n s i d e r
in the previous
Sub~ase_~a. some
the domain
we define:
In this duced
ease we enlarge
formulae
If
the f o r m u l a
might
have
been
intro-
cases.
mp_ 2 = i AI, A2,
F k = 3xB(x)
which
(and hence
F k = (ALVA 2)
then we define
Fk
~ r(m))
and if either
or for some p s e u d o - f o r m u l a
for
E.G.K.
Z(m*l)
=
0
r(m*~)
=
Y(m)
Dl(m*l)
=
D2(m*l)
=
in the case that for all
where
Lopez-Escobar,
W. V e l d m a n
21B
E(m*2)
=
0
r(m*~)
=
F(m)
Dl(m)
Dl(m*2)
=
Dl(m)
D2(m)
D2(m*2)
=
D2(m) ,
U
{A I}
F k = (ALVA2) ;
and in case
U
{A 2}
F k = 3xB(x)
we define
s > 0
tl, t2,..,
Z(m*~)
:
0
r(m*Z)
=
r(m) U {B(ts)}
Dl(m*s)
:
Dl(m)
D2(m*~)
=
D2(m) ,
is some
(previously
agreed
upon)
enumeration
of
Dl(m). Subcase
3b.
Failure
of subcase
3a.
Then we set:
Z(m*~) = 0 F(m*8) ='F(m) Dl(m*O)
= Dl(m)
D2(m*~) = n2(m). r(m), Dl(m) ,
Combining we obtain 9.1
the spread
REMARKS.
(A)
m
and
Dl(m) , mitted (B)
m*O
~
two conditions m*0
by
has been defined
= D2(m)
Z.
is admitted
law
~c
by
Z.
so that for any
m
are equivalent:
are admitted
D2(m*0)
into the complementary
~ = <x,xo>.
The spread
the following
D2(m)
by
Z,
F(m*O)
and for all
= F(m),
s > 0,
m*s
Dl(m*0)
=
is not ad-
E.G.K.
216
A node
m
such that
Thus
m
is a p r o c r a s t i n a t i o n
node. ~.
Or in terms
~p
w tion
for
8EZ
and not w o r t h
If
For a p r o o f parameter
theory
aEZ
then
Q
and i n d i v i d u a l 7.3,
namely
is an i m m e d i a t e
tion of
D.
of i0.i it s u f f i c e s
Condition
node
iff
~.
That
),
to o b s e r v e
parameter
that e a c h
consequence
~
F
satisfies Condition
for any
7.2 is
n-ary
Der~(Qcc...c (for
of c a ses
eondi-
i = i, 2.
that
e,
is a d m i t t e d by
8p = 0.
repeating.
= Par.(F
a procrastination
~(Z)
1 and
n Qcc...c).
is an e x p l i c i t
3 of the d e f i n i -
~.
Conditions consider
contains
7.4 - 7 . 7
are a l i t t l e m o r e
complicated
so we shall
them separately.
"v"
AND THE S P R E A D far too m u c h
LEMMA.
i : i, 2
following
properties:
SEA ~ >
~.
For any g iven
information
To every
P a r i ( F a)
(III)
we h a v e that
7.1 is i m m e d i a t e
LEMMA.
(II)
m*0
OF THE S P R E A D
i0.i
(I)
iff
SOME P R O P E R T I E S
as a lemma.
ii.i
node
is a p r o c r a s t i n a t i o n
stating
W. V e l d m a n
(A) h o l d s w i l l be c a l l e d
of f u n c t i o n s
worth
w
Lopez-Escobar,
a(E,
so we f irst
formulae
there corresponds
A, B
sEZ
we find t h a t
cut it down to size.
such that
a subfan
~
of
Par.(A) l
~
with the
AEr~,
SEA
> F a _c r 8 ,
8E~,
p = ~k,Z,l~,
8p # 0 ~ >
p = ~k,Z,3~,
8p ~ 0,
A = Fk
or
F(Sp)J--F k
or
FkEFa, (IV)
8(4,
F~ = 3xCtx) ~ >
<
there is an
E.G.K. individual
PROOF.
Lopez-Escobar,
constant
a
W. Veldman
217
C(a) ( F ( ~ p )
such that
and
(if
3xC(x)
(F(~p)
then
C(a) ~ F )
(if
3xC(x)
~ F(~p)
then
a ~ Parl(F ~) U Parl(A)
Let
q
or
U Parl(B).
be a natural number such that
Parl(A)
and
U Parl(B) ! %0 U ... U %q
Par2(A) U Par2(B) ! ~0 U ... U ~q. Let ~0
i
be such that
such that
A = F.1
and then determine a natural number
q.
The subfan
mined by the condition that (i)
B(ei,~o,l~)
(ii)
VkV~(e(~k,s
(iii) (iv)
of
~
is then deter-
iff
= i, i ~>
VkVs163
8(ek,~,l~)
= a(~k,s
VkV~([6(~k,~,3~) 8(~k,~,3~))]
= 0] v [6(~k,&,3m)
and
= fu(~(ek,~,3m),
= fl(~(~k,s
g' f0' fl
D2e ~ D28
= i),
+ g(8(~k,s
v [8((k,s
where the functions DIe ! DI8~
B(A
A
B({k,Z,&~))]),
are chosen so that
DI8
is sufficiently
(iii) ensures that
larger than
and so that (iv) ensures that the required instantiations junctions on existential
formulae are placed into
F B.
Dla,
of dis-
That such
functions can be found follows from the fact that the only time that constants from
Xp
are placed in
11.2
If
e(~
THEOREM.
Pari(A~B ) c Die ~
and
i = i, 2
Diy
(AnB)
(ii)
(AnB)
then the following
~ F ,
VBB~sEF e ~ r B
and
A(F B ~ >
is at the node
is a formula
equivalent. (i)
(y(Z)
BEF~].
of
~
~p.
such that
two conditions
are
218
E.G.K. Lopez-Escobar,
PROOF.
That
that
the
Let
A
Then
from
({) ~ >
FB's
are
be the
({i)
theories.
subfan
(ii)
and
is an
of
(II)
immediate
Thus
E
W. Veldman
assume
determined
of L e m m a
consequence
ll.1
(ii).
of the
We w i l l
according
fact
prove
to L e m m a
(i).
ll.1.
we o b t a i n
V B B ( A ( B ~ F B) and h e n c e
that
Using
the
monotonici•
there
is
a natural
of
r
number
and
PO
VB~EA(B We n e x t
prove,
If
A(m)
Basis
:
induction
0
can be
stated
Vs[~(m*~)
Furthermore F(m*s)
if
m
= F(m)
F , AI-~B.
Hence
k,
p = {k,~,r~. Case
1
Subcase
we c o n c l u d e
that
that
PO - l t h ( m )
~ PO
that
r(m), r , AI-~B.
then
B (r(m),
so
r(m),
p : lth(m).
Let
Then
F,
AI--RB.
the
induction
follows:
= 0 ~>
r(m*~),
induction
from now
theorem
(r(~P0)).
Then
as
fan
r ,
is a p r o c r a s t i n a t i o n
so the
procrastination Let
PO"
l t h ( m ) < PO"
step.
hypothesis
:
such
on
lth(m)
and
lth(m)
step.
Induction
by
the
node
hypothesis
on we
shall
AI-~B]. then
then
assume
A(m~O)
give
= 0
and
us that
that
m
F(m),
is not
a
node.
Z, r
be the u n i q u e
We p r o c e e d
natural
by cases
numbers
depending
such
that
on the v a l u e
of
r.
r = i. la.
A(m*~)
= O.
Then
the
construction
of
Z
tells
us t h a t
219
E.G.K. Lopez-Escobar, W~ Veldman F(m*l)
= F(m)
U {F k]
and
so the
induction
r(m),
Fk,
of
tells
hypothesis
gives
us t h e n
that:
But the
construction
Fk = A
or
F k E F e.
,A
Thus
as
lb.
A(m~2)
= 0.
Case
2
In this F(m).
case
3
Subcase C(a), ~uction
F(m)I--~F k
that
F(m*2)
= F(m)
and the
argument
is
node.
for
some
argument
s > 0
then
we h a v e
proceeds
as
= 0
A(m*s)
that
r(m*s) =
and
for a p r o c r a s t i n a t i o n
node.
r = 3. 3a.
For
some
F k = 3xC(x) hypothesis
s > 0,
(F(m),
Now u s i n g
condition
we o b t a i n
(IV)
A(m*s)
F(m~s)
= 0
= F(m)
and
for
some
U {C(b)}.
of L e m m a
r(m),
F r o m the
re,
AI-~B.
II.i
we c o n c l u d e
that
AI-~B
re,
or
r(m),
since
3xC(x)
E F(m)
3xC(x),
we h a v e
r(m),
Sub~ase_3b.
F k = (ClVC2).
formula
that
C(b),
F(m),.
But
or
r = 2.
The
Case
either
AI-~B.
Fe,
Then
for a p r o c r a s t i n a t i o n
us that
we o b t a i n
F(m),
Subease
AI--~B.
r,
re,
rei-~B.
that
in e i t h e r
case
A]--pB.
Analogous
to S u b c a s e
3a.
either
in-
E.G.K.
220
Lopez-Escobar,
We now consider the situation when
W. V e l d m a n m
is the empty sequence 9
Then
F ,A I-~B . From the latter it follows that theory we may conclude
w
that
"V" AND THE SPREAD
F I-R(AnB)
(AnB)
~,
any essential use of case 2 of the definition
second-order
quantifier
V(2);
F
is a
E F
In the case of
will be used for the quantifier
and since
"V".
,,n,, we did not make of the spread
~
We will only consider
the first-order
case being almost
As in Section ii we must first obtain an appropriate
fan of
(by essentially
12.1
LEMMA.
corresponds
To every a subfan
(I)
BEA ~ >
Q E D2B
(II)
B(A 2 >
F ~F 6
of
the same as ii.i
(III)
(IV)
the same as ii.i
(IV).
THEOREM.
such that
If
a E ~
and predicate p a r a m e t e r
~
(III)
12.2
sub-
the same method that was used in II.I).
a 6 ~ ~
It
%he
the same. ~
.
and
Q
there
with the f o l l o w i n g properties
vP(n)A(P (n))
Par.(vP(n)A(P (n) )) ~ Pari(F~) l
is a formula of
then the f o l l o w i n g
R two
conditions are equivalent:
(i)
vP(n)A(P (n))
(ii)
E F
VB~E~VT[F ~ i F B
and
term such that
Pari(T ) c Pari(F8)
T
PROOF.
The only interesting
Let
be an
Q
is an
n-ary elementary a b s t r a c t i o n
~>
case is (ii) ~ >
n-ary predicate
parameter
A(T) (i).
such that
E F8]. Thus assume Q ~ Par2(F a)
(ii).
221
E.G.K. Lopez-Escobar, W. Veldman and then let 12.1.
Then
A
constructed
be the subfan of
(ii) specializes
according
to Lemma
to
V 8 8 E A ( A ( I X l . . . X n Q X l . . . x n) E FS) , which in turn leads to
VSBEABP(A(Q) Proceeding
E F(Sp)).
as in the proof of Theorem
11.2 we arrive at
r~i-mA(Q). Then using the assumption
that
Q ~ Par2(F e)
we conclude
r I-~vPA(P) and then that
II
w
Deri~
in S e c t i o n
VPA(P)
II
E F
AND THE SPREAD
~,
Of the conditions
7.1-7.7
listed
7 the following one is the only one that remains to be
proven. 13.1
THEOREM.
For any sentence
S
of
]~
the f o l l o w i n g
two condi-
tions are equivalent:
(i)
Der~R(S )
VCCc~E;~(S E Fa).
(ii) PROOF.
Again the only interesting
case is
ter is proven with the help of the subfan (I)
SEA ~ >
(it) = >
~
of
(i)- and the lat~
such that
S ( r
(II),
(III)
Analogous
to (III) and
w
THE COMPLETENESS OF ~,
(IV) respectively,
of Lemma ii.i.
So far we have shown that for sen-
222
E.G.K.
tences
A
of
Leoez-Escobar,
W. V e l d m a n
I~:
14.1
Der]9(A) ~ >
Va~(A)
(see Theorem
14.2
Va~(A)
nerl~(A)
(see Corollary
However
=>
in order to have an honest
must show that it is plausible
14.3 where
completeness
"The sentence
A
~
theorem
for
~
we
Va~(A)
is an abbreviation of
8.3).
that
Valid~(A)--> "Valid(A)"
5.1)
for the
(informal)
statement
is logically valid from the inruitionistic
viewpoint" In order to avoid multitude
of realizations
the following 14.4
some of the problems associated
schematic
Given that then
~
for
~
with the
it is better to consider
14.3 in
form
is a r e a l i z a t i o n for
~
and that
Valid(A)
~I=A.
One way to show 14.4 would be to give the exact conditions under which:
a sentence
A
of our r e s t r i c t e d s e c o n d - o r d e r
language
i8 logically valid from the i n t u i t i o n i s t i c viewpoint.
Fortunately would Vali~
suffice
we do not need to know the exact conditions.
for our purposes
which would allow us to conclude
show that such properties In V a l i d ( A ) (2)
to specify
characterize
two important
Logical validity
The intuitionistic
viewpoint.
of
14.4, we do not need to ValidR(A).
concepts
and (~)
enough properties
are involved:
It
223
E.G.K. Lopez-Escobar, W. Veldman The e s s e n t i a l
(C1)
characteristics
of
(*) are that:
The validity of a compound sentence be reducible to the validity of simpler sentences,
(C2)
The validity of a sentence be independent of the interpretation of the non-logical symbols. On the o t h e r hand
to be in c o n f l i c t the t r a d i t i o n a l mathematical
with
view
"truth"
fied with
eventually
A possible
the
finding
compromise
as m e n t i o n e d
intuitionism
a proof
in the
is solely (and,
statement
of
(~)
concerned
in fact,
with
the intui-
is sometimes
identi-
of it).
is to satisfy
(**) as far as the
logical
consider
of an intuitionist.
approximation
is a sequence
(both by p r o d u c i n g The
seems
introduction,
In o r d e r to do the latter we must
world"
what he is doing).
characteristic
and t h e i r proofs
could be d e s c r i b e d
"There working
is that
are concerned.
As a first
for,
of a m a t h e m a t i c a l
"mathematical
tionist
(C2)
statements
tionistic
symbols
the e s s e n t i a l
the
"mathematical
world"
of an intui-
as follows: of m a t h e m a t i c a l
statements
new constructions
sequence
itself
on w h i c h he is
and by r e f l e c t i n g
can be thought
on
of as extend-
ing indefinitely". It w o u l d p r o b a b l y arranged
as an
flection
is an important
m-sequence
not a finitist) and
secondly,
experience m -seq u e n c e , person.
be a m i s t a k e
to c o n s i d e r
for two reasons:
part of i n t u i t i o n i s m
and as we all know h u m a n
even t h o u g h
of a given
it does not f o l l o w
Now we are
interested
that
may a p p e a r it w o u l d
in logical
sequence
Firstly,
because
to be re-
(the i n t u i t i o n i s t
reflection
it could be argued
intuitionist
the
that
is very erratic
the
subjective
to him as appear
validity,
is
(part of)
an
so to a n o t h e r that
is we wish
224
E.G.K.
characterize aZZ
those
sentences
intuitionists,
particular appears
creative
subjects
(i.e.,
are not
CI,
@
must
as true by
consider
proofs
any
the way
it
is progressing. for example:
(a) think
iff they
that
correctly,
do have
reduce @
and
Similar interest
both
considerations
assert
proofs
of
A,
struct
later on,
proofs)
the
(b) are and
(c)
@
ac-
of Divine
would
madness,
apply @
to
(AvB)
would
the c r e a t i v e
of
in the
[and
appear,
to us,
If we wish
but also
that he might
intuitionist
not as a
on w h i c h
sequence
(AnB).
subject,
iff
we
Of more
In view
of
see that he
of c o n v e r t i n g
the
that he might
con-
of an i n t u i t i o n i s t
statements
consider
he is w o r k i n g
to be m y s t i c i s m
include
should, c o n s i d e r
world
to u n d e r s t a n d
he a c t u a l l y
(linear)
3xA(x)].
or even those
of m a t h e m a t i c a l
statements
that we
Thus
B.
then we must
it is b e t t e r
accepted.
statements
assert
the m a t h e m a t i c a l
not only those those
was
be a c c e p t e d
in it.
into proofs
sequence
would
had also been
(A^B)
constructed
inspiration).
(apparent)
(A^B)
of m a t h e m a t i c a l
when
if we c o n s i d e r
actions
B
intuitionist
in view of a s s u m p t i o n
only when he had a m e t h o d
already
to be just the
and
concerning
(AnB)
(which
sentences).
see that
is to c o n s i d e r
Thus,
A
sequence
were
our i d e a l i z a t i o n s
then his
the
we w o u l d B
statements
to simpler
only when
was w o r k i n g A
h o w our p a r a d i g m a t i c a l
of c o m p o u n d
if we c o n s i d e r e d
would
intuitionist~
with
by
forgetful.
the truth
both
(i.e.,
that we imagine
world
as valid
too i n v o l v e d
idealizations,
c l a i m to have
Now we must cepts
It suffices
some
be r e c o g n i z e d
not become
to us that his m a t h e m a t i c a l to make
W. V e l d m a n
which w o u l d
so we should
intuitionist.
We do have
honest
Lopez-Escobar,
in his
considers later
on.
(or the result
the m e t h o d
in the
"mathematical
world"
at a given moment, Or in o t h e r words,
the m a t h e m a t i c a l
sequence
on,
but r a t h e r
world
of a
as a p a r t i a l
E.G.K. Lopez-Escobar,
W. Veldman
225
ordering. Our notion of r e a l i z a t i o n
for
called a "temporal" record of the
~
corresponds
to what might be
(possible) results of a creative
subject, w h e r e for simplicity we "record" only atomic acts or statements.
That is, suppose given
~
=.
corresponds to the structure of e v i d e n t i a l and e n v i s i o n e d by some creative subject. Dl(e) up
~;
D2(a)
Then given an
D2(a)
lection of rather simple species)
~EK,
objects e o n s t r u c t e d
is the c o l l e c t i o n of atomic
since species are properties,
situations e n c o u n t e r e d
gives us the c o l l e c t i o n of m a t h e m a t i c a l
to stage
Then
statements
(or
could be c o n s i d e r e d as a col-
c o n s i d e r e d up to stage
specifies those atomic statements v e r i f i e d by stage Now, granted such a reading for a r e a l i z a t i o n
a.
M(a)
a.
~
of our res-
tricted s e c o n d - o r d e r language, we obtain, by a simple induction on the logical c o m p l e x i t y of the sentences of
{A : A
is a sentence of
~
~,
and
that given
~EK:
~]=aA}
coincides with the c o l l e c t i o n of sentences c o n s i d e r e d as true sentences by the creative subject at stage In p a r t i c u l a r we obtain that ~I=A}
{A : A
~. is a sentence of
~
and
is the c o l l e c t i o n of sentences which are true for the given
creative subject at all stages of his m a t h e m a t i c a l world. Now we can return to 14.4. is a r e a l i z a t i o n of
~
Thus suppose that we are given that
and that V a l i d ( A ) .
Then
ered as true by one and by all of the intuitionist.
A
is consid-
Thus
appear in all the m a t h e m a t i c a l worlds of the intuitionist, A
is logically valid it would be in all the stages.
can be i n t e r p r e t e d as t h e ' ~ a t h e m a t i e a l world" then
A
would
and since
Thus if
~
of some i n t u i t i o n i s t
~I=A. Thus the truth of 14.4 is reduced to the question:
226
E.G.K.
Lopez-Escobar,
Can be an arbitrary r e a l i z a t i o n
W. V e l d m a n
~
of
~
be i n t e r p r e t e d as the
(possible) m a t h e m a t i c a l world of some i n t u i t i o n i s t ?
Actually,
in view
pleteness t h e o r e m tion.
answer
(?)
it suffices
Thus w h e t h e r
in fact
of Corollary
an honest
to c o n s i d e r
or not the
theorem
w
correctly
thinking
intuitionistic
restricted
second
realizations Furthermore of the
gives since
does
R
of
~
is
upon a p o s i t i v e
(see Section 8) be in-
the
intuitionists
and not forgetful) of them
(i.e.,
it seems
so as to have
AND KRIPKE MODELS,
predicate
order
logic
calculus ~,
us a (formal)
the notion
the r e a l i z a t i o n s
Kripke models
relationship for the
or not rests
shown
to be
not u n r e a s o n -
a positive
an-
(?).
R E A L I Z A T I O N S OF
of the
~
idealized
able to a l l o w our i d e a l i z a t i o n swer to
result we have
(possible) m a t h e m a t i c a l world of some i n t u i t i o n i s t ?
Since we have a l r e a d y honest,
realiza-
question:
Can the u n i v e r s a l r e a l i z a t i o n
terpreted as the
of the com-
just the u n i v e r s a l
completeness
completeness
to the f o l l o w i n g
8.3, for the p u r p o s e s
between
for
Since
has an a n a l o g u e
(formal)
of v a l i d i t y ~
specially
IPC via Kripke m o d e l s
leads
since
in the
for
~
vla
for the IPC.
are n a t u r a l
modifications
to c o n s i d e r
a completeness
to M a r k o v ' s
A
formula A~
validity
of IPC it may be of interest them;
every
principle
the
theorem and ours
not. It is now common
is e q u i v a l e n t "absurdity"
to s e c o n d - o r d e r
may be defined
that by further to all
knowledge
intents
restricting and purposes
that
second-order
intuitionistic
by
V(2)PP. ~2
minimal
logic
and that
What we have
we obtain
a conservative
logic
extension
in
observed
a calculus
~
~2 ~2 is
which
of IPC and
is in
E.G.K. w hich n e g a t i o n Because
tions even
(nor absurdity)
~
in view of our
of
~,
~J=l
verified)
ments
are
the v i e w p o i n t
15.1
such that
~ J=l.
treatment
The
(informal)
but r a t h e r
to the
considered
in
~
DEFINITIONS.
of
I
Or put
we obtain
appear
interpretation
for the r e a l i z a true
too few e l e m e n t a r y
in more
some
strange.
to the False being
fact that
~ .
227
rSle.
latter may
does not c o r r e s p o n d
from
W. Veldman
has no p r i v i l e g e d
of such n o n - p r e f e r e n t i a l
realizations However,
Lopez-Escobar,
picturesque
(or state-
terms,
is not very d i s c r i m i n a t i n g .
Let
~ =
be a r e a l i z a t i o n
and
e~K.
(i)
~
is a credulous
(2)
~
is a trivial r e a l i z a t i o n
iff every node
(3)
~
is a natural
iff there
for
Then node
of
realization
~
iff
~J=~ I. is credulous.
is a node
of
~
which
is
not credulous. (4)
is an ideal r e a l i z a t i o n
~
iff every node
of
~
is not credu-
lous.
The
ideal r e a l i z a t i o n s
that any
ideal r e a l i z a t i o n s
(as far as f i r s t - o r d e r versa.
15.2 (I)
tion
to Kripke
can be t r a n s f o r m e d
formulae
The t r a n s f o r m a t i o n s
are
concerned)
models
in the
sense
into an e q u i v a l e n t Kripke model
and vice
are as follows.
DEFINITIONS. Given
a realization
the Kripke m o d e l (2)
correspond
Given
a Kripke
predicate
let
Q
=
for
~
let
(~)i
be
. model
First
~
for
K =~
obtained
Then define
(K) 2
be the r e a l i z a -
as follows:
be a new p r o p o s i t i o n a l
parameter).
let
parameter
for all
~(K:
(i.e.,
a
0-ary
228
E.G.K.
The r e m a r k
If
~
:
D(~)
D2(~)
=
{Q} U M(a).
the e q u i v a l e n c e
is an ideal r e a l i z a t i o n such that
K
for
is a Kripke
~
itf
model
A
A
iff
f
is a Kripke
model
15.6
If
~
is an ideal
realization
node
(i.e.,
if we simply then, Let
if
~
above
then
all the
anything
is a
of IPC
is an ideal
realization
((K)2) 1 : K.
for
and
~,
then
for any
A
a credulous
node
~ s 8
credulous
is left,
be the result
A
(~)i
of IPC
formula
is credulous
delete
provided Red(~]
first-order
every node
then
(K)21= ~ A*.
If
Because
~
(K) 2
15.5
essentially
ideal r e a l i z a t i o n s
~l=e A*.
then
and for any formula K~
for
for any formula
(~)iI= ~ A
If
between
as follows:
Kripke m o d e l
15.4
W. V e l d m a n
DI(~)
concerning
can now be stated
15.3
Lopez-Escobar,
nodes
we obtain
of d e l e t i n g
the
is also
then
8
a credulous is credulous)
of a r e a l i z a t i o n an ideal
realization.
credulous
nodes
of
~.
Thus
15.7
If
~
is a n a t u r a l
realization
then
Red(~)
is an ideal
realization.
It is easily if
~
natural tence
proven
is a credulous realization, A:
by induction
node
of
~,
on the c o m p l e x i t y
then
~ =,
~ I= A.
we obtain
Then that
of if
A ~
for any
that is a sen-
E.G.K. Lopez-Escobar, W. Veldman 15.8
~ I=A = >
15.9
Red(~)l=A ~ >
229
Red(~) I=A,
Ve ( K ( 7 7 ~ I = a A ) .
In the case of the universal spread
~
: <~,S,...>,
Lemma 8.2
allows us to t r a n s f o r m 15.9 into:
15.10
Let
Red(e) l=A : >
MA
Ve (~ 7-13m(A ( F ( ~ m ) ) .
be an a b b r e v i a t i o n for the following:
Er(~m)) : >
MA : Va ( ~ 7 7 3 m ( A
Ve ( ~ 3 m ( A
E r(~m)).
Then from 15.10 and the p r o p e r t y of the u n i v e r s a l r e a l i z a t i o n we obtain:
MA =>
15.11
Since
Red(e)
15.3 - 1 5 . 6
Der~(A)).
is an ideal realization, we can apply the results
to conclude that for f i r s t - o r d e r sentences
MA, = >
(A
is valid in all Kripke models = >
Or in other words, recursive predicates
(~)
(Red(e)l=A = >
A
I--IPcA).
under the a s s u m p t i o n that for all p r i m i t i v e A(n):
V~ E~ u 7 3 n A ( a n )
~>
Va 6 ~ 3 n A ( ~ n )
we have r e c o v e r e d the completeness t h e o r e m for IPC w i t h respect to Kripke models which, as shown in Kreisel 1961, implies
Ve (B q 7 3 n A ( n , a )
=>
for primitive r e c u r s i v e predicates
V~ (B3nA(n,~).
A(n,~).
Since the nodes above a credulous node are also credulous~ we see that the r e a l i z a t i o n s up from ideal r e a l i z a t i o n s
for
~
can be c o n s i d e r e d as being built
(which c o r r e s p o n d to Kripke models
for
230
E.G.K.
Lopez-Escobar,
IPC) and trivial realizations
W. V e l d m a n
(in which every sentence
would appear that the reason why Markov's our completeness
theorem
a node in a realization
principle
is true).
is avoided
It
in
is because we do not have to decide w h e t h e r is credulous
or not.
Further evidence
to
this last remark is given by the following: 15.12
THEOREM.
If
~
is a natural realization for
=
and (D)
then for every sentence
S
of
l=S PROOF.
Let
complexity
K~ = {~K
iff
:7 ~ I =
Red(~)l=S.
I}.
Then we prove by induction
of the formula
(i)
for all
aEK*,
~ I = a A ~ Red(~)I: aA.
Let us consider the case when
A = (BnC).
Red(~)I= e
is straightforward.
(BnC)
and let
BEK
the assumption
that
@~K*,
then
B
B ~ a
Red(~)l= ~ (BnC)
by the induction hypothesis is a credulous
~
that
[ ~ I = 8B = >
That is ~ I=~ (BnC).
(D), that for all sentences Va E K[~I=aS]
S iff
~I=@B.
and
we conclude
node of v B~K*.
(i) has been established
thus that
gives us that
more we are assuming ~I= BC].
BEK*
That ~l=a (BnC)
Suppose
be such that
hence,
Once
on the
Red(~)I= ~ If
B(K*
Red(~)l= BC
then and
that
~ I=BC.
If
and thus
~ I = BC.
Further-
Thus we have that
we obtain,
VB {
again with the help of
of Va ( K*[Red(~)I=~S].
E.G.K. w 16. HISTORICAL The above lings
theorem
his book "The Foundations can be compared
structed
in w
231
seems to confirm E.W. Beth's
the way in which his theorem
One gets this impression
there
W. Veldman
REMARK
completeness
concerning
Lopez-Escobar,
from reading Section of Mathematics".
with the spread
The semi-models
could be saved. 145, last paragraph,
(The semi-model
~ of explicit
theories,
theories,
occuring
was right in thinking that defining a fan of models, some inproper
points,
and then applying
in
M mentioned con-
in M which are not proper models
be compared with the overeomplete
taining
own fee-
can
in ~ ). So Beth necessarily
the fan theorem,
conwould
give the desired result. It is a pity that he did not carry out this program and distorted Brouwer's
argument
for the fan theorem.
seen the crucial rSle
of negation.
paper are a continuation
of Beth's
He also does not seem to have
Nevertheless,
the ideas in this
and it is a matter
of historical
justice to mention his name here. (We are indebted
to Prof.
Troelstra
for asking our attention
for this.)
232
E.G.K.
Lopez-Escobar,
W. Veldman
REFERENCES BETH, E,W, 1949 Semantical considerations on intuitionistie m~thematics, Math., vol. 9.
Indag.
1955
Semantic entailment and formal derivability, Mededelingen der
1956
Semantic construction of Intuitionistic Logic, ibid., vol. 19,
Kon. Ned. Akad. v. Wet., new series, vol.
18, no. 13.
no. ii.
DYSON, V,H, AND KREISEL, G, 1961 Analysis of Beth's semantic construction of Intuitionistic Logic, Technical Report No. 3, Applied Mathematics and Statistics Laboratories, Stanford. GIRARD, J-Y, 1971 Une extension de l'interpretation de G~del ~ l'analyse, et son application ~ i'elimination des coupures dans l'analyse et la theorie des types, Proceedings of the Second Scandinavian Logic Symposium, (Fenstad, Editor) pp. 63-92. HEYTING~ A, 1966 Intuitionism, An Introduction. Amsterdam.
North-Holland
Publishing Co.,
KREISEL, G, 1962 On weak completeness of intuitionistic predicate Journal of Symbolic Logic, vol. 27, pp. 139-158.
logic, The
KRIPKE} S, 1965 Semantical analysis of Intuitionistic Logic, Formal Systems and Recursive Functions, (Crossley and Dummett, Editors), pp. 92-129. PRAWITZ, D, 1965 Natural Deduction, A Proof Theoretical Wiksell, Stockholm. 1971
Study.
Almquist and
Ideas and results of proof theory, Proceedings of the Second Scandinavian Symposium,
(Fenstad,
Editor) pp. 235-308.
TROELSTRA, A,S, 1973 Metamathematieal Investigation of Intuitionistic Arithmetic and Analysis. Lecture Notes in Mathematics, vol. 344. E.G.K. L6pez-Escobar Department of Mathematics University of Maryland College Park, Maryland 20742
W. Veldman Math. Instituut Katholieke Univ. Nijmegen
THE REAL E L E M E N T S
IN A C O N S I S T E N C Y TYPE T H E O R Y
Dedicated
to Kurt
SchHtte
PROOF FOR SIMPLE
I
on the o c c a s i o n
of his 65 th b i r t h d a y
Horst
Luckhardt
Abstract
This
is the first of two papers
of simple tion,
type theory mean,
the s t r o n g e s t
question
recursive plus
type
theory
functionals
results
one c o m p r e h e n s i o n
method
today.
by i n t e r p r e t a -
Concerning
to a r i t h m e t i c
the q u a n t i f i e r f r e e
over
consistency
the
first
are proved.
can be r e d u c e d plus
What does
and hew can it be a c h i e v e d
constructive
the f o l l o w i n g
(I) Simple
on the questions:
a special
over
the p r i m i t i v e
rule of e x t e n s i o n a l i t y
A-property
on objects
of type
O.
(2) C o n s i s t e n c y cator", types
is fully d e s c r i b a b l e
a generalization
and e x t e n s i o n a l i t y .
(3) A n e c e s s a r y primitive
recursive
types O(Oe)
based
it rather
instances
ency proof
has
of c o n t i n u i t y
for c o n s i s t e n c y contains
indi-
with r e s p e c t
to
is that
the theory T of the
no c o n t i n u i t y
indicator
for
points.
gives
to the negations
and to c o n t i n u i t y
to make
"continuity
Especially:
on c o n t i n u i t y
leads
prehension
of the m o d u l u s
functionals
at d e f i n a b l e
The ~ - o p e r a t o r ening;
condition
by the n o t i o n
very
no p r o o f - t h e o r e t i c
strength-
of the above m e n t i o n e d indicators.
fine distinctions.
Therefore
com-
a consist-
234
H.
(4) w - r u l e s
for
to t h e o r i e s
types
greater
of c l a s s i c a l
strength.
Moreover
AemiVa.e
= a imply
recursive
constructive
Markov's
incompatible
with
higher
second
paper makes
question
by
and
By G ~ d e ~ s w e l l - k n o w n creasing
strength
consistency
concepts
} O are
the
this
it f o l l o w s
intuitionistically
functional
functionals
which
Significant
by p u r s u i n g
which
was
interpretation have
only
results
Hilbert's
developed
in a p r e v i o u s
using [33.
than on the
But
this
proof
more
this
on
of
a descripon the
fundamental
in f u l l
paper
for
justified
for a r i t h -
the b a s i s
ordinal
segments
and
intuitionistic
means
properties
to the b a r - r e c u r s i v e
functionals.
of t h e s e
is shown.
For
today
in the
concentrates
to a q u a n t i f i e r f r e e
functionals
concept.
we have
more
notions
there
of
calculus
is v i a
their
- in the c a s e step
on
constructive
functional
first
con-
one way
inter-
ordinal
o n the c o n s t r u c t i v e The
in-
inductions
f o r m of v a r i a b l e
second
analysis
is o n l y
So the o n l y
of
and relations
into
transfinite
of p r o o f s .
In the
systems
by an i n s i g h t that
good
for
abstract
not mean
analysis
This method
and more
proofs
does
is a v e r y
combinatorial
in a r e d u c t i o n
consistency
can only be
say b y f i x e d
- although
pretation
From
of a n o n -
survey
require
meaning.
to do this, them
of
obtained
results
them which
structive
the e x i s t e n c e
absurd.
type
of a c e r t a i n
and K r e i s e l - T r o e l s t r a ' s
that
- explicit.
the a u t h o r
analysis
consistently
interpretation.
Introduction
between
are
be a d d e d
: a.
failure
means
above
and analysis
of
and A~Va.~
the
cannot
(MP)
is n o t
objects
of t h e r e a l e l e m e n t s ,
functional
O.
principle
- even with
intuitionistic
concept metic
(MP)
comprehensions
tion by
zero
intuitionistic
function
fo~ lawlike
second
than
intuitionistically
that w-rules
The
and
Luckhardt
step
content consists
of a n a l y s i s ,
the c o m p u t a b i l i t y
the n e w a b s t r a c t
notions
H.
used
The
are of
first
mainly
principle
is an i n d u c t i o n
computation the author
t i o n of B r o u w e r ' s
The
second
tool
computation ciple
according
consistency
constructively
of
to r e a l i z e
matical bert's
real
vantage
and
that
it g i v e s
the q u e s t i o n
The
answer
no!
classically
directly method
Can
in t h e
and
gives
of
of
us
this
the
opinion
- although
connexion
it c o n c e r n s
with
which
notions
the
are
prin-
for the theorem;
which
domains
occasion method
to s i m p l e simple
enable
of m a t h e on H i l -
has
a constructive
the adcontent.
type
type
theory?
theory
- the higher
is
types
cannot
be g r a s p e d
forced
to set u p a n e w
analysis.
treatment
the
that we dispose
analysis
So h e r e w e
that
way.
fact
theorem
objects
says
is n e e d e d
at a n o t h e r
of
species.
underivability
be e x t e n d e d
the p r i n c i p l e s
The
is the a d d i t i o n a l
interpretation
is t h a t
manner.
from
which
and uniform
classical
its p r o o f - t h e o r e t i c a l
author's
on the
explained
idealized
up
but which
second
-
is the g e n e r a l i z a -
such fundamental
this m e t h o d
The reason
new
rests
abstract
elements,
in an e f f e c t i v e
which
analysis
ics.
is:
is:
a generalization
produce
proof
to e a c h
built
the m e t a t h e o r y
as a n a l y s i s .
this
in a c o n s t r u c t i v e
the a u t h o r
ideal
Now
only
As
strong
This
the s y s t e m
the c o m p u t a b i l i t y
activity.
types.
to G ~ d e l ' s
sufficient
makes
principle,
in all
directly
as
to s p r e a d s
is n o t w i t h i n
it c a n b e v i s u a l i z e d
us
works
which
for a t t a i n i n g
bar-induction
proof
the
- locally
introduced
procedure
consistency
theory
principle
is a u n i f o r m i z a t i o n
here which
In short,
235
two k i n d s :
tool
the
Luckhardt
in E33
fundamental
The
significance
is s t r e s s e d domain
of
because
of m a t h e m a t -
H.
236
The
aim of
these
will
look
for n e w
to us
a further
two p a p e r s
and
notations
direction
type
theory.
be t r a c e d
the
this
properties.
situation.
These
At
first
then w i l l
we
indicate
of r e s e a r c h .
axiom
theory
systems
over
the
considered
functional
here
for a r i t h m e t i c ,
language
as w e l l
as
the
used.
w 2 is d e v o t e d type
is to a n a l y s e
characteristic
In w I we d e s c r i b e analysis
Luckhardt
to the p r o o f - t h e o r e t i c
By the m e t h o d s
back
o~ G ~ d e l
to c o m p r e h e n s i o n s .
ducible
from
one
simple
Finally
also
extensionality
reduction and C o h e n
Further,
comprehension
and the
each
(C)~
axioms
of
a pure
of
choice
comprehension
of
can be e l i m i n a t e d
simplification
can
is de-
A-property.
by r e l a t i v i z a t i o n
as
in [3].
Because
~((C)~
sistency
of type
represents
a generalization and
theory
is e x p r e s s e d
of the
"modulus
extensionality.
theory
T of
indicator that
for
the
types
the a d d i t i o n
strengthening
of
and
functionals
based
because
distinctions.
w 4 it is p r o v e d to c l a s s i c a l
the
~-operators of
consistency
definition.)
that
w-rules
or i n t u i t i o n i s t i c
points.
respect is,
that
the
no c o n t i n u i t y
does
the
not
shown
alter
formation
the
of c o n t i hand
- is a p r o o f - t h e o r e t i c
the p r e c e d i n g
level
a consistency
~ 0 cannot
theories
to type
(On the o t h e r
assumed
types
-
- It is f u r t h e r
~((C)~
of
con-
indicator"
contains
allow
Therefore
for
with
on c o n t i n u i t y
(C) a, ~ ~ 0 - c o n s i s t e n c y
make
ently
Such
in w 3 the
for e x a m p l e
at d e f i n a b l e
the d e r i v a t i o n
by t r u t h
In
of c o n t i n u i t y " condition
of ~ - o p e r a t o r s
be p r o v e d fine
O(0~)
strength.
indicators
the a d d i t i o n
A necessary
continuity,
by a " c o n t i n u i t y
the p r i m i t i v e - r e c u r s i v e
proof-theoretic nuity
a certain
which
can
proof
has
be a d d e d
contain
the
then to
consist-
H.
primitive-recursive theorem
and w h i c h
Moreover tive
principle
compute
(MP)
O with
of the e x i s t e n c e
the failure
comprehensions
means
elements,
and analysis,
is studied
on the c o n s t r u c t i o n
are c o n c r e t e
m e n t of type
hints
theory
Functional
the n o t a t i o n s
there
in this
statements
indicating
Axioms
language
of type
which
treat-
to formulate
Notations.
the theory of simple
only e q u a l i t y the n o t i o n
types
of type o. Besides "type degree"
~: m a x ( g s l , . . . , g ~ m ) + 1 , the c o r r e s p o n d i n g
an e x t e n s i o n a l i t y
rule
(ER)-qf for q u a n t i f i e r - f r e e
to that in [33,
for f i n i t e l y m a n y e l , . . . , ~ m for gs > g6.
be realized.
proof-theoretic
type ~ can be coded
~ 6 is short =
re-
proceed.
of type theory.
the type O s l . . . ~ m ( m ~ O)
>
to the case of
can a c t u a l l y
Each
>
for arith-
characteristic
set hierarchy.
analogous
of
concept
in [43
Contrary
case,
might
in [33 here we still use
for
Then H i l b e r t ' s
how a natural
E3J using
g ( O ~ 1 . . . s m) denotes
are,
it is s u i t a b l e
in the f u n c t i o n a l
holds
objects
the
can only be describ-
by the author
for type theory.
by i n t e r p r e t a t i o n
language.
For our p u r p o s e s
a manner
this
interpretation
which
explicit.
w h i c h was r e a l i z e d
strictions
T plus
construc-
with M a r k o v ' s
= a. From
for lawlike
functionals
- is m a d e
and analysis,
the power
together
of the f u n c t i o n a l
- even with
arithmetic
w hich
recursively.
of a n o n r e c u r s i v e
As~Va.s
of e-rules
and the fan-
(MP) and A s ~ l Va.s = a follows.
of the real
I.
arithmetic
constants
intuitionistically
incompatibility
ed by i n t u i t i o n i s t i c
These
function
and K r e i s e l - T r o e l s t r a ' s
In the second part
metic
237
intuitionistic
all their
can be p r o v e d
intuitionistic
higher
functionals,
the n o n a b s u r d i t y
function
Luckhardt
pp. and
24-26 6 with
level
in each type
provided
6 within
formulas,
g6 > ge;
in
in
the same
g6 ~ m a x ( g s l , . . . , g S m ) .
238
H. L u c k h a r d t
E s t i m a t i o n s of p r o o f - t h e o r e t i c
strengths - by this we u n d e r s t a n d the
imitation of proofs of one system by proofs of another system in a fixed manner -
(The insight in such facts has its own p r o o f - t h e o r e t i c
are e x p r e s s e d by <, _<, =. [] denotes a quantifier,
strength.)
and Ao,Bo,...
are
q u a n t i f i e r - f r e e expressions.
The following systems are given in the form e x p e d i e n t for functional i n t e r p r e t a t i o n [3~. These a x i o m a t i z a t i o n s differ from the usual ones but are p r o o f - t h e o r e t i c a l l y e q u i v a l e n t to them.
Arithmetic A C l a s s i c a l p r e d i c a t e logic E q u a l i t y axioms for type O Peano axioms Primitive-recursive
functionals T of all finite types
Q u a n t i f i e r - f r e e e x t e n s i o n a l i t y rule
(ER)-qf
Analysis A plus e-choice
(~AC) ~
Q u a n t i f i e r - f r e e choice
(AC)~'B-qf
In [3] it is shown: A $ A r i t h m e t i c ~ T and
Simple type theory A plus Extensionality Comprehension
(ER) (C) ~ for all
Q u a n t i f i e r - f r e e choice
(AC)a'B-qf
~
$ Analysis ^ = T U BR.
H.
is a g e n u i n e (mAC) e comes
part
of
7.
Luckhardt
(C) ~ and
289
(AC)~'~-qf
give
-->
9
~o~
-->
VF~~176
vz 1
(AC) c~,c~
0
AX , y a A ( x , y , z l y x ) ' ))
Gz~~
~
Gzn'
= z(Gzn)n
namely
F = Gz I
(Here and b e l o w
functional
It f o l l o w s
the r e s u l t s
~
from
contractions
in [3]
~ is r e d u c e d (c)O~-A
suppressed.)
and b e l o w
proof-theoretically
: vyO(O~)AxO~[yx
2.
Reduction
to
Our
first
is to s i m p l i f y
aim
the a c t u a l
a model
that
(C) ~ is not
provable
of
essence
of
(see [3],
side
chapter
(functionally
and
the
ing
the old
(ER),
extensionality II).
But
~ O].
of
observation
By the m e t h o d s
that
(C) e
the
(ER)
and
to b e c o m e of G ~ d e l
aware
and C o h e n
(C)~:
can be e l i m i n a t e d flexibility
(ER)-qf
comprehension
~ in o r d e r
(C) ~
for m o r e
interpretable)
instances
AZ~'XZ
to r e d u c e
in A plus
7 ~ A plus Now
~ 0 <-->
the p r o b l e m .
~ can be d e f i n e d
the r i g h t
to A plus
(C)~
~ A plus On
are
for ~ > O.
In w 2
of
from which
as f o l l o w s :
Ax~
in
(AC) ~'~
and for
are
thus
we
by r e l a t i v i z a t i o n add
again
the
simple
obtain:
all simplified.
introduction
We do this
of h i g h e r
types
by f o l l o w increases
240
H.
the s t r e n g t h pressed:
situations
of h i g h e r
Theorem all A
o
of p r i n c i p l e s
Luckhardt
well-known
can be s i m p l i f i e d
in lower
types or o t h e r w i s e
in a u n i f o r m m a n n e r
with
ex-
the use
types.
1(a)
F o r all types
B, 6 i, ~ ~ B, u
L 6i
(i = I ..... n) and
:
. ~ A o ( ~ - l y ~ ' ~-I Y I ' ' ' ' ' A (C)~_[]ly ~ "'" []nZn
~ lyn)
<-->
(C)B-DIX~I ... [ 3 n X ~ A o ( X B , X I ..... x n) where
~, ~-I
The p r o o f
are c o d e s and t h e i r
is o n l y an e x e r c i s e
[3x6B[x] f o l l o w s
in coding.
the r i g h t
the left side by s u b s t i t u t i n g
Theorem
I (b)
Proof:
By t h e o r e m
~-(C)~ A
r
<-->
from w h i c h
one gets back
to
for x.
for all B >
I (a) w i t h ~ - B, n = I
= o-->
Icl~
h e r e can also be w r i t t e n
But this f o l l o w s
I(c)
Because DyYB[~-ly]
side,
(C)~
Icl~
Theorem
types.
= O <--> D 1 X l . . . [ ] n X n A o (%-ly,xl ..... Xn)]
y by #x B y i e l d s
The p r e m i s e
of a p p r o p r i a t e
f r o m ~ - 1 ~ x = x the left side states
VzAya[zy Replacing
inverses
directly
from
F o r e a c h A there
(C)~
-->
as
= o nvB (IY'X(~-Iy))Yl (C) oe ---~I
= O.
(C)~
is an ~ such that for all
B >
(C)-A
A
Proof:
It s u f f i c e s
to p r o v e
the a s s e r t i o n
for p r e n e x
A ~ []iXl...~mXmAo[X I ,. . . , X m , X , U I ,. .. ,u n], w h e r e variables
of A. We use i n d u c t i o n
with respect
u I ,. ..,Un are all free
to m.
H.
I.
m=
Let
~ be
Luckhardt
241
O: the
characteristic
functional
of A
in T
([3],
p.
62) :
O
~ U l . . . U n u = O <--> In t h i s
case
take
e H: 0 a n d
A o [ U , U I ..... u n] ~ U l . . . u n as
the
desired
comprehension
functional.
II.
m+
O:
Case
1:
A
E Ax] D2x2...DmXmAo[Xl
= By
induction
Ax'~
B [ x 1 , x , u I ..... u n]
hypothesis
(C)~
-->
..... X m , X , U I ..... u n]
there
VyAx,x]
is a 6 w i t h
{yxx I = O < - - >
B[Xl,X,Ul,...,Un]}
Thus (c)O6-A^(c)OY-A
-->
VyAX
{AxI.yxx I = 0 <-->
VYIAuOY
(+)
-->
namely
z = l x . y I (yx).
~-:
gives
~ > 6,y
and
(C)~
{zx = 0 < - - >
A X ~ ' U X I = O}
A[X]}
6
if
Y
otherwise
all
g6
> gy
B >
--> (C)~
,, (c)~
Theorem
(+)
--> (C)-A Case
2:
According
A
-= Vx] to c a s e
(C)~
A
Putting
{
for
VZAX
{yl u = 0 < - - >
A[X]}
B[Xl,X,u1,...,u ] there
n]
is e s u c h
Vzmx
{zx = O < - - >
-->
VzAx
{~(zx)
-->
VyAx
{yx
that ~A[x]}
= 0 <-->
= O <-->
for
A[X]}
i[x]}
all
@ >
I (b)
H.
242
namely
y = Ix.s-g(zx).
Remark:
Similarly
For e a c h A there
one
theorem
paragraph
Theorem
I(c)
can be
prehension
can g i v e
is ~ such
(C)~
Using
Luckhardt
^^
(c)~
continued.
figure
is c a l l e d
for
all
-->
(C)-A.
there
reduced
of
Because
in a p r o o f
~ =^ ~r w h e r e
in a p r o o f
that
the r e d u c t i o n
instances
2
an i n t u i t i o n i s t i c
there
type
in the b e g i n n i n g
are o n l y
of A plus
is at m o s t
namely:
B >
~ given
~r H: A plus
reduction,
this
many
com-
(C) ~, we n o w h a v e
(C)~
one
finitely
of
with
(C)~
the r e s t r i c t i o n
that
axiom.
theory.
r
Remark:
From
equivalent
the n e g a t i v e
to its
"intuitionistic"
The
significance
the
impredicativity
meaningful the
special
asserts in our type. tivity
of this into
Thus
reduction two of
comprehension
(C)~
the the
normally
obstacle
considered
lies
in the the
the p r i m i t i v e which
- from we have
but
this
consequences
for
at once
that
~r is
version.
of a f u n c t i o n a l
zero-functional actual
it f o l l o w s
components:
impredicativity
the e x i s t e n c e case
translation
fact
natural
to o v e r c o m e
and
predicative
a fixed
the o t h e r
classically
functionals
formally
separating
it a n a l y s e s
and c o n s t r u c t i v e l y
recursive
looks
all
that
functional
objects
is not
accepted
the
of
the
and same
impredica-
separation
principle.
N o w we d i s c u s s
the
consistency,
which
can be e x p r e s s e d
H.
with
the use
of
Z
243
Luckhardt
as f o l l o w s :
r r
<=>
There
is an ~ w i t h :
~ v x ~176
Ay
{xy : O <-->
Az-yz
= O}
A (Deduction
<~>
There
is a n e w i t h :
~-~ V x O(O~) A + (ER) In this w a y w e a continuity
see
~-~
Here
o n the r i g h t
described
3.
<~>
Ax
For
all
by
Definition:
of
Ay(xy
= O-->
a separation
~: ~ - ~ Ax~176 A + (ER)
Az.yz
principle
indicator,
Consider '''''
u 6 <=>:
= O) }
involves
In g e n e r a l
the
continuity
such a situation
is
p-operator.
9n
%x is a n o t h e r
= O}.
of
^ yz#O)}
indicator".
8 ~ O8n...61
A x e6
Vy,z(xy=O
the u n d e r i v a b i l i t y
Vy ~ ~ . x y
a "continuity
'
In short:
the n e g a t i o n
side we have
{x~ = O -->
Continuity
the place
that
{xC~ = O ^
statement.
(~)
property
theorem)
(n ~ O).
is a c o n t i n u i t y {x(%x)~XU
argument
indicator
for
A % x ( ~ i x ) ... ( g n X ) ~
effectively
different
type
~6 a t
9 .. (~n x) }
from u with
the
same value.
The
"continuity
continuity".
indicator"
Every
modulus
is a g e n e r a l i z a t i o n
of
of
a continuity
continuity
gives
the
"modulus
of
indicator
244
H.
but not vice
versa.
Luckhardt
As an e x a m p l e
for
type
O(OO)
the m o d u l u s
of c o n t i n u -
ity X w i t h Ax ~ < • 1 7 6 1 7 6 1 7 6 1 7 6 1 7 6 gives
the
~ o,
following
B ~ o0,
Remark:
Of
indicator;
The
here
clarity
from
the
these
continuity
the p l a c e
the
above
of the n o t i o n
immediate
situations
insight
are
them
in g e n e r a l
[13,
71).
The
possible portant
to t r e a t example
N o w we r e t u r n indicator tained
r
in T,
"modulus into
continuity
intensional,
Ix~
of
~ (o (o~))
, 41
not
for
comes
there [23,
only
statements
be g i v e n
consistency
and
to a r b i t r a r y
continuity
into
the
continuity
constructively
of c o m p u t a t i o n s .
(see K r e i s e l
indicator"
this w i l l
to the
1 ~176~:
suffices.
of c o n t i n u i t y "
the
sequences
certain
(o (o~))
at p l a c e
incorporated
notion
extensional
from
for
u B can be
weaker
"continuity
lus of c o n t i n u i t y "
indicator
essential
to m a k e p.
uv s uw
Cu ~176176-.: i, X(u,I) , @i u ~: X(u,I)
B I ~ 0;
course
: w~176
is no h i n t p.
of h o w
154 and K l e e n e
generalizes
types,
However
the
but m a k e s
"moduit also
extensionally.
An
im-
below.
~ and
line
type O(O~)
(*).
If a c o n t i n u i t y
at the p l a c e
~o~
is con-
i.e. uO(~
(@u)
~ u~^
~U(@lU)
~ 0 ~ ~(@lU),
T then
for
be p r o v e d
this
~ the
continuity
statement
on the r i g h t
in A:
u~ = O--> -->
Consequently
u(~u)
= u~ = O ^
Vy,z(uy
Cu(r
: O A yz + O)
@ O
side
of
(*) c o u l d
245
H. L u c k h a r d t
Theorem (a)
3
Z ~+ ~
=>
There
for type O(O~) (b)
~r
~
<=>
is no c o n t i n u i t y
at the p l a c e ~o~
The e x i s t e n c e
at the p l a c e ~o~ form AxV~x,
(b) f o l l o w s equivalent VxAy
p l a c e ~o~. finable
of a c o n t i n u i t y
in A+(ER)
to
(*),
is, in A,
(ER), f i n a l l y
~ contains
afortiori
~o~ h e r e c h o s e n
no m o d u l u s
of c o n t i n u i t y
can be r e p l a c e d
at the
by any o t h e r de-
p l a c e of the same type.
o
(~I) u(l,~u)
~z o [u(i,z)
can be a d d e d
D-operator
(with l B -: lXnB" ...x~ 4 .I for
} ul
to a r i t h m e t i c ,
T and a n a l y s i s ,
the p r o o f - t h e o r e t i c
continuity
for type O(~0)
indicators
the f o l l o w i n g
T U BR c o n s i s t e n t l y
strength.
In these
can e a s i l y be e x h i b i t e d .
extensional
continuity
so
- : I,UU
u(@u)
o u(1,~u)
@U(@lU)
o
, @i u -: Wu
, further
So p p r o d u c e s
indicator
@i are a r b i t r a r y ,
= ul
= 1,~u(uu)
= s $ i = l(@lU)
(BI)
and
"inessential"
I: Cu o (ao)
B---OBI...B n)
by the a x i o m s
u(1,v)
changing
extensions
o ul]
extensional
o ul
v ~ < ~u-->
because
to
A z . y z = O}.
is c h a r a c t e r i z e d
without
(C)~
{xy = x ~ <--> A z - y z = O) and, w i t h
_ The p l a c e
puO(~O)
(~2)
for type O(O~)
or in ~ in the l o g i c a l
(a) from the fact that
In the n e x t t h e o r e m we s h o w t h a t the f o l l o w i n g
which
indicator
~i x.
to V x A y ~
Thus a c o n s i s t e n t
%oe(o(o~)) , ~I~(~176
in T for all ~.
c a n n o t be p r o v e d
analogous
{xy = x ~ -->
indicator
at p l a c e
246
H.
In c o n n e c t i o n for
simple
ries
of
with
type
equal
theorem
theory
3
Luckhardt
(e ~ O)
requires
proof-theoretic
this
fine
shows
that
distinctions,
strength
with
a consistency
namely
respect
between
to their
proof theo-
continuity
properties.
Remark:
The
place;
fixed
the c h o i c e
Theorem
place of
i is t e c h n i c a l l y
p ,
arithmetic
T U BR = T U BR p l u s
the
can be r e p l a c e d
by any o t h e r
definable
motivated.
4(a)
T $ T plus
Proof:
I above
It s u f f i c e s
statements
for
Z
,
~ arithmetic analysis
to p r o v e
plus
= analysis
the a s s e r t i o n s
arithmetic
and
analysis
plus
for then
the
functional
follow
domains;
by the r e s u l t s
in [3].
The
computability
can be g i v e n treatment local
by
given
there.
(~
continuity
arithmetic
Theorem ~(C)~
Proof:
p of
(T U BR p l u s
~) of
[3],
X, X I I I
chapter
Formalizing
proof
(see [3],
this p.
a closed
as in [3],
137)
for
term
analogous chapter
T plus
p
of
type
to the
XIV
statements strong
over
do not
theory.
T plus
For
increase deductive
the c o m b i n a t o r i a l
a p)
consequences
strength
we c o n s i d e r
p.
4(b) has
p-
gives
(T U BR plus
O
^ = T U BR).
a sufficiently
point
the m e t h o d
consistency
A in A = T
Thus
in T p l u s
an i n t u i t i o n i s t i c
By t h e o r e m
1(b)
is the c o n t i n u i t y
proof
it s u f f i c e s indicator
in A+~.
to s h o w
previously
~(C)~176 formed
The
with
~.
starting-
of the
H.
u~176176 = O-->
1,Bu(uu)
Ax ~176176 {Xl = 0 - - >
~Vx
{xy = O <-->
Hence
= 0 A Vz~
= 0 -->
Az.yz
ly~176176
Thus
Continuity
sions
(in the
consistently presence
Az.yz
trary
sential
~ I)}
= I)}
= I}
properties
form
of Z).
(C)).
And
does
to the
= ul = 0
Av.uv
= O} w i t h
~v~
u =:
I)-
on every
~(C)~
247
1 ^ u(1,Uu)
n V z ~ 1 7 6 1 7 6 ~176{zu = 0 < ~ >
x =:
of
vyOO(xy
{xl = 0 ^ A y ( x y
nVxAy
= O @
Luckhardt
Therefore
together
not
alter
with
imply
classical
constructive
situation
extension
constructively
type
functional theorem
negations
domain
4(a)
(C)~
as w e k n o w f r o m
is a d d e d theorem
I(c)
and
does
not work
not
that
strength; which
comprehen-
[e.g.
we have
the p r o o f - t h e o r e t i c
where
theory
of
in the
the addition
this
is c o n -
constitutes the
theory
a n es-
of
truth
definitions.
In c o n c l u s i o n
Theorem
the
"continuity
perspective"
c a n be
summarized
as f o l l o w s .
4 (c)
A
T = A plus
~
,
~(C)~
^
T U BR = ~plus All
systems
lar
they are
Proof:
~
, BR,
between
have
consistent
Arithmetic
~(C)~ the
same proof-theoretic
strength;
in p a r t i c u -
[3].
and analysis
are
treated
in p a r a l l e l .
A
T
(T U BR)
= A
(~
)
[3]
< A
(~
plus
BR)
plus
Z,
= A
(~
plus
BR)
plus
~
A
7 (c)Oe-A Theorem
4(b)
H.
248
4.
Invalidity
Also
T
(T U BR)
T
(T U BR)
of h i g h e r
connected
with
plus
Luckhardt
~
functional Theorem
observations
all
for
(total)
4[ u s ] taken
classically
(eR) ~ e x p r e s s e s "natural" also valid O(OO). ~
equals
initial the
0 or not.
segment
This
~
of
has
vy~176
functional
in c o n t r a d i c t i o n
VX O(OO)
follows
~-rules
f u n c t i o n a l c o n s t a n t s ~i
for
Take
case we - say of
length
n - which
so t h a t ~ ( ~ , n
Vy(xy
(~R) O(~
= 0 A Vz.yz
+ O)}
= 0 ~>
(C)~176
Az-yz
= O) }
whether
a finite
contains
all
* i) = 0 l i k e -
~ O)
c ~.
Is t h i s
~ of t y p e
steps
(~,n * l)n = I + 0
o(oo)
from
of
determine
to
immediately
constant
in ~:
= O ^ Vz~
{X~ = 0 ^ Ay(xy
and e v e r y
consistently.
number
can also
computation,
* i) = O ^
{x~ = 0-->
this rule
numbers,
in ~ a f u n c t i o n a l
expressed
constants
the n a t u r a l
in a f i n i t e
first
the
of
to a d m i t
the a r g u m e n t
needed
Ax O(OO)
which
In t h e
~(~,n
pO = 0 -->
for a l l
concept
types?
can be fully
= 0-->
are higher
(u v a r i a b l e )
it c a n be e s t a b l i s h e d
information
wise.
therefore
for h i g h e r
Then
4(a)
or c o n s t r u c t i v e l y .
a standard
theory
[33
e-rules
our previous
(~R) ~ A [ ~ O ] . A . ~ .I ] ,.
interpretation
now gives
H. L u c k h a r d t
249
Thus we have proved that an w-rule for type O(00)
cannot be added con-
sistently to theories over c o n s t r u c t i v e f u n c t i o n a l domains w h i c h contain i n t u i t i o n i s t i c logic and
(C)~176
This result can easily be extended to
all higher types because we have
T h e o r e m 5(a)
Proof:
(wR)~ ~ -- (wR) B T
Let AIr, I, A [ ~ ]
for 8 <
.... be p r o v a b l e for all functional constants
B Then also A [ ~ - I ~ ] A[~-I~] are p r o v a b l e for all f u n c t i o n a l ~i" ' '''" constants ~ Now
of type ~. Here % is a code for B in e and ~-I its inverse.
(wR) ~ gives A[~-lue];
so in p a r t i c u l a r we have A [ % - 1 r B] from w h i c h
A[v B] follows with the coding r e l a t i o n ~-1%u8 = u by
(ER)-qf.
Consequently
T h e o r e m 5(b) (wR) ~ cannot be added c o n s i s t e n t l y for ~ > 0(00)
to theories over con-
structive f u n c t i o n a l domains w h i c h contain T, i n t u i t i o n i s t i c
logic and
(C)~176
The case of n u m b e r - t h e o r e t i c functions
is not covered by this method.
But the result holds here too and is also valid for m u c h weaker theories.
T h e o r e m 5(c) w-rules for types d i f f e r e n t from zero cannot be added c o n s i s t e n t l y to theories in which all f u n c t i o n constants can be computed r e c u r s i v e l y and w h i c h contain functionals,
(relative to their language)
the p r i m i t i v e - r e c u r s i v e
i n t u i t i o n i s t i c a r i t h m e t i c and the fan-theorem.
For instance all
systems of i n t u i t i o n i s t i c
analysis
(with ~ i n
(wR) ~
250
ranging
over
(because
[I],
Proof:
By
leads
freely
the
Kleene
growing
fan-theorem p.
115)
theorem
tree w h i c h
all
Where
choice
objects)
come
5(a)
under
this
it s u f f i c e s This
([I],
to s h o w
for
p.
by
according
that
by u s i n g
the a d d i t i o n Kleene's
all r e c u r s i v e
W1(x~176
-: T1((X)o,X,y)
Az < Y ~ T 1 ( ( x )
, (u~
I are
the
R is d e c i d a b l e
projections
-: V t < x
and has
Vy<x
A y < w ~ " u y : v ~176 y-->
Let o o
be a r e c u r s i v e
consideration. so we
Then
of
can m a k e
I:
1,x,Z)
a primitive
=
[R(u;w)
~ is also
I <-->
have
<-->
recursive
pairing
constant
informally numbers
R(V;W) ]
of
fo,fl
VyT1(fo,t,y),
o
the
following
sg(~t)
,fl > , sgu ~
sg(~f)
= I
=:
derivation
Ix.sg
(ux) .
the
consistent
a well-defined
(Kleene
Case
but n o t
t-W sg (ut) (t,y)
function
effectively
sg(~t)
f -:
of a
the p r o p e r t y :
(i)
Put
(~R) ~176
R (LlO O ;x ~
(2)
of
112).
Az < y ~ T 1 ( ( x ) o, X t Z )
^
to
example
sequences
~: T1((x) 1,x t y) ^
function
N o w we
(AC) ~
of
theorem.
is done
is w e l l - f o u n d e d
sequences
here
as all e x t e n s i o n s
Wo(x~176
(U~
rithm;
as w e l l
can be p r o v e d
to a c o n t r a d i c t i o n .
binary for
Luckhardt
H.
theory
under
recursive
algo-
with
= O <-->
[O],
within
p.
the
VyTI (fl ,t,y )
281)
theory.
H. L u c k h a r d t
By
(2) ,I t h e r e f o r e
(2),2 gives
we have ~ with:
(4) From
to K l e e n e
Az < y (3) , (4)
A ...
[03,
Case
p.
197
W 1 (f,y)
and w i t h
then
also
x -: f+7+I
VxOR(sgg,x;x)
2:
sg(~f)
(I)
= 0
in the same way
Thus
function
for each
Ay~
Hence w i t h
< I ~>
oo
I.
of the t h e o r y
Ay.~y =
(sg~)y
Vx~
(5) , (I
(wR) OO
Az ~ and with
as in case
constant
-->
This
e166a
f. Wsg((sgq~)f ) (f,y)
(5) is p r o v e d
(6)
^ 7TI ((f)1,f,~)
7TI ((f)1,f,z)
Vf<x Vy<x (5)
T1((f)o,f,~)
in this case
7TI ((f)1,f,O) According
(3)
251
{Ay. zy < I -->
Vx~
the f a n - t h e o r e m Vm~
~176{Ay. zy < I -->
is n o w refuted.
Take
Vx < m . R ( z , x ; x ) }
~ as the f o l l o w i n g
primitive
tion: I I (<m~ t ~ ) -:
if t<m A V y < m
0
if t<m ^ V y < m -
0
otherwise
" t-W O(t,y) t.W I (t,y)
recurslve
func-
252
H.
Luckhardt
Then %0(<m,t>)
(7)
< I
x --< m A R ( l y . w ( < m , y > ) , x ; x )
-->
Vt<mVy<m-'t
-->
With
(7) , (8)
< m
Am~
Further,of
by
sequences
(MP)
[~(<m,t>)=O
-->
Wo(t,y) A %0(<m,t>):1] }
~ 1 7 6{ A y . z y
to
< I ^
A
can
the rule
that
variable in
the
([5],
last
{Ay.ay
< m
317).
is
lawlike
proof,
we
< I -->
the
choice
objects.
If w e
obtain
Vx~
For
gives
transformed
by
has
objects apply
the
same
is n o w this
re-
re-
only
}
of
this
(~R) s w h i c h
u s for
a result
But
this
can be
Kreisel-Troelstra's
repaired
theory
of
choice
Vx~
intuitionistic
into < I -->
rule
fan-theorem.
< I -->
{Ay.sy
the
the variable
As ~ { A y . s y
As
9 ~ R(z,x;x)}
interest
of
= a, p.
Ax
a s for
the premise
As ~ V a . s
be
,x;x)
(6).
intuitionistic
is n o t
using
which
W I (t,y) A ~ ( < m , t > ) : O ]
(t,y)
i t is p r o v e d
Aa OO
by
-->
(~R) s e x c e p t
stricted
which
{[~(<m,t>)=1
~ R(ly.~(<m,y>)
in c o n t r a d i c t i o n
placed
--< mVy<x'-t _< m-'t. W s g g ( < m , t > )
(7)
Ax
as
Vt<x
~
(8)
form
-->
Vx~
logic
and
Markov's
principle
H.
With
this
Theorem For
supplement
the a b o v e
proof
theories
in w h i c h
contain
functionals,
all
function
(relative
intuitionistic
incompatibility
holds
situation
the basic
thus
ideas.
Theorem
which
in f o r m of
(see K l e e n e
[O], Aa O O
(Ch T) O
p.
and
recursively
the p r i m i t i v e - r e c u r s i v e
the
fan-theorem
the
following
5(b)
(MP) ~
will
0 =
perhaps
is b a s e d
idea behind
become
clearer
o n the c o n t i n u i t y
theorem
the f o l l o w i n g
I
5(c), (d)
consequence
if w e r e v i e w of c o n s t r u c t i v e
is r e c u r s i v e n e s s
of C h u r c h ' s
thesis
281)
9 ar
with
(Ch T) ~ is c o n n e c t e d
with
~e~
~: V e A x
{~x=O
<-->
(~R) ~ 1 7 6in the f o l l o w i n g
VyT1(e,x,y)}
way.
5(e)
theories
contain
case.
c a n be c o m p u t e d
language)
arithmetic
explained
The basic
For
in this
for ~ ~ O:
functionals. enters
constants
to t h e i r
(~R) ~, As ~ 7 V a ( ~ = a),
Theorem
also works
5(d)
and which
The
253
Luckhardt
in w h i c h
the relevant
all function part
of T a n d
constants
are recursive
intuitionistic
logic
and which
the f o l l o w i n g
holds: (~R) ~
Proof:
A~ O O
By t h e o r e m
gether
with
within
the
From
~
this
5(a) :
theorem theory
9 ~e~
,
(~R) ~176 ~
(~R) OO ~
IV in K l e e n e
VeAx
the a s s e r t i o n
{px=O
[O],
<-->
follows
(~R) O p.
I
Aa O O
(~R) ~176~ 281
used
VyT1(e,x,y) } for
with
(~R) ~ 1 7 6resp.
9 ae~
(~R) O
(~R) O to-
informally every
gives
constant
(~R) ~176
oo
2~
H.
Remark: For
The
same
theories
which
necessarily tion
partial,
= {e}.
= {e} ~
A[{I}] ....
We n o w
system
tionistic the
with
we
arithmetic
and
AS
only
A[a]
to s h o w
(~R) ~
with
as far
Then
AaVe.a
the
first
=
thesis.
within
of T, part
form
AE{O}],
Church's
part
situa-
in the
as p o s s i b l e
the r e l e v a n t
fan-theorem.
thesis
-
now
to this
in p a r t i c u l a r
follows
5(a)
to c o n t a i n
the
(~R) ~176e x t e n d e d
A[~I] .... ; then
of t h e o r e m
ax = Uy}.
{i} of their,
(~R) ~ to C h u r c h ' s
we have
(~R) ~
suppose
(~R)~176
(9)
A[~o],
the p r o o f
which
over
{T1(e,x,y)^
enumeration
constants,
foregoing
and A [ { u ~
AaVeAxVy
a recursive
equivalent
Assume
formalize
(Ch T) :
function
By the
(~R) OO.
for
contain
is d e d u c t i v e l y
AaVe.a
the
holds
Luckhardt
intui(up to
gives
{~c~ ^
Ay-~y
< I -->
Vx~
Consequently
I
Ae.~c~-->
Ac
{Ay-~y ~
I -->
Vx~
Aa.ae~-->
Aa
{Ay.ay
< I -->
Vx~
(10)
The
second
part
(beginning
with
the a p p l i c a t i o n
of
the f a n - t h e o r e m )
becomes
(11)
As
{Ay.~y
N o w we
see
that
(I0),I
and
(11).
Applying
the
< I -->
theorem
additional
Vx~
5(c)
is a d i r e c t
argument
for
consequence
theorem
5(d)
to
of
theorem
(10).,2 and
5(e) ,
(11)
gives
(12)
Again
As n~ Va-~
theorem
5(d)
= a ,
follows
(MP)
~
~Aa~176
immediately
from
this w i t h
theorem
5(e).
H.
(12) of
can be i m p r o v e d
Luckhardt
by s t a r t i n g
255
a similar
argument
with
(11)
instead
(10) ,2:
~A~
77{Ay-~y
77V~
Together
~
{Ay.~y
with
(9)
I -->
Vx~
}
< I ^ ~Vx~
this
(11) , (MP)
}
implies
~V~o~ From
this
the n o n a b s u r d i t y
tive
function
of
the
existence
of a n o n r e c u r s i v e
construc-
7 7 Va-a#~
follows
with
Theorem
5(f)
For
theories
tionistic
As77Va'e
which
contain
arithmetic
(MP) ,
This
concludes
this
be g r a s p e d
constructive
part
= a
and
!
the p r i m i t i v e - r e c u r s i v e
intui-
the f a n - t h e o r e m :
A~77Va.~
I; p a r t
= a
II d e a l s
proof-theoretically
method
functions,
available
today?
~
77Va.a#C~
with
the p r o b l e m :
by i n t e r p r e t a t i o n ,
In w h a t the
way
can
strongest
256
H. Luckhardt
References
[03
Kleene,
S.C.:
Amsterdam [13
Kleene,
Introduction
S.C.
and Vesley,
Mathematics. [2]
Kreisel, logic.
[33
[43
Amsterdam
R.E.:
The F o u n d a t i o n s
Luckhardt,
logic 27
of intuitionistic
(1962),
H.: E x t e n s i o n a l
G~del Functional
Proof of Classical Analysis.
Lecture Notes
in Mathematics
Kreisel,
306,
H.: Uber Hilbert's
fur m a t h e m a t i s c h e
Annals math.
logic
Springer
Interpretation. 1973.
reale and ideale Elemente.
Logik und Grundlagenforschung,
G. und Troelstra,
of intuitionistic
predicate
139-158.
A Consistency
Luckhardt,
of Intuitionistic
1965.
G.: On weak completeness
J. symb.
Archiv [5]
to Metamathematics.
1962.
A.S.:
Formal
systems
to appear.
for some branches
analysis. I (1970),
229-387.
Postscript I If partial
functionals
are admitted
in w 4 then the connection between
the higher types and type O must be revised, 5(b), (c), (d) and Postscript
?
The m e t h o d
of w
(f) carry over under appropriate
also gives a nice f u n c t i o n a l
metical
(C) ~ : This reduces
pretable
via
p. 46, 74,
but our main theorems
(AC)O,O_Az o
88-91).
,
to
(C) ~ - Ax ~
(4 A 4)0
-
conditions.
interpretation
of a r i t h -
w h i c h is f u n c t i o n a l
VyOAz o
by
T u BR ~
(see
inter[3],
CHURCH ROSSER THEORE~ UNENDLICH Herrn Professor
iWOR A - K A L K U L E
MIT
LANGEN TER~EN Dr. Kurt SchGtte
seinem 65. Geburtstag
zu
gewidmet
W. Maa~
In dieser Arbeit wird mittels
Transfiniter
Theorem fdr einen typenfreien bewiesen.
Der Beweis
vorliegenden
l~Bt sich unmittelbar
~ -KalkGle
Induktion
mit unendlich
halten.
Wir geben am Schlu8 an, wie die auftretenden Anwendungen
und Schwichtenberg zahlen bewiesen Martin-LSf
zur Termbildung
dieser Art mit Typen hat Girard~
~
an
angegeben. entstehenden
das Church Rosser (unver~ffentlicht).
der Terme
I)
0 , S
und die Variablen
2)
Sind
a
3)
Sind
ai
und
b
fGr alle
k-KalkG1
haben Barendregt
entwickelten
Methoden
i)
lob = IsL : J il :~
2)
l~xal
3)
ll
lal+1
,
sind Terme.
so sind auoh Terme,
der L~n~e eines Terms
=
KalkG1
9
(~xa)
so ist auch
:
labl = max(lal,lbl)+1
= sup(Jail+l)
der
Wit benutzen beim Beweis die von
Xl,X2,..
i ~ ~
bei
in [I] einen Beweis mit
F~ir den durch Weglassen
:
Terme,
ent-
Ordinalzahlen
[~ ).
Definition
Definition
und Reduktion
Theorem mit Hilfe von Ordinal-
und Tait fur den endlichen
(siehe Stenlund
auf die meisten
genGgend klein gehalten werden kSnnen.
Hilfe von Fundierungspr~dikaten Reduktionsregel
langen Termen
(mit oder ohne
die noch zus~tzliche
Ftir A -KalkGle
Regeln
ausdehnen
langen Termen
Typen),
beweistheoretischen
das Church Rosser
l-Kalktil mit unendlich
und
(ab)
Terme.
ein Term.
W. IVIaaB
258
Mitteilun~szeichen
:
i,j,k,m,n fur natGrliche Zahlen) zahlen; a,b,c,d,e fGr Terme; n mit n-maligem Auftreten yon
a,~,~,6 ffir abzihlbare Ordinalfir Terme der Gestalt (S(S..(SO)..)
S 9
Unser Ziel ist, das Church Rosser Theorem fir den folgenden Reduktionskalk~l ~ zu beweisen :
I)
a~
a
2)
(Ixa) b ~
3)
~
4)
(b) o ~
5)
a ~
6)
ai ~
7)
a ~
ax[bU
~
an
a'
,
a!l b
,
b
b ~
b'
=>
ab ~
fir alle
i e ~
b ~
=>
a'
a'b'
=>
a ~
, ~xa
~
~ Xxa,
a'
Wir geben einen zu diesem Kalkil iquivalenten ReduktionskalkG1 an, bei dem den einzelnen Reduktionen Ordinalzahlen als "Reduktionsordnungen" zugeordnet sind. Das Church Rosser Theorem fir den KalkG1 list sich dann dutch Induktion iber diese Reduktionsordnungen beweisen.
Reduktionen mit Reduktionsordnun~en (wir schreiben
~
anstatt
~ a,1
(I)
a
~
a
fir alle
a
(2)
a
~
a'
und
~
(3)
~
(4)
~
(b) c (5)
a
~
a'
(6)
a
~
a'
=> ,
~ ,
=>
b
: )
b'
-nb
~
b'
=>
(~xa) b
~ ,
~
b'
~xa
~
=>
Xxa'
ab
a'x[b']
a'n c
~
c'
b' b
~
~
a'b'
=>
Maa~
W.
(7)
ai
(8)
a
>
fGr alle
a!1
~
b
259
i e ~
=>
nach elner der R e g e l n
~
(I),..,(7)
und
at
b ~,n
mit (9)
=>
~ <
a
~
jeweils
und
b
Bemerkun~en:
Lemma I.
b
Anstatt
(j)
a
~
~ ~,n
a'
a'
J ~ a' a,n
=>
a
~
a'
a,n+l
der Regel
eine anschlie~ende
Wir schreiben Regel
a
(8)
kSnnte man bei den Regeln
Reduktion
~ ~,n
um mitzuteilen,
mit dab
~ < ~ a
(I)-(7)
zulassen.
~ ~,n
a'
zuletzt nach
erschlossen wurde.
I :
c
> ~,n
c'
8~
d
2.
c
3.
kxa
4.
=>
c
~
c'
f~ir
~ ~,n
c'
g >~
a und
~ a,n
d
~xa'
~
=>
a
=>
mit
~ a,n
~ < a
C
=>
[~
C'
a'
ai
>
a!1
C
8~
c'
ffir alle
i E
Beweis: I.
C
I~
C
>
2.
c
8~
d
=>
Dann gilt nach
I.
=>
C' c
J~
e
e
c'
~ ~ ,m
d
mit
j < 8
mit
8 := max(~,N)
und < a
~ < ~ . und
k c
8,k =>
c
i--
~
>
e
3.,4.
c'
Induktion nach
Lemma 2 : 0rdinalzahl
=>
c
8>
c'
6,k
G.
a
~ a
a
~m+n
<=>
a
mit einer a b z ~ h l b a r e n
260
W.
Beweis:
"=>"
Zur B e h a n d l u n g benutzt ai
~
Induktion
nach
der Regel
6)
(Gber Lemma a~
I. I.)
fGr alle
Maa~
der D e f i n i t i o n wird
von
die Regel
a
(8)
~
a'
des KalkGls
:
i e ~
=>
(I.V.)
a~
ai
mit abz~hlbare
ai 0rdinalzahlen
~i
fGr alle
i E ~
=>
a~
ai
fGr
a
mit
a a > ai
fGr alle
"<="
Induktion
Lemma
3 :
Beweis:
c
i e nach
~ ~,n
also
Nebeninduktion
(Regel
(7))
~
a
c'
und
d
~
d'
=>
Cx[d]
~
Induktion
I) n = I ,
=>
nach
~+n
J~
c'
c
nach
Ic[
~ ~,n
C'x[d']
. mit
j e {I,..,8).
. Die F~lle
j = I,..,7
machen
keine
Schwierigkeiten. Sei j = 8 :
c
8~
o'
=>
c
k~
e
~
und
~ < a . Da die B e h a u p t u n g
hat man d'
>
ergibt 2~ F~ir
Cx[d] d'
:
sich n ~ I
Church Rosser es gibt
Beweis:
einen
~
ex[d'] Ox[d]
C'x[d']
c'
k~
e
mit
k ~ (I,..,73
schon b e w i e s e n
erh~lt
. Mit Regel
:
c
nach
c
mit
~ a,n c'
a ~ ~
Nebeninduktion
unmittelbar
(8)
und Lemma
aus der I.V.
c'
~,m ~ ~,m
8 ~,n
. nach
Icl
wurde,
man w e g e n
C'x[d']
die B e h a u p t u n g
Theorem Term
c
Aus der I.V.
~ ~ ,m
~
folgt
Induktion
11 n = m = I
exCd']
f~r
~ ~,m
1.2.
W. MaaS Zun~chst
wird
die Behauptung
j,k ~ {I,..,7}
mit Hilfe
Jeder einzelne
261
c J>
f~r die Fille
c'
,
der N e b e n i n d u k t i o n s v o r a u s s e t z u n g
Fall ist trivial,
soda~
einige Beispiele
k>
o
bewiesen.
gen~gen
:
~ = 2 ~ k = 5 : (Xxa) b
2>
a,x[b, ]
mit
a
(kxa) b
5>
(Xx~) ~
mit
(nach Lemma
a'
>.
~
~
. Aus der N.I.V.
folgt,
und
sowie
~
~
0
~
a'
und
b
1.3.)
> a
da~ es Terme
~
b' >
und
und
gibt mit . Es
a
ergibt sieh
(~xg) ~ 2~
ax[b ]
und nach Lemma 3
a'x[b']
&x[6]
9
~ = 3 r k = 5 :_n
3~
an'
mit
n
5~
<~i > n
mit
>
~
<~i >
0
Aus der N.I.V.
folgt:
es gibt
einen Term
, sodaZ
b
,
^
also nach Lemma
1.4. auch
al,o
~
~i =
<~i > ~ 4
,
k
=
5
4>
c
und
~
Zun~chst
~
k' = 8
&
n
, und
,
an
b' b
(b) c
und
k' ~
, dabei
e
kann man die Behauptung
der N.I.V. Ftir
~
:
(b) c ~
a' n
ist
ftir die F~lle
5>
mit
e
k' = 1,3,5,8 k' = 1,3,5
m~glich. sofort mit
beweisen. hat man dann
b
J'~
eI
,~,
e
mit
j' < 8
und
P < ~
. Aus der Gestalt
yon
b
folgt
j' ~ ~1,3,5}
9 Es ergibt
262
W. M a a ~
s i c h das
Diagramm
(
b)
wobei
c
(+)
aus
dem vor-
her B e w i e s e n e n (++) b'
~
Nachdem kSnnen
C
8>
~
:
=>
k !
C
der
<
I.V.
d
F~lle
~ k = 8
m @ ~
aus
folgt.
fur die F A l l e
die r e s t l i c h e n
E {Ir..r7}
eI
-
die B e h a u p t u n g
wegen
~ ~
und
~
e
j,k
behandelt
n
s {I,..,7}
bewiesen
ist,
werden.
~
mit
k'
~ {I,..,7]
und
~ <
P Es e r g i b t
sich
wobei
c
(+)
F~llen
~ =k=
aus
und
den v o r h e r
(++)
behandelten
aus d e r
I.V.
folgt
8
~an beweist
diesen
Fall
in a n a l o g e r
Weise
unter
Verwendung
des v o r i g e n
Falles. 2~ n >~ I ~ m = I
Nebeninduktion
nach
n
mit H i l f e
yon
3) n >~ I , m >~ I
Nebeninduktion
nach
m
mit
yon 2
Hilfe
I
W. Maa2 Fiir die Theorie formuliert angegebenen dadurch
der primitiv
mit unendlich
lisation
der Terme
duktionen Gber nur Terme mit
rekursiven
Funktionale
langen Termen nach Tait
Reduktionsregeln
sich ergebende
263
nur Ordinalzahlen
Reduktionsbegriff
durchf~hren
endlicher
[4] < ~o
zugelassen.
ist stark genug,
zu K~nnen.
um die Norma-
zu k~nnen,
feste Ordinalzahl
beschr~nkten
zu. Es ist dann beim Church Rosser Theorem mitzubeweisen,
L~nge
der zur Erg~nzung
nicht ~berschreitet. finite
Induktionen
Mit dieser Methode wurde bis
system mit unendlich lysis
interpretiert
benStigten
~o
Terme
in Maa2
das Church Rosser Theorem
langen Termen bewiesen,
In-
l~2t man
L~nge
der Diagramme
Der
Um auoh die verw~ndeten
die L~nge yon Termen einschr~nken einer durch eine geeignete
Typen,
, werden bei den
da2 die
diese Schranke [21 durch Trans-
f~r ein Funktional-
in dem die pr~dikative
Ana-
werden kann.
LITERATUR [I]
J.Y.Girard,
Theoreme
de Church-Rosser
infinis. [2]
W.Maa2,
pour un systeme
de termes
(unverSffentlicht)
Eine Funktionalinterpretation Analysis.
Dissertation
der pr~dikativen
an der Universit~t
MGnchen
(1974) [3]
S.Stenlund,
Combinators, D.Reidel
[4]
W.W.Tait,
k-Terms
Publishing
and Proof Theory. Company,
Dordrecht
Infinitely
long terms of transfinite
In: Formal
systems
and recursive
(ed. Crossley/Dummett),
(1972)
type.
functions
S. 176-185.
U B E R S K O L E M E R W E I T E R U N G E N IN DER I N T U I T I O N I S T I S C H E N L O G I K MIT G L E I C H H E I T H e r r n Professor Dr. Kurt SchGtte
zum
65. G e b u r t s t a g gewldmet H. O s s w a l d
Als erster hat G. Eo Mints in
(7) gezeigt,
da6 im G e g e n s a t z zur
k l a s s l s c h e n L o g l k mit oder ohne G l e l c h h e i t und im G e g e n s a t z
zur
i n t u l t l o n l s t l s c h e n Logik ohne G l e i c h h e l t in der i n t u l t l o n i s t i s c h e n L o g i k mit G l e i c h h e l t
(abgekGrzt ILG) S k o l e m e r w e i t e r u n g e n im a l l g e m e i -
h e n keine k o n s e r v a t l v e n E r w e l t e r u n g e n slnd. G. E. Mints hat eine Formel A angegeben,
die die e l n s t e l l i g e
F u n k t l o n s v a r l a b l e F nicht enth~lt und fGr die gilt: ILG + V x R x F x ~ A aber ILG + V x 3 y R x y ~ A. Mints definierte: Rxy
: = (Gy ^ x = a -~p) --,Mx,
w o b e l G, M b e l i e b l g e e l n s t e l l i g e R e l a t l o n s v a r l a b l e n und a, b O - s t e l l i g e Funktionsvarlablen A
sind
und p elne A u s s a g e n v a r l a b l e
ist, und setzte
: = 3 z ( ( G Z -~ p) --~Ma A Mb).
H i e r soll eln etwas e l n f a c h e r e s B e l s p l e l mit d e n s e l b e n Eigens c h a f t e n a n g e g e b e n werden. Sind @, I, QXy und
B
r O - s t e l l l g e F u n k t i o n s v a r i a b l e n und d e f i n i e r t m a n
: = (x = @ ^ y = ~) v (x = 9 ^ y = ~) v (x = e A y = c) : = 4@ = ~ ~ c = @ v ~c = @,
so gilt e b e n f a l l s (I) ILG + V x Q x F x ~ B aber
(x) ILG + Vx~yQxy ~ B, wie m a n durch eln K r l p k e - M o d e l l n a c h w e l s e n kann. D i e s e s Modell ~ l~Bt slch g r a p h l s c h so darstellen:
H. 0sswald
Q
265
9
|
0 Die Situatlonsmenge yon ~ hat 2 Elemente ~,8
mlt a ~ 8. Die
belden klasslschen Strukturen ~a' ~8 haben Jeweils 3 Elemente, die gleichzeitig O-stellige Funktionen sind, wobei in ~8 @ u n d c
gleich
sind. Die Pfeile beschreiben die Relationen Q ~ und Q~ . Also: %
= C(~,~),(~,~),(~,~)~,
In ~ ist Vx3yQxy g~ltig, B aber nlcht gGltlg. Die Aussage
(I) kann man ebenfalls semantisch zeigen:
Gilt ILG + VxQxFx ~ B, dann glbt es elne Kripke-Struktur B mlt Situationen ~, 8,~ r 8, so dad gilt:
(I) a ~ VxQxFx (2) ~ ~ ~
=
(3) a ~ =~ (~) ~ = @ . Aus (I), (~) folgt a~F@
=4
~Fr
= r
Hieraus folgt :
8~Fr Hieraus und aus (4) folgt
Widerspruchl Die Aussage ILG ~ Q|
Denn aus (2) folgt 6 W ~ @
= ~.
(I) bewelst man syntaktisch so: Zuerst zeigt man ^ QcF~ -~B.
266
H. 0 s s w a l d
D u r c h V - Q u a n t i f i z i e r u n g und S t r u k t u r s c h l u B erh~lt man die Behauptung. A.
So T r o e l s t r a machte darauf aufmerksam,
das u n a b h ~ n g i g yon der A r b e i t
dab dieses Beispiel,
(I) e n t s t a n d e n ist, in e n g e m Z u s a m m e n h a n g
mit a n a l o g e n B e i s p i e l e n in der i n t u i t i o n i s t i s c h e n A n a l y s i s Es sei @ die C a u c h y f o l g e der O, ~ die der folge yon r a t i o n a l e n Zahlen,
Iund
steht:
~ eine Cauchy-
ffir die c = @ nicht e n t s c h e i d b a r ist.
D a n n gilt zwar ffir [@,~,~] Vx3yQxy, w e n n man Q wie oben deflniert, gibt aber keine A u s w a h l f u n k t i o n F mit VxQxFx.
es
D e n n well man ~ = @
nicht v o r a u s s e t z e n darf, muB F@ = ~ und F~ = c sein,
well m a n aber
auch ~c = @ nicht v o r a u s s e t z e n darf, k a n n man nlcht zeigen, dab F eine F u n k t i o n isto
Literatur (I) Mints,
G. E.: Skolemts m e t h o d of e l i m i n a t i o n of positive q u a n t i f i e r s in sequential calculi, DOklo Akad. N a u k SSSR, 861 - 864, Bd
169,
1966.
EINE VARIANTE
DES BEZEICHNUNGSSYSTEMS
Herrn Prof.
W(X)
FOR O R D I N A L Z A H L E N
Dr. K. Schfitte zum 65. Geburtstag
gewidmet
H. Pfeiffer
In dem in [I I b e s c h r i e b e n e n der dort e i n g e f a h r t e n az = bz W(X)
sein.
System
Bezeichnung
W(X)
far Ordinalzahlen
fur verschiedene
Dadurch wird die Definition
und damit auch der Wohlordnungsbeweis
genden wird ein System Vergleich
zu W(X)
W~(X)
angegeben,
so eingeschr~nkt
W(X) wird in W*(X)
ordnungstreu
W*(X)
Ordinalzahlen
ebensoviele
Bei V e r a l l g e m e i n e r u n g e n ante W*(X)
Ordinalterme
a, b
der O r d n u n g s r e l a t i o n
< auf
recht kompliziert.
Im fol-
bei dem die Termbildung
wird,
dab aus
eingebettet,
a + b
womit
im
az # b z
gezeigt
folgt.
ist, dab in
b e s c h r i e b e n werden kSnnen wie in W(X).
yon W(X)
auszugehen,
kann mit
darfte
es natzlich
um den erforderlichen
sein, yon der Vari-
Wohlordnungsbeweis
zu ver-
einfachen. I. Definition Sei
X
und E i g e n s c h a f t e n
eine durch
kleinstes
Element.
eventuell
mit
<
wohlgeordnete
Als V a r i a b l e n
simultan
a ist Term yon
Wt(X),
Ga
(1.1)
Jedes
Element
Sind
al,
und a2 ein
S(al,a2)
:= Sa 2
Sind
sei
O
ihr bezaglich verwenden wir
x,y,z,
Induktion:
yon
X
sei
es
sei
Sx
:= x
yon
aus
Term yon
al,...,a
# ...
x
y
(al,a2)
aI
durch
a ist Hauptterm,
Terme
(ao,Y) , wobei
(1.5)
Menge und
fur ihre Elemente
die Stufe
Sa
yon a und den
yon a.
Hauptterm, (1.2)
Wt(X)
Indizes.
Wir definieren Grad
yon
~ ap
S(a I # ...
p
und
ist
# ap)
und und
W*(X),
und
G(al,a2)
Term yon := max
Term yon
und
W*(X) X
Hauptterme ein
ein
W*(X),
Gx := 0 hat
a I nicht
Sa 2 < y zwar
ein
ist
die
< S~ ~
p > I,
W*(X),
aber
kein
{Sajll
~ j
~ p}
zwar
ein
so
Form
gilt,
Hauptterm,
:= Ga I + Ga 2 + I
und
und
.
so und
sei es
sei
. sei
Hauptterm, und
und
es
268
H. P f e i f f e r
G(a I # ...# ap) Wir b e n O t z e n yon
W*(X).
Sind
al,
:= Ga I + ... + Gap + I
a, b, c , e v e n t u e l l Die M e n g e
...,
ap, bl,
..., bq
a := a I # ...
# ap
a I # ... ~ ap
# b I # ...
Jeder
T e r m yon
#
assoziativ
und
wird
mit
die D e f i n i t i o n und der M e n g e n
der
als V a r i a b l e n bezeichnen
for E l e m e n t e
wit mit
H*.
p, q a I ,
~ bq
, so b e z e i c h n e n
wir den T e r m
a # b
ist auch T e r m yon
for Terme
X E X
Hauptterme,
# bq
und k o m m u t a t i v
W*(X)
Indizes, yon W*(X)
b := b I # ...
W*(X)
Die G l e i c h h e i t
mit
der H a u p t t e r m e
aus
W(X).
W*(X)
ist, wie
so e i n g e f O h r t ,
in
W(X).
Ebenso
x-Koeffizientenmengen
daG die O p e r a t i o n kann
K* a
for
a
des Terms
aus i
for
X
Obernehmen bei
A
erkl~rte az
weiterhin
und
B
Yea aus
endliche
Relation
nur e r k l ~ r t
Die R e l a t i o n
auf
in
[I] for
W(X)
gegeben
[I] die S c h r e i b w e i s e n
A <* b
Teilmengen
und
W*(X)
sein mGge,
<*
wie
sei wie
ist,
yon
W*(X)
sowie
x~
w e n n nicht folgt
(2.1)
{a, b} c H * A Sa < Sb ~ a
(2.2)
Sb = x i x <* (a, b)
(2.3)
a = (al,a2)
[a z1 ~ * b z1 ^ Kx~a l
(2)
a 1 = b 1 A a 2 <* b 2
(3)
~ b zI <. a zI ^ a ~* K xbl
und
<*
a ~* B , wo-
die n a c h s t e h e n d
a z , wobei
a = (a1,Sa)
Wit
allerdings
^ z < Sa < Sa I
gilt.
erkl~rt. b
+ ^ x := Sa = Sb ^ z := xal,b I A
^ b = (bl,b2)
(1)
und
werden.
U {a2}
<* b
V
V U {b2}
]
a <* b
(2.4)
{ao,...,ap,b ~ .... ,bq} c H * A (I) Vi(i
E p + I A j E p + I A i + j ^ fi = fj m a i <* bfi)
(p = q ~ Sk(k = a ~ ~ ...
Bemerkungen: we(x)
und
+ I) - (p + I) A
E p + I = a i ~* bfi ) A
(2) V i V j ( i (3)
p + q > O ^ ~f(f:(p
E p + I A a k <* bfk))
~ ap
I. Zu ( 2 . 3 ) :
<* b ~ # . . . Sind
2. Zu (2.5):
Aus
Alternative
a T = b~
folgt dar,
~ bq
a = ( a l , a 2)
z = x: I ,b I j so sind
vollst~ndige
A
a zI
und
b~
und
at= b I ; daher
und es gilt
der
b = ( b l , b 2)
definiert
Terme yon
*
stellen
(I) bis
(3) eine
H.
Satz
1.
W*(X)
Bemerkung <
auf
sprechenden
<
dutch
<*
linear
3. Da die Definition yon
W(X)
Satz I u n d gen,
wird
Pfeiffer
269
geordnet.
<~
auf
Satzes ffir W(X). Dasselbe der Wohlordnungsbeweis
einfacher als die von
gilt ffir den Wohlordnungsbeweis.
ffir
(W~(X), <~)
in [I] bewiesenen S~tzen ffir W(X)
auf
W~(X)
ist, wird der Beweis yon Satz I einfacher als der des ent-
W~(X)
, weil
folgen aus den analo-
<*
die Beschr~nkung yon
ist.
Wie in [I] Lemma 7 zeigt man dutch Induktion fiber Ga Lemma I. Sb 2 ~ x ^ a ~* K*x bl U {b2} m a <* ( b l , b 2 )
2. Ordnungstreue
Einbettung yon
(W(X), <)
Um zu zeigen, dab das Teilsystem (W(X),<)
~hnlich in
(W*(X) ,
(W~(X),< t)
ist, definieren wir eine Funktion sie
in
r
ebenso stark wie
(W~(X),
abbildet.
(3.1)
x E X ~ ~ x := x
(3.2)
Sa 2 < y < Sa o ~ ~((ao,Y)
(3.3)
~[a I = (ao,Y) + n ^ Sa 2 < y < Suo] = ~(al,a2)
(3.4)
p > I ^ {al,
Dabei sei
a + n a
+ 0
+ n, a2)
:= (~(ao,Y)
..., ap} c H * ~ ~(a I # ... # ap)
ffir n E ~ :=
a
;
(W(X),<)
yon W(X) in W@(X) und zeigen, dab
+ n + I, r := (~ al, ~ a2) := ~ a I # ... # ~ ap
induktiv fiber n dutch a
+
( n
erkl~rt. Man zeigt leicht:induktiv,
*
1 )
dag ~
:=
( a
W(X)
+ n
)
# 0
injektiv in
W*(X) ab-
bildet. Dutch
Induktion
Lemma 2 . 1 .
fiber
Ga
Sa=
Scpa
2.
r
) c K~ r a
3.
y a -- y~cpa
4.
x a
+
5.
b
e
w(x)
c
man f f i r a l l e
r
a E W(X)
) U {O}
+
=
xcpa
-~(a = ( a o , Y )
Mit diesen Hilfsmitteln a,
beweist
+ n #% z < y < S a o )
,* r
(a z)
-- (cp a) z
ergibt sich durch Induktion fiber
Ga + Gb
ffir
270
H. P f e i f f e r
Satz
2.
a < b = r a <* r b
Wir ffihren den B e w e i s b = (bl,b2) (5.3)
und
hier nur
gebildet werden,
~(al,a2)
for den Fall
Sa = Sb =: x
= (~(ao,Y)
d.h.
ist,
daft
~ a
durch,
nach
a I = (ao,Y)
+ m + I, ~ a2)
und
da~
a = (al,a2)
(3.2) und
+ m
A
~(bl,b2)
r b
,
nach
x < y < Sa o ,
= (~ bl, ~ b2)
gilt.
+
Sei
z := x a l , b I
Wit s c h l i e ~ e n
9
Wegen
dutch
I. a < b
nach
[I],
^ Kxa 1 U {a2]
x < y
Induktion
Ya I U Yb I , die g r 6 ~ e r
=
als
Sa I
ist
x < z .
fiber die A n z a h l
x
sind.
Wegen
Z
(7.3-I):(a
n
der E l e m e n t e
x < y
ist
Z
Z
yon
n ~ I
< b I V a I = b I ^ Ga I < Gbl)
^
< b.
Aus a~ = b~ ^ Ga 1 < Gb 1 , d . h . b 1 = (al,z) A z < Sa 1 f o i g t e n t g e g e n der Voraussetzung, da$ ~ b nach (3.2) gebildet sein mfgte. Also gilt alZ < blZ
und nach
n
r
gilt
1.1.
Ist
Induktionsvoraussetzung
<* r
Nach Lemma
y ~ z , so gilt
a I + I
sein,
bezfglich
r
~ b
(~ bl) z = ~(ao,Y)
w~re
entgegen
Ga + Gb
oder
<* ( ~ bl)Z-
~ a I = ~ ( a o , Y ) + m <~ (r bl)Z.
, denn w ~ r e und
(I.V.)
2.5 ist
Dann
ist auch
+ m + I , dann m f ~ t e
der V o r a u s s e t z u n g
nach
(5.2)
gebildet.
1.2.
Ist
z < y , so gilt
Also ergibt (*)
sich in b e i d e n
(r
Ferner
~ (a~)
= (r
+ m + 1) z
F~llen
+ m § I) z <* (r bl)Z
gilt
nach
I.V.
~ (Kxal)
U {~ a2}
<* ~ b
und nach
Lemma 2 . 2
.
Kx(ep a 1 + 1) = K hauptung
liefert.
2.
~ a
Well
sind, 3.
ist
Gilt
I.V.
ep a 1 O {O] <* tp b , was m i t
und
~ b
a 1 + b 1 , also a < b
nach
~ a ~* ~(Kxbl)
a at K~
nicht kann
(7.3-3) O [~ b2]
b 1 O {~ b 2 ]
nach derselben
aus
a < b
nicht
[1],
so i s t
(*)nach
(2.3-1)
Bildungsregel nach
(7.3-2)
die
Be-
entstanden aus
a ~ Kxb 1 U {b2]
[1]
gelten.
und nach
, u n d w e g e n Lemma 2 . 2
<~ ~ b .
Literatur: [1] P f e i f f e r , H . :
Archiv
f. matn.
L o g i k u. Grundl.
1__3_3, 74 - 90
(1970)
AN UPPER BOUND FOR THE PROVABILITY OF TRANSFINITE INDUCTION IN SYSTEMS WITH N-TIMES ITERATED INDUCTIVE DEFINITIONS Dedicated to Kurt Sch~tte on occasion of his 65 th birthday. W. Pohlers Several people conjectured the ordinal of the system ID~ (i.e. the intultionistic theory of N-times iterated inductive definitions) to be -@)S~N+10 in the notation of [I][2], c.f. [5]. In this paper we will show that in fact this ordinal is an upper~ bound for IDN( i.e. the classical version) a ~ therefore also for IDa. In a paper in preparation we will show that vice versa all instances of transfinite induction up to ordi@
nals < @~2N+I0 are provable in IDa. The result of this paper here is es~ablished by a technically rather complicated method. We first introduce systems H N and prove their consistency using transfinite induction up to our ordinal by a combination
of methods due to Gentzen and Takeuti
(the
complicated part). H N corresponds to a system of 2nd order arithmetic I with N-times iterated HI-comprehension with no set-parameters in the comprehension formula and without induction. However H~j, i.e. H N + axiom of complete induction,
is relatively consistent to H N what is proved in
IIl by an inner model argument (following the ideas in [9]). The inter-" pretation of ID N in HNJ' done in IV, is a trivial consequence of 4the fact that an inductive definition corresponds to one application o~ H$-com~rehension. So consistency of ID N is proved by transfinite induction up to ~N+I0,
whence it follows that this induction is not provable in ID N.
Some people asked me to write this paper in English. As I have not much practice in doing so and had no chance to get the paper revised by somebody who does, I want to apologize for all the grammatical- and orthographical mistakes probably occuring in the typescript.
I. The systems H N I. Syntactical notations 1.1. Basic symbols are variables for individuals (denoted by x,y,z,...) and sets ( denoted by X,Y,Z,...), sive functions and -predicates
constants representing primitive-recur-
(among them 0 for zero, S for successor,
= for equality) and the logical symbols -~,V . 1.2. A n-place notational-form ~ is a string of symbols containing besides basic symbols at most the symbols * 1 ' ' ' " * n " strings ~[Ul,...
If Ul,...,u n are symbol-
Un] is obtained from U by substituting Ul,...,u n for
272
W. Pohlers
"I" " " "'*n respectively. I .3. Terms and formulae are defined as usual. A formula is called atomic if it is X(t) or R(tl,...,tn) with X a set-variable or R a n-place predicate constant. 1.4. By n w e
denote the numeral
S(S(...(S(O))...)). A term is called n-tlmes
numerical, if it does not contain any free variable. If all constants are interpreted in the intended manner each numerical term can be evaluated primitive-recursively. We call two numerical terms equivalent if their evaluations give the same value. Two formulae FI,F2 are called numerical equivalent if Fz is obtained from F2 by replacing numerical terms by equivalent terms. A formula containing neither free nor bound variablel is called numerical. The truth value of a numerical formula can be computed primitive-recursively~
We call a formula ~[x I, ...,x k] verifiable
(falsifi~ble) if for every sequence n I , . . . , ~ is a true (false) numerical formula. 1.5. We define ~F to be F - * 0 = I
of numerals U[n I , ...,~k ]
. In order not to have to deal with s t r u c
turals properties and difficulties arising by them
we introduce positive.
and negatlve-parts, due to SchGtte, which generalize the concept of anteand succedens in usual sequent calculus. To make this precise we define posltive- and negative-forms 1.5.1. *z is a posltive-form
as l-place notational-forms inductively by (p-form)
I .5.2. For each formula F ('i-*F) is a negative-form
(n-form)
1.5.3. If G is a p-form (n-form) (F-*G) and (~G-*F) are p-forms
(n-forms)
Every antecedent A~ of a sequent AI,..~ ~ is a n-part of the interpreting formtLla A~-*...-*A ~ B 4 ..~.~ B~. 0theTwlse if a negatlo~ appears in the antecedent the negated f6rmula is'"already a p-part of the interpreting formula. So for instance an inference 'negation left' is superflous etc. As syntactical variables we use ~ for p- and ~ for n-forms. A negative~positlve-form is a 2-place notatlonal-form of the shape ('i-~['2]), ( ~ - * ~ [ * 2 ] ) or (~['2]-*~). We denote n-p-forms by ~. A formula A is said to be a positive-part (p-part) or negative-part (n-part) of a formula F if there is a p-form $ resp. n-form ~ s.t. F is ~[A] or ~[A] (i.e. a ppart of F is always understood together with its place in F. We refer to p-parts and n-parts as parts. A part is called minimal if it does not contain any part except itself. By F = ~ t r G ( G is a structural consequence of F) we mean that every minimal part of F is a minimal part of G. A formula whose minimal parts are all atomic is said to be primitive. 1.6. By a n-place predicate
( n > 0 ) we understand a n-place notational-
form ~oB.t. ~[Xl,...,Xn] is a formula. For a n-place predicate
G and term~
tl,..,,t n we define @(tl,...,tn) to be @[tl,...,tn]. So if S[X] is a
2'73
W. Pohlers
formula U[@] is the formula where each denoted occurrence of X(t) is replaced by @(t) ioeo @[t]. I .7. A formula not containing bound set-variables is called arithmetical. I A formula VX~[X] with U[X] arithmetical is said to be H I . A formula F i s called arithmetical in a finite set of predicates
[@1,...,@n)if there is
an arithmetical formula ~[XI,...,X n] s.t. ~[@I,...,@ n] is F. We define collections T N of formulae inductively by I .7.1. T o contains all arithmetical formulae. I 1.7.2. T~+ I contains all formulae H I in (a finite subset of) T N. I .7.3. TN+ I contains all formulae arithmetical in ~N+I ~ The rank r(F) of a formula F is the least N s.t. F is in T N if such a N exists, otherwise it might oe ~. 2. Axioms and inference-schemata 2. I. Axioms 2.1 .I. Each formula ~[AI,Aa] with AI,Aa numerical-equivalent atomic formulae, is a logical axiom. 2.1.2. Each formula s =t-~@[~[s],U[t]]
, with ~[s]
atomic is an equality-
axiom. 2oi .3. Each formula ~[P]
(~[P]) where P is a verifiable
(falsifiable)
primitive formula is a mathematical axiom. 2.2. Inference-schemata (S~) ($2,o)
~[~A], ~[B] ~[~[x]]
II-
~[(A-~B)] ~[Yx~[x]] ~ if the eigenvariable x resp.X does
($2,1)
~[E[X]]
I-
@[VX~EX] ] J not occur in the conclusion
(s3,o) ~[t]-~[vx~[x] ] I-~[vx~[x] ] (S3,1)
~[@]-~[VX~[X]]~-~[VX~[X]]
~[X]is called the eigenformula of an
instance of (S3,1) ($4)
~[C], C-~F
I-
G
if ~ [ F ] = ~ str G. An instance of ($4) is said
to be a cut with cut-formula C. The underlined parts in the conclusions are called main-parts of the inferences. All main-parts are minimal. 2.3. Corresponding parts in an instance of an inference Let M be a minimal part in an instance J of an inference schema. We define the parts corresponding to M by 2.3.1. if J is a cut, each minimal part of t%~e premise which has the shape of M corresponds to M, 2.3.2. if J is an inference with main-part M, every minimal-part of the formulae denoted in the premises in our list ($I) - ($4) cogresponds to
M, 2.3.3. if J is no cut and M not the main-part of J M occurs in the pre-
274
W. Pohlers
mise(s) of J and this occurrence is to correspond to M. 2.4. By a derivation
we understand a finite tree with axioms at its top-
nodes being locally correct with respect to ($1) - ($4). We denote by %1-F that F is the endformula of the derivation %. 2.5. The fibre of a formula F in a derivation is defined inductively by 2.5.1. Each minimal-part of F belongs to the fibre of F. 2.5.2.
If a minimal-part M belongs to the fibre,
each minimal-part cor-
responding to M belongs to the fibre. 2.6. A minimal-par t is said to be explicit in a derivation % if it belong to the fibre of the endformula,
it is said to be implicit in % if it be-
longs to the fibre of a cut-formula in %. We call an inference explicit (implicit) in % if its main-part is explicit
(implicit)
in %.
2.7. Inductive definition of the endpiece 5 of a derivation %. 2.7.1. The endformula of ~ belongs to 5. 2.7.2.
If the conclusion of an inference which is not implicit in % be-
longs to 5its premise(s) 2.7.3. The premise(s)
also shall belong to 5.
of an inference implicit in % do not belong to 5.
We say that an inference is on the boundary of % if it is implicit and its conclusion belongs to the endpieee. 2.8. A formula is called pure if it contains no free set-variable. dicate U is called pure,
A pre-
if ~[t] is pure or a formula X(t).
2.9. A derivation is said to be a derivation of H N if each main-part of an implicit inference
(S3,1) is pure and of rank ! N.
II. Consistency of H N I. We introduce two new inference schemata:
(S) U[X]
I- ~[@]
if X does not occur in the conclusion. An instance of
(S) is called a substitution with eigenvariable X. The corresponding part of a minimal-part EK@] in the conclusion of a substitution is the minimal9 [X] at the same place in the premise. (Str) F I- G if F = ~ str G called structural rule. If M is a minimal-part in the conclusion each minimal-part of the premise which has the same shape as M corresponds to M.
(Both schemata are actually superflous in
H N but they are needed to make the reduction-proceedure work). 2. A modified derivation is defined as a derivation but with respect to the schemata
($I) ..... ($4), (S) and (Str). In a m.d. a minimal-part may
belong to more than one fibre. We call it implicit if it only belongs to fibres of cut-formulae,
explicit otherwise. A m.d. is said to be a m.d.
of H N if moreover holds: 2.1. Each main-part of an implicit
($3,1) is pure and of rank ~ N.
W. Pohlers
275
2.2. All substituted predicates are pure. If M is an implicit minimalpart in the premise of a substitution then r(M)< N. 3. Ordinals We assume familiarity with the ordinal notational systems of [8]. Z*(N) is obtained from Z(N) by the restriction Sa ~S~+I add 2 o to the terms of Z*(N), denote ~
O
for terms (m,~). We
by o, (o,o) by I etc. First we
define some special ordinal functions and list their properties. 3.1. 2[a,n]is defined inductively as 3.1oi. 2[~,n] = a if S ~ < n . 3.1 .2. ~[a,n] : (~[~,n+1],2n) if n < S a . Proposition I.
a) ~ cZ*(N) =~
~[a,n] ~Z*(N)
for all n.
b) S(2[a,n]) = min[Sa,n}. c) k < n = ~ Ek(~[~,n] ) = ~ . d) ~ < ~ , K k ~ < 2 [ ~ , k ] for n ! k ! S ~
=~[~,n]
<~[8,n].
3.2. We use a ~ 8 as an abbreviation for m < 8 and K k ~ < 2 [ G . k ] For s = o we omit the subscript o.
for s < k < S a .
3.3. We define ~[s,a] inductively by 3.3.1. ~[o,~] = (o,~) 3.3.2. *[s+1,~] = (o,~[S,~]). Proposition 2.
a) ~ ~Z*(N) =~ b)
~[n,~] ~Z*(N) for all n.
S(~[~,a])=Sa.
c) ~(~[n,~]) d) m < n
= ~
=~[m,~]
e) m ~ n ,
for k<S~. <~[n,~].
~<m[m-n,~]
=~
m[n,~] <m[m,~].
f) ~ c ( 1 , 2 N) = O m [ n , a ] c(1,2N)
for all n.
4. Height of a formula in a m.d. The k-de~ree of a formula ~[XI,...,X n] with XI,...,X n the only set-variables oceurringfree in it, is the length of the formula ~[@,...,@] where @ is a predicate of length k. In the following we assume a fixed m.d. We choose k s.t. all predicates used in inferences
($3,1) and (S) are of
length < k. We omit k and write degree instead of k-degree *) 4.1. By the de~ree of a cut we understand the degree of its cut-formula. 4.2.
The height of a formula F in a m.d. is the maximal degree of all
cuts below F in the m.d. 4.3. The potentialdifference of a cut in a m.d. is the difference bet*) At this place I wish to thank Prof. Sch~tte for discovering an error in a first version of this paper.
276
W. Pohlers
ween the height of its premises and its conclusion. 5. Order o f a m.d. of H N To each node of a m.d. 9 we assign an ordinal of Z*(N) by induction on the length of the tree above the node. 5.1. Every axiom in % gets the ordinal QN" 5.2. If a and 6 are the ordinals of the premises of an inference
($I)
(o,a@ 6) is the ordinal of its conclusion. 5.3. If a is the ordinal of the premise of an inference
($2) or ($3, o)
(o,a) is the ordinal of its conclusion. 5.4. If a is the ordinal of the premise of an inference eigenformula of degree n, the ordinal of its
(S3,z) with an
conclusion is ~[n+1,a].
5.5. If a and 6 are the ordinals of the premises of a cut with potentialdifference n, the ordinal of its conclusion is w [ n , a @ 6]. 5.6. Let a be the ordinal of the premise F of a substitution in ~. Then there is a i < N s.t. r ( M ) < i i
a rank
for all implicit minimal-parts of F. We call
of the substitution and assign the ordinal ~ [ a , i ] @ ~ N to its
conclusion. 5.7. The conclusion of an inference
(Str) gets the same ordinal as its
premis e. The order o(~) of a m.d. is the ordinal of its endformula.
It is clear
that for every m.d. ~ o ( % ) ~ (I,~N) holds. Proposition 3. Suppose al,...,a m to be a string of ordinals in a m.d. with no substitution of degree < i between a I and a m. If there is a k < i
s.t. 61 ~ a I for
an ordinal 61 and 61, ...,6 m is the string built up analogously starting with 61 instead of a I 6 m ~ a m holds. The proof is by induction on m. W e just indicate the only remarkable case that the last inference is a substitution. By induction hypothesis we have 6m_ I ~ am_ I, hence by proposition I d) 6 m = ~ [ 6 m _ I , j ] ~ N Further it is K ~ m =
< ~[am-1" j ] # ~ N = a m
[~[6m_1,j]}<~[am, r] for J < r
~[am_1,r ] <~[~m,r] for k < r <
j. So ~ m %
since k < j
is provided
and K r 6 m = K r 6 m _ 1
<
am"
6. Reducible modified derivations A m.d. %I-F is ca~led reducible if % is cut-free or there are m.ds. ~il- F i satisfying
~.I. o(~i) ~o(~), 6.2. if there are cut-free derivations of each F i there is a cut-free derivation of F. I~a
I.
If E is a numerical formula and % a m.d. E ~ E
~ is reducible.
W. Pohlers
277
Fromm lemma I we get b y induction on ~[o(%),o] lemma 2. If a numerical formula is provable in H N it is cut-free provable. So we have THEOREM I H N is consistent established b y a transfinite induction up to ~[(1,2N),O] which is, by work of Buchholz and Sch~tte[2],
equal to @ ~
+10. If we succeed in pro-
ving lemma I elementary enough we may conclud~ by (a modification of) G8del's second theorem that transfinite induction up to @ ~ provable in ~ .
+i0 is not
(This result also can be obtained without t~e use of
G6dels theorem by methods used by Gentzen for first order arithmetic.) Until otherwise mentioned let % and E be as in lemma I. We may assume that all individual-variables
of % which are not used as eigenvariables
are replaced by ~. Proposition 4. If % contains an explicit inference in its endpiece ~ is reducible. Proof. As E is numerical it contains no quantifier. So there are at most explicit inferences
($I) in the endpiece. Suppose % to be
m[~A]
re[B] Y ~[(A-*B)]
where
6 E (A-*B) is a n-part of E. We construct %~ and %2 as
[,A]
and
~[B]
(Str) ~ A-*~[ (A-*B) ]
6"
6' and ~ c y
resp.
B-*~[(A~B)]
IA-*E As ~ y
(str)
hold we get by prop.3
6, c 6
B-*E and 6" c8.
If there are
cut-free derivations of I A-*E and B-~E we get a cut-free derivation of (A-*B)-~E by an inference
($I). Condition 6.2. is then established by
Proposition 5. If G = ~ str E and %1-G is a aut-free derivation then there is a cut-free derivation %'I-E.
278
W. Pohlers
The proof is obtained easily by induction on the length of %. Proposition 6. If % contains a cut-formula A-~B in its endpiece Proof. Assume % to be
it is reducible.
n2
J~ Y G 6 GI
8
G2
n2
J2 GO
nl
E where J2 is J1 if J1 has a potentialdifference ~ O. Otherwise J2 is the first cut with potentialdifference + 0 below J1. First we remark that from
(A_,B~_~ F
we get m.ds. IA-~F ~i \I/
We define %' as
and B ~ F
(str) ~l
__B F (Sir)
~ G-~A-~B
--~(Str
)
62 ~
i ~-~GI
G2
B-*G s
~ GI-~A'-~B '
G2
iG
AS-*Bt
n2
~i
n2
s
B-~GI
G2 GO
n2
n4
~2
A--~G~ G
9 63
B 9
A~-*Go
with ~ i ~
n3 nl
0
E We want to show ~' c~. It is ~' =~[n 3- n ~ , ~ @ ~2] < ~ [ n 2 - n l , 6 @ s] = ~ if ~I, S 2 < ~ [ n a - n 3 , 6 @ s ] . Since degree(K) < d e g r e e ( A - ~ B ) < n 2 it is n2-n 3 > 0. ~ =w[n2-n3,6~@s] <~[na-n3,6@e]because 6~ s<6~s by prop.3 forit is ~ ~Y. ~m =~[n4-n3, e~@ o2] < ~ [ n ~ - n 3 , 6 @ s ] if c~,o~ < ~ [ n ~ - n 4 , 6 @ s ] s i n c e n~-n4 > 0 because degree (B') < n~. From 6 i < 6 (i =2,3) we get o~ =~[n~-n4,6~+ ~] <~[n~-n4,6+ s] and c~ =~[n~-n4,63# e] <~[n~-n4,6~ s]. Moreover Kk~, =Kk6 ~ UKk6 ~ U ~ 6 3
U~ks so ~' ~ .
W. Pohlers
If B is 0 =I
the proof is similar
279
(but %' is somewhat
Proposition 7. If 9 contains a logical axiom in its endpiece
9
simpler).
is reducible.
Proof. Assume % to be
[A~, A2 ] q/
j
~[A]
A-~F Y G
6 E cut below Q[AI,A2]. If neither Az nor Aa is in the for the derivation of ~[A] and as well we fibre of A A is inessential
with J the uppermost may derive
C~
~[F] (Str)
G 8' E Trivially 8' cSj
hence ~ is reducible.
of A A is an implication
If Al or A2 belongs
atomic and obtained from Aa by some substitutions. shape ~' [AI*,A] where A and ~~ are numerical
6 or A is
Then ~[A] is of the
equivalent.
@'[Aa,F] = ~ s t r G we have AI*-+F ==>str G. From A construct ~' as AI -~F G
to the fibre
and hence ~ reducible by proposition Since
F we get A
P and we
(Str)
8' E
Clearly 6' c 8 .
The proof is similar if A-~F is below the axiom.
Proposition 8. If ~ contains a mathematical Proof. Assume % to be
axiom in its endpiece
9
is reducible.
W. Pohlers
280
~' [P' ]
P' -~F Y G 6 E
where
J is the u p p e r m o s t
cut below ~[P]
. If P is not in the fibre of P'
we do as in the same case of the proof of p r o p o s i t i o n
7. Assume
P to be
in the fibre of P'. By p r o p o s i t i o n 4 we m a y assume that there only are substitutions
between ~[P] and ~'[P'].
that P' is atomic.
Since no variable
By p r o p o s i t i o n
P' too does not contain free variables. merical atomic formula.
6 we m a y provide
of P can be used as eigenvariable So P' m a y be regarded as a nu-
If P' is false we construct
E' as
q/ ' [P' ]
and as
P' -*F
(str)
(str) 7P' -~G
P' -~G
8' ~P' -+E
6' P' -*E
if P' is true. milary).
In both cases 6' c 8 .
(The remainig
cases are treated si-
Condition 6.2. is established by
Proposition 9. Suppose E to be a quantifierfree derivation 9
I-P-~E
closed formula.
(~ I - 1 P - ~ E ) w i t h
formula then there is a cut-free
P a true
derivation
If there is a cut-free
(false)numerical
atomic
9 ~E.
The p r o o f is by induction on the length of %. The only interesting
cases are that P-~E is a logical-
All other cases are trivial tical axiom because ference because numerical PI as a
If P-~E is a l o g i c a l - a x i o m
to P what means
of an in-
Q[P, PI] Px is
that PI is true too. But E contains
positive part and is therefore a m a t h e m a t i c a l
is an e q u a l i t y axiom
for a m a t h e m a -
it is true and it can not be the main-part
it is atomic.
equivalent
or e q u a l i t y axiom.
since P can not be essential
s =t-~Q[~[s],~[t]]
axiom.
we distinguish
If P-~E
the subcases~
P occurs in ~. Then E is also an equality axiom. -
P is s = t .
Then U[s] and ~[t] are n u m e r i c a l
equivalent
and E is a lo-
gical axiom. -
P is ~[s].
If s = t
is true E[s]
and ~[t] are numerical
equivalent.
Hence ~[t] is true. Since it is a p-part of E, E is a m a t h e m a t i c a l If s = t
is false E is a m a t h e m a t i c a l
axiom,
because
s =t
axiom.
is a n-part of E.
W. Pohlers Proposition
281
1o.
If the endpiece
of 9 contains
an equality axiom ~ is reducible.
Proof. Again we may assume that no free individual variable piece of ~. So if s = t - ~ [ U [ s ~ [ t ] ] merical formula.
s = t is a nu-
If s = t is true we have a logical axiom and ~ i s
ble by proposition
7. If s = t
is reducible by proposition
is false we have a mathematical
reduci-
axiom and
8.
6. A part in the conclusion of an inference weakening part
occurs in the end-
is in the endpiece
if none of its minlmal-parts
($4) or (Str) is called a has a corresponding
part
in the premise(s).
A cut in a modo 9 is called suitable
of its cut-formula
contain either a weakening part or the main-part
an inference Proposition
if both fibres of
on the boundary of ~. 11.
A m. d. without axioms in its endpiece
contains
a suitable
cut.
The proof is obtained easily by induction on the number of cuts in the endpiece. Corollary. contains
a suitable
cut or is reducible.
Lemma I. (repeated) is reducible. Proof. By the corollary above we may assume that 9 contains
a suitable
I. Suppose that one of the fibres of the cut-formula
contains
part. Say 9 is of the shape ~i Fi A T --~r !
9[A]
A~F Y G 8 E.
But then we may derive as well
(i=1,2
resp. i = I
)
cut.
a weakening
282
W.
Pohlers
F. l
F'
(str)
G
8' E with 6' c8. The proof is analogous if the weakening part is above Z[A]. 2. Now we assume that both fibres contain an inference on the boundary. If the cut-formula is an implication we are done by proposition 6. So we only have to consider the cases where the cut-formula is Vx~[x] or V xu Ix ] . 2.1. Suppose the cut-formula is Vx~[x].
Then Z is
C~ 0
~o
~l ' [~l IX]] (Z1
VxU2 [x ] -*F~
~ ' [vxU~ [x] ]
Vx~[x]-'F J1
n2
7 5 GI
G2
n2
Ja Go
nl
E
where J1 and J2 are defined as in the proof of proposition 6. We construct a m.d. ~' as pointed out on the next page. We only have to show that ~' c b holds.
Since s o c u l
and ~o C ~i we get 6 1 Q 8
and 82 c 6 by
proposition 3. It is ~'= ~[n3-ns,~1@ ~ 2 ] < ~ [ n 2 - n 1 , 6 5 e] if ~i and ~2 are smaller than ~[n2-n3,6@ e], because degree (~[t]')< n2 implies n2-n 3 > O. But it is ~i =~[n2-n 3,81~ s] < ~ [ n 2 - n 3,6~ e] and ~2 is equal to ~[n2-n3,62~ e] < ~ [ n e - n 3 , 6 ~ e]. Since Kk~' = Kk6z UKkS2 UKke we get
W. Pohlers
~83
(z
o S,'[UI It] ] 9
~
8o
(str)
[t]-~[
Vx~
[x]]
!
6'
B ~ ~ [t]-*~ [VxU[x] ]
[~xU [x ] ]
V x ~ [x]-*F
[t ] -~Vx~ [x] -~F
7 ~[t]-'G
I ~ [t ]'-*GI
G2
62 ~[t ]'-~i
n2
Om
n2
b2
U [t ]'-*0o
[t ]'-~G~
n3 nl
Go E 2.2. Suppose
the cut-formula is
VX~[X].
Then % is
6o
0
~i [~l [x ] ] Z~ [ V X ~ [X ] ]
V X~2 IX ] "~F~
n3
6 [V X~ [X ] ]
V X~ IX ] -~F
n2
$ G2
n2
Y G
6 GI v G
~o
n~
G' E
Suppose r ( V X ~ [ X ] )
=i+I. Then G'Is defined to be the premise of the
first substitution below G whose
rank
is < i, if such a substitution
exists. Otherwise G' is E. (The case where G' is between G and G~ is treated like the case we are going to show). By proposition 4 ~
may
assume that there are no inferences with maln-part below 61. So VXU2[X]
284
W. Pohlers
only could be a l t e r e d by a substitution. set-variable
occurs free in V X ~ 2 [ X ]
Since it is p r o v i d e d that no
we m a y conclude
that ~2 is ~. We now
construct ~'. 0
~J. [~:t Ix] ] (Str)
~I [x] ~ i [vx~ [x] ] C~W
Vx~ [x ] --~F
~[x] ~ $ [vx~ [x] ]
n4
7 G"~U IX]
n4 ~t
8'
G2
7 G1 "~U IX ]
n4
V !
~Oo~U[X]
n4
~r -~G, ~ [ x ]
J
~ , - . . . P~ [Q ]
[@ ] "~ VX~~ ] "~'F1
n4
~l t
Vx~ [x ] ---.G' --'-F~
n3
VX~ Ix ] -~TG' "~F
n2
C~
IVX~ IX ] ]
ne I G' -'G
n2
6" I G' "+GI
G2
n2
V t!
7G' -~G
W/
nl
o
IG' -~G' G !
E Before we start to show that ~' is a reductum of ~ (I) If m>n
We mention:
~ is a part of a m.d. and we replace the height n of F b y n the ordinal of F becomes G' c ~ .
The p r o o f is an easy i n d u c t i o n
on the length of the s u b d e r i v a t i o n
above
F. First we have to show that ~' again is a m.d. ference
of H N. ~' contains
(S3,1) which is not in ~. So 2.1. is fullfilled. Because
no inof
W. Pohlers
285
r(~[X]) < i %' is a m.d. of H N . It is an easy consideration that it is possible to assign to each substitution, different from J, the same rank as in %. Since all implicit minimal-parts of G' are provided to be of rank < i < N we may give the rank i to J. It remains to show ~"c ~. It is nl < n2 < n 3 < n 4. By (I) we get m~c ml, hence by proposition 3
(2) ~ ' ~ . It is ~i' = ~[n4-n3, P@ ~o'] and ~l =~[s+1,~ o] with s the degree of ~[X]. So s < n 4. If s < n 4 it is n 4 = n 3. Hence n4-n 3 < s + I . By (I) we have ~o 'c_ ~o < (O,~o). It is p = ~ [ ~ ' , i ] @ ~ N < (O,~o) since S ~ o = N . So we have ~i' <~l. For i < k it is ~ i ' = [~[~',i]}UEk~ o' < ~[~1,k]. Hence
(3) ~l'c ~i. i+I Every substitution between ~i' and ~" contains VXU[X] as implicit min~mal-part and therefore is of rank > i+I o By definition there is no substitution of rank _
(~) ~"~
i+1 ~
.
Suppose k < i+I and ~ e Eke". It is easy to realize that (in the set theoretical sense). Jf ~ ~ ' it is ~ e ~
KkG~' c Kk~L" holds < ~[~,k]. So
assume ~ e ~ ' = KkPU~ o' . If ~ eKk8 o' < ~ [ ~ , k ] < ~[~,k] we are done. For k < i it is KkP = Kk~' < G[~,k] by (3) and KiP = [~[~',i]] < ~[~,i] also by (3). So we have (~) ~ " < ~[~,~] f o r k < i + ~ . By (4) and (5) we get ~"c ~ and the proof of lemma I is completed. III. Consistency of HN} I. Let ~V be the l-place notational-form VX(X(O)-~Vy(X(y)-*X(Sy)) -~X(*I)). The axiom ~ of complete induction then is expressed by the formula V x ~ ( x ) . HN$ is H N with ~ as additional axiom. We shall show that if ~N is consistent H N ~ is consistent too. By F ~ we denote the formula F with all first order quantifiers restricted to J~. Proposition 1 2 . The following formulae are provable in H N for N > I . (I) ~ ( 0 ) (2) J~(t) -~ J~(St)
(3) s = t (4) ~ ( t I ~
(5) } d
Y(s)~N(t) ... -+~(tn) -~ ~/(f(tl,. ..,tn) ) for each n-place function constant
286
W. Pohlers
Proof. (I), (2) and (3) are trivial. (4) is proved in the usual way by induction on the definition of primitive-recursive functions. We shall give the formalized proof of (5) in order to show that ($3,1) is used in the right manner. For X a set-variable we define 8[X] to be q (~(*x)-+IX(*x)) i.e. ~ ( * z ) ^X(*z). ~[X] is an abbreviation for X(0) ^Vx(X(x)-+X(Sx)). In H N we get: I) Vy(~/(y)-+X(y)-+X(Sy))-+ ~(x)-+X(x)-~X(Sx). From I) and (2) we get 2) " -~ 8[X](x)-~8[X](Sx). Since we have 3) X(0)-+8[X](0) by (I), we get 4) B[X]~-+B[8[X]]. By 4) and the axiom 8[X](x)-~X(x) it follows 5) (B[~[X]]-*~[X](x))-+~[X]~-+X(x). From 5) we get 6) g~(x)-+ B[X]Mr-+X(x) with an inference (S3, z) whose main-part is~/(x). ~/(x) is pure and has rank I. From 6) we get ~ ~ provable in H N (N>I). Lemma 3. If U[x I ..... Xn] is a formula resp. ~[xl,...,Xn]a term with only the indicated individual variables occuring in it it holds
b) I-IN9 I-
=:>
HN
I- ~/~(XI )
"'~ "~d'['/(Y~s " ~[XI-''~ ~ n]
The proofs are straight foreward by induction on the length of ~[Xl, .... Xn] resp~ the length of a derivation of U[x I .... ,Xn]. Lemma 4. If H N is consistent so is H N ~ . THEOREM 2 H N ~ i s consistent.
IV. Embeddin 5 of ID N int O HN%. I. Derivation properties of HN~. Let ~ N be the collection of formulae arithmetical in a finite set of pure
formulae of rank
i N.
Propositon 13.
a) b) c)
HN~ I-~[F,F] i f F ~ N HN~ I- VxU[x]-+U[t] i f HN~ I- VXU[X]-+~[ @] i f
VxU[x] c ~ N. VXU[X] and ~[6] are i n ~-N"
Proof. I t is easy to prove a) by i n d u c t i o n on the l e n g t h of F. b) and c) are immediate consequences of a ) . Corollary.
d)
HN~ ~ ~X~/x(X(x) -*U [ x ] )
for U[x]G TN, i . e .
HN~-- (H;-CA)N ~
"
287
W. Pohlers
[ by (H~-CA)N ~ we mean the theory of N-times iterated HI comprehensiona x i o m n o t containing free set-variables with axiom of complete induction~
e)
IIN~ I- ~[0], U[x]-*U[Sx]
I-~[t]
i f U[O]
~.
Rema rk. In [6] it is proved that the ordinal of (HII-CA)~ i.e. H11 comprehension with axiom of complete induction is ~ 0 . By d) and our result below it follows that the ordinal of (HI-CA)~ is ~ 0 too. I conjecture that this does not hold for the theories with induction schema~ There the ordinal of (HI-CA) - might be ~ 0 whereas the ordinal of (]11-CA) seems to be @(@~(~+I ))0 i.e. the first ~-critical orinal of the ~-th number class collapsed to the 2nd number class. 2. Let IDN(~ I , ...,~ ) be the theory of N-tlmes iterated inductive definitions, i.e. first order arithmetic augmented by the set constants Qi and their 'defining axioms' Qi.1)
~i[Q1,...,Qi,x] -+ Qi(x)
Qi.2)
u176
~ i=1,..o,N,
-* Vy(Qi(y)-+~(y)) 3
where ~i is a i+1-place notational form, s.t.~i[X1,...,Xi,x] is an arithmetical formula satisfying the 'monotonicity condition' for X i. For details see [3] [12]. To each formula F o f the language of IDN (the union of all IDN(~ I,...,~N )) we define a formula ~ in the language of H N ~ by substituting for all occurrences of Qi the predicate ~i~WX (Vx (~i [~I "''' ~i-I' X, x] -* X (x))-+X ('i)) Proposition I4. If F is a formula in the language of IDN ~
eTN holds.
Proof. Clearly each formula in the language of IDN is arithmetical in QI"" "' QN so we only have to show that all formulae ~i(x) are pure and of rank i which is immediate by induction on i. THEOREM 3 ID N
I-
F
=~
HN~
I-
F@o
Proof. The proof is by induction on the length of a derivation of F. By propositions 13 and 14 all cases concerning first order arithmetic are clear. We are done if we succeed in proving Qi.1 )* and Qi.2)* in HN%. First we prove Qi.2)*. 1) Vx(~i[~ 1,...,~i_1,~,x]-~(x)) -+ Vx(~i[~ 1,...,~i_1,~,x]-~(x)) and 2)
S(y)-~(y)
are provable by prop.13 for ~(x) ~ N "
288
W. Pohlers
3) Vx(~ i [~I,''. ,~i-I ,~,x] - ~ (x)) - (Vx(~ i [~I " ' " ~ i-I ,~,x] -~(x) ) -~(y) ) - ~ (y) follows by I ) and 2). From 3) we get by an inference (S3,z) 4)
Vx(~i[ ~1,..,~i_1,~,x]-*~(x))-,~i(y)-~(y).
Since ~i(y ) is pure and of rank i this clause is permissible in HN}. From 4) Qi.2)* is immediate. Similar to 4) we get: 5) Vx(~i[~ I , .... ~i-I ,X,x] -+X(x)) -~Vy(~i(y ) -*X(y) ). By monotonieity condition (which is provable without using (S3,z)) we get 6) Vx(~i[~1,...,~i_1,X,x]-+X(x))-+~i[~1,...~i,y]-~i[~1,..., Hence
~i_1,X,y].
7)
~i[~I ..... ~i,Y]-+ Vx(~i[~ I .... ,~i_I,X,x]-+X(x))-+X(y) and from 7)
8)
~i[~1,...,~i,y]-+ ~i(y)
by an inference (S2,z)o
Conclusion. The consistency of ID N is provable in ID N + transfinite induction up to ~e~N+10., hence this induction is not provable in ID N ( and therefore also not provable in IDNZ, the intuitionistic version. ) This follows from Theorems 1,2 and 3 since besides transfinite induction (and the provability predicate of H N which is formalisable in ID N and even replaceable by a primitive recursive one) we only used finitary methods in our consistency proof. Remark: @ ~ N + I 0 is the least upper bound for the derivability of transfinite induction. In a paper in preparation we will show that induction up to any ordinal smaller than ~ N + I 0 is provable in IDiN which implies that the ordinals of the intuitionistical - and classical theories of iterated inductive definitions coincide. Lit e rature K1]
Buchholz~W.: Normalfunktionen und konstruktive Systeme yon Ordinalzahlen. This volume.
[2]
BuchholztW. Sch~tteIK.: Die Beziehungen zwischen den Ordinalzahlbezeichnungssytemen 7 und @ ( ~ ) . To appear in Archiv f~r mathemati~ sche Logik und Grundlagenforschung. Feferman, S. Forms/ theories for transfinite iterations of gernerallzed inductive definitions and some subsystems of analysis. Intuitionism and proof-theory. Proceedings of the summer conference at Buffalo N.Y. 1968. A.Kino, J.Myhill, R.E.Vesley (editors)Amsterdam-
[3]
London (North. Holland Publ.Co)197o.
W. Pohlers [4] [5]
[6]
[7] [8]
[9] [lo] [11] [12]
289
Gentzen~G. : Die Widerspruchsfreiheit der reinen Zahlentheorie. Mathematische Annalen 112 (1936) Martin-L8ftP. : Hauptsatz for the intuitionistic theory of iterated inductive definitions. Proceedings of the Second Scandinavian Logic Symposium. J.Fenstad (editor). Amsterdam-London (North Holl. Publ. Co) 1971 Pohlers~W.: Eine Grenze f~r die Herleitbarkeit der transfiniten InI duktion in einem schwachen [II Fragment der klassischen Analysis. Dissertation Mttnchen 1973 Schdtte,K.: Beweistheorieo Berlin-G~ttingen-Heidelberg (Springer) 196o SchGtte~K.: Ein konstruktives System yon Ordinalzahlen I u n d II. Archiv f~r mathematische Logik und Grundlagenforschung 11 (1968) und 12 (1969). Takeuti~G.: On a generalized logic calculus. The Japanese Journal of Mathematics 23 (1953) Takeuti~G.: On the fundamental conjecture of GLC V. The Journal of the Mathematical Society of Japan Io (1958) Takeuti~G.: Consistency proofs of subsystems of classical analysis. Annals of Mathematics 86 (1967) ZuckertJ. : Iterated inductive definitions, trees and ordinals. Metamathematical investigation of intuitionistlc arithmetic and analysis. A.S.Troelstra (editor). Berlin-Heldelberg-New York (Springer) 1973
COMMENTS
ON GENTZEN-TYPE PROCEDURES
AND THE CLASSICAL NOTION OF TRUTH Dedicated
to Kurt Sch~tte
on occasion
of his 65 t birthday. Dag Prawitz
Gentzen probably tion to Hilbert's cal work
thought
program.
(Gentzen 1934-35)
of his work as essentially
But it appears
that he also understood
tion to the general understanding argued elsewhere
(Prawitz
this latter aspect Hauptsatz
of Gentzen's
is formulated,
did himself,
of the structure
1965 and 1971)
a contribu-
from his first main logiit as a contribuof proofs.
I have
that the significance
work is better brought
not for the calculus
but for the system of natural
out when his
of sequents
deduction
of
as Gentzen
or equivalently
for a typed lambda calculus. However,
there is a third aspect
missed by this reformulation
of Gentzen's work which is
and which concerns
ence between the rules of his calculus the semantical
definition of truth.
with this third aspect
the close correspond-
of sequents
this aspect himself.
to stress the correspondence
between Gentzen-type
semantical
rules of truth was perhaps Beth
1957, and SchGtte completeness
has much contributed logic
(SchGtte
(see also Kreisel
to obtain very natural
theorem for first order logic.
Gentzen-procedures
The first logician inference
was utilized by Beth 1955, Hintikka 1956,
of
of Gentzen's work.
Gentzen never mentioned
The correspondence
and the clauses
The present paper is concerned
proofs
rules and 1958).
1955, Kanger of G~del's
In particular,
SchGtte
to the theme with which I am here concerned.
were used by him in connection with infinitary
1951) and higher
latter case the correspondence
order logic
(SchGtte
1960);
in the
mentioned was the subject of special
attention. I shall here try to give a systematic of this third aspect
of the kind of procedures
into logic by Gentzen. briefly mentioning
exposition
of some features
that were introduced
I shall start by treating first order logic,
also infinitary
propositional
logic,
and shall
D. Prawitz
then show how this treatment second order logic. in which Gentzen's natural
Among
can be extended
other
calculus
In contrast recursion
logical
Valuations
seml-formal
in some respects
1.
as the
truths.
and related notions of truth which
of formulas,
is by a
the rules of a formal
or
system may in general be understood as defining certain inductively.
An obvious way of approaching
notion of truth by the use of a Gentzen-like notion
to
I want to emphasize a sense
to the usual definition
over the complexity
predicates
things,
of sequents may be understood
system for generating
I.
291
of inductive valuation
the classical
procedure
is via the
as defined below.
Definitions 1.1
Bipartitions
of sentences.
I shall call bipartitions
of sentences written
is the set of all (individual) individual and
~
constants
an~ parameters
terms belong to
~
of as a set of true sentences
.
and that
~
(e,~,~)
over
O
~
belongs
to
of atomic
sentences.
9
or
, i.e.
A
sentences
(|
where
of
@
and
is said to be included to be an extension ~'~. A bipartition the inductive
~
over
~'
and
(symbolically
@
$
(|
~'
$
and
and
$
is meant
are sets the bi-
are the sets of atomic A bipartition ~ ~'
= (@,$,*) ~ ~)
if
~' or
= (~,@',~') ~
is said
@' ~ ~
and
is called a base when it is used as the base for
said to be atomically
Let
A E ~ 9
(in the given
~ , and atomic when
in a b i p a r t i t i o n
definition
over
is to be thought
|
of another bipartition.
for bipartitions may also be used about bases. sentences
sentences whose
@
is said to be consistent when
respectively.
of ~ '
and
is true in the bipartition when
By the atomic ~art o f a b i p a r t i t i o n partition
where
as a set of false sentences,
total when every sentence
language)
,
symbols,
is false in the bipartition when
A bipartition are disjoint,
and function
Intuitively,
and
and I shall say that a sentence A E ~
(8,@,~)
terms that can be built up from some
are two sets of sentences
individual
We shall be dealing with what
complete when
~ U ~
The notions A base
defined
(|
is
is the set of atomic
| .
be the set of all terms that can be formed from numerals
292
D. P r a w i t z
and s y m b o l s for p r i m i t i v e
recursive
set of all true s e n t e n c e s ences
t = u
where
is an e x a m p l e
u
belongs
and
complete.
W h e n there |
and
of a c o n s i s t e n t
more a t o m i c a l l y
terms
t
writing (~,!)
individual
parameters
or in some s e n t e n c e
.
Then,
~=
sent-
(8,~,~)
the a r i t h m e t i c a l
I may
suppress
of
base.
the set of
it simply as a pair
(~,~)
over
~
;
|
or
I
.
8s
A
,
or
is
terms that can be b u i l t up from the together with
the s y m b o l s
occurring
to be first
order
s , respectively.
the s e n t e n c e s
s e n t e n c e s u s i n g the l o g i c a l symbols
~
as a b i p a r t i t i o n
as, a 2 ...
In this section,
range
to
w h i c h m a y be w r i t t e n
the set of i n d i v i d u a l
A
be the
terms d e t e r m i n e d by a s e n t e n c e
understood
in
~
and atomic b i p a r t i t i o n w h i c h is f u r t h e r -
By the set of (individual) of sentences,
and let
the set of all f a l s e
It w i l l be called
may then also speak about
s
~
is no risk of c o n f u s i o n ,
in a b i p a r t i t i o n ,
by a set
functions,
t = u
are s u p p o s e d
constants
from some g i v e n language.
~, ^ ,
and
V
A
and
The letters
and p r e d i c a t e B
are to
over sentences. 1.2.
Inductive
b_~ a base (T, P)
~
valuations.
= (~,~,$)
, also w r i t t e n
generalized
The i n d u c t i v e
, sometimes written (T~, F~)
V~
valuation
V
induced
, is the b i p a r t i t i o n
, d e f i n e d by the f o l l o w i n g
simultaneous
induction:
(I)
@ c T
(2a)
If
A E T , then ~ A E F .
(2b)
If
A E F , then ~ A E T .
(3a)
If
A E T
and
(3b)
If
A E F
or
(4a)
If
At
E T
for each
t
in
|
, then
YxAx
E T .
(4b)
If
At E F
for some
t
in
8
, then
~xAx
E F .
A, B, and The
and
At
converses
by r e c u r s i o n
$ ~ ~
.
B E T , then B E F , then
are here s u p p o s e d of (2) - (4)
over the c o m p l e x i t y
A ^ B E T . A ^ B E F .
to be s e n t e n c e s
(used in the u s u a l of f o r m u l a s )
over
@
definition
are taken notice
. of truth of in
the f o l l o w i n g d e f i n i t i o n . 1.3.
Semi-valuations.
ing
($,,))
and)
the c o n v e r s e s
valuation.
if
V
V
is a s e m ~ v a l u a t i o n
is a b i p a r t i t i o n of clauses
(T, F)
over
satisfying
(2) - (4) in the d e f i n i t i o n
8
(contain-
(clause
(I)
of i n d u c t i v e
D. Prawitz
1.4.
!Total) valuations.
1960, a (total) valuation consistent b i p a r t i t i o n (4) in the definition 2.
293
Following
over a set
(T, F)
~
over
the terminology
~
satisfying
the clauses
(2) -
1.2. and their converses.
Remarks 2.1.
While
is defined
in the usual definition
simply as the complement
of truth by recursion
the consistency
is in a way the prototype
Hauptsatz when formulated for a classical The clauses be understood
in the definition
and
be "from formal
A ET
B
is in
and
A E T
infer
~A
E F" ~
etc.
and
~
valuations may also
and
of the fact that
A E T
or
clearly in many respects by Gentzen.
B E F
in the rule corresponding (e.g~
respectively.
the subformula property
initiated of the rules
of any cut rule. also some of the questions
one immediately
or the corresponding
system are typical for questions
this kind.
There are three such natural questions
about Gentzen system of (with w e l l - k n o w n
that I shall consider next in section 3: one concerns
the relationship
and semivaluations
between inductive valuations
(invertibility
a second one concerns valuations
of the Gentzen rules);
the consistency
of the inductive
(Gentzen's Hauptsatz);
(iii) a third one concerns
their completeness.
The inductive valuations
system may be understood depending
valuation,
in tree form)
The system is
semi-formal
2.2.
is
The system is of course semi-
to ask about the inductive valuations
(il)
A
in the base are decidable)
similar to the kind of calculi
As already remarked,
(1)
where
and the inference rules would
a derivation
A E F
Note in particular
and the absence
answers)
the expres-
of the system
With each true and false formula in an inductive
we can in an obvious way associate
wants
of sequentso
As axioms
A E T
of the infinitely many premisses
to (4a).
calculus
A E F .
~ , respectively,
(provided the sets
because
of the proof of a Gentzen
of inductive
we would thus have all expressions ~
The proof of
as the rules of a semi-formal system, where
sions are of the form
falsity
of truth, we have here to prove
the consistency and totality of inductive valuations.
in
of SchGtte
is defined as a (total and)
or the corresponding
either classically
on how we read the various
clauses
semi-formal
or intuitionistically in 1.2.
It is to be noted that the problem of proving an expression may not be the problem of finding or effectively
describing a configura-
294
D. P r a w i t z
tion that
is a d e r i v a t i o n
a derivation
or not.
but
to decide w h e t h e r
For instance,
we can easily and e f f e c t i v e l y such that
if
A
arithmetical
is true
base,
then
is to know w h e t h e r
T
if
describe
A
a tree
in the inductive ~
T
of
last
induced
A ~ T .
or not.
is
theorem,
ending w i t h
valuation
is a d e r i v a t i o n
is a d e r i v a t i o n
the c o n f i g u r a t i o n
is F e r m a t ' s
A E T
by the
The problem
See f u r t h e r
section
3.3 and II.3.b. 2.3.
While
tive v a l u a t i o n we accept
the e x t e n s i o n s
as sound,
tion induced
of the sets
may be open to d i s p u t e s the existence
by a base
of the
is an i m m e d i a t e
of g e n e r a l i z e d
inductive
definitions
of. t r a n s f i n i t e
induction
over the
to extend
the results
far as possible, bility first have
complexity
order less
case
just because
clause
The e x i s t e n c e
system.
containing
(~,~)
I. sequence (i.e.
is c o n s i d e r e d
Va
case
are
to yield
of all
footnote
Since order
I want
logic
as
use of the possi-
by an ordinary
induction
to be p o s s i b l e
in the clauses
in the
(2) - (4)
class)
in the case
in section
in h i g h e r where
to be s a t i s f i e d
Va+1)
and Vwl
Va
IV.
3.1.b.
rules
of
of a
semi-valuations
in ~S
logic)
V 0 = (~,i), Va
an induc-
to a t r a n s f i n i t e
the clauses
is then an i n d u c t i v e
to section
of the premiss
order
rise
for limit
(where
of its
4.
of course
by a p p l y i n g
a bipartition
the c o n c l u s i o n
of the p o s s i b l e
(and e.g.
V B, e < a .
containing
Since
be taken as inference
1.2 give
obtained
the premisses
in the third n u m b e r Cf.
viz.
The g e n e r a t i o n
of the kind
of
sions are added the u n i o n
matter.
V0, VI, V2,... , V , V +1,...
the e x t e n s i o n Va
I shall not make
of s e m i - v a l u a t i o n s
they cannot
In the g e n e r a l
tive d e f i n i t i o n
valua-
of the p r i n c i p l e
class.
to second
which happens
disjunctive,
(4b),
semi-formal
inductive
of the a c c e p t a n c e
or e q u i v a l e n t l y
the p r e m i s s e s
different
is s o m e t i m e s (3b) and
(one unique)
induction
in an induc-
than the c o n c l u s i o n I.
is a s l i g h t l y
clauses
transfinite
F
on what r e a s o n i n g
consequence
order logic
of formulas,
complexity
2.4. (@,i)
this
and
second n u m b e r
w h e n not n e c e s s a r y
to replace
over the
for first
T
depending
V~+ I
is
(2) - (4) to
and the concluordinals
is the first valuation.
a
is
ordinal
D. P r a w i t z
3.
Immediate 3.1.a.
results Inversion
by an a t o m i c base
principle.
the t r u t h and f a l s i t y the c o n c l u s i o n
by an i n d u c t i o n
of the f o r m u l a s .
of the c l a u s e s
A semi-valuation but
3.1.b.
valuation
over
.
Then
The p r o o f
of the c l a u s e s and
B E F
V = (T, F)
A E T
and
(I) - (4),
.
proof
hence
u
E T
of G e n t z e n ' s T h e n by 3 . 1 . a and
there are s h o r t e r
derivations
B
valuation
i n d u c e d by an
is total. over the c o m p l e x i t y
t E |
, either
of the i n d u c t i v e At
E T
or
c l a s s i c a l l y but not i n t u i t i o n i s t i c a l l y
, At
of f o r m u l a s .
i n d u c e d by a con-
A E ~ 9
The i n d u c t i v e
complete base
on a c l a s s i c a l u n d e r s t a n d i n g
t E 8
be a
in the i n d u c t i v e
from the i n v e r s i o n p r i n c i p l e
for some s e n t e n c e
Completeness.
G i v e n that for each
and hence
Y xAx E T
of f o r m u l a s and valuations:
At E F
, we can
that e i t h e r for each
or for some
t E e
, At
E F
and
E F .
This
situation
is a c l a s s i c a l
introduction
l o g i c a l matters:
of formal
while
to see that s o m e t h i n g bound
(@,V)
of the u s u a l
The proof goes by i n d u c t i o n
Frege's
Let
valuation
immediately
that b o t h
inspection
3.3.
to be an i n d u c t i v e
is c o n t a i n e d
part of
a prototype
of
atomically
is
in the p r e m i s s . )
The i n d u c t i v e
is o b t a i n e d
Assume
B E T
in
is c o n s i s t e n t .
and is as m e n t i o n e d Hauptsatz:
required
of s e m i - v a l u a t i o n
on an i n d u c t i o n over the c o m p l e x i t y
Consistency.
of
to such a v a l u a t i o n :
V
i n d u c e d by the atomic
3~2.
conclude
(The d e r i v a t i o n
of s e m i - v a l u a t i o n s .
|
The p r o o f d e p e n d s
sistent base
induced
over the d e r i v a t i o n s
does o b v i o u s l y not n e e d
it can be e x t e n d e d
Embedding
semi-valuation
valuation
of the d e f i n i t i o n
s h o r t e r t h a n the one a s s u m e d
valuation,
depends
An inductive
is a l s o a s e m i - v a l u a t i o n .
The p r o o f is i m m e d i a t e
of course
295
troversies
in a s e m i - f o r m a l
instance,
of r e a s o n i n g
sound.
For
example
in the proof above)
by c l a s s i c a l
c o m p l e t e but this r e a s o n i n g
that
a reduction
logical reasoning
in a f o r m a l
of the r e a s o n i n g
over what k i n d
of the sense in w h i c h
constitutes
only r u d i m e n t a r y is a d e r i v a t i o n
in the c o m p l e x i t y
t h i n g is a d e r i v a t i o n
illustration
systems
required
system,
of
is r e q u i r e d there is no
to see that some-
system with accompanying
con-
that is to be a c c e p t e d as
reasoning, ~(VxAx
we see
^ ~WxAx)
is not a c c e p t a b l e
(continuing
E T ~ if ~
the
is atomica~y
intuitionistically.
296
D. Prawitz
It was
one of G e n t z e n ' s
we get a b e t t e r truth
if we replace
finition vely.
A E F
A E T
understood
II, we can c o n s t r u c t i v e l y
A E T
or
conscious
A E F , although
or not)
of the classical and
of such e x p r e s s i o n s
In such a d i s j u n c t i v e l y
that
(whether
approximation
the e x p r e s s i o n s
1.2 by a sequence
in section
A E F
represent
of
in the de-
understood
sequence,
that
notion
disjuncti-
to be c o n s i d e r e d
the c l a s s i c a l
we can prove n e i t h e r
fact
A E T
nor
constructively. But from a s t r i c t l y
that
insights
constructive
the i n d u c t i v e
complete
base
constitute
of truth:
Each
such that
A E T
and only if
A
is false
simple
3.4.
V'
part
because
The inductive
complete
base
V = (T, F)
induced of 1.4.).
it is clear
total
and c o n s i s t e n t
that
V' c V
and
A E F
each model
M
if
determines
the form of the sup-
about
generation
of total
in total valuations: valuation
i n d u c e d by a
is a total valuation. over
~
is i d e n t i c a l
by the atomic The second
by the d e f i n i t i o n s
ation
M
has
of 3.4 is just a summary
the t e r m i n o l o g y
diately
in
notion
of course a m o d e l
of G e n t z e n - t y p e .
of s e m i - v a l u a t i o n s
total v a l u a t i o n
The first (using
is true
system
conclude
and a t o m i c a l l y
of the c l a s s i c a l
section by a result
and a t o m i c a l l y valuation
A
we must
determines
M; and conversely,
Total valuations.
each
inductive
if
semi-formal
and e m b e d d i n g
consistent more,
in
(T, F)
this r e p r e s e n t a t i o n
We may sum up this valuations
of view,
by a c o n s i s t e n t
a representation
if and only And
point
induced
such v a l u a t i o n
such a valuation. p o s e d l y most
classical
valuations
part
of
of v a l u a t i o n
to the
(e,T,F)
of 3.1.a, part
Further-
3.2 and 3.3
then follows
and i n d u c t i v e
, but then since both
.
V
and
immevaluV'
are
B
3.5.
Embedding
consistent total lemma
extending
Lemma. if 3.6. valuations
i.e.
~
and
Each and
to a total valuation. first
3.1.b,
to a c o n s i s t e n t
~'
be two bases with
V~ _c V~,
truth.
and truth
valuations.
to a c o n s i s t e n t
then the f o l l o w i n g and a t o m i c a l l y
com-
3.4.
, also
Lpgical
each e q u i v a l e n t
V',
the atomic base
Let
in total
can be extended
by a p p l y i n g
and f i n a l l y
~ _ c ~'
V
valuation
is i m m e d i a t e
plete base,
Then,
of s e m i - v a l u a t i o n s
semi-valuation
inductive This
V = V'
In view of the e q u i v a l e n c e
in models,
to logical
the same set of terms.
.
the f o l l o w i n g
truth:
three
between
conditions
total are
D. P r a w i t z
is true
(i)
A
consistent
and
complete
base
~=
(ii) (iii)
A
is true
in each total v a l u a t i o n
over
|
A
is false
The e q u i v a l e n c e equivalence
between
total v a l u a t i o n
4.
Generation
set
in view
~
in the i n d u c t i v e
between
(i) and
(ii) and
(iii)
of finding
a valuation
this n o t i o n
(i.e.,
them to
applications
and so on.
is the e x t e n s i o n
of
(Ti,Fi)
(Ti,F i)
in the case
disjunctive.
If we want
containing
(|
of (3.b)
can
a consistent
to g e n e r a t e , we have
apply
of clauses
I shall r e p r e s e n t of the form
and
by one a p p l i c a t i o n
the
(2) - (4) obtained
(Ti+1,Fi+ I)
of the
2.3), m a k i n g
(4.b) where
some arbi-
the c o n c l u s i o n
systematically
to c o n s i d e r
(T,F)~
of the sentences
successively
, F0 = ~
I to sect.
of pairs
I may write
of
are false
, then to the results
TO = ~
and
The c o n s t r u c t i o n
sequences
~
of f i n d i n g
the c o n v e r s e s
obtained
choices.
by an e x p r e s s i o n
the formulas
set
I.e. we form a sequence
possible i.e.
"
from 3.4 and the
in w h i c h
we may
(@,~)
where
(cf. footnote
choice
tions
@A
(~,~)
We first a p p l y
to
over
(and the fact that a
of some
such a s e m i - v a l u a t i o n
(To,Fo) , (TI,FI) , (T2,F2),...
trary
from 3.5
.
semi-valuation).
to the problem
containing
defining
clauses
(~A,~,~) "
semi-valuation
(ii) follows
are true and the formulas
To g e n e r a t e
from these
induced by any
of s e m i - v a l u a t i o n s
semi-valuation
in 1.2).
valuation
in no c o n s i s t e n t
of 3.5 be r e d u c e d
clauses
atomically
is a c o n s i s t e n t
The p r o b l e m some
297
instead
all these
may then be r e p r e s e n t e d the pairs r => A
in the sets
T
(T,F)
where and
is
all s e m i - v a l u a -
r
as a tree
by sequents, and
A
are
F , respectively.
e.g. E,F=>A,A
to denote
a sequent w h o s e
of the s e n t e n c e s is the sequence In the figures sentences. finite able
in
E
consisting below
I shall
E
@~d~
"
of the
is the sequence
sentences
is always
restrict myself
sets and shall
set
antecedent
and then by the ones
consider
to denote
in in
F A
consisting and whose
followed
a sequence
to the case when
only s e m i - v a l u a t i o n s
~
first
succedent
by
A .
of atomic and
~
are
over the d e n u m e r -
298
D. P r a w i t z 4.1.
and let
Definition. ~
valuation sequent
over
~
and
order)
the
(a)- and
(b)-rule
by w r i t i n g of the
whose
(b)-rules
immediately
(b)-rule.
applicable applicable,
below
If there
(a)-rule. F => A
upwards
obtained
F => a
with the
in
a sequent
~
any
(b)-rule, (b)-rule
is to be a top sequent
and
In other F => &
above
we apply
nor any
in
then we apply
the same form as
the sequent(s)
is no such
If n e i t h e r
semi-
by a l t e r n a t i v e l y
of an (a)-rule,
(a)-rules
F => A
the line
instead
(a)-rule
in the final
the
is
tree.
(b)-rules (b2)
~,A,F
=>
A
E,F
=>
4,mA
=>
F
~,A,A
"IA,F =>
(a3)
E,A
(b3) E,A,B,F
=> A
E,AAB,F
=> i
F
=>
=,A,A F
(a4)
=>
1~
=> E,B,A
E,AAB,A
(b4) E,At,r,YxAx E,YxAx,r
=> A
(taken in some fixed antecedent
in some
tl, t2,.., 4.2.
order)
sequent
The m i n i m a l T = UD. i I
a (minimal) be g e n e r a t e d V
is a
t
is to be the first
such that below
by
and
~
.
of
over of)
generated
occur in the
@
F0 => AO '
Let
containing
the tree.
containing
(~,~)
over
In (b4),
|
As is easily
is any s e m i - v a l u a t i o n V'
does not
of the s e m i - v a l u a t i o n
F = ~Ai
(the branch over
At
...
term in
the one to be constructed.
semi-valuations.
of a b r a n c h
semi-valuation
seml-valuations if
of (a4),
are to be all the terms
be the sequents and let
F => ~,At I,A 1~ => ~ , A t 2 , A r => E , Y x A x , A
=>
In an a p p l i c a t i o n
there
The
of sequents
the line has
(a2)
i.e.
of sentences
below as far as possible.
stage have
above
sets
~ U ~ .
the sentences
is c o n s t r u c t e d
sequent
by
is the tree
the top of the tree by the a p p l i c a t i o n the
be finite
(i.e. we take
which
if we at a certain
~
determined
~ = (|
as origin
fixed
applying words,
tree @ => ~
in some
Let
be the set of terms
e
tree
seen, (~,~)
Furthermore,
by the s e m i - v a l u a t i o n
=> AI' (@,~,~)
(T,F)
is then
"''
and is said to all m i n i m a l
are g e n e r a t e d containing
s
over
this way;
(~,i)
tree over
, then
(~,~,$)
299
D. P r a w i t z
such that
V' ~ V
4.3.
.
The s i m p l i f i e d
tree c o n s t r u c t e d
semi-valuation
tree
in the same w a y as in 4.1
over
except
(|
is the
that rule
(b4)
is
r e p l a c e d by: (b4') r => ~,Aa,A r => ~ x A x , where
a
is the first
A
parameter
ai
in
|
that does not
o c c u r in
s e q u e n t s b e l o w the one to be c o n s t r u c t e d . It is e a s i l y seen that if the s e m i - v a l u a t i o n V2
tree over
V1
(~,~,~)
g e n e r a t e d by the s i m p l i f i e d
such that
VI
cance
|
V2
branches
never
by some
for the c o n s t a n t s
of the s w i t c h to s i m p l i f i e d split in m o r e
g e n e r a t e d by
, then there is a s e m i - v a l u a t i o n
semi-valuation
is o b t a i n e d f r o m
tion of terms in
is a s e m i - v a l u e d
as,
trees
tree
over
(|
(simultaneous) a2,
...
The s i g n i f i -
is of course
than two b r a n c h e s
substitu-
that their
at one and the same
point. 4.4.
Consistent
semi-valuations.
It is e a s i l y
the s e m i - v a l u a t i o n g e n e r a t e d by the b r a n c h valuation where
tree is i n c o n s i s t e n t ,
some a t o m i c
succedent.
sentence
semi-valuations.
semi-valuation truncated
closed.
semi-valuation
ends in a s e q u e n t s
If all the b r a n c h e s
4.5. over
said above
for f i n i t e
only if the t r u n c a t e d
to g e n e r a t e
of the
exists
sets
@
(simplified)
and
A branch
of a t r u n c a t e d
tree are closed,
also
and f r o m w h a t
conclude:
a consistent ~
(simplified)
I shall speak about a
W i t h this t e r m i n o l o g y
There
and the
in q u e s t i o n is said to be
of a t r u n c a t e d
in this s e c t i o n we may
C l o s e d trees.
(e,~,~)
tree.
of the k i n d
the tree is said to be closed. has b e e n
if we only w a n t
in this way,
semi-
a sequent
of a b r a n c h w h e n we
W h e n the c o n s t r u c t i o n
tree is m o d i f i e d
(simplified)
tree w h i c h
of this kind,
contains
in the a n t e c e d e n t
We can b r e a k off the c o n s t r u c t i o n
have r e a c h e d a s e q u e n t consistent
t h e n the b r a n c h
occurs b o t h
seen that if
of a ( s i m p l i f i e d )
where
semi-valuation
semi-valuation
e = ~0~
if and
tree over
(~,~,*)
is not closed.
4.6. base
Semi-valuations
(~,~,~)
c o n s i s t e n t w i t h a base
I shall say that a s e m i - v a l u a t i o n
~.
Let
~
be a
tree is t r u n c a t e d
S00
D. P r a w i t z
with respect to
~
when
is m o d i f i e d by b r e a k i n g a sequent to
$
F => a
or some
the c o n s t r u c t i o n
is r e a c h e d w h e r e
either
belongs
to
~
said to be c l o s e d w i t h r e s p e c t
to
~
branch
A E a
is stopped
The g e n e r a l
Calculi
the f o r m u l a s
of f o r m u l a s , remarked
was
these
completeness
F
belongs tree is of e a c h
of s e q u ~ o t ~
in the i n f e r e n c e
important
sequents make
of the i n d u c t i v e
As a l r e a d y
to r e p r e s e n t
valuations
constructive
thus
or s e q u e n t s
inventions.
it p o s s i b l e
rules,
the
constructively
approximation
of the c l a s s i c a l
of truth.
Let a c o n d i t i o n or
in
in the d e r i v a t i o n s by s e q u e n c e s
and thus gives us a b e t t e r notion
A
w h e n the c o n s t r u c t i o n
side-formulas
one of G e n t z e n ' s
in 1.3.4,
classical
some
A semi-valuation
idea of s i d e - f o r m u l a s
The idea to i n t r o d u c e replacing
.
tree
of a b r a n c h as soon as
in that way.
II.
I.
of the s e m i - v a l u a t i o n
off the c o n s t r u c t i o n
A E F .
on
V = (T, F)
in the d e f i n i t i o n I.I.2 clauses
(2) - ( 4 )
(2a')
If e i t h e r
then e i t h e r
~A
side-formulas
of i n d u c t i v e v a l u a t i o n s
A E T
can then be e f f e c t e d by r e p l a c i n g
the
in the f o l l o w i n g way:
~ F
A ~ T or
An so on for the other
V
on the scope
or
V
satisfies
the c o n d i t i o n
X
,
satisfies
clauses.
From a constructive depends
be a c o n d i t i o n of the form
The idea to i n t r o d u c e
point
of the
of view,
(informal)
the i m p o r t a n c e quantifier
of this change
in (4a').
This
clause n o w reads: (4a')
If for each
the c o n d i t i o n Given
•
that for each
n o w c o n c l u d e by YxAx
t E @
, then e i t h e r
(4b')
t E |
, either YxAx
At E T
E T
or
V
, either
At
E T
that for e a c h
E F ,
from which
f o l l o w s by
VxAx
E F
Furthermore
get
YxAx ^ ~ YxAx E F , which
when
the i n d u c t i v e v a l u a t i o n s were
t E 8
(4a')
or
V
satisfies
satisfies or
, either
that e i t h e r
by (2a') and two a p p l i c a t i o n s
At At
E F E T
~xAx of
X
E T
, we can or or
(3b'), we
could not be c o n c l u d e d c o n s t r u c t i v e l y d e f i n e d as in s e c t i o n I (cf.I.3.3
and 1.3.4).
2.
The n o t a t i o n
of s e q u e n t s and the i n f i n i t e
The new c l a u s e s
calculi
(2') - (4') o b t a i n e d by i n t r o d u c i n g
side-formula8
D. P r a w i t z
301
in the d e f i n i t i o n of i n d u c t i v e v a l u a t i o n s rules
in a s e m i - f o r m a l
read c o n s t r u c t i v e l y , occurring
is p o i n t l e s s
elements
of this s e m i - f o r m a l
understood
sentences
A
and the s e n t e n c e s
occurring
in a s e q u e n c e
as in s e c t i o n
ence
F
~=
1.4.
But
is false (e,~,~)
(R I) of
~
of s i d e - f o r m u l a s
B
A E F
i.e.
understood
obtains.
of the form
in c o n t r a s t
S
A 6 F
the r u l e s
in
A
the
in a seof the form
as a s e q u e n t
to the s i t u a t i o n there,
(with a c l a s s i c a l
sequences as a s s e r t -
If we c o l l e c t
o c c u r r i n g in c o n d i t i o n s
or some s e n t e n c e
F => A this
"or"):
e i t h e r some sentI is true.
of the s e m i - f o r m a l
system with a
become:
For all
or some
of c o n d i t i o n s ,
and
A , we can r e p r e s e n t
In this n o t a t i o n , base
sequences A E T
in c o n d i t i o n s
is now to be r e a d
in
the "or"
disjunction;
s y s t e m can now be taken to
one of these c o n d i t i o n s
quence
sequent
tension between
( 2 ' ) - (4') and c o n s t r u c t i v e
of the form
ing that at least
B E T
a strange
definition
(see s e c t i o n 3 below).
be " d i s j u n c t i v e l y "
F
taken as i n f e r e n c e
point of v i e w the i n t r o d u c t i o n
The e x p r e s s i o n s
with
is b e t t e r
U n d e r s t o o d as an i n d u c t i v e
there a r i s e s
in the c l a u s e s
and f r o m a c l a s s i c a l
S
system.
F, A
B E A
such that e i t h e r some
is e l e m e n t
of
~
A 6 F
, the sequent
is e l e m e n t F => A
is to
be an axiom.
(R 2a)
r => ~,A
(R2b)
A, r =>
-I A , F => A (R 3a)
F => A,A
F => &, -IA
F => A, B
(R3b)
A, B, F => A
F => A, A A B
(R ~a)
F => A,At I
AAB,
F => A, At~
(R 4 b )
...
At,
F => A, Y x A x
to d i f f e r e n t
in the a n t e c e d e n t
the ones u s e d in 1.4.1
I.
Alternatively,
the i n f e r e n c e r u l e s s h o w n by $chGtte. main
idea m o r e
orderings
and succedent.
to operate
As seen,
the rules
that we pay
are the same as
semi-valuation
sequents
on c e r t a i n parts
of the
of s e q u e n t s
F => A
of the f o r m u l a s
of i n t r o d u c i n g
Since the n o t a t i o n
clearly,
the c o n v e n t i o n
or r e p e t i t i o n s
for c o n s t r u c t i n g
instead
F => A
YxAx,
The rules are to be u n d e r s t o o d w i t h no a t t e n t i o n
F => A
trees except
one may a l l o w sentences
as
seems to show the
I shall use this n o t a t i o n here.
302
D. Prawitz
that we there the risk these
prescribed
of a c e r t a i n
different
rules
coincide,
The calculi w i t h
the rules
calculus
of sequents
sequent
S
calculus
determined
is provable differences) 1951
3.1
F =7 A B
in
~
When
I shall write base
the infinite
= (| ~
infinite
in L o r e n z e n
between
the inductive
~S
.
is (except induction
the
The for some
introduced
1951.
calculi
A
belongs
valuations
and the
of sequents:
to
some
A in
r
belongs
to
F~
T~ .
is obtained
immediately
by an i n d u c t i o n
over the
of derivations.
Thus, calculus
from a s t r i c t l y
of sequents
inductive
valuation
3.2. sequents
induced
determined
above:
point ~
of view,
yields
new above
the
~.
The Hauptsatz for the ~
either
, or both
for some A E T@
by 1.3.2,
infinite
is a c c o r d i n g l y
~9 F => A , A
is e x c l u d e d
the infinite
nothing
and
calculus
a triviality
~& A, F =7 A ,
B E r , B E F@
and
A E F~ .
we have
of
or for
Since
the
again by the result
~& r => A .
3.3.
Restrictions
derivations
on the rule
in the infinite
are to be e f f e c t i v e l y complete
by
If
above,
B E A , B E T~
last p o s s i b i l i t y
by
by a c o n s i s t e n t
of view:
then by the result
classical
determined
The Hauptsatz.
from this point some
, I shall call
contained
if and only if either
This result length
system,
the system with
The e q u i v a l e n c e
or some
(For
the fact that
result
infinite ~
(R I)- (R4)
in this
and also
the formulas.
concerning
by the base
by the a r i t h m e t i c a l
by SchGtte
Classical
order b e t w e e n
see 6.3).
determined
notational
3.
a certain
oversimplification
calculus
described,
bases ~ = ( @ , ~ , ~ )
(R 4 a ) .
with
If we require
of sequents
we obtain
determined
for c o n s i s t e n t
denumerable
|
that
the
by
and atomically
and d e c i d a b l e
~
and
the result: If
A E T~
by the f o l l o w i n g (|
],[A})
scribed,
~
=7 A
observations.
truncated
The
semi-valuation
tree
over
with respect
to
~
can be e f f e c t i v e l y
and if closed w i t h r e s p e c t
to
~
, it is a d e r i v a t i o n
In the c a l c u l u s respect
, then
to
determined
~ ) is not
by
~
.
closed with
If the tree respect
to
(truncated @
deof
--~A
with
, the tree g e n e r a t e s
D. Prawitz a seml-valuation and
$
V'
with an atomic
are disjoint
atomically
complete
and
~'
base,
and
then
V'
, A
cannot
(using 1.3.2). of theorems
4.
and
effectively,
obtain two notions
" ~
is an
A
is false
is consistent
of the fact that the set that the derivations
are to
1959.
pointed
=> A"
the interest
out above:
definition
and
" ~
the equivalence
to left,
the questions in 1.3. have
3.1 holds
about inductive
The answers
T~
base)
F ~ , we
that constitute
than obtained
valuations
by
a
T ~ and
P~.
only from right
raised
in 1.2 and
anew for the infinite
(first given by SchGtte
is the arithmetical
and
constructively
to be considered
of the present
By the introduction
of
A =7"
of truth and falsity
answered ~
~
proved by Shoenfield
in the inductive
better approximation
when
~'
if ~
~ . Hence by 1.3.1.b 3ince
provided
point of view,
is the fact already
of side-formulas
of sequente.
V~
by requiring
first
such that Hence
results
From a constructive
Since
~'~
V' ~ V~ .
This gives a simple proof
Constructive
approach
(@,~',~')
are disjoint.
@'~
then be true in
is not affected
be described
part
~
and the lemma in the proof of 1.3.5, in
303
1951
follow the general
calculi
for the case pattern
of 1.3
with some slight modifications. 4.1. rules
Inversion
(R2),
miss(es). provable,
principle.
If the conclusion
(R3) and (R4b) is provable, If the conclusion
~xAx,
r => A
then there exists a sequence
Atl,r I => At; r',r => A,A,
At2,r 2 =7 A2;
...
can be obtained
r',r 2 =7 A2,A';
of any of the
then so is (are) of the rule
of provable
the pre(R4a)
such that a derivation
from derivations
of
is
sequents of
r',r I => AI,A';
...
The proof is immediate
by induction
over the length
of deriva-
tions. 4.2. ~
Hauptsatz.
The Hauptsatz result
yields
1.3.2 and follows
principle
~
r => A,A
and
~A,
F => A , then
and is a generalization like this result
now using induction
is a negation ~
If
r => A .
r => A ~
~ B , then if
A
~
directly
over the complexity B,r => A
is a conjunction
and
~
of the consistency from the inversion of formulas:
F =7 A,B
B ^ C, then
~
If
and hence
r =7 A,B
and
A
304 ~
D. P r a w i t z F => A,C
and
~
B,C,F => A
is a d e r i v e d
rule and)
~
If
F => A .
for every and ~
t
~Ai} i
A
is a u n i v e r s a l
and
~
...
Constructively, that either section
5.
A
that
~
A => A
are true
but
introduced
(i)
complete
,
F,F i => Ai,A;
If
~
is complete,
basis
indicated
in
Since
for
not to describe
the
truths of sequents valuation
induced by some
truths,
i.e.
valuations
induced
by c o n s i s t e n t
purpose,
different
the sentences
which
and
A,F
=> A,A
individual not
changes
section: about
the base
is its
are axioms.
terms now have
occurring
of a term
two obvious
(R I) by:
given a d e r i v a t i o n
is a p a r a m e t e r F => A,At
thing known
we replace
sequents
substitution
there are
(R I)- (R4) of the last
in the axioms,
t
for
for every
t .
(R4b) is now e q u i v a l e n t
(R4b')
then it holds
the logical
this latter
(RI') All
rule
~
for complete
But as a l r e a d y
Since the only positive
of
F => A,Bt
bases.
in the rules
a
prove
.
in the inductive
completeness,
(ii)
~
[ti]i,~Fi)i
F =~ A .
in g e n e r a l
his calculus
to generate
When we have to be made
thinning hypothesis
.
in all inductive
atomically
Hence
A =>
of the logical
and f a l s i t i e s ~
~
~
then
sequence
in 4.1.
also
that
we have:
The g e n e r a t i o n
truths
or
T e r t i u m non datur.
Gentzen base
and hence
VxBx,
for some
stated
we cannot
=> A
I above,
4.3. every
~
(by the fact
of the i n d u c t i o n
sentence
B t i , F i => A i
with the p r o p e r t y
F,F 2 = > A 2 , ~ ;
and hence
by two a p p l i c a t i o n s
in
of
the same
F => A,Aa r
or
status
, where
A , we can by
a , obtain d e r i v a t i o n s In other words,
to the finite
the infinite
rule:
F ~-> A,Aa F --~-> A,VxAx
where 6.
Completeness 6.1.
logically
PFoof true
a
does not
occur
of the calculus of c o m p l e t e n e s s .
if and only if
in
F
or
A .
of sequents The proof
~ => A
of the fact that
is now immediate
A
from the
is
D. Prawit z
equivalences simplified of
=> A
1.3.6 and 1.4.5 and the fact that a closed truncated
semi-valuation
tree over
(|
in the calculus of sequents.
sequents
305
in general
~ },~A))
is a derivation
The same proof holds for
if we define logical
truth for them in the
obvious way. 6.2.
Remarks.
predicate plete.
There is nothing
calculus which makes
In contrast,
sequents,
truth I.
this calculus generate
one expect
that the calculus
from the very construction
it is immediately
to logical
in the usual formulation
obvious
the first order logical
of a system intended to
truths. of construction
culus of sequents which makes the calculus
(i)
of
that it is complete with respect
is the natural formulation
to logical
is com-
of the calculus
One seems thus to be justified in saying that
To state in summary this principle respect
of
obviously
of the cal-
complete with
truths, we may recall the following
facts:
Studying an inductively defined notion of truth, we saw that the problem of finding a total valuation a sentence
A
is true is equivalent
sistent semi-valuation
in which
that the non-existence
of a consistent
which
A
is false,
A
is equivalent
in which
to finding a conis true, and hence semi-valuation
to logical
in
truth of
A
(section 1.3). (ii)
Furthermore,
we found a construction
semi-valuations consistent itself
that generates
all
in such a way that the non-existence
semi-valuation
in which
A
of
is false shows
in the fact that the construction becomes
closed
in a certain way (section 1.4). (iii)
Hence,
we just take these closed constructions
as deriva-
tions and are sure to derive exactly all the logical 6.3. calculus
I.
Additional of sequents
remark.
Although
is sufficient
truths.
this way of describing the
to account for its completeness,
This point has also been stressed by Kreisel and I am grate-
ful to him for much stimulating See also his contribution
of the theme treated h e r e . _
tc the present volume which became known
to me only after I had completed manuscript.
discussions
the present revision of my lecture
306
D. Prawitz
it does not explain why the inference the converses just the rules side-formulas
of inductive valuations
after the introduction after all,
existence
of certain semi-valuations, sentences
of those defining
the (a)-rules
in the construction
of the (a)-clauses
defining
of sequents;
valuations.
semi-valuations of these
are the
On the contrary,
of the semi-valuation
trees applied
satisfy the converses
the inductive valuations
coincide with the generalization side-formulas,
may here be appro-
the semi-valuations
the inductive
in order to make the generated
introducing
induced
does not simply depend on the defi-
fact that the clauses defining
converses
the truth
valuations
complete bases.
The colncidence mentioned
nitional
In other
(simplified)
but also as asserting
in all inductive
A warning against a certain oversimplification priate:
of
trees can be looked upon not only as stating the non-
and falsity of certain by atomically
are
the calculus
in the latter way.
it should be explained why the closed truncated
semi-valuation
i.e.
the semi-valuations,
and some obvious modifications;
of sequents was first described above words,
rules of the calculus,
of the rules for generating
(a)-clauses
i.e. with the (a)-rules
do not at all obtained by
of the calculus
instead they coincide with the g e n e r a l i z a t i o n
of the
(b)-clauses. It is to be recalled are interpreted valuation
that the sequents
trees.
When the upwards
tree is broken off because read downwards
in the calculus
as just the negation of the sequents construction
of inconsistencies
instead with the opposite
of sequents
used in the semi-
of the semi-valuation
in all branches and is
interpretation
of the se-
quents, we are in effect replacing
the semi-valuation
clauses by the
transpositions
replacing
and
by
" EF"
and
of their converses, " E T"
.
Thus,
wanted to explain depends (with the replacement
the coincidence
on the essential
just described)
defining the inductive valuations (b)-clause
Infinite
The notions and results sentential
in question
. ~ F" that we
fact that the transposition
of the converse of an (a)-clause
is identical
to the corresponding
and vice versa.
~!~
to infinite
" ~ T"
sentential parameters,
logic.
sentential
of sections
I and II are easily extended
Given an infinite
we consider sentences
of negation and conjunction
logic
~ Ai iEI
denumerable
set of
formed by the operations
of sentences
~Ai)iE I
with a
D. Prawitz
denumerable
index
set
I ; when
modifications
are necessary.
I.
valuations
Inductive Leaving
the clauses
The
has a h i g h e r
out the set of c o n s t a n t s
from the bases,
(3) - (4) in the d e f i n i t i o n
of i n d u c t i v e
we now replace
Ai 6 T
for all
i E I , then
~ Ai 6 T . iEI
(3b')
If
Ai E F
for some i E I , then
~ Ai A F . iEI
other n o t i o n s
of section
and the r e s u l t s
are then defined
in s e c t i o n
in the same way
1.3 then i m m e d i a t e l y
extend
to
case.
Generation
of s e m i - v a l u a t i o n s
In the c o n s t r u c t i o n for finite
I.I
some
v a l u a t i o n by
If
the present
conjunction
of s e m i - v a l u a t i o n
and u n i v e r s a l
trees,
we replace
quantification
the rules
by the rules
(b3')
(a3') ~,Aj,r, ~ A i => A iEI
r => ~,A 1,A
In the a p p l i c a t i o n s that
Aj does not occur
to be constructed,
At, A2,
...
trees,
containing
valuation
tree
over
semi-valuation determined
~
r => A,A I
.
below
are g e n e r a t e d
of a c o n s i s t e n t
to the t r u n c a t e d
closed.
the
of (b3'),
by semi-
semi-
But we now make no use of
trees.
as in s e c t i o n
we may now i n t r o d u c e
m i n e d by bases
sequent
such
[Ai)iE I .
semi-valuations
is e q u i v a l e n t being
j E I
by a base
For the same r e a s o n s way as there,
of
and the n o n - e x i s t e n c e
(~,~) (~,~)
of some
in the a p p l i c a t i o n s
I, all m i n i m a l
valuation
(R3a')
and
are to be all the s e n t e n c e s
As in section
Calculi
is to be the first
in the a n t e c e d e n t
sequent
simplified
Aj
of (a3'),
the s e m i - v a l u a t i o n
r => ~,A2,A
r => ~, ~ Ai,A iEI
~, ~ A i , F => A iEI
3.
cardinality,
(3a')
as there
2.
I
807
In place
of the rules
r => A,A 2 ...
r => A, ~ A. iEI i
II and in the same general
infinite
calculi
of sequents
(R3)-(R4), (R3b')
Aj,
deter-
we now have r => A
~ Ai,r=> iEI
A
308
D. Prawitz
where and
AI, A2, j
...
in (R3a') are to be all the sentences
in (R3b') is to belong
The results
Calculus
to generate
To generate
sentential
extend to
logic.
the logical t r u t h s
the logically
logic, we replace
~Ai}iE I
I
of section II.3 and II.4 now immediately
these calculi for infinite 4.
to
of
true sentences
the axioms in the calculi
in infinite
sentential
determined by a base
by the axioms
(El') as in first order logic but leave the other rules
(R2a),
(R3a'),
remains
(R2b),
infinite.
and (R3b')
as they are.
Except for notational
The calculus
differences,
thus
this calculus
is the one studied by Tait 1968. Since the rules of the calculus in the construction
that a closed truncated tion of logical
F => A
are the same as the ones used
of the semi-valuation semi-valuation
in the calculus,
truth is immediate
IV.
tree over
(r,A)
is a deriva-
the completeness with respect
in the same way as in section III.6
that we now make use of semi-valuation semi-valuation
trees and it is thus clear
trees instead
to (except
of simplified
trees).
Second
order valuations
and related notions
The notion of truth for second order sentences
that we could
hope to approach by extending the notions and results I and II is the notion of truth in generalized
of sections
second order models
in the sense of Henkin. Let a domain sequence be a sequence D0
is a non-empty
empty set of in
~
set of individuals
n-ary relations
of the descriptive
language,
in
~0
and "
constants
~
= ~0' ~1' ~ 2 '
~n'
n > 0 t
and parameters
individual variables variables
over
~n
"
order domain sequence
of formulas
range over
T
in
(~,I)
(~,I)
and the n-ary ~
to ~ n
(il,i2,...,in)
such that
I'
except for assigning
is like
I
by recur-
(n > 0)
predicate
is said to be a (normal)
if for each second order term
belongs
I
of a second order
G
~ x l x 2 . . . X n A ( X l , X 2 , . . . , X n ) and for each interpretation tion of
is a non-
in the usual way, letting the
~0
The sequence
where
Given an interpretation
we can then define the notion of truth in
sion over the complexity
"'~
' (i.e.
A(al,a2,...,an) ij
of the form I, the denota-
the set of all n-tuples
is true in to
second
aj
(9,I')
belongs
where
to ~ n )
9
D. P r a w i t z
It is truth second
in second
order d o m a i n
in all such m o d e l s
I.
sequence
(~,I)
and logical
where
~
is a (normal)
truth in the sense
that shall be c o n s i d e r e d
of truth
here.
Definitions 1.1.
Second
is u n d e r s t o o d 80' ~I' 82' that
order m o d e l s
309
order b i p a r t i t i o n s .
a triple
"'"
(8,q0,$)
such that
By a second
where
80
|
is a sequence
dicate
symbols
constants
of second to
The logical also
to
order)
80
'
constants,
over
case
, i.e.
are s u p p o s e d quantifier
order v a r i a b l e s
the term is atomic,
formula
at most
containing
case
the term is molecular.
and
G
is an
obtained
n-ary
second
by s u b s t i t u t i n g
and then,
If order
G
~
to use
if the term was m o l e c u l a r ,
or c o n s t a n t s and
now being a l l o w e d
A(G)
eliminating
(second
in
8n ,
is a second free,
is a second
then
as
of the form
..., x n
occurrences
u
to bind
n-ary
is either a symbol
x I, x2,
term,
two sets
4, ^, An
pre-
individual
parameters
A(Xl,X2,...,Xn)
for free
n-ary
are
whose
(or xn)~
~ x n A ( x n)
terms
and p a r a m e t e r s
and
or an e x p r e s s i o n
where
e
X
8 of ranges
Xxl,x2...XnA(Xl,X2,...,Xn ) over
@
sentences
n-ary predicate
the u n i v e r s a l
second
constants
n > 0 , is a set of
and where 8
sentences
term over a sequence
in w h i c h
for
and whose
8 n " second order
(n-ary)
Cn
and parameters,
order s e n t e n c e s
terms b e l o n g belong
and
of ranges
is the set of all i n d i v i d u a l
can be built up from some i n d i v i d u a l
and f u n c t i o n
order b i p a r t i t i o n
order
in w h i c h
order sentence
is the sentence of
Xn
the
in
A ( X n)
X-symbols
in
the usual way by conversion. The sequence of s e n t e n c e s such that and
en
s
~0
of ranges d e t e r m i n e d written
or
and the p r e d i c a t e
or in s e n t e n c e s
of
The t e r m i n o l o g y also used for second 1.2. order base
The ~=
(4) by
9s
bE a sentence
of the
terms n-ary
parameters
or by a set 80' ~I' 82'
determined predicate
and c o n s t a n t s
by
A
"'"
or
s
parameters occurring
in
A
s , respectively. introduced
for b i p a r t i t i o n s
in section
I.I.1
is
order b i p a r t i t i o n s .
(second
order)
(| | )
A
is the sequence
is the set of i n d i v i d u a l
is the set c o n s i s t i n g
P~, P~,...
clauses
~A
inductive
is defined with
valuation
as in I.I.2
the a d d i t i o n a l
induced by a second
(replacing
clauses:
|
in the
310
D. Prawitz
(5a) If
A(G) E T
for all n-ary terms
G
over
0, then
YX~A(X n) E T.
(5b) If
A(G) E F
for some n-ary terms G
over
|
yxnA(x n) 6 P.
1.3.
Quasi-valuations.
then
I shall also consider the quasi-valuation
induced by a second order base
~
, which is the pair
as in second order inductive valuations
except
(T,F)
defined
that the clauses
(5)
are replaced by: (5a') If
A(P)6 T
for all predicates
P
in
~n
'
then
yXnA(xn) 6 T.
(5b') If
A(P) 6 F
for some predicate
P
in
|
'
then
y X n A ( x n ) 6 F.
1.4.
Semi-valuations
order semi-valuations
and
order inductive valuations corresponding
(total)
in a way analogous (T,F)
over
|
We define
valuations
second
from second
to the one in which the
first order notions were defined.
sider semi-quasi-valuations verses
(total) valuations.
and second order
In addition,
we con-
that satisfy the con-
of the clauses defining the quasi-valuations,
i.e.
(2)- (4)
and (5'). 2.
Remarks
and further definitions
The notions
of truth and falsity in second
all the properties showing
defining the second
that these valuations
can be generated
the one obtained for first order logic thus have an approach similar
and the consistency
satisfy
in a way analogous
(result I.I.3.4),
for first order sentences.
we find that the inversion
Investiga-
principle
3.1.a
3.2 in section I are proved in the second order
valuations
3.1.b and in particular
valuations
3.3 now fail, also when classical
valuation
the completeness
of semi-
of inductive
reasoning
is accepted.
it is easily seen that no second order inductive
induced by an atomic base
showing e.g.
to
one would
case in the same way as in section I, but that the embedding
Indeed,
By
to the notion of truth of second order sentences
to the one established
ting this possibility,
order models
order total valuations.
that
is total;
VX I~ (xlt A ~ X ~ )
belongs
Any derivation to
T
or
F
already have to contain such a derivation as a proper part.
woul~ It can
be shown that only a quite special kind of second order sentences get a value in second order inductive that these valuations for second
valuations.
cannot be used to represent
order formulas
It is thus clear the notion of truth
that we are concerned with here.
In the case of quasi-valuatlons ever, all the basic results
and semi-quasi-valuations,
of section 1.3.
how-
- i.e. 3 . 1 - 3.3 - imme-
D. Prawitz diately
carry over without
valuation converse
change.
311
But it is clear that a quasi-
is not n e c e s s a r i l y a valuation of clause
since clause
(5a) in 1.1 will not be satisfied
(5b) and the in the general
case. We want the second order variables finable
to range
by second order terms as expressed
definition
of inductive
one hand, we cannot
valuations,
in the inductive
definition
valuations
of quanti-
because
the exten-
sions of the second order terms may depend on the meaning
of quanti-
fication
of the
(a fact formally reflected
inductive
valuations),
de-
(5) in the
but the dilemma is that on the
take this as an inductive
fication as attempted
over the relations
in the clauses
in the incompleteness
and on the other hand,
the quasi-valuations,
which are inductively defined and are total when the base is atomically complete,
give the variables
with our original
a range that is too small to accord
intention.
Any straightforward
extension
of the Gentzen-like
procedures
of first order logic to second order logic is therefore However,
by paying attention
be valuations,
to the quasi-valuations
we can get some solution
with in the end of section 1.3: tions and that of embedding In fact,
that happen to
of the two problems
that of generating
given semi-valuations
dealt
the total valuain total valuations.
given a total valuation V, it is easily seen that by an
appropriate
choice of an atomic base
quasi-valuation
induced by
~ .
also a given semi-valuation end, I make 2.1. and let
the following
be an G
~
,
V
can be embedded
Less easily,
can be embedded
in this way.
Let
~
n-tuples
To this
be a second order base
n-ary second order term over
relative
in the
it can be seen that
definitions.
Possible values. G
value of of all
excluded.
to
~
is a partition
(t1,T2,...,tn)
~
.
A possible
R = (RI,R 2)
of terms in
e0
(e,~,~)
of the set
such that
(1)
if
Gtlt2...t n E ~ , then
(tl,t2,...,t n) E R I ; and
(il)
if
Gtlt2...t n E ~ , then
(tl,t2,...,t n) E R 2 9
2.2.
By a representation
= (e,~,~)
is understood
for each molecular
n-ary
and for each possible unique contain
predicate
a sequence (n > 0)
value
symbol
of possible values relative
R
of
PG,R E e*n
|
= ~,
~* e2,
to
... such that
second order term
G
G
there is one
relative
to
and such that
Just these symbols and no others.
~
@~' |
over @
"'"
e ,
312
D. Prawitz 2.3.
Atomic
closure
of a bipartition.
Let
05
...
be a r e p r e s e n t a t i o n
* 82,
order b i p a r t i t i o n
(~,s
and let
|
of possible
relative
to
By the atomic
usin~
values
~*
is understood
i)
@0 = |
ii)
|
the atomic
and
|
(t 1,t 2 ,. .~ ~'
(using
A second
the atomic some
in
second
if the b i p a r t i t i o n ~
is closed
of
(~,v)
2.5.
order
and
where
written
~5 = (|
of possible
then already
order base
(|
over
of the form
as in 2.3,
in
~ = (8,%o,~)
where
V
, in other words,
atomic
values
occur
and
R = (R I,R2).
is closed
is a consistent
|
sentences
sentences
relative
05
of
to
~
,
| is said to be normal
is the q u a s i - v a l u a t i o n if
if
closure
is an atomic
induced
closure
and total,
term relative
symbol
that when
there to
68 , and
n-ary
P E 8n
a bipartition
is exactly term
~ G
~
one possible is then over
@
= (e,T,F) value
closed there
if and
Gtlt2...t n E T ,
then
Ptlt2...t n E T ; and
(ii)
if
Gtlt2...t n E F ,
then
Ptlt2...t n E F .
is consistent
possible
the atomic
closure
but not
total,
values
of a second
of
may a c c o r d i n g l y
~
there may be n o n - d e n u m e r -
order
term relative
have
record
the immediate
results
to
~
and
to be non-denumerable.
Results I first
n-ary
such that
if
~
of a
is an
(i)
ably many
~0
(tl,t2,...,t n) E R 2 ; |
~
It is to be noted
When
3.
in
of
only if to each m o l e c u l a r predicate
in
.
is consistent second
all atomic
order b i p a r t i t i o n part
such that
sentences
PG,Rtlt2"''tn
and all
where
of which must
An atomic by
~
representation
the symbols
all atomic
is the set containing
are the symbols
2.4. already
of
E R I ; and
PG,Rtlt2...t n PG,R
closure
~3' = (~',~0',~')
0 O*n ;
of the form
n)
that are not
here
= |
base
is the set containing
all sentences
lii)
~.
be a second
mentioned
above.
D. Prawit z
3.1.a and 3.2. results order"
Inversion
principle
3.1.a and 3.2 in section before
"inductive
3.1'-3.3'.
for q u a s i - v a l u a t i o n s .
as in section
3.1.b. valuation
Embedding
logic,
a semi-valuation
the second
order
trivially
V
by the 3.2
valuation,
(a)
Every
consistent
semi-
I fails
for second
order
e
if
to an i n d u c t i v e
base
V
that
can of course
(@,T,F)
V'
valuation
valuation .
Since
is and by 3.1.a
V'
since
induced
furthermore
V'
is a v a l u a t i o n
also in
by
is always
if
a semiI is consistent.
V
Total v a l u a t i o n s . The q u a s i - v a l u a t i o n base
i n d u c e d by a c o n s i s t e n t
is a total v a l u a t i o n
the d e f i n i t i o n (b)
over
in the i n d u c t i v e
non-atomic)
it follows
3.4.
in section
V = (T,F)
case be e x t e n d e d
is c o n s i s t e n t
- 3.3
to a v a l u a t i o n . 3.1.b
is i n c l u d e d
(possibly
V'
3.1
" i n d u c t i v e valuation" by by " s e m i - q u a s i - v a l u a t i o n " .
of s e m i - v a l u a t i o n s .
the result
The results
I.
can be e x t e n d e d
Although
The "second
and " s e m i - v a l u a t l"o n " .
in section I hold also when we r e p l a c e " q u a s i - v a l u a tlon " and " s e m i - v a l u a t i o n " Proofs
and consistency.
I hold also w h e n we insert
valuation"
Results
313
A closed
1.1
of inductive
quasi-valuation
quasi-valuation
if it s a t i s f i e s
and a complete
clause
(5b)
in
valuation.
is a total valuation.
Hence
the
induced by a normal
atomic
base
is a total
be a total v a l u a t i o n
over
~ .
Then
valuation. (c)
Let
V = (T,F)
included closure Hence,
in the q u a s i - v a l u a t i o n of
if
(~,T,F) (@,T,F)
quasi-valuation Proof. a consistent complete, tion
1.2 if it s a t i s f i e s
I. there
satisfy
3.1.b
complete
is also a s s e r t e d
is i n s u f f i c i e n t ;
i n d u c e d by an atomic
then
V
is i d e n t i c a l
(5b)
base
is itself
of clause
1960 but
to sect.
1.2.3.
of
to the (~,T,F).
i n d u c e d by
consistent
(5a)
and
in the d e f i n i -
in that definition.
by SchGtte
cf. f o o t n o t e
part
the q u a s i - v a l u a t i o n
the converse clause
V'
is
is also a total valuation.
i n d u c e d by the atomic
and 3.3'
and a t o m i c a l l y
V'
is closed,
V'
Since by 3.2'
it must
and
V
Hence, by
the p r o o f given
314
D. P r a w i t z
the d e f i n i t i o n
of total valuation,
To prove easily
Let
atomic base
Then
over
V = (T,F)
(@,@,$),
term over
~
assertion
(b), we note
proved by i n d u c t i o n
Lemma.
n-ary
the part
(a) follows.
the f o l l o w i n g
lemma,
P
be a p r e d i c a t e
such that for all
tl, t2,
induced
in
|
in
Gtlt2.~
n E T , then
P t l t 2 . . . t n E T , and
(ii)
if
Gtlt2.~
n E F , then
Ptlt2...t n E F .
for every
(ill)
if
A(G)
E T, then
A(P)
E T , and
(iv)
if
A(G)
E F, then
A(P)
E F .
To use
the lemma
quasi-valuation
there
P E |
lemma hold
(see s e c t i o n
inductive
because closed
The second
trivially
is a total v a l u a t i o n
V
Embedding
it follows
over
semi-valuation let
I.
3~
V = (T,F)
V'
3.5, we make use
second
order
logic
1960.
of
an
n-ary
in the c o n c l u s i o n 1.2 of
by part
case
(a).
of the next
from the first
one
that
is
(|
(|
and hence
V ~ V'
and that
V'
in total v a l u a t i o n s I.
can be extended
to a total valuation.
be a c o n s i s t e n t induced
1968.
by an atomic
(see V.2) with respect
lemma,
1967.
calculus
That
of
.
which
is proved
technique
used
the result
of sequents
to l o g i c a l
over
closure
V ~ V'
The e s s e n t i a l
in T a k a h a s h i
of the c u t - f r e e
semi-valuation
such that
of the f o l l o w i n g
is proved by P r a w i t z
the c o m p l e t e n e s s
by SchGtte
closure
is a total v a l u a t i o n
in the proof was also present implies
(iv)
~
(ii) of the
in the d e f i n i t i o n
that also
be the q u a s i - v a l u a t i o n
To prove
over
in (c) follows
of s e m i - v a l u a t i o n s
consistent
Then
is a closed
~ .
precisely, V'
G
V
(i) and
in (c) is only a special
Each
.
if
is a total v a l u a t i o n
assertion
More
(e,T,F)
term
Hence by (5a)
part is an atomic
by the first a s s e r t i o n
and let
that
V' ~ V; and by the a s s u m p t i o n
its atomic
3.5.
clause
Thus,
first a s s e r t i o n
3.5.
n-ary
2.5 above).
satisfies
valuation.
The result
(b) we note
such that the c o n d i t i o n s
V
an :
that
is to each
predicate
of the lemma,
A(X n)
G @O
if
it follows
by an
and
..., t n
(i)
to prove
is
A:
be the q u a s i - v a l u a t i o n
and let
which
for
truth was proved
D. Prawitz by induction over the length of Lemmao tion over
As in 3.5, let |
and let
by the atomic closure tion
8"
A :
V = (T,F)
be a consistent semi-valua-
V' = (T',F')
be the quasi-valuation induced
(~',~,$)
(8,T,F)
of
of possible values relative to
formula
A(XI,X2,...,Xn)
G I, G 2, ~
Gn
respectively, GI, G2,
315
,
n ~ 0 ,
using some representa-
(8,T,F)
~
of the same number of arguments as
and for all possible values
.~., G n
relative to
Then, for each
for all second order terms
(8,T,F)
XI, X2,
RI, R2,.o. , R n
..., X n,
of
it holds:
(i)
If
A ( G 1 , G 2 ~ . . , G n) E T, then
A(PGI,RI,PG2,R2,...,PGn,R n) E T';
(ii)
If
A(GI,G2,o..,G n) E F, then
A(PGI,R1,PG2,R2,...,PGn,Rn)E
When
n = 0, the lemma asserts that
To see that
V'
V ~ V' ~
is also a total valuation,
using the notation
of the lemma it suffices by (b) of 3.4 to show that closed, over
i.e. by the observation 2.5, that to each
8'
fled.
there is a
Let
R
occur in
m
(|
is
m-ary term
G
such that (i) and (li) of 2.5 is satis-
be the partition
([(t 1,t2,.,~ and let
P E |
F'.
Gtlt2...tm~ T'}, ((t I,t2, .... tm): G t l t 2 . . . t m E P'])
PGI,RI, PG2,R2,~
be the predicates from
8*
that
G , which therefore may be written G(PGI,RI'PG2,R2'~
Applying the lemma to
n)
G(Xi,X2,...,Xn)tlt2..~
A(XI,X2,~
n) , it follows that
G(GI,G2,~
n)
relative to
R
(8,T,F)
m
in place of
is a possible value to Hence, we can take the
predicate PG(G1,G2,...,Gn),R which belongs to 3~
8*m
and thus to
Logical truth.
~m
as the
P
required in 2.5~
By the results above, the following three
conditions are equivalent to logical truth in the sense of truth in all second order models:
316
D. P r a w i t z
(i)
A
is true
atomic
order base
(ii)
A
is true in all total
(iii)
A
is false
over
|
The e q u i v a l e n c e tional
fact
order
valuation second
struction
logic
(iii)
second
follows
follows
valuation
order
over
@A"
semi-valuation
from 3.4.c and 3.4.b. 3.5
(and the defini-
is a c o n s i s t e n t
semi-valuation).
of s e m i - v a l u a t i o n s (8,~,~)
@ U ~
,
from
sentences,
~
in s e c t i o n
1.4 is e x t e n d e d
In the d e f i n i t i o n
where
and where
@ ~
are now finite two rules
to
of a semi-
is now the sequence
and
we add the f o l l o w i n g
of ranges
sets of
for the con-
of the tree:
(a5)
E,
(ii)
straightforwardly.
over
by
order
order v a l u a t i o n s
of s e m i - v a l u a t i o n s
tree
determined
of (i) and
of (ii) and
The g e n e r a t i o n second
second
in no c o n s i s t e n t
that a total
Generation
i n d u c e d by any normal
(@A,@,~)
"
The e q u i v a l e n c e
4.
in the q u a s i - v a l u a t i o n
second
(b5)
VXnA(X n)
A(G), r,
VXnA(xn), r
~,
In a p p l i c a t i o n term over occur
~
=>
=> A
of (a5),
G
fixed
in some
of (b5),
order)
sequent
GI,G2,...
r
A
=>
~, A(G2),
E, y X n A ( x n ) ,
=>
is to be the first
(taken in some
In a p p l i c a t i o n over
r
a
in the a n t e c e d e n t
terms
r => ~ , A(GI),
n-ary
such that
below
a
A
second A(G)
order
does not
the one to be constructed.
are to be all
n-ary
second
order
~ .
As before,
all m i n i m a l
semi-valuations
are g e n e r a t e d
by such
trees. The s i m p l i f i e d 1.4.3,
semi-valuation
now also r e p l a c i n g
trees are defined
analogously
to
(b5) by
(b5 ' ) r => ~, A(P),
A
r => z, ~ X n A ( x n ) , in the a p p l i c a t i o n parameter
among
of w h i c h
PI' n p~,
one to be constructed.
..
P
a
is to be the first not
As before,
occurring
n-ary
predicate
in the sequents
the m i n i m a l
below
semi-valuations
can
the
D. Prawitz be obtained
from the ones generated
trees by substitutions, second
by simplified
now also substituting
the result
definition
second
Calculi
Infinite
and closed
of sequents
order case and the
trees need no change.
for second
order log!~
of the infinite
order logic
calculi
of sequents
is obtained by adding
(R5b)
F => A, A(GI)
F => A, A(G 2)
A(G),
r => A, ~XnA(X n) in (RSb)
G
are to be all the However,
is to be any n-ary
n-ary
term and in (R5a)
terms over the sequence
not only from a classical
II.3 still holds,
or false
~
=> A
or
in the second
upon classically). are never complete. (also when
~
A =>
Furthermore,
true sentences
point
valuation
also
~
complete
A => A
The finite
(looked
fails
in general
base). A, F = ,
would be provable
A,A
as axioms,
then
(as will be seen in
that are not logically
The finite
A
these valuations
in such a calculus
2.
of view,
induced by ~
IV.2,
but it seems that the set of provable
to warrant
...
in question.
The only sentences
the section below) interesting
GI~ G2,
point of view for which the
And as we saw in section
is an atomically
r => A
can hold are the ones that are true
order inductive
If we should add all the sequents all logically
of terms
but also from a constructive ~
F => A
yXnA(xn),
there seems to be no point in such calculi. for which
in section II
the two rules
(R5a)
result
order terms for
calculi
An extension to second
1.4.4 hold in the second
of truncated
V.
where
semi-valuation
order parameters.
Also
I.
317
sentences
true is not sufficiently
this kind of calculi.
calculus calculus
of sequents
for second
order logic is ob-
tained from the one for first order logic by adding rule the preceding section and the rule
(R5a') r => A, A(P) F => A, VXnA(X n)
(R5b)
of
318
D. P r a w i t z
where or
P
A .
is to be an The rules
truncated
the calculus. in the first is p r o v a b l e
predicate
of the c a l c u l u s
ones for c o n s t r u c t i n g a closed
n-ary
simplified
simplified
Hence
p a r a m e t e r not
are thus a g a i n
semi-valuation
semi-valuation
o c c u r r i n g in
identical
trees.
A
in the calculus.
is l o g i c a l l y
to the
In p a r t i c u l a r ,
tree is a d e r i v a t i o n
from IV.3.6 and IV.4, we i m m e d i a t e l y
order case:
r
in
o b t a i n as
true if and only if
=~ A
D. Prawitz Bibliographical
319
references
Beth 1955, E.W., Semantic entailment and formal derivability, Mededelingen der Kon. Nederlandes Akademie van Wetenschappen, Afd. letterkunde, n.s., 18, 309-542, Amsterdam. Gentzen 1934, Gerhard, Untersuchungen ~ber das logische Schliessen, Mathematische Zeitschrift, 39, 176-210. Hintlkka 1955, Jaakko, Form and content in quantification theory, Two papers in symbolic logic, Acta Philosophica ?ennica, no. 8, 7-55, Helsinki. Kanger,
1957, Stig,
Provability in logic, Stockholm.
Kreisel 1958, Georg, Review of Beth, La crise de la raison et la logique, J. Symbolic Logic, 23, 35-37. Lorenzen 1951, Paul, Algebraische und logische Untersuchungen Gber frei Verb~nde, J. Symbolic Logic, 16, 81-106. Prawitz 1965, Dag, Natural deduction, A proof-theoretical Stockholm.
study,
1968,
-
Hauptsatz for higher order logic, J. Symbolic Logic, 33, 452-457.
1971,
- Ideas and results in proof theory, in: Proceedings of the Second Scandinavian Logic Symposium (ed. J.E. Fenstad), 235-307, Amsterdam.
Sch~tte 1951, Kurt, Beweistheoretische Erfassung der unendlichen Indukticn in der Zahlentheorie, Mathematische Annalen 122, 369-389. 1956,
1960,
Shoenfield
-
-
Ein System des verknGpfenden Schliessens, Archiv fGr mathematische Logik und Grundlagenforschung, 2, 55-67. Syntactical and semantical properties of simple type theory, the Journal of Symbolic Logic, 25, 305-326.
1959, Joseph, On a restricted ~-rule, Bulletin de l~cademie Polonaise des Sciences, 7, 405-407.
Tait 1968, William, Normal derivability in classical logic, in: The syntax and semantics of Infinitary languages, Lecture notes in mathematics (ed. J. Barwise), 72, 204-236. Takahashi 1967, Moto-o, A proof of cut-elimlnation in simple type theory, Journal of the Mathematical Society of Japan, 19, 399-410.
BEMERIiUNGEN
ZU
B.
REGEL
UND
SCHEMA
Scarpellini
Einleitung In
dieser
f~lle
Arbeit
yon
(konsistente) zwei
mit
schaft
:
E)
t ein
Die
Frage
Ist
T zudem
T
(VX)
+
Wir und
setzen
durch
~ uns
der
An
: ~)
Fragen
haben
es
i)
allgemeine
als
deren Die
kann
2
wird +
S[tze
es
konsistent : 2)
~ B(x)
beantworten
durch
B(t)
.
?
Besitzt
?
wo
A(x)
sein, ) eine
befasse]
eine
positive
Frage
2)
zu
er-
konsistente,(eVtlo
?
nat~rlich
Teil
T ~
Theorien
Fall,
Primerweiterung
zum
)
fragen
zweckm~ssig
(yx)(A(x)
sind
gilt
intuitionistischen dem
B(x) Eigen-
so
Eige.schaften
) mit
A(x),
nur
die
sehr
zur
spezielle
Verf~gung
For-
stehende
wordeno Beispiele
nichttriviale Der Form
Grund
der
hat
gibt ist
S~tze
auch
Grundstock
igkeit
&
noch
B(x)
eine
zu (u.a.
ausser
ist,
die
den
S~tzen
2,
3 weggelassen,
versehiedene
Platzmangel yon
einem
zeigen,
dass
Formen
der,
dass
gewissen
yon
ob-
Refle-
es d o c h
Interesse
eher ist~
Anwendung.
Arbeit
gewissen
T
illustrierende
durehaus
man
T
folgender
gilt, ~
Sei und
mit
A(t)
Spezial-
lautet.
verf~gt,
(Vx)(A(x)
mit 1
und
T ~
so w i r d
disjunktive
- 3)
xionsprinzipien)o die
T +
Stellen
Besitzt
Terme
disjunktiven
Anhang
einigen
suggeriert
Wir
Ist
Methoden
folgt
Variablen,
den
ausschliesslich yon
zu b e w e i s e n d e n
Technik
f~r
) die
wie
konstante freien
Term,
B(x)
aufz~Lhlbare)
Die
~ber
: i)
beweistheoretischen typische
einziger
damn
Ausnahme
rekursiv
die als
mit
eine
intuitionistiseh,
werden
ist.
wir
denen
konstanter
lautet
Formel
wohl
x
(A(x)
(mit
men
yon
Theorie,
Formeln
Ist
behandeln
Fragen,
Resultate
an Hber
den
Zweck
zu
beweistheoretiseher die
Struktur
man,
Technik,
ausgehend mit
intuitionistischer
yon
relativer Theorien
einem Leicht-
erzielen
o 0hne
nern Formeln, dutch
Zweifel
. Hingegen n~mlich
lassen
sich
beschreiben die
Primer~veiterun~en
, dass
die
bier
sie
vermutlich
sich
erzwingen
die
bewiesenen eine
S~tze
disjunktiven
lassen
(Frage
stark
Eigent[imlichkeit Eigenschaften 3).
verallgemeipositivel schon
B.
I. S p e z i a l f ~ l l e , (a) Sei
Regel
und
in welchen
S ein k o n s i s t e n t e s , A, B zwei
dass
folgende
man
Sch~tte,[ o)
Dann
S
~
trivialerweise
S , A~B
Uber
Ferner
Funktionen
Funktionen. primitiv
Fall
S ~
seien
gibt
Sei n u n G d i e
lelchte
sich
Gehen wir
der Begriffe
aus
o) g e s n h l o s s e n
S klasslsch,
S ~A
Situation
und
so ist d i e
S ~B
weniger
(, d i e m i t
viele
ebenso
~i
folgen,
einfaoh,
wie
fol--
Zahlvariablen
' ~i
Konstanten
die
Formel
Insbesondere
zugehSri~en
' ... bezeich--
fLtr p r i m i t i v
re-
definierenden
fur n i c h t - p r i m l t i v - r e k u r s i v e Prim(T ) , d i e
ausdr~ckt:
ist darn% fttr j e d e n k o n s t a n t e n
~
ist
~h/nktor
~
Prim(F)
.
Formel
Rechnung
~
flndet
G ~ ~(~ ~ ) q P r l m ( ( )
ist also
ist,
die ~blichen
wen~
dass
(S~)(Prim(~)A-Prim(~))
offenbar
.
konsistent
in Z i + ~G
die
Rege~
gilt
~ (~)~Prlm(~)
,
z i + ~G
~
^P=i=(~)
aber yon der Theorie
Im Verlaufe
dieser
(3~)(~Prim(~)
Regel
zum
Schema
Uber,
Z.x + w G + G . Arbeit werden wir uns mit
Problem
besch~ftigen.
Gegeben
welches
d i e disju/iktive
ist
Eigenschaft
u n d hat t d a
Eigenschaften.
dlsjunktiven
Zi + nG
konsistente
.
man
Z i + IG
sofort~
2)
Ist
wUrden
Konstanten
Zi ~
~ibt
: Kann
?
verfUgt
(~$)qPrim(~)
Die Theorie
Fallp
B
ist
gen~gend
es eine
Zi
Harrop-Formel
(wegen
der
:
l)
Durch
oder klassisches es h H u f i g
Zahlentheorie t die nebst
vorhanden~
rekursiv.
F beweisbar
hat
Frage
ist d i e
aber keinerlei Dann
ist
zeigt.
Funktionsvariabeln
Glelchungen~
impliziert.
).
intuitionistische
net w e r d e n ) . kursive
, so
ja (, sonst
zu o)
Belspiel
auch
Schlussregel
konsistent
Im intuitionistischen
Zi die
Formeln.Dann
sich die naheliegende
Antwort
Sei
Schema
40 ) :
A
dass
im W i d e r s p r u c h
das
formales,intuitionistisches
zul~ssige
werden,
gendes
Regel
geschlossene
S S , pg.
wenn
stellt
die
Schema
System t
siehe
321
Scarpellini
er-
:
so ) .
so e r h a l t e n
Varianten
und
wit die
yon
ein intuitionistisches besitzt,
~ G eine Daraus
Formeln
in-
s
System A
~ B
S~ , fur
322
B.
welche
die Regel
dingungen
gilt
gefunden
: " wenn
werden,
ist u n d
die disjunktive
gewisse
einfache,
oretischen heit
mlt
] , ~S
3 und
Beispiel
IndlviduenDefinition
ist
Formel
2
Satz
1
: Seien ~(x
mit keinen
andern
schwach
positiv.
liebige
Terme
El)
eine
werden,dass
der beweisthe-
gewisse
Vertraut-
yon
positiv,
wenn
positiv,
A ~ V
, ~
M heisst
solchen
S kan~
enthalten.
schwach
vem~ge
Pr~dikatenkalkGl;
sie w e d e r
wenn
, V
7
sie aus
alleine
noch
aufgebaut
P r i m b a s i s t werLn sie n u r
enth~it.
Wir nennen
D
Primformeln ist. Prim-
M konslstent,
I ... x n k ) , B k ( X 1 ... X n k ) , k = 1 , 2 , . . . freien
Variabeln
als d e n
Sei M e i n e k o n s i s t e n t e
t I , ...
S~MvE
, tnk
gelte
endliche
~ ( t
S ~ ~ B k ( t
die
Es w i r d
F~lle
konslstent
ist.
ist E e i n e
I s t d a n n M'
soll gezeigt
triviale
intuitionistische
Formelmenge
konsistent
Es
es s o l l e n B e -
S + A mB
[7 ~vorausgesetzt.
heisst
heisst
Primformeln : Eine
;
dass
Pr~dikatenkalk~l
S der
t[nd N e g a t i o n e n
SuM
hat.
v~llig
L 6 ~oder
1 : Eine Formel
Eine
Definition formeln
nicht
und FunktionskonstaLnten
und negierten
wenn
garantieren,
Eigenschaft
aber doch
(b) D e r i n t u i t i o n i s t l s c h e
enth~it.
S ~ A ~ so S ~ B "
welche
Behandltuag zug~Lnglich sind.
~4
Im ersten
Scarpellini
angegebenen
Primbasis.
FUr
Formeln
| die ~'s
seien
Jedes k und be-
:
Primbasis,
I ...
t
) gilt
1 ...
t
) .
eine maximalkonsistente
sodass , d~n~
Primbasis,
gilt
auch
die M umfasst,
so ist
Theorie T = SuM,
u U (V ~)(Ak(~) ~Bk(~) ) k und hat die Ubllchen disjunktiven Eigenschaften.
konslstent Beweis:
Der Einfachheit
ger nk'S behandelt
man
halber analog.
theorie
intuitionlstischer
wlesen.
Wit
I ) S e i M'
nehmen
Primformel,
Sequenzenkalk~l
wie
SequenzenkalkUls
----~ p
~ falls
kUl nennen
wlr
; den Fall beliebi-
die Terminologie
fGr Details
Primbasis,
sel z.B.
auf
die M umfasst.
so p 9 M t o d e r ~ p e M ' . W i r
folgt:
schen
= 1 an
der Beweis[5]
ver-
in Schritten.
elne maximalkonsistente
Ist p e l n e
nk
Wir verwenden
Systeme;
fiihren d e n B e w e i s
wit
zu den Regeln
(GS) a d d i e r e n
wit
Gbertragen
und Axiomen fur jede
des
Wit bemerken: S~M
t in den
intuitlonisti-
Primformel
p das Axiom
p & Mttttnd ~p-----@ t f a l l s n p ~ M t . D e n r e s u l t l e r e n d e n G S M I . Zt~m K a l k ~ l
GSM'
addleren
wlr nun
alle Regeln
Kal-
B.
der
Scarpellini
Form
,
~(t),V WO
t eln b e l l e b i g e r
Formelfolge,
zu zelgen,
dass
den Theorien
Term
GST ~ Der
A
ist
Formel
bezelchnen
bzw. ---~A
und
also bewiesen,
Wit
GST ~
----* A
wit
disJunktiven
Wenn
GST~A
II)
AvB
9 so
Im f o l ~ e n d e n
diese
die
Ublichen
Konsistenzbeweis Begriffe:
usw~
Da nun ist,
noch,
f~r alle
k~nnen
eine H~he
wit
h(s)
H~hensprungs III)
eine
schlUsse
die
jeder
Sequenz
im E n d s t U c k
ordnen
wir
zu.
Schnitte) Ausnahmeo
~
Es b l e i b t
~ -Schlusses
~(S)
= ~(Sl)
ncch der Fall
setzen
: 0(S)
~d(~)
haben
=~d
(s
wir
die
Ordinalzahl
IV)
Fttr B e w e i s e
duktionsschritte
P aus
st~ck,
3)Elimination
SI,S 2 / S gleich
eines
Schlusses
B k ( t ) in d e r
Pr~mlsse~
gleich
0 ist~
eingeftihrt
worden
P aus
GST in bekannter
Weise
dann
auch der Begriff
des
werden.
S in P induktiv SchlUsse
sich alles
glelch
SI/S
' wo d = h ( S l ) Als
yon
und
oben nach
Struktur-
wie bei
Gentzen~
zu d i s k u % i e r e n .
- h(S)
0rdinalzahl
ist; ~ o ~(P)
yon
Wir
' @ ' P nehmen
SE yon P .
wir nun die bekannten
Gentzenschen
Re-
als d a sind
, 2) E l i m i n a t i o m
eines
in d e r E n d s e q u e n z ,
Primformel
eingeftthrt
der Endsequenz
GST kSnnen
i) V e r d ~ h l n u n ~ s r e d u k t i o n
die
eines
SI/S 2 ordnen wit jeder Konklusion S -I (statt 0(SI) ~ 1 w i e b e i G e n t z e n ) .
Bedeutung~
definieren
kann
zwei~en
also
Hauptformel
"Komplexit~t"
~-Schlusses
@ O(SI))
O(SE)
einer
fiir d e n
~ 2 zu
eines
die ~bllche
Formel
Sequenz
blelbt
:
ftihren 2qlr
Wir haben
Schnittes
F~r Axiome,logische
(inklusive
eines
jeder
GST konsistent
ein 9 d i e
slnd~
Hauptformel
S im B e w e i s
Insbesondere
f~Ir S c h n l t t e
0rdinalzahl
der
bzw.
als d a s i n d
GST und
Komplexit~t
Komplexit~t
A
usw.
c sines
der Begriff
hat,
im E n d s t ~ r
~
sind mit
GSM v ~
k~nnen t dass
Begriffe
solchen
die Komplexit~t dass
eine
schwierig
~quivalen% kurz
P im S y s t e m
Schluss
~ch~ittformel
zuordnen.
0rdinalzahl
Beweise
einer
auf eine kleine
Im Falle
GST~B
Kcmplexit~t
Sohl~sse
Als n ~ c h s t e s
unten
die
der
per Definition
Man beachte
Bild
ist d i e
~
.
eharakteristisch
kritiseher
Schlusses,
Wie ~blich,
der Komplexit~t ist
Gentzen
, und
Es ist n i c h t
schreiben
zeigen
beweistheoretischen
yon
EndstUek,
kritischen
wir
GST.
Eigenschaften
oder
betrachten
ist
GST vollkommen
Satz.
wenn
, Bk(X)
enth~It.
w l r mit
GSM Iund
Tim
ist u n d d i e ~ b l i c h e n GST ~
.o.
fiir x in ~ ( x )
eine
die Kalk~le
f~tr G S M t ~
Satz
frwi
Kalk~l
SuM'
k = 1,2,
~
die h~ohstens
Den resultlerenden
bls
323
logischen
logischer
Zeichens
Axiome
aus d e m E n d -
aus d e m E n d s t ~ c k .
324
B.
Reduktiensschritte
y o n d e r Art
Reduktionsschritte
bezeichnen.
Wir mUssen
nun noch
~-SchlGsse
aus d e m
Sei a l s o
ein kritischer
A)
EndstGck
P wie
wir kurz
einfUhren,
die
als v o r b e r e i t e n d e
es
uns g e s t a t t e n ~
zu e l i m i n i e r e n .
,~ in P ; s e i n e
sei
ein B e w e i s
~hldern w i r
2) w o l l e n
"~
~-Schluss
Voraussetzung
Es l i e g e
Dann
i),
Reduktionsschri~e
Bk(t),U Ak(t),r
Folgende
Scarpellini
Konklusion
liegt
also
im E n d s t U c k o
erf~llt:
Po y o n folgt
~ B k ( t ) im S y s t e m
ab
G S M ! vet.
:
p
9 o
__---~Bk(t)
,
Bk(t),
r
9
, Schnitt,
Verdiinnung
e
Wir
sagen,der
resultierende
Beweis
P'
folge
aus
P durch
einen
~-Reduk--
iogischer
Zeichen
wollen
tionsschritt. ~-Reduktionsschritte s~mmenfassend
als
haben
einen Beweis
schritte
Reduktionsschritte
folgende
P k~nnen
angewandt
2) v o r b e r e i t e n d e
Elimination
eigentliche
Reduktionsschritte i) A u f
und
nur
werden
fundamentale endlich
viele
wir
zu-
bezeichnen. Eigenschaften
:
vorbereitende
Reduktions-
,
Reduktionsschritte
vergr~ssern
die
Ordinalzahl
yon
P
nicht, 3)
eigentliche
Im F a l l e in
eines
~d(~O
eines
~-Schlusses
dies
Form
ersetzt
Po aus GSM'
einen ~
wird
; die
Beweis
oder
a.wendbar ( [ 6 ]
~7])
BL)
eine
Ist
folgt
die
Ordinalzahl
aus d e r T a t s a c h e , dureh
Ubrigen
die
F~lle
dass
y o n P. das
~o
Ordinalzahl
~ < ~o aus d e n U e b e r -
folgen
yon Gentzen.
y) W i r n e n n e n die
verkleinern
@ O(SI) ) jetzt
Beweises
legungen
Reduktionsschritte
A,r---*~
ein Beweis
P aus G S T N o r m a l b e w e i s ,
~ F
hat.
oberste
Po aus G S T y o n
---~A
wenn
Auf Normalbeweise
Sequenz
des
Endst~ckes
, fur w e l c h e n
seine
ist das
~(Po ) ~
yon
P,
~(p)
Endsequenz
Basislemma
so e x i s t i e r t gilt.
B.
VI)
Wir
beweisen
nun
durch
Scarpellini
transfinite
325
Induktion
Gber
O(P)
folgende
Be-
hauptung: B)
Sei
P ein
wo
F schwach
ein
Beweis
Ist
~(P) in
Beweis
aus
positivist P
<
aus
o ~o
' so
P vorkommt
(und
wir
Einschr~nkung
bereitenden
dann
ein
kommt
GSM'
der
existiert
~(P1)
< O(P). M'
Da
~ndererseits
eine
yon
: Die
ist
anwendbar.
9 yon
und
ein
J
M'
Axiom
yon
Also
yon
Dann
P ist
~
transformiert gleicher
Dann
P vorhanden
Satzes GSM'.
halten wieder mMss
muss
und
F.
der
kein
Schluss
O(P)
~ GO
Ferner P keine
sich
folgende
Form
aus
GST
muss
yon
~ und
Gem~ss
, fQr
darns a b e t
k~nnen vor-
p
F~lle
~ ,
yon
das
~ p
: GSM' ~
impliziert ~p
~ M'
Axiom
mit
p ,
dies
sein
p
p ~
, was
M~
der
~ , p prim,
somit
ein
Beweis
Sequenzens Wir
nun k~nnen Beweis
im
der )
Existenz einen
P' y o n
9
F 3 notwendigerweise
kleinerer einen
ist
~-Schluss
existiert
dann
ein
Beweis
Induktionsvoraussetzung
und
eines
aus
und
GSM'.
wegen
Beweises
P
o
a(P') Die
die
< 0(P)
Endsequenz Form
>
Wegen
der
der
Hypothese
yon
~
~-Reduktionsschritt mit
aber
Reduktionssehritt
kritischer
~(t)
Pr~dikatenkalkUl
die
P1 yon
ergibt
P
o
gilt.
und
somit
. Das
logischer
(y)
~(P)
Pl y o n
Reduktionsschrit~
Beweis
9
Basislemma
Induktionsvoraussetzung. bis
kein ein
Hypothesen
loglscher einen
~
, ~
den O(Po)<
alle
ein
P in
notwendigerweise
sein.
~(t)
einen
F I
existiert
dass
der
ist,
kein
Endsequenz
Bk(t)
aus
P1
GST,
sich
Beweis
des
also
bedeutet
kann
ein
yon
es
zusammen.
Axiom
Primbasis
ist
widerspricht.
aber
: P ist
existiert
dann
lassen
Bowels
F
valenz
Nun
ein
Induktionsvoraussetzung.
erf~llt
~ F,
Pvor.
zur
Po
Sei
annehmen,
zul~sst.
Widerspruch
-
ist).
da
EndstGck
Allgemeinheit
mazimalkonsistente
Endsequenz
Ordnungszahl,
in
odor
P vorkommen.
F 2
yon
9
9 % Fund
C bewiesen,
seinem
Baslslemma
p
yon
anwendbar.
Form
Nach Induktionsvoraussetzung
aber
~
Form
.
konsistent
mit
yon
Endst~ck
gem~ss
Da
Konsistenz
Beweis
im
Dann
EndstQck
da
S die
F
Behauptun
P nicht
S der
.
: P ist
p prim,
die
Sequenz
hat
9~
Reduktionsschrltte
unterscheiden F i
yon
ist
f~llt
einer
; dann
GSM'
speziell ohne
GST
ausf~hren . Das yon
B
Aequi-
P muss mit
E i)
(t) k und er-
widerspricht
F haben,
es
also
aber ge-
positivem
326
B.
F ~
: Die
Endsequenz
wendigerweise
und
P in
ein
einen
Dann
gilt
F ~
: Die
aber
Beweis
Pl y o n
GSM'
p
der
~
Endsequenz
nicht
P hat
die
eigentlicher
Reduktionsschritt Wenn
yon
Scarpellini
, so m u s s
9
nach
yon
auf
Form
w
p
, p prim 9
Reduktionsschritt
p mit
auf
~(Pl)
<
Dann
muss
P anwendbar
O(P)
not-
sein,
transformieren.
Induktionsvoraussetzung.
P hat
die
Form
P anwendbar,
so
~
~p
. Ist
schliessen
notwendigerweise
ein
wir
ein
wie
eigentlicher
unter
kritischer
)
F 4
.
~-Schluss
Form
p,P C--*~p in
P vorkommen p
Wir
ein
(,den
Bild
behaupten
W~re
dem
~
folgt
ab~Ludern
kurz
mit
Sl/S
in der
Endsequenz
hat.
: p
~
Axiom
nicht
weise
wir
so,
p ein
ist
ein
so m G s s t e Axiom
wegen
yon
GSM'
bezeichnen),
yon
der
sein.
GSM'
dessen
, und
somlt
Maximalit~t
yon
Wlr
dann
kSnnten
Hauptformel
GSM'
~
mp
.
M I notwendigerden
Beweis
P wie
:
5
p
, p
, ~
P
Das
Resultat
hat
S die
w~re
die
Komplexit~t
ein
Beweis
Ordinalzahl
~(S1)
0 hat
benutzt
raten
somit
Also F
GSM'
6
: Die
nicht
yon
der
Form
F
(3~)
=
wird
~
in
der
Form
A v
B
L~sst
P einen
unter
F
geben
:
in
P1
~(P1)
hingegen
<
O(P)
0(S1)
nur
: denn
in
@ 1
Da p
9
P
h(S)
= 0
Axiome
einen
die
, also
O~d(O(S1)
Ordinalzahl
Widerspruch
mit
~
1)
= O(S1)
1 erhalten
der
.)
~ Wir
1
,
ge-
Induktionsvoraussetzung 9
.
Endsequenz
A(~)
4
~p
-
, dass
wieder
,
, ist
d = h(S1) wobei
,it
P1 y o n
@ 2
yon p oder
, A A B an
P hat
Form
, p prim 9
, (~x)
A(x)
~
F
Dann
, (Vx)
, F
muss
A(x)
schwach
positiv
und
F notwendigerweise sein
; wir
nehmen
yon z.B.
9
eigentlichen
. Wenn
-~p
die
nicht,
Reduktionsschritt
so m u s s
es
einen
zu,
so
kritischen
schliesst ~
~
man wie - Schluss
B.
>
P
dessen
nauptformel
(~x)A(x)
Schluss
so
diesen 0(PI
)<
Analog
Sx)A(x)
verf~hrt
man
Behauptung Liegt
Form
)
A v B
Basislemma,
dass
kommt,
nun
einen
k~nnen.
und
- (~x) Indem von hat
der
Satz Es
M" Ak
Diese
wir
B
P aus
von
Lassen
)
also
, A V
GST
vor,
A(t)
: GSM'
~
B
wit mit
A(t)
, ~x)A(x)
mit
so
~
,
.
Endsequenz
aus
- Schluss
h~chstens
Reduktionsschritten
B)
bei
finden t)
einem mehr
wir
einen
. Analoges
gilt
Eigenschaften
und
aus
in
P vor-
Beweis
eigentP1 y o n
anwendbar
im
de.
anwenden
vielen
~eweis
ist
der
und
, der
endlich
Reduktionsschritt
gewisses
elner
folgt
Rk-Reduktionsschritt yon
aber
, die = 1
, 2
wird
naheliegende ersetzt
in
beantwortet
: FGr
seien
Voraussetzungen
die
ist.
P
Fall
aus A v
GST
B
konsistent,
.
womit
Satz
~ ~)
konsistent
.
die
Primbasis
l)
p ~ M1
genau
2)
-~ p e M I g e n a u
die
wenn
damn
vielleicht
ist,
Formeln
1 erfGllt.
M l wie
dann
1
eine
die
maximalkon-
vorausgestzt
aufz~thlbarer
(V~)(Ak(~)~
k
Satz
Weise
kleinere dass
gegeben
M
,
sind.
die
M und
yon
in
dutch
aufz~hlbar
durch
Primbasis
ob
kann
rekursiv
To = Su~
: Definiere
Frage,
werden
rekursiv
, ...
FolKerung
Bewels
hat.
ist.
M")
und
P1
gilt
= A A
A(x)
AusUben
kein
sin
M'
disjunktiv
Endsequenz
Beweis
kritischen
disjunktiven
die
die
(~x)
F 6 schllessen,
sich
Frage
F
Beweis
jeden
den
(fUr die
der
einen
F~llen
nach
Primbasis
, k
in
bewiesen.
~
auf
, auf
bewiesen
(M ~ , Bk
wir
A(t)
stellt
wir
beliebiger
unter
somit
sistente
den
somit
vorbereitenden
wie
)
GST
in
odor
enden
A(x)
wit
erhalten
Bild
Ordinalzahl-verkleinernden
Also
lichen
ein
ein
t
(~x)A(x)
9
.
B) i s t
VII)
A(t)
Induktionsvoraussetzung
. Nach
GSM, ~
also
Die
weg,
0(P)
327
Scarpellini
folgt
T
~
, Bk ist
, k die
= i
, 2
, ...
Theorie
Bk(~ ) )
:
T O ~- p
wenn
~ Dann
,
Ip
.
O
Die
Theorie
gleichen
T1 = S u
The0reme
Wir
k~nnen
den
entsprechenden
dlsjunktlven
nun
wie
dutch
MI v
k.J ( ~ x ) ( ~ ( ~ ) ~ Bk(~ ) ) hat dann dis k T o und ist semit konsistent nach Satz i .
Repetition
des
SequenzenkalkUl
Eigenschaften
hat
Beweises
yon
GST 1 beweisen,
; dabel
vereinfacht
Satz
i
dass slch
, angewandt schon der
auf
T 1 die Beweis
etwas,
328
B.
well
die
wsrden Satz bei
Konsistenz
muss
1 und
nur
Formeln
die
noch
verlangt
mit
die
Platzgr~nden Belspiel
konstanten sistente die
man
wird~
Primbasis,
die
und
nicht
mehr
bewiesen
so g i l t
~ts
hat
man
positiv wie
der
der
nur
endlich
viele
Funktionskonstanten
jeder
endlichen
aus
Harropim Zu -
erweiterten
implikatlven
Theorie
Axiome
erh~it,
nicht
eingehen.
Individuen-
, Ist
Erweiterung
zu t
werden~und
meistens
Verallgemeinerungen
S enthalte
in
zugelassen
sich dann,
Konsistenz
HinzufUgen
keine
Verallgemeinerungen
Axiome
die
auf diese
yon
~ cn und
M
eine
S v M
kon-
u E
:
so folgt
aus
der
S V M u E ~ (Jr)
konsistent freien
und
Folgerung
disjunktiv
Variabeln
A(x,y)
s ~ M ~ E ~ A(~,~ 1) v zu
Satz
S U M U ~,/ ( V x ) ( ( ~ y ) A ( x , y )
zwei
wird
verschiedene
zus~tzliche dass
durch wit
Sprache
~ ...
R) w e n n
Also
lassen als
Allerdings
kSnnen
: Die cI
Regel
1 garantiert
Harrop-Formeln,
vorauszusetzen~ Aus
Folgerung
Harrop-Formeln
zusammensetzen.
sammenhang
Satz
,
welchen
wobei
durch
Scarpellini
ist,
wo
durchl~uft.
...
i, d a s s
D A(X,r A(x,y)
alle
v A(~,o n)
die
~ A(~,~).
Theorie
V ... Formeln
V A(X,C n) V A ( x , x ) ) yon
S mit
genau
B.
II.
Die
(a)
intxrltionistische
Zahlentheorie
Bemerkun~en
Versucht
man,
die
Methode~
nlsfiische
Zahlent~eorie
lich
die
nur
oder
zu
nicht-konstruktiv) man
Schwierigkeiten,
herrfihreno
Zus~tzliche
auf
der
finitiren
etwas
dem
auf
rekursiv
die
nStigen
(b)
Der
Erg~nzungen
Hinblick
D~
auf
fur
ste~ige
die
Analysis
~
wie
bei
hilt;
(kurz [
wit
Funktor
. I s.t
~e~
f(i)
..~
,Xp
heisst sine
= ai
eine die
Folge
Belegung
uns
mit
vielleicht
Zusatzbetrachtung
den
ist.
fn'gl .....
wir
' i ~
yon mit
sine
zuordnen~mit
: Sind
u 9
~ f
und
N
....
2,
, an_l~
Zeiehen
freien
yon
"'"
Funktoren nehmen
fULhrt
wit
an
,fUr
vorhanden
Abbildungen
er k e i n e
n
~ , x t(f,g,~)
; u ~ v , so
sin
.
. Bin
Variabeln
gm'Ul .....
ist
. Jedem
% 0
, so
die
1
ent-
St
und
yon ist
Up
m
(fi,gj
Jedem
t(~
yon f
'
, ~l ....
e NN
so , U k G N)
, ~ ,x)
und
sin
Zahl-
Weiss Term
Sirule
gleicher t(f,g,~)
u ~
Zahlvariabeln,
Term
(im
Verkettung wit
. . . . .
in nat~rlicher
bezeichnen.
F01genzahlen
~(v,w,~)
'
n Punktions~
gilt
ist
schreiben n
' ~
. . . .
Ststigkeitsfunktion v,w
,
' i = i, und
seien
Zahle~.Ferner
=
halber
f~r
vorkommen.
Z s t ~
fl
~
~ Ui
stetige
wenn
~l WP-Konstanten,
[ ~ ])
g
rekursive
f e NN
sioh
, w ~
>
Zahlentheorie
Konstanten
= nat~rliche
Terms
u,v,w
fl ......
induktiv
Nt
.
formalen
( N
Einfachheit
Zahlen-
aus
) und
Folge
und
f,g,u
der
Folge
~ an_l>
lisst
v
sich,werna wlr
aufz~hlbaren
einer
.... N
mit
n-i)
yon
in
wir
,wenn u i ~
zugeordnet,
wert
sich
intuitionistische
endliche
. Der
d~rfen
yon
Folge
Belegung
jeder
~
'~
abzihlbare
auf
aber
(i
N t
konstant,
bezeiehnen
u. = .< a o ,
X
~ :(NN) s x
uns
WF-Konstanten
Folgenzahlen u,v
ergeben woli~n
Funktionen
WF-Konstanten).
] sin
heisst
die •
' ~i
jede
sine
Abbildung
sehr~nken bzw.
fur
Zahlen
Kleene
stetige
Spiter
ergeben
rekursiv
in
Zeichen
' ~2
(NN)s
und
Wahlfolgenkonstanten
Ublichen
(~I yon
wir
Kons~anten
den
vorhanden,
nat~rlichen
Term
zul~sst.
mathematisch
den
eigent-
.
statten
und
Funktionen
Abbildungen
das
jede
M~thede
Art
diese
erst
, dass
(konstruktiv
, so
Trotzdem
wir
, und
intuitio-
Schnittelimination
technischer
wollen
lassen
sich
der
einer
, da
anbringen
Z.m mit Variabeln wir haben neben
Variabeln
man
Fehlen
verharrt.
Zuerst
es
~-Regel
Zahlentheorie
vom
befassen
Acht
zeigt
einer
die
Formalismus
theorie
yon
finit~re die
auf
Uebertragung
Komplikationen
ist.
ausser
mit
direkte
die
Zahlentheorie
vSllig
Abschnittes , so
versehen
Aufz~hlbaren
interessanter
Asp@kt
ist
sich
letzten
~bertragen
sine
technische
man
des
Zahlentheorie
Beschr~nkt
Im
329
Scarpellini
yon
L~nge,
t(~,~,x) [ 6 ] , gilt
:~t(v,w,u)
- i.
330
B.
Sind nur
Gt
rekursive
rekursiv
f~r
Konstanten
w~thlen.
jede Bele~ung
beim
Primformeln f,g,~
gilt
Ein konstanter
Term
t(~
) heisst
( ~ ~ f , ~ ~ g) g i l t
(c)
der
Form
Systeme
0
der
Z.m bezeichnen theorie, welches
~
f~r
wir
0"
das
neben
saturiert,
= t(g)
,an_l>
sich wenn
wenn
f~r alle Belegungen
f,g
numerische
Terme~
System
der
WF-Konstanten
nur
Funktionale seien
intuitionistischen Zeichen
(und
enth~it.
zudem
Zahlen-
F~r
die Axiome ~
quantorenfreie jede
(k)
WF-Konstante
= ak
f~r
zu Z l a l l e w a h r e n P r i m f o r m e l n (im S i n n e d e s l e t z t e n A b s e h n i t t e s ) ~ so e r h ~ i t m a n d a s S y s t e m Z ; mit Z bezeichlI 12 n e t m a n d a s U n t e r s y s t e m y o n Z. , in w e l c h e m n u r r e k u r s i v e K o n s t a n t e n 11 beim Aufbau yon Termen und Funktoren verwendet werden. Uebertr> man die
Systeme
Zi,Zil,Zi2 dabei
Induktionsaxiome
delnd),
so e r h ~ i t
noch mit
Addiert
wahr,
Zahlentheorie
rekursive
...
vorhanden.
so l ~ s s t
,~,x) h e i s s e n
.
, ..o h e i s s e n
formale
den
primitiv
wo XI = ~ao,
k ~ n-i
,
t beteiligt,
= q(~
:
intuitionistischen
Mit
Axiome)
O'
,
vom
= q(f,g,~)
t(f)
Terme
Aufbau
t(~n,~,x) :
t(f,g,~)
yon ~
Scarpellini
man
man
in bekannter naeh
dem
Weise
Vorbild
Kalk~le
in den
yon Gentzen
GZi,GZiI,GZi2
einer Umsetzungsregel
Sequenzenkalk~l
versehen~
die
I l l als R e ~ e l
. Jeder es
( die
dieser
behan-
Kalk~le
ist
gestattet,numerisehe
T e r m e d u r c h a n d e r e y o n g l e i c h e m Weft zu e r s e t z e n (siehe ~ I , C ~ J )" F ~ r das F o l g e n d e b e n ~ t i g e n w i r e i n e n H i l f s s a t z f~r p o s i t i v e F o r m e l n : Hilfssatz
: F~r
A* i n p r ~ n e x e r ist,
angeben,
jede positive Normalform,
Formel
deren
A lisst
sich
effektiv Tell
quantorenfreier
eine
eine
Form~l
Primformel
sodass Z.
I-- A ~ A *
(k
= ~,1,2)
1k gilt.
Hinweis:
Man benutze,
~quivalent
ist
zu
dass
jede
(~x)(~)((x-i
Formel
A v
(~)B(~)
V A)A
(x%l ~ B ( ~ ) )
intuitionistisch
)
.
B.
(d)
Wahre
Im
pr~nexe
Fulgenden
@
Ist
~
und
"wahr"
ist
A
(~y)
,~,x,f,g
durch
@
) ist
wahr,
wer~n
"Stetlg
und
wahr"
oder
auch
definiert. so
heisst
wahr
wird~
A wahr,
A(~t~,~,x,y
wernq
) wahr
A
f~ir
jede
iSto
mit ahr,
@
(~)
A(~t~,~,x,~
dass
A(~
Lassen
wir
nur wahr"~ ist
Da
H
4 unten
unter
sieh
1 haben 4
Die
:
]
A(~u) ist
und
(
zu~
so
stetigen
stetigen
sprechen
wir
Furmktionale
von
vorhan-
Funktional-lnterpretation
d o c h m ~ s s e n wir die
1
ohne
freie V a r i a b l e n
f~r ~
(v) @ 0
den
= leere
Beweise
yon
H
lem-FurLktionen"~ , so
) pr~nex
, ~ elne
0( u 's verschieden Folge)
sagen
dann
konstruktive nur
e
sie erf~lle
Gegenstiicke eines
: Es
, H
3
deren lassen
als
existiert
Sei einer
e(v)
A(~u)
und ~
A(~u,v)
Zi2~- A.
eine
w a h r ist.
wahr.
A(~u,~,x
yon
Formel
, sodass
und A rekursiv wahr,so
Folge
sind
yon
und
WF-Konstanten,
alle
den
4
verlaufen
Existenz sich
eine
durch durch
A.
Mit
zu
H
Illustration
eine
die
unteren
Index
o
A(Cgu,~,x ) ist wahr, g e n a u d a n n w e n n A ( ~ u , ~ , n ) w ~ h r
kursiv
existiert,so-
so Z
eine p r ~ n e x e
ist
Darm
~
alle
zur
F
sind l e i c h t b e w e i s b a r ,
Stetigkeitsfunktion
wir
Sprache
aequivalent
Hilfss~tze
H ~ : Sei A(~u)
wir
Furmktoren
~- A . II : Sind alle K o n s t a n t e n in A r e k u r s i v ,
H 2
FgOgl~,~,x
aus Platzgr~inden w e g l a s s e n .
H i : Ist A wahr,
In
Funktor
sod ss
ist.
t und
unserer
exi,tiert,
].
Die n ~ c h s t e n Beweise
in
ein
) wahr Terme
"wahr"
yon Kreisel ~
Term
wahr,we~m
rekursive
"rekursiv sind~
) ist
,~,x,F[~u,~,x]
den
H
bis
aussagenlogisch
A(~i~,~,x,y
WF-Konstanten,Furaktions-
Furh
Q
@
H
yon
) quantorenfrei,
f~g~
@
Folgen
Zahlentheoretische darn~
= A(~t~,~,x
Belegung
331
Formeln
sind
Zahlvariabeln kurz
Scarpellini
~hnlich
wie
in
~
~
garantiert
eine Hilfe
3,
H
4
f~ir
einzige dieses
Stetigkeitsfur~ktion~
- i die F o r m e l A ( ~ u . v )
Begriffes
sp~teren
pr~nexe sodass
.Dann
mlt
Formel gilt
Sind
die
"Sko-
werden,
re-
e kodifizieren; lassen
(und
Gebraueh ~
].
GSdelnummer
formulieren
Partialrekursive konstante
bis
E~
f~r alle n iSto
sich
leicht
beweisen),von
denen
erw~en.
folgenden
Eigenschafteno
und
OSdelnummer
: Ist
erf~llt
e die e(v)
~ 0
~ so
erf~llt
~(e) die F o r m e l A(~u).
B.
332
(e)
Regel
und
Schema
in
Scarpellini
Z l1
Satz als
2
: Seien
einzigen
liebige
%(~,x)
freien
nume~ische
Terme
E) g i l t
Z.11~ % ( F , n )
Dam i s t
die Theorie
und
, Bk(~,x ) Formeln
Variabeln;
ohne
die % ' s
n und k o n s t a n t e
, so g i l t
~
positiv.
Funktoren
Zil ~
T = ZilU
WF-Konstanten,
seien
F~r
mit alle
F gelte
~ , x k
, be-
:
Bk(F,n ) .
(V~,~)(%(~,~)
~ ~k(~,~)
) kon~ist~.t
disjunktiv.
Bemerkun~en
:
i) Bei
in Z. m U s s e n die V a r i a b l e n in d e n P r i m a x i o m e n n i c h t lI g e h a l t e n W e r d e n ; i n s b e s o n d e r e sind die u n i v e r s e l l e n Abschl~s--
Herleitungen
konstant se d e r 2)
Primaxiome
beweisbar.
Unter
"disjurLktiven
Eigenschaften"
etwas
Allgemeineres
als
in I)~
einer
Theorie
Wir verlangen
T verstehen
jetzt
z~
wit
nun
:
Gilt T ~ ( ~ ) A(~u,~) (mit (~)A(mu,~) konstant), so gibt es e i ~ e Stetigkeitsfunktion ff , s o d a s s f u r ~ ( v ) @ 0 e i n k o n s t a n t e r Funktor Fv e x i s t i e r t m i t T ~ A v( ~ u) , v. , FAehnliche Formulierungen gelten in den Fillen m A(~ u) V B(~ ) , T ~
~d
yon der
Satz
2
fUhren
Wir
: Auf
wit
Zuerst
den
Teil
eine
Beweis
im
also
( s i e h e z . ~ . [ ~ ] ~heorem ~ yon
annehmen,
iibersetzen
addieren
,
Grund
Allgemeinheit
quantorenfreier
I)
A(~ , ~ )
i~ rekur~i.e~ Fall
Beweis kung
(~)
wir zu
Primformel
Theorie
alle
Regeln
1 kSnnen
%'s
seien
ist.
Wie
SequenzenkalkUl die
GZil
Hilfssatz die
und
~
Bk(r, t ) , P
beim
Satzes der
~
F,t
frei
fur
mit
GT.
II)
Fiir Beweise
griffe aus delt wird. Die
[l]
Komplexit~t
eines
~,x
P
aus
%,
GT
ein,wobei
eines
Beweis
in
Bk
fiihrem die
. Das
wir
den
Induktionsschlusses
die
Induktionsregel
Strukturschlusses,
definieren
System
Satz
in
1
Schritte.
SequenzenkalkNl.
wie
...
bezeichnen
beweistheoretischen gleich
eines
wir
deren
%
resultierende
wieder
Einschr~hl-
yon
ihn
k:l,2~
in
ohne Formeln~
Form
%
wo
wir
pr~h~exe
zerlegen
T des
[ ~ ] ).
~
wie
logischen
in
in
[i
Be] behan-
Schlusses
[ I ]~ w ~ h r e n d
wir
die
oder
Kom-
B. Scarpellini plexit~t
eines
~-Schlusses
Die B e g r i f f e H~he, wie
per Definition
Endst~ck~
kritiseher
333
die
Komplexit~t
Schluss
usw.
y o n B k iSto
sind w i e d e r
gleieh
in [ ~ ].
III)
Jeder
Sequenz
oben nach unten verfahren setzen
wir
gleieh
setzen wieder
IV)
F~r Beweise
Beweises
P aus
GT o r d n e n
0rdinalzahl 0(S)
zu 9 Mit
wie
[ 5 ]. F ~ r
wir wieder 0(S)
Wir
[ I J:
Seines
eine
0(P)
in [ I
],bzw~
wir
Ausnahme einen
induktiv
~-Sehluss
=~d(~o
~ ~ ( S I ) ) , wo d = h ( S l )
- h(S)
= r
wo
P ist.
S die
Endsequenz
yon
yon ~-Schl~ssen
yon
SI/S ist~
GT d e f i n i e r e n w i r n u n R e d u k t i o n s s e h r i t t e w i e
P aus
2) E l i m i n a t i o n
in
Reduktionsschritte,
i) V o r b e r e i t e n d e
logischer
Zeiehen,
~) I n d u k t i o n s r e d u k t i o n . Hinzu
kommt
In P liege
die
~-Reduktion~
ein k r i t i s c h e r
die wie
%-Sehluss
Der Funktor
saturiert,
mit
GZil yon
Fund
der Term
P
B k ( F , n ) vor.
#
t seien
Wert n
numerischem ~
Dann
ist.
Form
Bk(;,t) , V %(F,t) vor.
definiert
folgt der
konstant;
zudem
. Schliesslich ~ndere
man
liege
P wie
sei d e r T e r m ein B e w e i s
folgt
P
o
t aus
ab:
o
Bk(F,n)
Umsetzung
Bk( ,t)
Bk(F,t)
Schnitt~ Verd%innung
Ak(F,t) ,~ a
Reduktionsschritte eigentliche liche
der
Sorte
Reduktionsschritte
Abschnitt V) V o n n u n
IV)
des
deren
freien
Variabeln
haben
Beweises
an w o l l e n
weise,
9
die
enthilt.
gesagte
Ein Normalbeweis
wieder Satz
hat,
Form
Alles
gilt
P heisst F
yon
, 3) u n d genannte
wit nut noeh
Endsequenz
und Basislemma
oder
2)
Reduktionssehritte
%-Reduktionsschritte werden.
die
Eigensehaften
und
i) bis
eigent-
2) im
1 .
Normalbeweise m
oder
im B e w e i s
P betrachten~ 9
yon
Satz
F
hat,
d.ho
Be-
wo F k e i n e
1 ~ber Normalbeweise
auch hier.
Standardbeweis, wo
sollen
Vorbereitende
F eine
pr~nexe
wenn
seine
Formel
Endsequenz
ist,
deren
die F o r m
quantoren-
B.
334
freier VI)
Teil
eine
Primformel
FUr Normalbeweise
sei
saturiert,
yon
P satnriert
wit
~
)
~mit + 0
der Eigenschaft
, so
ist
P
VII)
Wir beweisen
nun durch
0hne
mit
seinem
trifft F 1
also
:P ist
zusammen. Es
und
P
tim
A(~u)
P
~hldst~ck
. Erse~zen
einen Normalbeweis eine man
Kontraktion. eine
folge
v
gilt
so h a t
kSnnen
zu,
Auf
ergeben
saturiert
sagen,
Stetig-
0(Pv) ~ber
seine
a~s
=
P d~roh
O(P).
O(P)
:
Lhudsequenz d i e
Form
ist w a h r .
der Allgemeinheit
B) zu.
sagen,
Induktion
i n GT,
Reduktionsschritte Endst~ck
Wit
transfinite
) und A(~U)
Einsehr~nkung
vorbereitende
Term
~
so f i n d e t
0ffensichtlich
P ein Standardbeweis
Wir
:
s~turiert.
v Kontraktion.
A(~
yon
Normalbeweis,
saturierende
>
konstante)
, so e r h a l t e n wit u~ v s a g e n ~ P v f o l g e aus P d u t c h
Wit
.
Begriffe.
dutch ~
eine
B) Xst
zwei
Sei P e i n N o r m a l b e w e i s
A(~u~v)
~(~)
wir noch
jeder(notwendigerweise
ist.
P ein nichtsaturierter
keitsfunktion Ist
ist.
P benStigen
i m gs_nzen B e w e i s
Pv y o n Ist n u n
wenn
Scarpellini
2)
wit
0(P)
arunehmen: ~
~
.
o Standardbeweise
alle
s i e h darn% f o l g e n d e
enth~it
keinen
Dar~n m u s s
P notwendi~erweise
einen
oder
Induktionsreduktion
i) P l ~ s s t
P f~llt
kein(
dann nicht
P' m i t O ( P ' )
< 0(P I
Fille:
kritischen
logischen
Schluss
in der Endsequenz.
eine
logischen
zulassen
oder
Reduktionsschritt einen kritischen
Sehluss
%(F,t)
,~
;
enthalten.
In
den
ersten
logischen einen
Beweis
Pl
A u f Pl t r i f f t B)
beiden
F~llen
transformiert
Reduktionsschrittes
bzw~
derselben
somit
Endsequenz,
die
sich
einer
P vermSge
eines
lnduktionsreduktion
aber
mit
Induktionsvoraussetzung
in
~(Pl)
~
zu~
also
~(P)
.
trifft
a u f P zu.
Xm letzten
Fall
~(F,t) Term f~gen 9
gibt
es g e m ~ s s
mit 0(PI) <
t saturiert
und hat
sit
erhalten
O(P2) < O(P)
tionsvoraussetzung
zu,
Zil ~
%(F,n)
. Auf
Zil ~
Bk(F,n ) , also
also
Grund GZ
. Auf
wit P2
ist ~ ( F , n )
der Bedingung ~- B k ( F , n ) .
mI
einen Beweis
o Da P saturiert
einen numerischen
einer Umsetzungsregel %(F,n)
Basislemma
~(P)
Wert
wahr.
ist d e r
no D u r c h
einen Beweis trifft
P1 y o n
ist,
somit
P
Hinzuyon
2 die Induk-
Es g i l t
somit
Satzes
folgt
dann
somit
einen
~-
E) d e s
Wir kSnnen
B.
Reduktionsschritt einen Beweis P' F 2
trifft
Dann
saturiert
mit
Bild
existiert
und
in der
sine
enth~it
zu~
keinen
Stetigkeitsfunktion
yon
F 1 ~ nach
Endsequenz Also
9
A(~u)
, P1 y o n 0 ( P l) < 0 ( P ) wahr 9 also
Auf
folgt
~unktiven
~
einen
a u f Pl zu
analog wie
D)
in ~6 ]~pg.
B)
# 0 , so
somit
den Vor-
auf P
zu. v f~r ~(v)
, und
Somit ~ 0 ist
mit
Bild
erh~it
gswis~en
in der End-
. Unterl~sst
man
einen Beweis
Funktor
F[~11 ]_
o B ( ~ u , F [ ~ u ] ) is% F~lle
o Da somit
aufgezihlt
sind,
B) hat.
folgt~
enth~it~
: Ist 5 (v)
(~)B(~u,~)
) . Da alle m6glichen
Eigenschaft
Eigenschaft
a u f P zu.
logischen
gen~gt
y
also
Schluss,so
B)
P
Schluss
A(~u)
ist
~-Schluss
B)
Auf
~ach H 3 .
B(~u,F[
~ mit
A(~)
io~ischen z.B.
logischen
P die
GIu/nd y o n
einen kritischen
v
A(~ u) ~ r
. Sei
aush A~u
dass
trifft
kritischen
saturiert.
Form ~
einen kritischen
man den kritischen
folgt,
der
ist
also
P in
< ~(P).
~ndsequenz.
F 1 trifft
P sine
sich dann
aber mit O(P')
aussetzungen
sequenz
(f)
transformiert
P
: P enth~it
Daraus
Es
Endsequenz,
die Induktionsvoraussetzung
A W u . ~) w ~ r .
VIII)
335
ist d e r k o n t r a h i e r t e B e w e i s
hat
F ~
auf P aus~ben.
P' d e r s e l b e n
: P ist n i c h t Schluss
Scarpellini
dass
sin Normalbeweis
P, d e r
einen %-Reduktionsschritt ll0
, dass
GT und
somit
zulisst.
T die dis-
Ei~enschaften haben.
Regel und
Schema
i n Z.•
D a Z. n u r T e r m s u n d F u n k t o r e n e n t h ~ i t , d i e aus r e k u r s i v e n K o n s t a n t e n a u f 12 ~ e b a u t sind, ! ~ s s t s i c h j e d e r B e w e i s aus Z. durch sine einzi~e G6delnum12 m e t e k o d i f i z i e r e n . S c h l i e s s l i e h w o l l e n w i r sine T h e o r i e T k u r z r e k u r s i v aufz/hlbar
nerunen~
T rekursiv
aufz~hlbar
Dann
gilt:
Satz
2
meln
falls
Z. ~ i ist.
: Sei % ( ~ , x )
T ~ Z. und m2
falls
, B k ( ~ , x ) k = 1,2 . . . .
die Menge
der Axioms
sine rekursive
Men~e
yon
yon For-
ten
aus Z , m i t ~ ~ x als e i n z i g e n f r e i e n V a r i a b e l n u n d o h n e W F - K o n s t a n 1 ~ die %'s s e i e n p o s i t i V o F e r n e r sei ~ sine p a r t i a l r e k u r s i v e Funktion
mit
der Eigenschaft ER)
Ist
e die
G6delnummer
sin konstanter die Dann hat
sines
Funktor
G~delnummer sines
die Theorie
aufz~hlbare
: aus
Zi2 y o n % ( F , n )
und n sin numerischer Beweises
T = ZiU
Primerweiterung
Beweises
ist,
so ist ~ ( e )
y o n B k ( F , n ).
(~,x)(%(~,x)
T', w e l c h e
Term
, wo F
=
Bk(~,x ) )
die disjunktiven
sine
rekursiv
Ei~enschaften hat.
336
B.
Aus
Platzgriinden
weitert Form
man
sind
F,t
Hilfe
eine
aus
der
Flhnktionen
einem
Zi2
uns
Kalk~l
und
mit
einem
GT durch
Hinweis
begn~gen.
Hinzuf~gen
aller
Zuerst
Re~eln
er-
~
der
in den
die
f~llt ~(e)
die
Die
Definition
mit
Hilfe
nannten
> ~
frei
ffir ~,x ~
eines
Formel
ist
Menge
Satz
%
mit
der
, Bk
3 und
Gegenst~cken ~
" . Darau
der
definiert
man
partialrekursiven
zu H 3
~ H
4
(z.B. ~ i n H
5)
Eigenschaft: P yon
~
A aus
GT,
so e r -
A. unter
Fixpunktsatzes,
von~
in
in
Standardbeweises
erfolgt
Eigensehaften
aufz/hlbare
%(F,t),r
Funktion
von~
des
Hilfe
~
konstruktiven
GSdelnlnnmer
r
~k(F,t),r
partialrekursiven
partialrekursive
Ist'e
Mit
mfissen w i r
zu
:
Dabei mit
GZi2
Scarpellini
und
Bezugnahme der
auf
Beweis,
die
dass
Reduktionssehritte
~ wirklieh
hat,
erfolgt
durch
transfinite
Induktion
es d a m n
nieht
sehwer
, ausgehend
yon
M yon wahren
Primformeln
aus
Z
zu
die
genannt
~ber
T eine
erzeugen,
~(P).
rekursiv welche
12 die
Eigenschaften i)
Ist
A
Z.
1
2) Auf
Qilt
T U
es
zeigt Die
sich
konstant M
U
~
z.u i
M
karhn sich
klassischen
nicht
und
2,3,4
pr~nex
k
Analysis bietet
trivial,
zwar und
gilt
T U
(~ konst~t)
(F,n)
nun
den man
Beweis hie
lassen
fHr
und
M
~
A
, so
gilt
schon
.
M~A man
:
yon
genStigt
sich
intuitionistische
Einschr~nkungen
Systeme
A,
, dass
S~tze auf
hat
%'s
hat) keine f%thrt
und . Die
2
ist
Hber
der Bk'S
den
Rahmen
fast M
Z.U
M~k(F,n)
1
wSrtlich
, falls der
~bertragen das
fibertragen
&
lassen
sie
(mit
System
S~tze
Sehwierigkeiten, dieser
.
hinauszugehen~
. Insbesondere
Analysis
Ausdehnung
prinzipiellen Hber
Satz
verallgemeinern
Systeme die
, so ~ilt
Arbeit
die
2,3,4
gewissen
St~rke auf
ist hinaus.
der
diese aber
doch
B.
Scarpellini
337
Anhan 6 1 Es
gibt
noch
kationen Wir
weitere
M~glichkeiten,disjunktive
zu untersuchen.
begn~gen
uns
der
Eine
meln
. Verallgemeinerungen
sind
( unter
geeigneten
Satz
: Seien
A~
beweisbar. Beweis
: Wir
GZ. die 1
auf
sei
kurz
halber den
Fall
~eschlossene
dem
yon
Formeln
Z. + A m l
B
Impli-
geschlossener mit
und
(A D
freien
For-
Variabeln
schwierig. B)
~
A in
Z. n i c h t 1
disjunktiv.
Zi + A ~
Hbersetzen
Fall
) nicht
Formeln
yon
besprocheno
mit
Zusatzvoraussetzungen
B zwei
Darn% i s t
davon
Einfachheit
Eigenschaften
B in den
SequenzeD~alkNl
, indem
wir
zu
Regel
R hinzuf~gen. kann
Ein
dann
mit
Hilfe
Das
w/irde
spruch
des
F
~
Z. + A ~ 1 2)
Sei
Modell
B
~
einen
und yon
Beispiele:l)
so
+
A ~
B
kSnnen
~
die
und
Ist
ZK
eine
so
sind
die
somit
P
dem
yon
m
kSnnte
man
in
Muster
yon
klassische
GZ.R
Andernfalls
GZ.R yon 9 A finden~ o 1 , Mithin Z. ~ (A ~ B) ~ A ~ im Widerm auf Beweise P in GZ.R mit einer Endm klassische Gentzensche Reduktions-
A
nach
Theorie
enthalteno
somit
daraus
GZ.R 1
ernaltenen
Beweis
uneingeschr~nkt
anwenden,
A
der
R-Schluss
: Z. 1 Wir
Annahme.
Eigenschaften
~
in
Basis-Lemmas
9
theorie
P
kritischen
bedeuten
zur
sequenz
ZK
Beweis
keinen
Z. 1
+
yon A
~
B
[
Zahlentheorie~
Voraussetzungen
des
] die
disjunktiven
herleiten~ nicht
Satzes
ZK
erf~llt
~
B~
abet
~ mithin
B disjunktiv.
~)Z(~) hat
eine
. Setze
B
Formel,die
ausdr~ekt,
=
, A
~I
)Z(~)
=~
B
dass
Zermelo-Fraenkel
. Nach
Beispiel
ein
iist
dann
Z. + A D B d i s j u n k t i v . Addiert man zu dieser Theorie die Harrop-Formel l , so b l e i b t d i e ~ r w e i t e r u n g konsistent, verliert aber die disjunktive
A
Ei~enschaft. ~)Sei die
wieder
B
Theorie
Formel
T
B nicht
" werun T ~ ~ B T
+ IB
D B
=
(~)Z(~)
= Z
+
, A hin~egen
(~B ~
B)m
1 entscheidbar
so T ~
B
ist
" gilt
, so b l e i b t
diese
sei IB
D
B disjurLktiv. in
ZK
also
, gilt
B
Da
o Naeh T ~
nicht
trivialerweise.
Beispiel
ZK
, und
T ~ nB Bilden
konsistent, v e r l i e r t
aber
1 ist
da
die
. Die wir
die
Re~el
die
Theorie
disjunktiven
Eigenschaften. Es
ist
Z. ~ I ~) A
A
uns so
nicht Z. ~ 1
Sehliesslieh =
(~)qPrim(~)
sicher
nicht
B
gelungen, , 2) sei
Z
i
ein + A D
nochmals
, B
beweisbar
=
Formelpaar B
das
ist
Beispiel
(~)(qPrim(~)9 in
ZK
, also
A~B
zu
konsistent in
~(a)
Prim(~)) nicht
in
Z
finden
mit
, aber
nieht
. Somit
ist ist
Wenn
disjunktiv.
betrachtet
. Dann i
: i)
Setze (A D
B)D
Z. + A D 1
A D
338
B.
disjunktiv
nach
es
nicht
zu
beweiseno
Anhang
Satz
mSglich,
konsistent).
leicht
lassen
Realisierbarkeitsbegriff
spiel
kurz
theorie
skizziert.
ohne
yon
verstanden.
Satz
: A(x), E)
Sei
Dann
B(x)
Ist
T
man
leicht
Zi + A ~
verifiziert~
B mit
Hilfe
ist
yon
Satz
Beweis
gew~hnliehe
T
~
beruht
auf
werun e i n
Dies
sei
an
intuitionistische sei
im
passen-
einem
Bei-
Zahlen-
Sinne
yon
Kleene
Satz
Formeln
rekursiv
gilt
erzielen,
steht.
"Realisierbar"
der
mit
folgender
aufzihlbare, A(n)
(Vx)(A(x) = B(x)
i~t
Resultate
Verf~gung
ZI d i e
gilt
seien
eine
ZI u n d Da~
sich zur
Funktionsvariabeln.
~]
Der
Wie
Disju~tivit~t
2
verh~Itnismissig der
(und
die
Scarpellini
, so
Eigenschaft
disjunktive
gilt
T
m
: Erweiterung
yon
B(n)
realisierbar
einem
technisehen
Lemma,
das
wlr
ohne
Beweis
an-
geben. Lemma: A).
: Sei
Dann
terung
A
eine
l~sst
T yon
gesehlossene
sieh
ZI +
effektiv A
Formel
eine
angeben
und
gelte
rekursiv
mit
der
e / A
(d.h.
aufz/hlbare,
Eigensehaft
e realisiere
disjunkte
: Jedes
Axiom
Erweiyon
T ist
realisierbar. Beweisskizze effektiv terung T
T n B(n).
~
n alles
(~x)(A(x) Erweiterung auch
schon
eine
sich
B(x) : Aus
=
B(x)
+ A(n) es
ist
die
folgt
[Z]
die
rekursiv
mit
der
des
gilt
yon
folgt
ohne
im
funktioniert
E) f/
dass
Partialre-
ZI
+
aus
dem
Lemma
positiven T
A(x) yon
Verfeinerungen
kombinieren, erzielen,
aueh
:
disju~tiv~
automatisch.
Kombinationen
schliesslich
auch
. Da
yon
eines
fast
zu
Satz
f vermittelt
weiteres
ihnliehe
diese
im
aufz/hlbare,
und
man Erwei-
B(n)
eine
sofort,
Methode
Satzes
finder
Primerweiterun~
Realisierbarkeit
auf
e zu
Falle
aufz~hlbare
Hinweise
nach
f mit
Realisierbarkeit
so
Diese
Lemma
realisierbare
sehwierig
~ rekursiv
dass
letzten
und
ein
nieht
nieht
) existiert.
selbst
A(n)
in
Lemma
beweistheoretischen auch
Beweistheorie
dem
Beweis),
Versehirfungen
~--
realisierbare
seinem
. Gemiss
Ueberg~ig
damn
Hingegen
B(x)
der
den
sich
und
Satz
eine
n 62
prinzipiell
dem
disjunktive~ D
es
P
Satz
.
)
e / A(n)
disjunktive
. Wegen
) sofort
aus
bei
etwa
naeh
ergibt
T hat. nicht
Platzgr~nden Die
: Sei
aufzihlbare~
finden,
Beweistheorie
sentliche aus
ist~
(~x)(A(x)
geben die
gibt
~
B emerkun~en
+
ZI
Also
~ Daraus
(Vx)(A(x)
ZI
yon
9(x,y) zu
) = f
(und
Satzes
rekursiv
effektiv
kursive ~(nTe
des
eine
dann
um
doeh
Man
erkann
ganz
we-
m~ssen
wir
weglassen. noch,
wenn
kein
B.
passender
Realisierbarkeitsbegriff
St~rke
der
Die der
Scarpellini
Analysis
beiden
und
Methoden
3~
verhanden
fur
Theorien
mit
~berlappen
sich
ist
: fur
Funktionalen also,
Theorien
yon
hSheren
aber
sie
der
Typs.
implizieren
einan-
nicht.
Anhan~ In d i e s e r A r b e i t w u r d e unter
Benutzung
von
Allgemein Lemma
so
man
dass
die
eine
durchgef~hrt
in dass
Schliessens
die An
Teile
dafOr sich
dieser
urn
'starke
Stelle
sei
in
[$
und
] dank
etwa
das
Schema. nicht
Systemen
des
Basis-
dem fur
Trotzdem vom
dem
nach
Reduktionstheorie
Arbeit
in
heurisdie
ist
Regel~
anzunehmen
2
beweistheoretischen
nat~rlichen
Schliessens
kSnnen. in
dieser
yon
Richtung
aber
sich Das
ein
auf
gesteigerter
R.
vQn
Herrn getan.
Systeme
Basis-Lemma
Normalisation" Herrn
wird
Dissertation
~6~
lassen. ist
Methoden
eine
befindliehen
Obertragen
Erscheinung, dig,
Schritt
Vorbereitung weite
dieser
.
nahestehen
man f~r
durchaus werden
] die
recht Hat
in
und
wichtiger
zeigt,
Problem
Untersuchungen
h~tten Ein
~6
dass
Reduktionstheorie
abh~ngen
einer
aus
sagen,
funktionieren:
Rahmen
in
sich
Regel-Schema Motto
hat
Techniken
l~sSt
dem
tischen
a u s s c h l i e s s l i c h mit dem Sequenzenkalkfil g e a r b e i t e t I
des tritt
gruppieren
Haberth~r
ffir
HaberthNr wird
ge-
natNrlichen dabei
Aufwand
R. Deft
an
nicht
mehr
Begriffen
in
notwen-
. wertvolle
Diskussionen
gedankt. Insbesondere tippte,
und
sei ohne
auch
dessen
Herrn Hilfe
Ch.
0ppliger
diese
Arbeit
gedankt, nicht
der zustande
dieses gekommen
Manuskript w~re.
B.
340
i] G . G e n t z e n
:
Neue
Scarpellini
Fassung
die reine
des W i d e r s p r u c h f r e i h e i t s b e w e i s e s
Forschungen
zur Logik
Wissenschaften, [2]
S.C.Kleene
:
Introduction
Neue
S.C.Kleene-R.E.Vesley
: The
(1962)
:
(1965)
Interpretation Functionals in
[6]
[7]
B.Searpellini:
"
Proof
. Ann.
181
[8]
in M a t h .
Notes
K. S c h ~ t t e
:
Second
Consistency
o
Intuitionistic 212
and T r a n s f i n i t e
Systems
, Springer Induction
.
(1971)
.
in Intuitionis-
.
of M a t h . L o g i c ,
Vol.
of Intuitionistic
Commentarii [4
(1969) and
A Model
"
of G e n t z e n s
Lecture
tie S y s t e m s
of Constructive
in Mathematics"
Proof Theory
Annals
by means
Type
(1959)o
Some Applications
Induction
,,
of Analysis
"Constructivity
Math.
of I n t u i t i o n i s t i e
.
of Finite
North-Holland ~]
exakten
.
North-Holland G.Kreisel
der
(1938).
o
Foundations
Mathematics
~4]
& Grundlegung Folge
to M e t a m a t h e m a t i c s .
North-Holland, 3]
fGr
Zahlentheorie.
Math.
Beweistheori~
.
Helv.
4 , No.2
, (1972).
Analysis. ,Vol.
Springer
45
(1960)
, Fasc. o
4
(1970)o
Infinite
Terms and Recursion
H.
Schwichtenberg
Systems of infinite
in Higher
and S.S.
Wainer
terms defining functionals
were first considered by Tait [10] and further Feferman
[3] initially
unpublished
constants
in a proof-theoretic
notes Feferman
terms inductively
introduced
generated
abstraction
context. o
Later in of infinite
of all finite
recursive
and autonomous
of finite type
developed by
the system T
from variables
for the ordinary primitive
application,
Types
types and
functions by
enumeration:
if for each
n, f(n) codes a term tn~ T o and f is itself defined by a term of T O then the t e r m < t n > n ~
N is in T o .
to an arbitrary f u n c t i o n a l ~ a n d denoted by To(~)
.
the resulting
Feferman proved
functions
definable
in~(This
also follows
This definition
that i f ~ i s
in To(~) are precisely
This immediately poses used to characterize
can be relativized
system of terms is of type 2 then the
the functions
recursive
from our results here together with [11]) . the problem of whether
full Kleene recursion
specifically whether, f o r ~ o f
infinite
terms can be
in higher types and more
type n+2, To(T) gives a characterization
of the n+1 - section of ~. We show in w
that for arbitrary ~ o f
of types ~ n+1 definable appearing
in T O (~) are just those functionals
in a naturally- constructed
based on ~, which generalizes recursively "jump").
type n+2 the functionals
[11].
though not necessarily
K l e e n e - type hierarchy (This hierarchy
recursively
The proof of this equivalence
expands primitive
since~may
not be a
uses normalization
for To(~)-
342
H. S c h w i c h t e n b e r g ,
S.S. W a i n e r
As a consequence we obtain a n e g a t i v e answer to the second p r o b l e m above as follows . are p r e c i s e l y
The type n+l
the functionals
Ha n+1 , a E 0 U + 1 [ 5 ] . Ha , aE
0
functionals d e f i n a b l e
in To(n+2E)
o b t a i n e d in K l e e n e ' s hierarchy
But M o s c h o v a k i s
[7] has shovm
that the h i e r a r c h y
does not exhaust the 2 - s e c t i o n of 3E .
In c o n n e c t i o n w i t h the first p r o b l e m m e n t i o n e d above F e ~ e r m a n [4] has r e c e n t l y o b t a i n e d a new d e f i n i t i o n of full r e c u r s i o n in higher
types w h i c h , a l t h o u g h not f o r m u l a t e d as a system of terms, is
n e v e r t h e l e s s m o t i v a t e d by the idea of autonomous we investigate ways of g e n e r a l i z i n g
the autonomous
so as to obtain complete c h a r a c t e r i z a t i o n s (The obvious idea is first to allow d e f i n a b l e functionals f u n c t i o n s as in T o.
In w
sequencing scheme,
of h i g h e r - type r e c u r s i o n
"long" s e q u e n c e s , e n u m e r a t e d b y
of a r b i t r a r y pure But
enumeration.
type, rather
than just
this is i n s u f f i c i e n t as it stands, and n e e d s
to be m o d i f i e d further.)
This leads to a h i e r a r c h y of systems of
terms To,TI,T 2 and Long Partial
Terms,
the last one of which turns
out to be nothing other than a r e f o r m u l a t i o n of F e f e r m a n ' s d e f i n i t i o n
[4] 9 w
The System To(~) of Infinite
Terms.
Type symbols are 0 and w i t h o-,T also ~
T for
(o-1-*(o-2-*...(0-n~V)..)).
symbols are denoted b y ~ , ~ defined)
Let M
A
M
= M
(T -~ ~"
, i.e.
be
the class of all
Mo = N ,
(set-theoretic)
the natural n u m b e r s , and
M , the set of all mappings from M T
into M
0-
U Mqr are denoted b y ~, F , G, H, ~,# and finite
. E l e m e n t s of q-
sequences of them
etc.
We fix a f u n c t i o n a l ~ o f will
a (canonically
T
~-
F,G,~
As usual we write
Finite s e q u e n c e s of type
etc. and we let ~ I b e
code n u m b e r of T .
f u n c t i o n a l s of type T
(0--~7) 9
be b u i l t
up
a r b i t r a r y type ~ .
from v a r i a b l e s
'
x% ,
The t e r m s ......
.
for
of
To(~)
each
H.
type ~ , the s y m b o l ~ , k-th primitive abstraction
Schwichtenberg,
~3
Wainer
and for each k ~ 0 a constant
recurslve
f u n c t i o n , by means
and autonomous
the introduction.
S.S.
formation
Pk for the
of application,
of sequences
as described
Each term will have only finitely-many
in
free
variables.
We define
inductively
(i) a set C ~
N of codes,
term ta denoted by the code a ~ C ~ , (iii) a function for each a g C ~ , Typ a sequence
of variables
for each a ~ C ~ t h e assignment by Typ
(a) determines containing
value
all variables
in t a, (iv)
a type-preserving
of variables
determined
(a).
type of t a and ~--TI,...,T n is to be
the form ~ , ~ w h e r e
thought
sequence
~ = Xl,...,x n of free variables
variable
x~i if T i is the J-th occurrence
this ~ we also write
be clear
that ~ contains
variables
~=~-1,...,~-n
of as determining
in t a (i.e.
ta(~) for t a .
~ is t h e the
x i is to be the
of that type symbol
in ~ ) .
From the definition
it will
all of the free and none of the bound
of t a .
I (Variables)
a = e
.
II (Application)
Typ (a) : ~ , ~ - ~
C ~ if I ~< i ~< n and
' ta : t a ( ~ )
Let a 1,a 2 ~ C % where Typ
Typ(a2 ) =~[,or ~ . Then a = ~2,a I,a2~ ~ C ~ , Typ and
free
of t a under
to the sequence
For each a ~ C ~ Typ (a) will have
With
Typ such that
the type of t a and furthermore
[a] E (in ~ ~ )
of F = F I , F 2 , . . . , F n
(ii) the
[a] ~F = [al]-F[a2] -F "
= ~ i and [ a ] ~ = F i
(a I ) = ~ ~ ,0--~~ and (a)= ~,p~, t a = (tal
H. Schwichtenberg,
344
S.S. Wainer
III (Abstraction) Let a I E C~ and Typ(a I) =r~=m,p~
.
Then a = < 3 , a I ~ C ~ , ta(X)=~y0 ta1(X,y) jSyp Ca).= r~,~_.p~ and [ a ] E G =
[al]E'G for all G ~ M
.
IV (Autonomous Sequences) Let a I E C$, Typ(a I )=r0,0~and for all n, [ a l ] n = b n E C ~ a n d a=
,al>
C%
Typ (bn)=CT,0n.
ta = <
,
Typ
Then (a)=r,r,O -~ -
0"1
and
[aSFn = [bnSE for all n e ~ . V (Primitive Recurslon) Let al,...,anE 0~ where n ~ 0 is the number of arguments of the k-th. primitive recurslve function Pk and for I ( i ( n Typ (ai)=r~,0~~
Then a = < 5 , k , ~ ,
al,...,an> E C~,
ta=Pk(tas,...,tan) , Typ (a)=T,0~and
[a]E=pk([a 11E,..t.Ian] ~)
VI (The Constant ~) a=<6,r1~ e C~ , ta= the symbolS, Typ(a) = rZ, ~ and [alE= ~.
Obviously Typ can be chosen as a primitive recurslve func ti on. To(Y) is the set of all terms ta, a~ C$ .
We want to
normalize the terms of To(~), that is eliminate all subterms of the form (kxt)s.
For a system of nonconstructive infinite
terms (in the sense that no restriction is imposed on the formation of infinite sequences) this was done by Tait [10], extending earlier work of Lorenzen, Novikov and ~ch{Jtte concerning infinite proofs. Tait's paper.
We assume here some knowledge of
Now it is moreor less standard how such operations
on nonconstructive infinite terms can be paralleled by operations on their constructive counterparts such as terms in To(~ ) or, more precisely, codes in C ~ (see e.g. Feferman [41, Lopez-Escobar [6]~Schwichtenberg [9])
9
Hence we do not give proofs but
merely state the proper lemmata, following mainly Feferman [4].
H. Schwichtenberg, S.S. Wainer
345
Most of them (Lemmas I-4) are proved using the primitive recursion theorem.
The type level L T of a type symbol T is defined by L o = 0 , L(~ ~ T ) = max (Lr+ I, L r),
The rank of a code
aE C~is defined as the supremum of the type levels of all subcodes of the form a1=<3,.,.> occurring in a context ~2,al,a2>.
Mo~e
precisely we inductively define Ra for a ~ C ~ a s follows R =
0
R <2,ala2> = m a x (RaI,Ra2,La I) if a I has the form <3,.,.>. = max (Ra1,Ra 2) otherwise. R <3,aI>
= Ra I
R <4,~u, al>= max (Ra I, s~p R[a 1]n) R < 5 , k ~ l, al,...,a n > =
max (Ral,..,,Ra n)
R <6,F~>= 0 . -~,~. Here La I = L ~ where Typ(al)-C
C!early we have Ra ~ ~.
A code a e C~ (and the corresponding term ta) is called irreducible or normal if and only if R a = 0 . Lemma I (Extension) There is a primitive recursive function Ext such that for all a E C}and all types ~ the following holds. Let Typ(a)=~ ,p~. r~,~p~
Then Ext (a,r~~) g C~, Typ(Ext(a,r~)) =
R Ext(a#~)= Ra and for all G,~ of the appropriate
types, [a]E= [Ext(a,ro~] G'E .
Lemma 2 (Interchange)
There are primitive recursive functions PiJ such that for all a E 0 ~ the following holds. Let Typ(a~)=r~ , ~-~ . Then Pij(a) e G~ , Typ (plj(a))= r~lj(~) ,o-q , R P l j ( a ) = R a and
346
H. Schwichtenberg,
for all ~ of the appropriate where ~lJ interchanges
S.S. Wainer
types ~ [a]~= [Pij(a)]~lJ(~),
the i - t ~ a n d
J-th. components in the
respective n-tuple.
Lemma ~
(Substitution)
There is a primitive recursive function Sub such that for all a, b E C~ with Typ (a) =r~ , X ' PN and Typ (b) =r~, ~w the following holds. Sub ( a , b ) E C ~, Typ (Sub(a,b)) = Z~ ' P .I, R 8ub(a,b) ~< max(Ra,Rb,Lb) for all Z of the appropriate
L emma 4
types,
[a][b]F'~= [Sub(a,b)] E 9
(Reduction)
There is a primitive recurslve function Red such that for all m and all a g C~wlth Ra ~< m+l the following holds~ Red ( a , m ) = a ~
~, Typ(a')= Typ (a), Ra ~ ~< m and for a l l F
the appropriate
types, [a I ]F=
~orma~ization
[a]E
of
.
Theorem I
There is a primitive
recursive function N such that for
all a g C~the following holds , N(a) = a ~
C~, Typ(a ~) = Typ(a) ,
a ~ is in normal form, i.e. R a * = 0, and for all ~ of the appropriate
types, [a~]E= [a] -F .
Each term ta in To(~) defines a functional, namely kF.[a] -F, whose arguments occurring in the term. characterization
correspond
to the free variables
We wish to give a recursion-theoretic
of the functionals
definable in To(~), and
since arbitrary finite types can be canonically coded into pure types it will henceforth be more convenient restrict attention ~=a1'''''am
for us to
to those functionals h whose arguments
are of pure types ~< n and whose values are of
type 0. ~ is now assumed
to be an arbitrary but fixed type
and
H. Schwichtenberg,
347
S.S. Wainer
n+2 object.
If h(~1,...,~ m) is definable in To(~) then it is defined by a normal term of type O.
Such a term can only be either
a variable of type 0 or a term of the form Pk(Sl,...,sr) where sl,...s r are normal terms of type 0, or else a term of the form st where s and t are normal.
In this latter case s
cannot be of the form ((SoSl)...)s k with k ~ I since s o would then have to be a variable of impure type, so s must be either ~ o r
a variable of pure type ~ I or a term of the form
where kx.a x is defined by a normal term. Hence t must be either of type 0 or else of the form ky.t' where t ~ is of type 0 (If t were of the form then we could replace it by ky. Y) 9
Thu~ it
is clear that each of the functionals h(a1,...,~ m) definable in To(~) can be generated by means of the schemes 4,...,7 below. The converse,
that the functlonals generated by schemas I,~..,7
are all definable in To(~), should be clear and can easily be proved by a simple application of the primitive recursio~ theorem. Each scheme defines a functional h e where the index e codes up (in the usual way) all relevant details of th~ particular being applied.
We now let ~ = xl,...,x k denote variables of
type 0, ~ = ~ 1 , . . . , ~ m
variables of pure types ~ n and # a variable
of the appropriate pure type ~ n 9
1.
he(~,~)= Pk(~)
2.
he(~)
=r
3.
he(S)_
= ~ j ( ~ . h e l (~,~))_ where type of ~ j > 1,
4. he(g) 5.
scheme
where type ~i = I .
= }(X~.he1(a,~)) )
provided that for each x, he1(X) is an index for a functional with arguments ~ .
~8
H. Schwichtenberg,
S.S. Wainer
6.
he(s):~ he1(he2(g), 5)
7.
he(~)= he1(~' ) where ~' is some permutation of ~ . To be precise,
the above schemes should be interpreted as a
simultaneous inductive definition of a set of indices e, and for each index e a functional h
We believe however that the
e
intention is clear. w
The~-hierarchy. We now develop a recursion-theoretic hierarchy based on a
fixed but completely arbitrary type n+2 object ~, and prove that the functionals of type ~ n+1 appearing in the hierarchy are precisely those functionals definable in To(~) 9
The hierarchy is
just a generalization of [11] to higher types. Let lelF(~) , e < ~ , be a standard enumeration of all functionals (with arguments ~ of type ~ n) primitive recumsive in a type n+1 object F (in the sense of Kleene [5]).
We assume ~elF(~)= 0 if e
is not an index for a functional of the appropriate string of variables. We associate w i t h ~ a n ~(F)
(<x,~>)=
operator~defined
as follows
<[x~F(~,on),~(X#.[x~F(~,#))>
The ~-hierarchy is then obtained by iterating ~ o v e r a simultaneously generated set of ordinal notations.
Note however that the word
"hierarchy" is used in a rather broad sense here, since ~ m a y not be a jump operator in the usual sense (and a l t h o u g h ~ r a i s e s recursive degree" it need not raise "degree").
"primitive
As a result of this
our hierarchies will not in general have the uniqueness property. Definition.
I 1
and
for a c O ' a r e
inductively defined as
follows, where ~,# are variables of type n. will usually drop the superscript ~ )
(Since ~ i s
fixed we
H. Schwichtenberg, S . S . (1) I E 0 ,
"/(b
Wainer
~9
9
(ll) If a E O t h e n
2 a E o , b) =<~x]a(=, on),~(kp.~x| a(~,#))~
where On here denotes the zero type n object. (Ill) If a E 0 a n d , = ~e~ Fa is a function such that , ( 0 ) = a , , ( m ) e and $(m)) = F~ (x) (=) " Clearly if a
Examples (I)
I f ~ i s of type 2 then the above hierarchy exhausts the
l-section of ~(see (2)
[11]
)9
If ~ is the functional n+2E which introduces quantification
over type n then the above definition gives an alternative version of Kleene's proposed hierarchy of hT"per-order n+1 predicates ~ ] . Our definition differs from Kleene's particularly in the formation of limit levels, where we insist that fundamental sequences ~ b~ primitive recursive (rather than Just recursive) in previous levels ~ However standard methods show that the two definitions give rise to the same class of predicates and functlonals (and coincide at limit stages).
~9schovakls KT~ has shown that~ for n = I , the
hierarchy does not exhaust the 2-sectlon of 3E (nor the l-section of 3E). (3)
If~Is
the superJump functional we obtain an alternative
version of Platek's hierarchy [82 but again, Aczel and Hinman 11~ have shown that this does not exhaust the 1-sectlon of the super Jump.
0
350
H. Schwichtenberg, S.S. Wainer
Limit Property. There are primitive recursive functions M and N such that if for each m,k~.G(m,g)= ~$(m)~ =r F F , , = ~e~ and a=r ~o r 3asM(e)E0, r
where
then
Proof Choosa
F F M so that [M(e)~ a(0)=a, [M(e) 1 a(m+1)= 2r
.
Let <~a>n denote a standard primitive recursive coding of a sequence Z as a single type n object, and lets o be a primitive recursive function such that ~So(J)IF(n, on ) = (F2r (m)(<8o (~ (m)) '<~>n>))o = (F3asM(e)(<m+1,<So(~(m)),>)) 0 ~ NOw let m n denote the type n object with constant value m and let S I be a primitive recursive function such that IS1(J)JF(mn,o n)= ~J]F(m) for any type n+1 F object F. Then ~(m)= li~ a(m)= ~S1(i)~Fa(mn,0 n) = (F2a(<Sl(1),mn>))o = (F3asM(e)(<1,<S1(i),m~>>)) o.
We therefore have
S(m,~) = (F3asM(e) (<m+ I ,<8 o (F3aSM (e ) (> ) )o'n>> ) )o and it remains to choose N so that N(i) is an index of this expression as a function of m and ~, primitive recumsive in FSasM(e)" Lemma ~. There are primitive recursive functions I and C such that if e is an index of a functional h e defined by schemes I,...,7 then for any b E 0 ,
C(e,b)g0,b
Proof First note that the arbitrary bg0appears because in order to deal with scheme 5 we need to locate k~.he(X+l,m) above
H. Schwichtenberg, k~.he(X, ~) i n O s o
S.S. Wainer
351
that the Limit Property can then be used
to piece together the whole functional kx,~.he(X,~).
Also in
dealing with scheme 6 we will need to locate h
above h . eI e2 These complications arise because there is no corresponding
Uniqueness
Property for an arbitrary~-hierarchy,
since
Uniqueness requires quantification and we do not in general have 2E recursive i n ~ .
I and C will be defined simultaneously by the primitive recursion theorem, with induction on the definition of h e by schemes I,...,7. Suppose h e is defined by I, so from e we can find k so that he(X,~) = pk(~) . Clearly there is a primitive rgcursive function fl such that fop any F of type n+1 , ~f1(k)]~(x,~)=pk(~).
Thus
we only need to put I(e,b)= f1(k) and C(e,b)= 2b in this case. Suppose h e is defined from hel by 2,3, or 7.
By induction
hypothesis we can assume b
If h e ( ~ ) = ~ k ~ . h e 1 ( ~ , ~ ) ) by scheme 4 then again by hypothesis we can assume bn,~) . Therefore he( ~ ) = (k~. If2( e, I(e I ,b) ) ~FC(el ,b) (<~>n,~)) = (F2C(e I ,b) (
352
H. Schwichtenberg, S~
Now put C(e,b)= 2
C(el,b)
Wainer
and h e is clearly primitive recursive
in FC(e,b) with index I(e,b~ primitive recursively computable from e and I(el,b).
Next suppose h e is defined by scheme 5 9 Then he(X,~)--hhe~ ) where, by the induction hypothesis, I
b
Define ,(0)=C(el,b)
and , (m+1) -- C(hel (m) ,, (m)) , and define ~(0)= O, ~(m+1)= I(h e (m),,(m)) . Then for each IF m,ka.he(m,~) = [~(m+1)~ ,(m+1) where , and ~ a~re primitive recursive in FC(el,b) with indices z and i primitive recurslvely computable from C(el,b), I(el,b) and primitive recursive indices of C and I.
Also ~(0)= C(el,b) ~
every m, by hypothesis.
Therefore by the Limit Property, h e
is primitive recursive in FC(e,b) with Index I(e,b) where C(e1'b?sM(Z ) C(e,b)= 3 and I(e,b) is given by a simple primitive recurslve function of N(i).
Finally suppose he(~)= hel(he2(~),~) by scheme 6. By induction hypothesis we can assume b < o C(e2,b), he2= |l(e2,b)l
C(e2'b) and for all d~O, d
he1= ~I(el,d)~C(el 'd) .
Define ,(0)=C(e2,b),r
and $(m+2)= 2~(m+1) . Then ~ is primitive recurslve (and hence primitive recu~sive in FC(e2,b ) with an index u primitive
H. Schwichtenberg,
S.S. Wainer
353
recursively computable from e,b, and a primitive recur sive index of C 3C(e2'b)su
Also r 0"
<
o
r
c(
Put C(e,b) = 3
by hypothesis and so e2'b)5u.
Then b < o C(e2,b)< o C(e,b)
and since FC(e2,b )=k~.FC(e,b)(<0,~ >) it follows that he2 is primitive recursive in FC(e,b) with an index primitive recursively computable from I(e2,b).
Now for some fixed
primitive recursive function f3 we have
hot(X,~) = II(el,C(e2,~))~F~(1)(x,~) = ~f3(l(e1,C(e2,b)))~Fr
n, O n )
= (F,(2)(n>)) o = (FC(e,b) (<2' n>>) )o Thus hel is also primitive recursive in FC(e,b) with an index primitive recursively computable from e,b and primitive recursive indices of I and C.
Hence h e is primitive recursive
in FC(e,b) by Kleene's scheme S&, with index I(e,b) given as a primitive recurslve function of I(e2,b), e , b , and primitive recurslve indices of I and C.
We give I and C the value 0 if none of the above cases applies. Inspection of the above cases shows that C(e,b) and I(e,b) are defined simultaneously from C(el,b) C(e2,b), I(el,b) I(e2,b),e,b and primitive recursive indices of C and I.
Since el,e 2 < e the
simultaneous definition is a primitive recursion on e.
Therefore
by the simultaneous primitive recursion theorem (e.g, Lemma 2.1 of [2]) we can indeed find primitive recurslve indices of C and I which satisfy this definition ~ This completes the proof.
354
H. Schwichtenberg,
S.S. Wainer
Next we show that every functional G(~) , with arguments ~ of pure types ~ n and with values of type O, which appears in the~-hierarehy, is definable by a term of To(~) 9 Lemma 6 There are primitive recursive functions p and Pl such that if the type n+1 functional F is defined by a term t c of To(~) thenle] F is defined by the term tp(c,e) of To(~) and x,~.
Ix~F(~) is defined by the term tp1(c) of To(~) .
proof We first define p by the primitive recursion theorem with cases corresponding
to the schemes 8o,...,$8 by which ~e~ F
is defined. In this proof and the next, u,v will be used to denote ~ariables of To(~) of the appropriate types ~ If ~e~ F is defined by $I,$2,$3 then [e~ F is Just a primitive recursive function of its numerical
arguments and
so p(c.e) is given explicitly as a function of e .
If ~elF=k~.~eIDF(le2~F(~),~) assume inductively that tp(c,el ) defines ~e2 ~F.
through S~ then we can defines le1~F~and tp(c,e2 )
Therefore ~e~ F is defined by the term
~ . tp(c,el ) (tp(c,e2)~)~ and we can clearly compute p(c,e) as a primitive recursive function of p(c,e I) , p(c,e 2) and e.
If ~el F is defined by $5 then lelF(0,~)= ~e1~F(~) and lelF(x+1,g)= le2~F(le~F(x,~),x,~) inductively that tp(c,el )
where again we can assume
defines ~e1~F and tp(c,e2 ) defines
le2 ~F . Now let r(O)=p(c,e I) and r(x+1)= the code for the term ~ .
tp(c,e2 ) (tr(x) ~) x ~ ~
Then for each x, tr(x)
H. Schwichtenberg,
S.S. Wainer
defines k~.~e~F(x,~) and therefore < t r ( x ~ x ~
355
defines le~ F .
But r is primitive recursive, with index i primitive recursively computable from p(c,el)
p(c,e2) and e.
Hence we can primitive
recursively compute from i, first a code for the term defining r, and then the code p(c,e) for the term < t r ( x ) > x E g
which defines ~e~ F.
The cases where ~e~ F is defined by S6 and $7, corresponding to permutation of arguments and function application, are trivial.
If lelF(g)=~i(k#.~e1~F(~,#))
through $8 then it is easy
to define p(c,e) primitive recursively from e and p(c,e 1) such that t p ( c , e ) = k ~ ,
ui(kV.tp(c,el)~V)
9 The case S0 is treated
similarly, replacing ~i by F and u i by t c 9 It is clear from the above cases that p is primitive recursive, as required.
To define Pl simply note that k x~. defined by the term <tp(c,x)> x E g ,
Ix~F(~) can now be
whose code is given as
a primitive recursive function of c.
Lemma
7
There is a primitive recursive function q such that if a~ O ~ t h e n
q ( a ) E C ~ and tq(a) defines F ~ a
q
Proof Again by the primitive recurslon theorem. so that t q ( 1 ) = k t ~ O
.
Define q(1)
Now assume tq(a) defines F a.
Since x = < x , ~ > o (0) and ~ = < x , ~ > 1
there are terms tk and t$
356
H. Schwichtenberg, S.S. Wainer
which define the decoding functions k~.~o(O) and k~.a I 9 Fa P But F 2 a = k a . < l ~ o ( O ) ~ (~I,0 n),~(k#.I~o(O) ~ a(a1,#))> and so F2a is defined by the term ku. <tpl (q(a))(tku)(~u)0n' ~(kv. t l
(q(a))(tku)(t~u)v)> whose code q(2a) is clearly given as
a primitive recursive function of q(a~. ks. F e~Fa(~o(O)) (~i) , so if r inductively that Fr
If 3a5 e g O t h e n F3ase=
~e~ a we can assume
is defined by tq(r
) for each x and
therefore F3ase is defined by the term ku.
~ (tkU)(~u) 9
Now r is defined by the term tp(q(a),e ) and so kx.q(r
is
defined by a term whose code is primitive recurslvely computable from q(a), e and a primitive recurslve index of q .
Thus we can
compute q(3a5 e) primitive recursively from q(a), e, and a primitive recursive index of q, so that tq(3ase ) is the term ku.
which defines F3a5e ~
The
primitive recursion theorem then provides an index of q satisfying the above definltlon, and this completes the proof.
Putting the above results together we have Theorem 2 A functional with arguments of pure types ~ n and values of type 0 is definable in To(B) if and only if it is primitive recursive in F~a
for some a ~ 0 ~
Corollary If~Is
of type ~ 2 then the functions definable in T0(~)
are precisely the functions recursive i n ~ .
But for ~ o f type >i 3 the functions definable in TO(~) do not, in general, exhaust the 1-sectlon of ~ .
H. Schwichtenberg, @3.
Extensions
S.S. Wainer
of To(~)
The reason why To(~) f o r ~ o f Kleene recursion
in'seems
type level ~ 3 does not give full
to be that sequences used to build up
terms in To(~) are indexed
by natural numbers and so each term
can be regarded as a countable well-fotmded Kleene-computations Thus it is tempting
in types ~
3
tree , whereas
are in general uncountable.
to allow sequences
objects and to consider defined
$57
indexed
by h i g h e r - t y p e
a system T I(B) of infinite
just as To(~) in w
except
terms which is
that clause IV is now generalized
to read as follows IV ~ (Long autonomous
sequences)
r~ 01 and for all ~F~ M Then a - ~ 4 , a l " ~
, [a I ]F = b F ~
But if t a = ~ t b ~ F ~ M ranges over M of b F
and Typ(D F)=Z,r 0 u
F~ G types, [a]~F=
[bF]~G
is a term formed by IV * then as
there can still only be countably- many different .
Thus the "depths"
terms in TI(~) remain countable, adequate
c~
C~ , t - a = ~ t b ~ p ~ M - ~ T ~ p ( a ) = ~ , ~ ~ 01 and f o r
all F,G of the appropriate
values
Assume a I ~ C ~ , Typ(a I ) =
of the trees corresponding so we cannot expect TI(~)
to define all functions recursive
in~.
to
to be
In fact for the
case ~ = 3E we have : Theorem The functionals those definable
of type ~ 2 definable
in TI(3E ) are just
in To(3E).
Proof For i = 0,1 we let C ~ b e and for each a ~ C~ we denote XE.
[a]~ .
the set of codes for terms in Ti(~), the corresoonding
functional by
We show that there is a primitive recursive function p 3E such that if a ~ C I is normal and Typ(a)= r ~ , ~ where ~ is a 3E sequence of types 0 or I, then p ( a ) ~ C o and for all
~8 ~e~
H. Schwichtenbe~g, , [a]~ : [p(a)]~ .
The only non-trivial
a = 4 4 , a l b , Typ(a)= r~,z-, 0 ~.
E
E
S.S. Wainer case is when
Then for all
9
3E
roceeding by induction on
we can then assume that [a]~ ~ = [p[p(al)]~] I .~ow using the function-quantifier
3E we can primitive
p(a I) and a primitive
~n.[b] n
enumerates
[c]~ = # n
recursive
all the values of }~
= [~4,b~]~[c]~
compute,
p([P(al)]o~)
and
Then for all ~,# ,
, and so from b and c we can primitive
recursively compute p a5 such that We finally obtain
Eal
the required p by the primitive
Clearly this Theorem will hold for a n y ~ s u c h definable
in To(~) , and it will also generalize
when relativized
from
index of p, codes b and c such that
([hi n = p([P(al)]~o)).
[a]~_
recursively
recursion
theorem.
that 3E is to higher
types
to 4E,SE etc.
-~tb~ F M
The depth of a term t~4,ai~-
the obvious way by depth(t~4,al>)=
formed by IV ~ is given in
supF(depth(tal)+l,depth(tbF)+1)
and since each b F = [a I ]~g C ~ we are here only taking the supr~mum of c o u n t a b l y - m a n y
(countable)
terms with uncountable
ordinals.
depth is to allow the ~F's to be used as
constants
in t bF, i.e. to let bF~_ C }'-F .
suggestion
of Feferman,
give new systems
we further
extend our systems
types,
define,
simultaneously
fixed ~ = ~ , . . . , F n .
F 1,...,Fn)
for all F of the
[a;F] ~ defined by that term.
[a;~] ~ in order to make explicit
and VI as before
of terms to
a set C ~ of codes and for each a ~ C ~ a term
ta ~ T2(F) and a total functional ~ . We write
Thus, following a
T2(~) as follows.
This time we inductively appropriate
Now a natural way to get
The clauses
in the definition
(but with VI introducing
together with
the relativization
to the
are I,II,III,V
each of the constants
H. Schwichtenberg,
IV ** (Long relativized autonomous
sequence s )
type Z , Type (a I)= ro-,0~ and for all G~ ~ , and Typ(b G)=
oI .
r p_,
Then a = ~ 4 , a l > ~
~9
S.S. Wainer
C -F, ta
Assume a I ~ C-F, F of [a I;F] ~ = b G ~ C~F'~
:
,
Typ(a)= rp~,~ _. O~and for all G e M , HeMp~ [a;~]~
: [ [ a ~ ; F ] -G ; ~,G] ~
.
With F the empty sequence we thus obtain C and T2 , so if we denote the depth of a term ta in T2(~) by laI ~ then the depth lal of a term ta= ~
Ial where
in T 2 is given by
~E~
= sup
(la~l
+ ~ ,
IbGI ~ may now, of course, have uncountably many different
values,
and so lal will in general be uncountable
1,2 in Moschovakis
(cs
definitions
[7]).
We shall show (Theorems 4 and 5) that for arbitrary a with Typ(a)=~Uthe
partial functionals k~.[a;~]
are just the Kleene
partial recursive functionals k~.lel(~) by the lemma below,
It then follows
that the total functionals
exhaust the functionals recursive
[a;~] with a ~ C~ ,
in~.
Lemma For each ~,~
there is a primitive recursive function f such
that (with F,~ ranging over ~ , ~ , (i) ~ 2 ( a ~ C ~'g)
respectively)
~-~ f(a)e C ~
(ii) ~G(a~ C~'~)
~ [a;~,G]~= [f(a);F] ~'~ & lal~'~ < If(a)1z
Proof Given ~,~ , we can easily find a primitive recursive function q such that for a l l ~ , [a;F,G] ~
=
~Mo~,[q(a);F]~= a .
Hence
[[q(a);~]~; F,~]~ [~4,q(a)> ; ~ ] ~ [f(a);Z]~'~
by IV *~
360
H. Schwichtenberg,
with f(a) lemma
depending
primitive
S.S. Wainer
recursively
on q(a).
The proof
of the
is now obvious.
Theorem 4 There
is a primitive
(i) (ii)
recursive
function
g such that
Iet(~)~ ~ g(e)~ C~ Ie](~)$ ~
[g(e);~]=
tel(g)
Proof We shall define the primitive is by cases
g from its own primitive
recursion
depending
The implication proved by induction
theorem
in the usual manner.
for left to right on lel(~)~ w .
to left in (i) is by induction
after
the definition
Case
function
(ii) are
The proof of the implication on Ig(e)I ~ and will
to the cases
= le11(~21(~), ~)
that as in w
Sub such
in (i) together with
from
be clear
$4, $8 and $9, the other
or similar.
84; lel(~)
First note
The definition
is completed.
ourselves
cases being obvious
index using
on the form of e.
right
We restrict
recursive
we can easily
9 obtain a primitive
recursive
that b ~ C ~ implies
(i) [a;[b;2]~;F] ~ ~ [Sub(a,b) ; 2 ~
(il)
lal[b:= ]e '=< [Sub(a,b)lZ and Ibis< ISub(a,b)~.
(However, usual ~2,~2,
note
that if SUbo(a,b)
to t e r m - s u b s t i t u t i o n
is the function
we have to put
corresponding
Sub(a,b)
=
Co, SUbo(a,b)~ , b ~ with C o ~ C -F such that [ C o ; F ] G H I H f = H I )
We now obtain
tel(c~)= [ g ( e l ) ; [ g ( e 2) ; ~],_~] = [Sub(g(el),g(e2)) Hence
it suffices
by ind.hyp.
; ~]
to put g ( e ) = Sub(g(es),g(ef) )
9
as
9
H. Schwichtenberg, Case
88:
S.S.
~el(~) ~- ~j(X~e11(_~,~))
9
g ( e l ) E C ~'~ and le11(~,/~)= [g(el);Z,#] lemma~le 1 ~ ( ~ , # ) = from g(e I).
By ind. hyp.
recursively
It remains
to set g(e) 89:
= [a2;~_]
9
compute
= [a};~]
= a3 9
lel(x,~)
= [g(x);z]
can easily
-~ Ixl(~)
9
By ind.
Now from a primitive
hyp. we can assume
recursive
But then an application
index of g we
of IV m* yields < 4 , a ~
; ~]=
Ixl(~)
C x'~ such that
x , ~ ] = [[a2;x,G] ; x,~ ] = Ixl(_~) and it then remains
to put g(e)=~/4,a2~2
simply
9
~.
There is a primitive (i)
that
a code a I e C x'~ such that [ a l ; x , ~ ] = g(x) and then
a code a 2 E C x'~ such that [[a2;x,z] ; x , Z ] = [[al;x, ~]
Theorem
computable
from a I such that
~j(k#le11(a,E) )
[~4,a2>;
we have
for all # , and hence by the
[a lIZ] p with a I primitive
X~eI~(Z;~)
Ixl(~)
361
It is now easy to obtain a2,a } also primitive
recursively
Case
Wainer
recursive
function
h such that
a e C~ ~* l h ( a ) l ( ~ ) ~
(ii)
a e C~
-* ~h(a)l(~)
= [a;~]
It is fairly straightforward primitive
recursion
Since functionals method terms.
the treatment anyway,
scheme
involved
codes as values.
the
first
under which
of partial
for a more direct
in the context
this is to return
functional
a discussion
to look
recursion
the conditions
such anh using
the details.
IV * may be applied,
that the enumerating
general be partial.
of T2(~)
partial
One way of doing
sequencing
we omit
it seems natural
of introducing
T I (~) and then relax
defined
theorem;
to define
of infinite
to the system the autonomous
by not requiring
any longer
given by a I has only previously
The functionals
so defined
But not only [a] ~ as a function
will now in of G will be
9
362
H. Schwichtenberg.
partial
S.S. Wainer
(as we would like) but also the values [a]~ for certain
fixed ~ may be partial functionals and as such will not even be objects of our underlying domain U T ~ we instead let ta be the term<%2F
To avoid this difficulty M
with
a sequence
of
variables of type ~ , so that the values of [a] ~ , when defined, are natural numbers
(i.e. total objects of type 0) .
This leads to a
system of infinite "partial" terms ta,aE C defined by I, II, III, V and IV *** below (We no longer relativize t o , s i n c e necessary here.
it is not really
One can easily show, for this new system,
is a primitive recursive function ka.a'
that there
such that if a E C~then
a ' ~ C and for all F [a'] ~'~ = [a] ~) . IV ~**
(Long partial autonomous sequences) Assume a I ~ C and
Typ(al)=
fT.,On.
Then a = < 4 , a 1 > ~
C and ta = < t b ~ F ~
b F = [a I]~ and tbF is undefined if bF r C.
where
Furthermore T y p ( a ) = } ,
and [a] E is defi~ed with value m if and only if (i) [al]~ is defined, (ii) [al]~ : b F ~ C
with T y p ( b F ) = } , 0 u, and (iii) [bF] ~ is defined
with value m. Now in what sense do I,II,III,IV ***, V constitute a definition of the concepts a ~ C,t a and [a]~?
The formerly critical point in
the inductive definition of C was the use of quantification over M
in IV (with ~ = 0) and IV*,IV ** (with T arbitrary), which meant T I that C was "at least" a complete H 1 set. But this clause has now
been removed to give IV * ~
and so the new C can be defined independently
of ta and [a] ~ , and is simply primitive recursive of indices for partial recursive functionals).
(as is the set
Incidentally the
primitive recursive function Typ also needs to be redefined so that Typ(<4,a1> )= Typ(al).
We next consider [a] ~.
Since [a] ~ may
now be undefined we need to give a definition of the relation [a] ~ ~ G, to be read "[a] ~ is defined with value G".
This relation
is clearly analogous to Kleene's lel~(~) ~ z and is given by the
H. Schwichtenberg, following
induction
(I)
Variables.
(2)
Application. G2~ M
(3)
S.S.
Wainer
S~
:
r ~ , F: FI,. ..FnE M
[a] E -~ F i if a = ~ 1 , i ,
[a2 IF -~ G 2 where
If [a I]F ~ GI and _
and
_~p and
G I~M
then [a] ~ -" GIG 2 where a = ~ 2 , a l , a 2 ~ .
Abstraction.
If [a I]F'G _~ HG for all G ~ M
then [a] ~-~ H where
a = ~ , a1"~ .
(4)
Long partial
autonomous
then [a] ~ -- m where (5)
Primitive
sequences.
a--~4,a1~.
Recursion.
If [ai]~F -~ m i for I ~ i ~ n k then
[a] ~ -~Pk(ml,...,mnk ) where
a = ~ 5 , k , r ~ ~, al,...,a_~,nk _F~M~
Pk is the k - the. primitive For details.
the "partial"
Notice
a=~4'a1~and
quite arbitrary in particular may have
$9).
the problems
x where b F - [al]~
which
t a starting
(where
Since a I is
the values bF
tree
;
the structure
occurs
of [a] -F from given a,F node.
In such
in the case of
in the case of sequencing
of t a in this case has an infinite branching).
at an inductive mentioned.
definition
functionals .
(I)...(5)
This definition
point of [4].
(as is done in [4]) or by reduction
functionals
to
of ta has only a l-fold branching),
occurs
and is the starting
that the partial
(analogous
ta
through Kleene's
from the outermost
the structure
terms are not explicitly
recursive
about
can arise
an infinite b r a n c h i n g
We have arrived
directly
can arise when
~
but only a 2-fold branching
Feferman,
which
One can think of a computation
a computation,
(whereas
C we omit corresponding
of a non-well-founded
computations
and
function.
t a for a ~
we do not know anything
through
abstraction
terms
F~M
the structure
as working
recursive
we may have bF = a for some F and so in general
the undefined scheme
however,
ta : ~
If [all -F -~ b and [b] -F- : m
X~.[a] -a exhaust
in which
is due to
One can show either to Theorems
4 and 5,
the Kleene partial
3~
H. Schwichtenberg,
S.S. Wainer
REFERENCES. [I]
P.Aczel and P.G. Hinman, Generalized Recursion North-Holland
[2]
S. Feferman,
"Recursion in the Superjump",
in
Theory (Eds. Fenstad and Hinman),
(1974). "Classifications
of Hierarchies",
of Recursive Functions by means
Trans. Amer. Math. Soc. vol 104 (1962)
pp. 101-122. [3]
S. Feferman,
"Ordinals and ~kmctionals
in Proof Theory", Proc.
of Int. Congress of Mathematicians Nice (1970), pp. 229-233. [4]
S. Feferman,
"Recursion in Total Functionals
of Finite Type",
to appear. [5]
S.C. Kleene,
"Recursive Functionals and Quantifiers
of Finite
Types I, II", Trans. Amer. Math. Soe. vol 91 (1959) pp. 1-52, vol 108 (1963) pp. 106-142. [6]
E.G.K. Lopez-Escobar,"Remarks Constructive
Formulas",
on an Infinitary Language with
Journ.
Symb. Logic vol 32 (1967)
PP. 305-319. [7]
Y.N. Moschovakis,
"Hyperanalytic
Predicates",
Trans. Amer Math.
Soc. vol 129 (1967) pp. 249-282. [8]
R.A. Platek,
"A Countable Hierarchy for the Superjump",
Logic Colloquium (1971)
[9]
'69 (Eds. Gandy and Yates) North-Holland
9
H. Schwichtenberg, "Elimination of Higher Type Levels in Definitions of Primitive Recursive Fnls.by Transfinite Recursion to appear in Proc. of Bristol Logic Colloquium 1973 (Eds. Rose and Shepherdson),
[lo]
North-Holland.
W.W. Tait, "Infinitely Long Terms of Transfinite Formal Systems and Recursive Functions Dummett) North-Holland
[11]
in
S.S. Wainer,
Type", in
(Eds. Crossley and
(1965).
"A Hierarchy for the l-Section of Any Type Two
Object", Journ.
Symb. Logic vol 39 (1972) pp. 88-94.
Consistency Dedicated
Proofs and Ordinals
to Kurt Sch~tte
on the occasion
of his 65th birthday Gaisl Takeuti
Proof theory began with Hilbert; Hilbert's program.
more precisely,
it began with
Much has been written about Hilbert's program,
by Hilbert and by others, and a variety of interpretations objectives
are possible.
school was doing, Hilbert's
Nevertheless,
from Hilbert's
reaction
influence
to the mathematical
its consistency
Hilbert's
on other people,
events of his day, it is my view
a +b
= b +a
finite standpoint.
sive operation.'
1 + 1 + l,
given before us, and
A consistency
proof,
For example,
we
form.
in the finite standpoint,
"Suppose a proof-figure
interpret
'finite operation'
usu-
to a
By a 'general
statement'
free deduction.'
to mean 'primitive
However,
his finite standpoint his standpoint
Such people
there is reason to believe to encompass Ackermann's
is beyond quantifier-free
and lies close to Gentzen's
standpoint.
this interesting historical
recur-
of a Gedaaken experiment,
finite standpoint with quantifler-free
investigate
Ne can finitely
for arbitrary numerals a and b from Hilbert's
they mean a 'quantifier
arithmetic.
by
is given before us."
Many people
Hilbert's
for mathematics
as a Gedanken experiment.
ally starts with the following contradiction
is the following:
figure like
infer a general statement can infer
and from
from his finite standpoint.
finite standpoint
operate on a concrete
of Hilbert's
Judging from what Hilbert's
that his main objective was to provide a foundation proving
both
then identify
primitive
recursive
that Hilbert
function.
primitive
intended
If so, then
recursive
arithmetic
I would hope that someone might question
concerning Hilbert's
366
G. Takeuti
standpoint
and Ackermann's
function.
From a modern viewpoint Hilbert's program sounds rather strange and there are at least two reasons for this: 1.
In Hilbert's
set theory. confidence
day people
really worried about contradictions
But today people have great confidence is based partly on their experience
in set theory.
in This
and partly on habit and
simply not thinking about the subject. 2.
GSdel's
incompleteness
Hilbert's program completely. proofs now require a method is nevertheless impossible
theorem has changed the meaning Because
that is finite
(or constructive)
very strong when formalized.
lies completely within Hilbert's of the accessibility
from transfinite
not transfinite
of Peano's arithmetic.
finite standpoint
induction,
sequences
of ordinals.
to extend that of Hilbert's.
standpoint
in the following way: figures,
experiment
Thus by Gentzen's
S,
is
it was necessary
for his
Let me formulate Gentzen's
We can finitely operate on a concrete
and
..-.
standpoint
The part
S
we can finitely operate
that is given before us.
be primitive
the order relation,
addition,
state-
... will be explained
further we need the following
Let a system of ordinals i.e. the set
Accessibility
given before us, and infer a general
initial part of a sequence standpoint
e 0.
that we need for consistency proofs.
standpoint
ment as a Gedanken
except for his proof
induction and it is usually accessibility,
Since Gentzen had to deal with sequences
sequence of concrete
His proof
of the ordinals we mean that there are no
strictly decreasing
Gentzen's
The situation
of the ordinals up to the first ~-number,
By the accessibility
finite
think this is
the continuum hypothesis.
Gentzen proved the consistency
later.
People
but which
similar to that of finding a new axiom that carries convic-
tion and decides
different
result consistency
or at least unlikely and extremely difficult.
is somewhat
infinite
of GSdel's
of
on any To discuss
definitions.
recursively presented, multiplication,
and
G. Takeuti exponentiation,
etc. are primitive
be a concretely given sequence Definition. al,a2 9
,
867
recursively
of elements
described.
of
Let
al,a 2, ....
S.
An element a n is a terminal point of the sequence
if and only if
l)
the sequence
2)
a n ~ an+ 1
terminals at
Definition.
A sequence
if and only if
~1,~2,...
an ,
or
~1,~2,...
is a fundamental
is a strictly
increasing
sequence
for
sequence and
is the limit of the sequence. By a p r o o f a method
M
of the a c c e s s i b i l i t y
of the ordinals If
F r o m the following less than
a I < ~,
Mn
Let
with
al < ~n"
then it is sufficient
to check
of
concretly
the a c c e s s i b i l i t y
a I +l
initial
al,a 2, ....
be a fundamental
sequence
for
~
and let
for finding a terminal point
for every sequence
Then we can easily construct
a method
terminal point for every sequence then we can easily find a fundamental
for every sequence
sequence
for
of
~0"
~l < ~2 < "''
be a method
we mean the c o n s t r u c t i o n
facts we can prove
terms in order to find a terminal point 2)
S
for finding a terminal point
g i v e n before us.
l)
of
~n ~.
al,a2 9
with
with
al < ~n
We then apply
to find a
a I < ~:
because Mn
M
to
al,a2,---
If
a I <~,
~1,~2,... al,a2,..,
is a and
find a terminal point. 3)
There
fundamental
is a very simple
sequence
systematic method
for any limit ordinal less than
A proof of the a c c e s s i b i l i t y as follows:
for c o n s t r u c t i n g
By l) we prove
that
of the ordinals ~
~0"
less than
is accessible.
r
Repeating
finite number of times 9 we prove
that
~, ~ + m, J ~ + ~ + m, ...
accessible.
that
m2
this
9
Then by 2) we prove
we prove
less than
e0
that
~2
is accessible.
a~, ..., ~ ~ ,...,~ ~ ,-..
is accessible.
and
a
goes l) a are each
Repeating
hence each ordinal
G. Takeuti
368
This proof is very clear and transparent the primitive theless,
recursive
structure
if one is familiar with
of the ordinals
~0"
Never-
there are many people who argue that Gentzen's proof does not
increase our confidence
in the consistency
of Peano arithmetic.
one believes from the outset that the consistency is obvious,
then of course this is the case.
one that I share for the following definition l, l + l ,
less than
of the natural numbers.
l+l+l,.--.
reason.
of Peano arithmetic
This view is however not There are two choices for the
One is a recursive
The other is a definition
definition
With the first definition we are lead naturally of quantities
problem of foundations
proof, which consists
of the elimination
bility proof for the ordinals does add to my confidence
difficulties.
of what the
Anyway since I a m a
am very familiar with the magic of quantiflers
less than
Gentzen's
of quantifiers
logician and consistency
and are accessi-
e0' is greatly reassuring.
in the consistency
Nith
to a conscientious
and here we become conscious
really entails.
i.e.
inside set theorem.
the latter definition we encounter many deep foundational
interpretation
If
It
and truth of Peano arithme-
tic. Although malization
the accessibility
is not simple.
The same method
and the number of repetitions becomes
larger and larger.
the 'furthermore' finitely operate operations
proof described
is repeated many, many times
becomes higher and higher as the ordinal
Analyzing
part of Gentzen's on concrete
above is clear its for-
this situation we can now complete
standpoint:
operations,
etc. and infer a general
"Furthermore
concrete
operations
we can on concrete
statement about them, as a Gedanken
experiment." About 25 years ago, I first sought to carry out Hilbert's program through cut elimination the following 1.
in type theory.
This effort was organized around
ideas:
I noted that combinatorial
been very successful
reductions
in all my examples.
to eliminate
Consequently
cuts had
I hoped to obtain
G. Takeuti a proof of the cut elimination
369
theorem that would be very combinatorial
and hence constructive. 2.
Cut elimination
interesting mathematical Sch~tte
implies the consistency problem
investigation
theorem that lead to nonconstructive
case.
of the cut elimination
solutions by Tait, for the second
by Takahashi
It should also be mentioned
and Prawltz
that Sch~tte
for the general
introduced
duction which makes the meaning of cut elimination cases.
and is an
in its own right.
started a semantical
order case, and independently,
of analysis
infinite
in-
clearer in several
Later Girard found an Ingeneous proof of the cut elimination
theorem and Martin-LSf
and Prawitz,
to prove a normalization
theorey,
independently,
used Girard's
idea
which is a sort of elegant variant
of cut elimination. Much earlier than this I introduced cut elimination of
theory for
1 ~l-comprehension
proof,
ordinal diagrams and proved a
~-comprehension
axioms and the consistency
axioms with generalized
like Gentzen's,
inductive
lies completely within Hilbert's
point except for the proof of the accessibility Systems of ordinal diagrams are complicated. have clarified
definitions. finltist
My
stand-
of ordinal diagrams.
Sch~tte and his school
several relations between ordinal diagrams and other
ordinal notations. The ideas involved in the proof of the accessibility diagrams are, at this time, much more abstract the proof of the accessibility and I are improving
of ordinals
than those involved
less than
the proof of the accessibility
but we have not yet succeeded
in reducing
that of the proof of the accessibility
of ordinal
eO"
Marlko Yasugl
of ordinal diagrams
its abstraction
of ordinals
in
to the level of
less than
e O.
MARKOV'S PRINCIPLE AND MARKOV'S RULE FOR THEORIES
OF
CHOICE SEQUENCES
Dedicated to Kurt Sch~tte on occasion of his 65 th birthday.
A. S. Troelstra
w I.
Introduction.
[K,T],
IT4].
We shall freely use notation and terminology from
The following forms of Markov's principle will be discussed:
MpR
Vxy(m ~ S z T ( x , y , z )
~ SzT(x,y,z))
~R
Vn(V~ ~ - S X R n ( ~ , x )
~
~
Vn(~y.1 ~XRn(~,~)-~i~.I
where of
R
V~XRn(~,x)) ,
and its weakening
~XRn(~,x)),
is a standard enumeration of all primitive recursive predicates
n
~,x
M
Va[~mZx(ax=O)
Mc
v~[~x(=x=
MI
V~[VS(A(~,8) V ~A(~,8))
c
,
-- ~ x ( a x = O ) ]
o) - ~ ( ~ x = o)]
,
,
& ~8A(~,8)
"
~8A(~,8)]
9
Assuming the lawlike functions to contain all primitive recursive functions, and all lawlike functions to be equal to choice sequences (i.e.
VaZ~(a= ~))
it follows immediately that
MI~ C
M
C
-~ M-~ Mp
as we shall show, in
R ~
c
C~S also
M
c
~ MI . c
The principal observations and results of this little note are as follows : I~ ) for any notion of choice sequence satisfying a few simple axioms relative to lawlike sequences,
Mc ~ MpR * - ~ -1 CT
A.S.
where
CT
(Church's
thesis for lawlike
CT
VaSxVySz(Txyz & ay = Uz)
(see
~ction
2).
Troelstra
As a corollary,
Mc
sequences)
2 ~) M c [K,T]
relative
can be extended to
in
as
C~S.
between completeness
logic, Markov's principle
is consistent
is formulated
is underivable
In section 6 we discuss the relationship istic predicate
371
of intuition-
and Church's thesis.
C~S; in fact, the elimination result of
C~S+ M c
on the one hand and
ID.~BI + M + K ~ = K
on
the other hand, where Koa ~def VbSx(a(~x) ~ 0 ) & V n m ( a n ~ O ~ a n = a(n.m)) . 3 ~ ) The underivability underivability
of
realizability".
of
MpR.
Mc
in
~
is of course a weaker result than the
To show the latter Kleene introduced his "special
We describe ~ simplified proof of this result in section 3.
4 ~ ) We use the elimination under Markov's
of choice sequences
rule
MR c
~V~x(A(~)
for
FIM,
v ~A(~,x)),
~V~X~A(~)
the system of intuitionistic
"choice-part"
to give a proof of closure
of
C S,
since
C~S
= bV~xA(~,x) analysis
of [K,V]
is conservative
over
(which is the
FIMM by [T3])
(section 4).
5 ~ ) We show
CS ~ M c -- M1 c
w 2. Incompatibility for
of
CT
(section
and
5).
M c , elimination
of choice sequences
CS + M c .
2.1. Let us assume lawlike and choice functions in", and let the
to be closed under "recursive
K - variables be supposed to range over a class
K*
of
neighbourhood
functions which is not assumed to coincide with the inductively
defined
Let
K.
KI
be defined by
K1a ~ Vb~zVc~b(a(~z) JO) & V n m ( a n J O ~ a n = a(n.m)) .
872
A.S.
Troelstra
We shall make the following assumptions about the properties of lawlike and choice sequences: (I)
V6Sx(a(~x) /0) & V n m ( a n / O ~ an= a(n*m)) ~ K*a
(2)
Va~(a= ~), ~ a ( ~ =
(3)
K*a ~
a)
K1a.
We start with two simple propositions with almost trivial proofs:
,, A. Proposition Proof.
Mc
Assume
aE
Ko
~
Ko,
Vb~x(a(~x) / 0 ) ,
c K* ~ KI" --
i.e.
V n m ( a n / 0 ~ an= a ( n . m ) ) .
By (2) V~~ ~ ~ ( ~ ( ~ x ) / o ) . Applying
Mc
with
kx.(1 : a(~x))
for
4,
it follows that
v~(~(~x) / o), hence with (1)
a 6 K*,
Proposition B. Proof.
~
and with (3)
a6 KI .
~CT .
Kleene constructed a binary tree with primitive recursive character-
istic function which is well-founded w.r.t, recursive sequences but not uniformly bounded (of. [K,V], lemma 9.8). a primitive recursive function
~
So, on assumption of
CT , there is
such that
Va~x(~(~x) / O) , Vnm(tm'l/O ~ C a = ~(n.m)) , but
~/K I .
On the other hand, as in the proof of proposition A, if
Va~tx(~(Tx) / O) , ~{
i t f o l l o w s by (2) t h a t
~V~Zx(~(~x)
~{K I Remark.
so
~x(~(~x) / 0), and t h e r e f o r e wi th
/0) ; but this is obviously false since
~V~Sx(~(~x) /0) . Note that in fact
Corollary.
V~ ~
C~S+ M c ~ ~ CT
proved in [K,T], 6.2.1).
~
Thus
K ~ K I , but
MpR ~ ~ C T .
for particular
n
refutes
CT.
(since (I), (2), (5) are provable in
CSS;
(2) is
A.S. 2.2. Theorem.
Let
for formulae
A
T(A) ~ A
(ii)
~
(iii)
CS + M c ~ A
Then,
for formulae of
ID~BI ,
~ T(A) ~-eA
As a corollary
* ID~BI+ ( K o = K ) + M ~ T ( A
Mc
M
c
T(Mc)
.
C~S but underivable in
C~+M.
IDBBI + (K ~ : K) + M
is derivable in
as
v~[~ v ~ ( ~ J o ) ~
:x(~=o)]
Below we shall regard sg(tx) = I.
)
is consistent relative
We first show that
We rewrite
(I)
be the translation as defined in [K,T], 7.1.
not containing choice parameters
(i)
Proof.
9
373
Troelstra
tx ~ 0
Application of
. as an abbreviation of the prime formula
T
to (I) yields
Ve[ Vf ~ VxVa( (e I (fl a) )x I O) S e ' V n ( e ' n % 0 ~ Va((el(nla))(e'n~1 ) = 0))] which is equivalent to (2)
V e K V f ~ V x V a ( ( e l ( f l a ) ) x ~ O ) ~ ~e'Va((ela)(e'(a)) =0)] .
Now note that Vf~VxVa((e1(fla))x~O ) ~ (For
~,
substitute
V b ~ V x ( ( e l b ) x ~0)
fb
satisfying
implies
Vf~a~Vx((el(fla))x ~0) .)
Ve[~Vx((elb)x~ which with (3) With
M
o) ~
~Vx(el(fl~y.1)x~O) Therefore
for
f,
for
*
note that
, hence
(2) is equivalent to
~e'Va((ela)(e'(a))
=
0)]
-- ~ e ' W ( ( e t b ) ( e ' ( b ) )
=0)] .
it follows that (3) is true.
On the other hand, sequences of
VC(fblc= c)
reduces to
Ve[~x((elb)x=O) K =K o
Vb~Vx((elb)xfO ) .
C~S
The consistency of
Mc
implies
M,
and in view of t ~
fact that the choice
satisfy (I), (2), (3) of Proposition A in 2.1, also Mc
relative
C~S then follows by the remark that
K o=K.
374
A.S.
(i)
FI~M+M
Troelstra
is consistent relative
B+M,
where
B
is Kleene's "basic
system" obtained by omission of the continuity axiom from
FIM;
this is
shown by realizability (see [K3, Part II), (ll)
in
FI~M one can prove the choice sequences to be a model for
IDB I + (Ko= KI) satisfying
(assuming the lawlike variables to range over choice sequences
FI~M) - this is implicit in [K,T~, 6.3, noting that
in the model indicated (ILi)
C~
Remark. C$.
FIM
and
IDB I .
Variants of this theorem hold for the weaker systems C$
but
K
is defined as
C So(C~SI)
and
IDB I + (Ko= K) + M
Also it follows that
Mc
by
w.r.t, arithmetical sentences for
CSS
E~L+AC-NF+M(E~L+AC-NF+
is underivable in
2.3. Corollar2 to the proof of 2.2.
C So, C S I
K. , and we add an axiom
V~Sx(e(~x) ~0) ; then the assertion of the theorem holds with by
and
K = K. o
is conservative over
is similar to
K~KI,
replaced (Ko=KI) +M) .
CS I + M .
H+M+CONT
I
is conservative ever
H -z EL, E L + ( K o = K 1 ) ,
ID B I + K o = K .
H+M Here
VaSbA(a,b) -~ Sc ~ KoVaA(a,cla )
CONT I
where of course Proof.
cla= b =def Vx~y(c(<x>*~y) = bx+1) .
Combine the proof of the preceding theorem with the method for
proving theorem I of ~T4~, where
H, CS H
in IT43, w 3 are to be replaced by
H+M,
C~H+M c
respectively.
w 3.
A new proof of the underivability of
3.1.
Kleene introduced in [K,V~ "special realizability" to show
underivable in
FI~M and
underivable in
ID~BI
MpR
i_n ~
and
CS.
C~S. To do so it is sufficient to show
since
C~S
is conservative over
Note first that the validity of the rule
IPR
IPR
~ (~A ~ SxB) = ~ ~x(~A ~ B)
not free in
for
ID~BI
(x
to be MpR
to be
ID~BI .
A)
is sufficient to yield the required underivability : for assume
A.S.
(I)
IDB 1 b V x ( ~ V ~ . T x x z
then with
Troelstra
375
~ ~ Txx~)
IPR
IDB I ~ V x ~ y ( ~ V ~ T x x z
~ Txxy)
and hence (2)
~DBBI ~ a V x ( . V z ~ T x x z
Now IDB I + CT IDB I + C T
~ T(x,x,ax)) .
is consistent relative
the recursiveness of
hence (I) underivable in
3.2. Theorem. Proof.
I DB I
IDB1,
~Txxy
and we can derive from (2) in
; this is obviously contradictory,
IDB I . So it remains to establish
is closed under IPR.
We use de Jongh's variant
5.1.13, [J], w 5 ~ read to a formalization of
"P" ID~BI
for with
E IA
of Kleene's
"D"
(see ITS],
in (i) of 5.1.15 of ITS]), relative
K-variables
(i.e. all prime formulae are of the form it is more convenient to axiomatize
F IC
IDB I
but without the constant
t = s).
Instead of the schema
K K5
with the equivalent schema of
induction over unsecured sequences :
(i)
W ( e n ~ o ~ Qn) ~ Vn(VyQ(n*~) -- Qn) ~ QO.
The definition of
E IA
is extended to
(vii)
E IVaAa m Va(E]Aa)
(viii)
EI~aA a E ~a(EIA a & ( E ~ A a ) )
(ix)
E Iveje
(x)
EI~eA e m ~e(E]Ae & (E--Ae)) .
(If
a, e
ID~BI
by
m Ve(EIAe )
do occur in
E
Now it is routine to show
we must rename IDB I + E
~A
a, e
on the right hand side.)
= IDB I + E I E
this by the verification for instances of
(I).
Assume (2)
ElVn(en~O
-~ Qn) & Vn(VyQ(n.#) -~ Qn)
,
~EIA.
Let us illustrate
378
A.S.
Troelstra
From (2) we obtain
V~(El(en/O ~ (E ~ e~/O)) -- EIQn) which is equivalent to V~((E ~ e n # O ) ~ EIQn ) which in turn implies (4)
~ n ( e n / 0 -- Elqn ) .
From (2) also Vn(Vy(EIQ(n.#) ) & (E ~ vyQ(n.~)) ~ EIQn ) which implies
(5)
~(~y(EIQ(n*~)) - EIQ~).
From (4), (5) with (I) ~pplied to
EIQ~
instead of
Q~,
EIQO. Then, as in [TS~, 3.1.14- 3.1.15, we obtain ~losure under IPR.
3.3. Remark.
The underivability of
ed from the underivability of
imply
IDB I
on the other hand is readily obtain-
M c : assume
it would already be derivable in C.~S form a model of
M
M
to be derivable in
C~S, then
IDB I ; but since the choice sequences of
(provably in
C~S, of. [K,T], 6.3), this would
C~S ~ M c , which we have refuted already.
A.S.
w
4.
Markov's rule for
4.1.
IDB 1
FI~
Troelstra
(from Markov's
377
rule for
I DB I
).
can be proved to be closed under Narkov's rule by extending the
Dialectica interpretation to
ID~BI ; this is done by first embedding
in a theory
WE-ID~B m
W~E-IDBW+AC-NF,
showing that
WE-IDB w + A C - N F
as described in [TS], 1.9.25, then
has a Dialectica interpretation into itself ;
the details are practically the same as for the interpretation of V*
described in
[H], w 2.
Then, after the manner of
obtain closure under Markov's rule for Next, we construct in fortiori a model for
IDB I
IDB I
a model
U
into
IT5], 3.8.5 we can
W~m-IDB m + AC-NF . K-ECF
WE-IDBm+AC-NF)
for
similar to
~-IDBW-AC-NF ECF
(hence a
itself, but simply
taking W~(a) E Ka,
IK(a,b ) m Vx(ax= bx)
otherwise as for
ECF,
of the constants
I, 91, 92, 93
of
E-HEO
and constructing in the obvious way representations of
E-IDB ~
(cf. the analogous construction
in 2.9.5 of IT5] ! a representation of
~
is found by applica-
tion of the recursion theorem analogue 1.9.16 of IT5] ; cf. again the construction of
[$ ]
in 2.9.5.
From this construction it follows that IDBI,
hence the closure of
closure of 4.2.
I DB 1
E-IDB w + A C - N F
~-ID~BW+AC-NF
is conservative over
under Markov's rule implies
under Markov's rule.
Now assume
(1) Since in
FI~Varx[A(~,x) V ~ A ( ~ , x ) ] , F I ~ ~ V ~ x A ( ~ , x ) C~
Va~(a=
.
~) ,
C~S ~ Va,x(A(a,x) V ~A(a,x)) , and if we write
A*(a,x)
for
T(A(a,x))
by the first elimination theorem of C~ ~Va,x[A*(a,x) V ~A*(a,x)
(T
[K,T], 7.2
],
as in w 7.1 of [K,T]) alsO,
378
A.S.
hence by the second elimination IDB I ~Va,x[A*(a,x) Also,
since
Troelstra
theorem of [K,T], 7.3
V
~A*(a,x)].
C SS ~ V~ ~ x A ( ~ , x )
by (I)
~D2B~ ~ V a ~ A * ( a , x ) and applying Markov's rule for
Since
A(~,x)
quantifiers
IDB I
was assumed to be a formula of Sa, ~e , hence the proof of
CS ~ A * ( a , x ) ~
A(a,x)
does not make use of the axiom assume the lawlike variables the
K - variables
V ~ x ( ~ ( ~ x ) / 0) ,
over
(cf. IT3]).
BC-F
V n m ( ~ n / O ~ om= ~(n*m)) ,
~,~
Therefore,
**
Now we note that (I ~ ) if we
~
if we let
K*
of
CSS,
A**(a,x)
T(B) <--~ B
and the
I DB I
hold except
be obtained from
to choice variables,
and
such that
then we obtain a model of
all axioms needed to prove
ranging over
CS ~
CSS.
being the class of
by changing all lawlike variables by variables
of
to range over the choice sequences
K* , K*
(cf. [K,T], 6.3) and (2 ~ BC-F
FI}~, it does not contain
A*(~,x)
K-variables
then
(~,x)
(from (21 and (I~ just mentioned),
hence
CA 5 v=~A(=,x) (by (2o)).
Thisestablishes
Markov's rule for
FI~.
Thus we have established Theorem.
FIM
is closed under
MR O
4.3.
For a direct proof of
MR
for
FIM,
O
theory of finite types containing the schema
we have to embed BR ~
FIM
for bar recursion
in
a
of type
0 , and constants for moduli-of-continuity ; this theory does not have an
A.S.
extensional model analogous to 4.4.
Troelstra
ECF
(cf.
379
[T5~, 2.6.7).
If we could establish, for all closed terms of type 2 in
(I)
WE-IDB 1 ~ZeVx1(e(x
I) =
t2X 1)
then we could establish closure of the whole of Markov's rule as follows.
WE-IDBI:
C~S (not only
The argument runs as before for
FIM
FIN)
under
till we
arrive at
~DB I ~ Va~xA*(a,x). By
m q-realizability,
one then establishes closure of
under the rule of choice (of. [T5], 3.7.2), W~E-IDB W + A C . N F
~-IDB ~+AC-NF
so for suitable
t2
~A*(a,t2a) ,
and then, by (I)
~-~DB ~ § AC-NF b ~eV~*(a,e(a)), IDB I ] - ~ e V a A * ( a , e ( a ) ) ,
and thus
hence
cs 5 ~eV~(a,e(a)). In
CSS V~(A(~,e(~))V ~A(~,e(~)))-~
(VaA(a,e(a)) ~-~ V~A(~,e(~)))
(cf. [K,T],
6.2.2), therefore CSS ~ SeV~A(~,e(~))
and so
CS ~V~xA(~,x).
A possible way to establish (I) might be the extension of the technique developed in IS], to establish closure of recursion.
~w
under the rule of bar
380
A.8.
w 5.
Troelstra
Markov's principle for functions.
Because of the presence of continuity axioms in
M1
CSS,
Va(A(~,6) v ~A(a,B)) & -7~S~A(~,~)-~ S~A(a,B)
C
is derivable from 5 1 9
Theorem.
MC :
CS ~ M c ~ M I
9
~
Proof.
C
Abbreviate
9
A(~,elB )
as
Ae(~,B) , and assume
VaS(Ae(a, S) V~Ae(a,6)). Then there is an
fEK
such that
e
(here
e
f(c~,6) abbreviates
f(J(~,S))).
Therefore
e
and thus
~Z~Ae(~,~
) .~ ~Ae(~,~)
is equivalent to (I)
~Z~(f(~,B)
Now let
=0) -- Z~(f(~,~) =0)
rn= kx.g(n,x) , (g
(in the notation of IT4]).
the notation of ~K,T]) Therefore
since
or
r n = ~x.(n) x
~ ( f ( ~ , ~ ) = O *-4
(lth(n) = 1) *-* Sn(fn,~ ) = O) , (I) is equivalent to
~n(f(~x.j((n)x,2X)) ~n(f(r
.
n,B) =0) -* Wn(f(rn, B) = O)
which is a consequence of V~B[Ae(~,B ) V ~ A
e
M C . Therefore, for all
(~,~)] & V~ ~ A
e
e
(~,6) -~ V B ~ A e ( ~ , ~ )
which by the schema of analytic data implies V2(V~[A(~,~) V-~A(~,~)] & - ~ A ( ~ , 6 )
-* Z~A(~,6))
.
A.S.
w 6.
Troelstra
881
Completeness of intuitionistic first order predicate logic
In [Krl] Kreisel defines validity
VaI(A)
for a formula
I~PC.
~(PI' .... Pk )
of pure first order intuitionistic predicate logic, constructed from the ri-placed
relation symbols
_ ~def VD~.. Val(A) where of
D
D ri
9
Pi ' I < i < k
. ~k* AD(P~, =
by
,P~) ,
""
ranges over arbitrary species, and (the
Pi
and
D
*
Pi
over arbitrary subspecies
may contain choice parameters) ; ~D
by restriction of the individual variables of
A=
to
is obtained
D.
Let "Proof" denote one of the usual arithmetized proof_predicates for intuitionistic predicate logic, then completeness is defined as VA(VaI(A) ~ Sp Proof(p,
A ))
and weak completeness as VA(VaI(A) - - ~ p Let us use
"Valc"
Proof(p,
for validity relative to species and relations not con-
taining choice parameters,
"Val"
also contain choice parameters. completeness corresponding to Comp', Comp~
~ )) .
for validity where Let
ComPc , Comp
Val, Val c
P' P~' "''' Pk*
may
denote the notions of
respectively, and, similarly,
the respective notions of weak completeness.
From theorem I of [Krl] we see that ComPc ~ C o m p ~
Comp' ~ ~
and by Proposition B in 2.~ (cf. also [Kr4], page IV- 4) CT ~ ~ C o m p ' ~ ~ Ccmp ~ ~ C o m P c . This also follows from the result in [Kr2] (for exposition see [D]) ~ but there a much stronger result is established, namely that not recursively enumerable, in other words
IPC
IA : Valc(A)}
does not even have an r.e.
axiomatizable extension which is complete. In footnote 16 of [Kr2] (p. 147) it is noted that permits to extend the result obtained there from principle is of course a consequence of
V~Sa(~
is
ComPc = a) .
Va ~ A a ~ V ~ A ~ to
Comp ; this
A.S.
382
Troelstra
w 7.
Miscellaneous additional remarks.
(i)
As is well known, Markov's schema
Mc
is false for lawless sequences.
It is worth noting, however, that it is also false for two kinds of projections of lawless sequences: (a)
it is false for the collection of all e E K, u E N
lawless, independent, contains all
n*(~)n
elVu(~ I ..... ~u ) ' ~I ..... ~u
considered in [TI], since this collection
which form a model for the theory of lawless sequences
(cf. [~2]), (b)
it iS alsO false for the universes
~
:
{el~: e6K},
~
lawless, for
the same reason (one is led to consider validity in all such universes in relating validity in topological models to intuitive validity (cf. [Kr2],
p. 137). (ii)
~n [Kr3], p. 126- 127 it is pointed out that the Dialectica inter-
pretation of bar recursion conflicts with that
CTo + ~ R
+ BRo + constructive
CT ; in ILl, 9.3 it is shown
~- rule
are inconsistent relative
to elementary analysis.
References
[D]
D. van Dalen, Lectures on intultionism, in : A.R.D. Mathias, H. Rogers (editors), Cambridge Summer School in Mathematical Logic. Berlin (Springer-Verlag) 1973, pp. I ~ 94.
[HI
W.A. Howard, A system of abstract constructive ordinals. J. S. L. 37 (1972), pp. 355- 374.
[J]
D. H. J. de Jongh, Formulas of one propositional variable in intuitionistic arithmetic. Report 7 5 - 0 5 of the Dept of Mathematics, University of Amsterdam.
[K]
S. C. Kleene, Formalized recursive functlonals and formalized realizability. Memoirs of the American Math. Soc. nr 89. Providence, Rh. I. 1969.
[Krl]
G. Kreisel, On weak completeness of intuitlonistic predicate logic. J. S. L. 27 (1962), pp. ~39- 158.
[Kr2]
- , Church's thesis : A kind of reducibility axiom for constructive mathematics, in : A. Kino, J. Myhill, R.E. Vesley (editors), Intuitionism and Proof Theory, Amsterdam (North-Holland) 1970, pp. 121 - 150.
[~r3]
G. Kreisel, A survey of proof theory II, in : J. E. Fenstad (editor), Proceedings of the Second Scandinavian Logic Symposium, Amsterdam (North-Holland) 1971, pp. 109- 170.
A.S. [Kr4]
[~,~]
Troelstra
383
- , Theory of free choice sequences of natural numbers, in : Stanford report on the foundations of analysis, Stanford 1963 (mimeographed, privately circulated only), section IV.
G. Kreisel and A. S. Troelstra, Formal systems for some branches of intuitionistic analysis. Annals of Mathematical Logic I (1970), pp. 229- 387.
[K,vl
S. C. Kleene and R. E. Vesley, Foundations of intuitionistic mathematics, especially in relation to recursive functions, Amsterdam (North-Holland) 1965.
ILl
H. Luckhardt, Extensional Godel functional interpretation. A consistency proof of classical analysis. Springer Lecture Notes Vol. 306, Berlin (Springer) 1973.
[s]
H. Schwichtenberg, Einige Anwendungen yon unendlichen Termen und Wertfunktionalen. Habilitationsschrift, ~unster (Westf.) 1973.
[TI]
A. S. Troelstra, Notes on the intuitionistic theory of sequences I, Indag. Math. 31 (1969), pp. 4 3 0 - 4 4 0 .
[m2]
- , Notes on the intuitionistlc theory of sequences III, Indag. Math. 32 (1970), pp. 245 - 252.
[T3] [T4] [~5]
- ,
An addendum.
- ,
Note On the fan theorem.
Annals of Mathematical Logic 4 (1971), PP. 437-439. J. S. L.
To appear.
(editor), Metamathematical Investigation of Intuitionistic Arithmetic and Analysis. Springer Lecture Notes Vol. 344, Berlin (Springer Verlag) 1973.