Transfinite Ordinals in Recursive Number Theory

Author(s): R. L. Goodstein
Source: The Journal of Symbolic Logic, Vol. 12, No. 4 (Dec., 1947), pp. 123-129
Published by: Association for Symbolic Logic
Stable URL: http://www.jstor.org/stable/2266486 .
Accessed: 20/06/2014 16:20

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Association for Symbolic Logic is collaborating with JSTOR to digitize, preserve and extend access to The
Journal of Symbolic Logic.

http://www.jstor.org

This content downloaded from 185.2.32.141 on Fri, 20 Jun 2014 16:20:35 PM

All use subject to JSTOR Terms and Conditions
THE JounNAL OF STMBOLIC LOGIC
Volume 12, Number 4, December 1947

TRANSFINITE ORDINALS IN RECURSIVE NUMBER THEORY

R. L. GOODSTEIN

The possibility of constructing a numerical equivalent of a system of trans-

finite ordinals, in recursive number theory, was briefly indicated in a previous
paper,1 where consideration was confined to ordinals less than e (the first to
satisfy e = w'). In the present paper we construct a representation, by func-
tions of number-theoretic variables, for ordinals of any type.
In addition to definite numerals, and numeral variables, we introduce majorant
variablesa, W,)r , r > 1. A relation containing a single majorant variable is a

required to hold, not necessarily for all non-negative integral values of o, but for
allvaluesgreaterthan someassignedconstant. A relationR(w, Wi , (2, W*, COr)
between the majorant variables w, wrX (1 < r _ m), holds if there is a constant co
and recursive functions cr(no, ni, *** , n7.-) such that R(no, nli, n2, ***, n,)
holds for all non-negativeintegers no, nli , n. such that no ? co and
*

nr+i > cr+,(no , ni, ... , n.), 0 < r _ m - 1. For instance,if a, b, c, aredefinite
b
numerals then a' > ao + c, since n' > anb + c if n > max(a, b, c); in particular,
for any definite numeral a, a > a. An example of a relation between two
majorant variables is wc'Wl' + 3w2 + 4 > co WI + 7w2 + 11 which holds since
nnNn + 3n2 + 4 > n3nN2+ 7n2 + 11 when n _ 12 and N n3.
Majorant variables have the same algebra as numeral variables, that is to say,
they obey the same rules of addition, multiplication, and exponentiation, for if a
relation R(no , nI, *2 ** , n) holds for all non-negative integers n. , 0 < r ? m,
without restriction, then a fortiori, R(w, co, * , Wn)holds.
We commence with a generalization of the concept of a numeral expressed
with assigned digits in a scale of notation. If f(x) is any function of a numeral
variable x (and possibly of other numeral or majorant variables as well), such
that f(O) = 0, f(l) = 1 and f(x + 1) _ f(x) + 1 for all x, we define, for any
numerals k > 1, n _ 0 and (a numeral or majorant variable) a > f(k), the
function 4f( )(n) recursively as follows:
If (k + 1)' is the greatest power of k + 1 not exceeding n, and c(k + 1)' is
the greatest multiple of (k + 1)' not exceeding n, then
(= (i)

= f(c)o*'f, (a) + gjz)(n - ctk + 11a), n > 1. (i)

For given k and n the equations (i), (ii) determine the 'value' of of () (n) as a
function of a and f(O), f(1), * , f(k) which we call the representation of n with
digits f(r), 0 _ r _ k, and basea. For instance 3,(0(561) = f(2) o' + f(3)off2' + 1.
We shall write either x or y for the variable in the affix function f(x), and when
f contains more than one argument the operative variable will be written in the
last argument place; thus the operative variable in h(p, q, x) is x, in the last argu-
ment place. We note that since f(O) = 0 and f(x + 1) > f(x) + 1 therefore

Received March 7, 1947.

1In this JOURNAL, vol. 9(1944), pp. 33-41.
123

This content downloaded from 185.2.32.141 on Fri, 20 Jun 2014 16:20:35 PM

All use subject to JSTOR Terms and Conditions
124 R. L. GOODSTEIN

f(x) > x for any numeral x. The 'value' of A m.+(n) for a given n is of course the
familiar representation of n in the scale m + 1, with digits 0, 1, 2, ... , m.
THEOREM 1. f (o )(r) = f(r), (r < k).
For the exponent of the greatest power of k + 1, not exceeding r, is zero, and
the greatest multiple of (k + 1)0, not exceeding r, is r itself. Hence

'Of,,)r) = f(r)qo + ,fgT)(O) = f(r).

Hence

(kk,
(k + 1) = f(1)a*f h + Ok, (0) =-A (k > 1).
THEOREM 2. For a fixed numeral k 2 1, and a definite numeral (or majorant
variable) s satisfying s > f(k) + 1, of )(n) is strictly monotonic increasing with n.
For brevity we write A/(n) for ok, (n). We have to prove, for all n > 0,
0(n + 1) 2 (n) + 1.
By theorem 1, since f(n + 1) _ f(n) + 1, the inequality holds for n < k;
suppose that it holds for n < m - 1, where m _ k + 1. Let c be the greatest
integer such that (k + 1)c < m and b the greatest integer such that b(k + 1)c < m,
and let a = m-b(k + 1)c, so that 1 < c, 1 < b < k + 1 and 0 _ a < (k + 1)c.
Then A(m) = f(b)s4(c) + {(a).
If a = 0, m + 1 = b(k +1)c + 1 and so(m + 1) = f(b)s(c) + 1 = (m)+1,
since 40) = 0.
If a > Oanda + 1 < (k+ 1)cthen

4/i(m+ 1) = f(b)s (c) + {k(a + 1) 2: f(b)so(c) + A1(a)+ 1 = {(m) + 1,

for a + 1 < (k + 1)c < m - 1, and so, by hypothesis, 4(a + 1) > {k(a) + 1.
If a + 1 = (k + 1)c, then m + 1 = (b + 1) (k + 1)c.
Consider first the case b + 1 < k + 1. Since a < a + 1 = (k + 1)c <in,
c
therefore {1(a) + 1 < i(a + 1) = Ak + 1 1 = s (c) and so

4t(m + 1) = f(b + 1)s (C) > f(b)so(c) + 0(c)>?f(b)s (C) + )/(a) + 1 = 4,(m) + 1.

If b + 1 = k + 1, then m + 1 = (k + 1)c+1. Since m > (R + 1)c ?

1 + ck > 1 + c, therefore, by hypothesis, 4(c + 1) _ ,6(c) + 1.
Moreover s > f(b) + 1 and a + 1 = (k + 1)c < m, so that {s -f(b) }s (C) >
80(c) = 4p(a+ 1) _ {(a) + 1; hence

4t(m + 1) = >4d+1) > s4c)s _ f(b)s*() + A(a) + 1 = A(m) + 1,

which completes the proof of the inequality for n = m, and so, by induction,
theorem 2 holds for all n.

We notice two special cases of theorem 2:

2.1 Taking f(x) = x, we have, if a > k > 1, 4k,,(n) is strictly increasing
with n.

2.2. Takingf(x) = 4? ,q+i(x),q

> p 2 1, we have, if a > p (k)), (n)
is strictly increasing with n.

This content downloaded from 185.2.32.141 on Fri, 20 Jun 2014 16:20:35 PM

All use subject to JSTOR Terms and Conditions
 TRANSFINITE ORDINALS IN RECURSIVE NUMBER THEORY 125

THEOREM 3. If f(x), g(x) are strictly monotonic increasing, and p 2(q),

and 8 > f(p) (where s is a numeral or a majorant variable), then, for all n,
and q 2 1
g
Xf(X sv) I (n) } = Of
q 9))v
(n)
For n = 0, each expression takes the value zero; if the equality holds for
n = 0, 1, 2, *-, m -1 then we deduce that it holds for n m ?: 1. Let
(q + 1)8 be the greatest power of q + 1 and c(q + 1)8 the greatest multiple of
(q + 1)' not exceeding m, and let b = m - c(q + 1)', so that c < q + 1,
b < (q + 1), and m < (q + 1)a+ . Then

44,wi (in) = g (c) (p + *)*Q,P+1(8) + , (b)

< + + (P +
g(c)(P
by theorem 2, since b < (q + 1)',
1)+1(8)

9g(c) + 1} (p + 1) q,,+1(a) < (p + 1)*q'v+1(6)+

since g(c) < g(q) < p + 1.

Thus +p1(a) is the exponent of the greatest power of p + 1 and
g(c)(p + 1){ 0?l-1 is the greatest multiple of (p + 1)#gI~1(g) not exceeding

Therefore

44x){ q.(m)i I = f(g(c))s'f #gq(I(a) + of l (b) }

= f(g(c))s+' a + 4("'"(b)
(by hypothesis, since a < m, b < m)
= <*,f (in),

which completes the inductive proof.

In terms of two functions p(n), p'(n) we define, for k 2 0, n 2 0, (anticipating
theorem 4 below)

X),,,(0, k) = k
Xpp(n + 1, k) = +pl't;(n) (k)
provided p(n) ? 1 and p'(n) ? Xp ,,(n, p(n)), for all n.

THEOREM 4. For all n 2 0, Xp,,p(n, k) is strictly monotonic increasing with k.

The theorem holds for n = 0, by definition of Xp,,'(0, k), and if it holds for
n = m, then it holds for n = m + 1 by theorem 2, whence by induction, the
theorem is true for all n.

Since X,,p'(n, 0) = 0, it follows that, for all k > 0, X.,,,(n, k) > k.

This content downloaded from 185.2.32.141 on Fri, 20 Jun 2014 16:20:35 PM

All use subject to JSTOR Terms and Conditions
126 R. L. GOODSTEIN

THEOREM 5. If p(r), q(r) > 1, and if p'(r) _ max {Xp p(r, p(r)),Xqp'(rq(r)))
and q(r) _ Xq,,(r, p(r)), 0 < r < n - 1, then X,,, (n, k) = X,,, (n, X,,(n, k)),
n > 0 k > 0.
The theorem holds for n = 0, since Xx,,,(0, k) = k for any X, ,s; if it holds for
n = m, then
Xq#p'(m+ 1, Xp,q(m + 1, k)) = 'q(rn)+l(Xptq(m + 1. k))

- 'Pq(m)[p'(m)+1n \Pm),q(m)+1)J

= ?tto;(y)) (kc) by theorem 3,

= &Pj;(mY) fZ.
since, by hypothesis, theorem 5 holds for n =m,
= X,,,,(m + 1, k)
which proves the theorem for n = m + 1, and so for all values of n.
THEOREM 6. If p(k) _ 1, q(k) _ 1, and if r(k) _ max {Xp,7(k,p(k)),
Xqr(k, q(k)) I and s(k) > max {Xp (k, p(k)), Xqs(k, q(k)) }, n > k _
0, then
Xp,7(n, p(n)) < Xqr(n, q(n)) according as Xp,,(n, p(n)) < Xq,,(n, q(n)).

ForO _ k <n - 1, let

t(k) = max {Xpt(k, p(k)), Xq,t(k, q(k)), Xr,t(k, r(k)), X8t(k, s(k)))
then by theorem 5,
Xrt(n) Xp,(n, p(n)) = Xt(n, p(n))

and
xrt(n, Xq,r(n, q(n))) = Xq,t(n, q(n)),

and so, by theorem 4,

Xp t(n, p(n)) <Xqt (n, q(n))
according as

Xp,,r(n, p(n)) <Xe,,(n, q(n))-

Similarly
Xp t(n, p(n)) < X,,,,(n, q(n))

according as

X ,(n, p(n)) <X ,,(n, q(n)),

whence theorem 6 follows.

TEOREM 7. If p(n) _ k > Oandp(r) > 1, p'(r) _ X,,,p(r, p(r)) forO < r < n,
then X,,,p(n, k) = Xpp(n + 1, k), n > 0.

This content downloaded from 185.2.32.141 on Fri, 20 Jun 2014 16:20:35 PM

All use subject to JSTOR Terms and Conditions
 TRANSFINITE ORDINALS IN RECURSIVE NUMBER THEORY 127

For Xp, (n + 1, k) = ( )+i(k) = Xpp(n, k), by theorem 1. It follows

that if N > n and p(m) _ k for n < m < N, then X.,,(n, k) =-X, (N. k).
If co, co,(r > 0) are majorant variables, we define, for all n > 0, p(n) _ 1,
k _ 0 (anticipating theorem 9 below)
Qp(O,k) = k (i)
Qlp(n+ 1, kIX)= (k) o = c. (ii)
For given k, n and p(r), 0 < r _ n -1, the equations (i) and (ii) determine
the 'value' of Qp(n,k) as a function of cw,w,, 1 _ r ? n - 1, which we call a
transfiniteordinal of type n.
THEOREM 8. A transfinite ordinal of type n is also of type n + 1.
For, by theorem 1, if p(n) _ k,
Qp(n + 1, k) = (nO)n(k) = Up(n, k).

It follows that, if p(m) > k for n < m < N - 1, then

Qp(n,k) = Qp(N, k).
THEOREM 9. For all n _ 0, Qp(n,k) is monotonic strictly increasing with k.
For Qp(O,k) = k, and if the theorem holds for n = m, then by theorem 2 it
holds for n = m + 1.
It follows that if k > p (n) _ X, Qp(n + 1, k) > Qp(n + 1, X) = Qp(n,X), by
theorem 8, and so there are ordinals of type n + 1 which are not of lower type.
THEOREM 10. If p(r) > 1 and q(r) > Xp q(r,p(r)), 0 < r < n - 1, then for
all k ! 0, n _ 0 Qp(n,k) = q(jn,Xpq(n, k)).
The theorem holds for n = 0 since Qq(0, Xpq(O,k)) = Qp(0, k). If it holds
for n = m then
Qq(m + 1, Xpq(m + 1, k)) =0q(m) P
(inq(m) (k))
= 0p1P(()c)m(k)

by hypothesis (and using theorems 3, 4 and 9)

= Qp(m + 1,k)
which proves theorem 10 for n = m + 1, and so for all values of n.
THEOREM 11. If p(r) _ 1, q(r) _ 1, 0 < r ?; n, then, for all n > 0
Qp(n, p (n)) < Qq(n, q(n))
according as Xp,,(n, p(n)) < Xq,r(n, q(n)) (i)
where r(k) _ max {Xp~r(k,p(k)), Xqr(I, q(k))}, n > k _ 0. (ii)
By theorem 10, Qp(n, p(n)) = Q,(n, Xp,,(n, p(n))) and U?(n, q(n)) =
Q4(n,X0,,(n, q(n))), whence theorem 11 follows from theorem 9. By theorem 6,

This content downloaded from 185.2.32.141 on Fri, 20 Jun 2014 16:20:35 PM

All use subject to JSTOR Terms and Conditions
 128 R. L. GOODSTEIN

the relation (i) is independentof the choice of the numbers r(k), provided condi-
tion (ii) is satisfied.
Theorem 11 shows that the decreasing ordinal theorem,which asserts that a
sequence of decreasing ordinals necessarily terminates, is, for ordinals of any
type, equivalent to a number-theoreticproposition.
For the sequence of ordinals fl,.(i, p,(i)), n = 0, 1, 2, * is steadily decreasing
if Xpn+l F%(i)Pn+l(i)) < Xp,,r,(i) Pn(i))

where
rn(k) _ max {Xpr,?(k,pn(k))I Xp,+l,rf(k, pn+1(k))}, n > 0 0 < k < i.
Thus the decreasing ordinal theorem is equivalent to the proposition:
If for all n andO < k < i, p,(k) > 1 and p,,(i) _ O,and if
r*(k) = max IX,,,,,R(k, Rl ?r(kypnt+(k))}
p,(k)), Xp%+
and
X, +i t"(ilpnl+(i)) < Xp,,r.(ilp,(i)) when p,1(i)> 0,
then there is a value of n for which p,1(i) = 0.
EXAMPLE. If p(O) = 1, p(1) = 4 and q(O)= 2, q(1) = 99 then Qp(2,116) =
flq(2,27.1003+ 4.100 + 1) = cc' + (w + 1) Wi + 1.

For 116 4.52 + 3.5 + 1 and so

Up(2, 116) - (4)Xo(2) + .w(3)Xo+ iw(1)

- co c4' + (co+ 1)wo + 1,
and
Qq(2,27.100 + 4.100 + 1) = (27) + ;,w(4)+2,;1) + i(1) etc.
There are recursive functions, with transfinite ordinal 'values,' of greater
generality than fp(n, k). The function f is based on the three processes of addi-
tion, multiplication, and exponentiation, but obviously functions based on a
wider range of processes may be defined recursively. If we confine our attention
just to two processes, addition and multiplication, then ordinals of type one will
be polynomials in w, each term having the form axn... w, and so only ordinals
less than w' will be determined; introducing exponentiation greatly extends the
class of ordinal functions, and correspondingly the introduction of additional
processes effects further extensions.
We define, for k > 0,a >_ 0, n > 0
G(O, a, n) = n + 1, G(1, a, 0) = a, G(2, a, O) = O. G(k + 3, a, 0) = 1,
G(k + 1, a, n + 1) = G(k, a, G(k + 1, a, n)).
Then G(1, a, n) = a + n, G(2, a, n) = na, G(3, a, n) = a' (proof by induction),
so that for k = 1, 2,3 the function G(k, a, n) defines addition, multiplication, and

This content downloaded from 185.2.32.141 on Fri, 20 Jun 2014 16:20:35 PM

All use subject to JSTOR Terms and Conditions
 TRANSFINITE ORDINALS IN RECURSIVE NUMBER THEORY 129

exponentiation. For k ? 4, 0(k, a, n) defines successive new processes (which

we may call tetration, pentation, hexation, and so on). For instance
a

G(4, a, n) = a

the expression containing n a's.2 It is convenient to extend the use of the term
index so that n is called the index of tetration in G(4, a, n), the index of penta-
tion in G(5, a, n), and so on. Simple notations for G(4, a, n), G(5, a, n), G(6, a, n)
are (by analogy with exponentiation) na, na, an , respectively, though this admits
of no further extension.
If a, b, c are chosen in turn to be the greatest integers such that a(k + 1) <
n, {a(k + 1) lb < n,c {a(k+ 1) lb < n, then, for k > 1, we define

(i) 04"k!,(n) = f(n), O< n k) (

f
(ii) t4,5(n) = c4,(c(C) 4fkE (a) 40kb) + 4,k, (n - Cla(k + 1)}b)
I n ik + 1.

We observe that c < k + 1 and, for a > 1, since (ax)n = a+Ix if xn(aG1z) = x(Gz),
that is, if n = aX/a-iX, therefore b < a(k + 1)/a-1(k + 1).
For given k, n the 'value' of 4f,(') (n), determined by the repeated application
of equations (i) and (ii), is a function of of and f(r), 0 < r < k which we call the
representation of n with base a and digits f(O) = 0, f(l) = 1, f(2), ...* f(k),
using the four operations, addition, multiplication, exponentiation, and tetration.
For example

4,l (103) = (I + 1){w+1 + (cL + 1){ } {(W} + W{ `W1} + w{"w}.

To determine the representation of n using X operations, we define 4(5)(n)

recursively as follows: For a given sequence x(n) and for k > 1, we have
g',O(1) = G(X, k + 1, x(l)),

9X~k(M +1) =G( - m, gX k(M), X(M +)), l m k1

and for n > k + 1 we choose a(r) so that for 1 < r < X, a(r) is the greatest
integer such that gX (S) < n, 1 < s ? r, (so that a(r) < n for 1 < r ? X).
Then we define
YXGk,,(1) = G(X, a, Ao )

Y(fkq(m + 1)=G(X -m, Yfk?,(m), ff(kT)(a(m+ 1))), 1 < m < X-1,

and 0f,(n) = f(n), n < k,

0)((5(n) = Yfku(X), n > k + 1.

THE UNIVERSITY, READING, ENGLAND

2Cf. W. Ackermann, Zum Hilbertschen Aufbau der reelen Zahlen, Mathematische

Annalen, vol. 99(1928), p. 120.

This content downloaded from 185.2.32.141 on Fri, 20 Jun 2014 16:20:35 PM

All use subject to JSTOR Terms and Conditions