b y , y by
Michael G. Wasserman B.A., Williams College
(1968)
SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE
DEGREE OF
DOCTOR OF PHILOSOPHY at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY September 1971
Signature of
Certified by
Accepted by.
Author.
Department o "--iath tis, July, 971
'jThesis upervisor
Chairman, partmental Committee Archives on Graduate Students
SEP 30 1971
4-8R AR\ESSignature redacted
Signature redacted
Signature redacted
C.
Categories in Recursion Theory by
Michael G. Wasserman
Submitted to the Department of Mathematics in July, 1971 in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Mathematics.
ABSTRACT
This dissertation presents an extensive snapshot of elementary recursion theory taken through the lens of category theory. It is divided into five chapters.
In the first chapter, several basic categories
implicit in recursion theory are made explicit,, and their properties are explored. In the second chapter, recursion theoretic notions are expressed in terms of properties of the basic categories. In the third
chapter, some important recursion theoretic results are proved in strictly categorical language, using several simple axioms about the basic categories.
The fourth chapter extends the results of the first three chapters to the analogous categories implicit in generalized recursion theory. The basic categorical distinction seems to be in ordinary recursion theory, the notions "finite," "la-finite," and "bounded" coincide, while in generalized recursion theory they are distinct.
Finally, in Chapter 5, a categorical formulationcf Turing reducibility is given, and, within this formulation,
a basic theorem about the jump is proved.
Thesis Supervisor: Hartley Rogers, Jr. Title: Professor of Mathematics
Acknowledgements
Mathematical acknowledgements go to my colleagues for their intellectual stimulation, and to Hartley Rogers for suggesting this topic and otherwise helping to get the ball rolling. My deepest thanks go to my wife, however, for standing by me through the darkest days,
4.
Table of Contents
Chapter
I. The Basic Categories and Their Properties
1.
Rl,
R
2 andR1
2. Coproducts, Products, and Pullbacks 3. A Question of Balance
4. Images, Coimages, Reversible Monomorphisms, Quasi-Unions, and Essential Domains
5. Singletons 6. Complements 7. The Ordering <C
II. Recursion Theoretic Properties Expressed
Categorically
1. Characterizations of
N
2. Facts About R. E. Objects 3. Recursive Monomorphisms4. The Notions Simple, Hyperhypersimple, and Maximal
5. Indices; Productive Objects 6. The Halting Problem
III. Categorical Proofs
1. More About Singletons
2. The Pullback Criterion
Page 7
7
9
23 30 38 41 4)458
58 66 7076
83 93 96 96 108..- , a
Chapter
III. 3. Product Maps
4. Essential Domains and R. E. Objects 5. More About Recursive Monomorphisms 6. Many-one Reducibility
7. Three Theorems
8. The Recursion Theorem, Myhill's Theorem, and Creative Monomorphisms
IV. Extension to Admissible Ordinals 1. The Categories and Simple Axioms
2. The Notion of a-R. E. 3. The Remaining Axioms
4. Projectibility and Boundedness
5. Extended Notions: a-Cohesive, a-Immune V. Reducibility Systems
1. Definition 2. The Functors I
3. A Completeness Theorem 4. S-Reducibility
5. Least Upper Bounds 6. A Theorem on Jumps Appendix Page 110 115 140 156 167 175 189 189 199 207 211 218 228 228 231 236 238 239 245 251
Categories in Recursion Theory
Note: The first four chapters of this dissertation build
up progressively to the definition of "recursive category"
actually made at the beginning of Chapter V. The development
proceeds by mentioning various categorical axioms;
eventually a recursive category is defined to be one
that satisfies all the axioms mentioned.
At the same time, certain very specific categories,
arising out of recursion theory, are under study. The
axioms were designed to capture the properties of these
categories, and likewise, the categories in question had
better satisfy the axioms; occasionally this required proof.
Thus there arises in this dissertation the curious syntax:
Axiom-proof. The proof in such a case is a proof not of the axiom, of course, but of the statement: "The category under study satisfies this axiom."
Finally, many of the results in the first four chapters are labelled with a star. Such results are proven using only the axioms and general facts about categories; that is, they hold pot only of the specific category in question, but in general for all recursive categories.
Chapter V defines and examines the notion of "reducibility
system." The treatment is totally abstract with no specific
7.
Chapter I
The Basic Categories and Their Properties
1.
Rl,
R
2,
andR'.
1.1. Notation. We shall work, at the outset, with the natural numbers N = {0, 1, 2,...} plus a special symbol c. Our basic objects will be subsets of
N I) {co}. In general, such subsets will be denoted
A, B, ... A will denote A - {oo}. On the other hand,
00
we may wish to begin with a subset of N. Such subsets will in general be denoted X, Y, ... ; X will denote
X U {oo}. We shall also use the symbol N to denote
00
N = N ( {a}.
We adopt the following notations for special morphisms:
C-* monomorphism
++ epimorphism +~ isomorphism.
1.2. Definition. Category
Rl*
The objects ofR
1
are subsets of
I,
containing G. A morphism a : A+ B, inRi
is afunction satisfying:8.
()
3
a partial recursive function $ such that= n) E B if $(n) is convergent
for all n e A,
a(n)
Oif $ is divergent at n.
(i) a(cO) = 00.
We shall say that $ induces a. Given a partial recursive function 4, define 4* : N + N via
$ (a) if convergent S=1if divergent.
We note that
R,
is a small cetagory, though this is not essential to our purposes. Note also that a morphism inRl
is a monomorphism iff it is 1-1, andan epimorphism iff it is onto.
1.3. Axiom.
R,
has a 0 object.Proof: {w} clearly serves. We note in passing that the 0 of
R,
is unique, and not merely unique up to isomorphism,1.4. Axiom.
R,
has kernels.Proof: Let a : A -+ B be given. Let
9.
inclusion. Clearly i serves as ker (a).
1.5. Definition. Categories
R
2,
Rt.
R
2 is thesubcategory
R,
including all objects, but only morphisms with kernel 0.The objects of
R
are subsets ofa : X + Y is a function such that
3
recursive function $ with X C domainN. A morphism a partial
(4) and
$(x) = a(x) e Y for all x e X.
Clearly the categories
R
2 and R are naturallyisomorphic. Note that
R2
is definable fromRl,
soall statements about R and R will be made for R
though proofs may occasionally be carried out for
R'.
1.6. Axiom. 0 serves as a conull object for
R2;
R2
also has a null object.Proof: Any singleton {n) serves as a null
object for
R',
i.e. {n, o} is null for R2.
We shallrefer at times to R2-null objects {n, co} as singletons.
2. Coproducts, Products, and Pullbacks.
in fact, given A and B, the same pair of morphisms A + C, B -+ C serves as the coproduct in
R
1 and in R2' Proof: Recall that, for X, Y C N,
X join Y = {2xlx e X}
U
{2x+l|x E Y}. Given Al, A2, objects inRi,
let C = (A1 join A2)w. We have00 : natural injections u i A AiC-* C via: 2x if x w u(x) f x 2x+l if x
$
w and u2(W =t = 00 Suppose$ni (i1, 2) induce maps Ai + B. Then the unique compatible map C -+ B is induced by
n (c/2) if c is even 0(c)= nl
$n (c ) if c is odd. s 2 A@
Thus (C, ul, u2)
serves as A 1 A2 in Ri. Note that ul and u2
have kernel 0, and that if the maps A -+ B have
ker 0, then so does the map C -+ B, so (C, u , u2)
also serves as A1 @ A2 in
R
2.
2.2. Lemma. If A++ B, then B is recursively
00
enumerable in A.
IL i *
Proof: Say A -+ B is induced by $n. Define
e(x,
A) via: dovetail computations for n (y)W0
for all
y s A. If one eventually converges to x, give output; otherwise diverge. Then B =
{x|6(x,
A) convergent}.00 00
2.3. Remark.
R,
does not admit arbitrary infinite coproducts; similarly for R2*Proof: Let Ai = {i, cc} for i e N. Suppose
C = @ Ai, with injections f : Ai C. Let x be any infinite subset of N, say X = {x1 , x22 *...}
Let g :Ai +X via f 1 Then 3!h : C + X
such that for all i, hfi = g In particular,
C0
h : C++ X, so X is recursively enumerable in C by (2.2). Also, any finite set is certainly
recursively enumerable in C, so every set is
recursively enumerable in C. In particular, C"
00 00
is recursively enumerable in C, so C" < C1.
00 00 -M 00
2.4. Notation. Let f : A -+ B, g : C -+ E.
3
!y making the following diagram commute:C<1 2 A '--) A $ C( C f+4 i +-Y 12
denote that y by f @ g If B and E are the same,
we also have a unique map y making
B
f +y g
A C--3A ( CE- C
1 i
2
commutative. Denote that y by f + g.
2.5. Remark.
R
does not admit arbitrary finite products.Proof: Suppose P = N x N with projections p1, P2 induced by 11 , 2. Consider the following
diagram, where -3 is the constant function
1 if x Wo 0o if x = 00 ,x if x s K , and a(x) = W if x K or x = (x) = B c---+4B (0 EC E
CO N +px 00 P-N P1 P2 Claim 1: Proof: ker (y) = 0. If c # a e ker (y), a(a) = 1
#
C = p1 (CO) then = p (y(a)) *.Claim 2: If a, b e R, then y(a) = y(b).
L+ Le t 0 :N -+P if x a1 b if x = b if be defined
. Then 0 also makes the x = a
diagram commute, for pl(O(a))
6(b) = 1 = (a) and similarly
= p1 (y(b)) =
p 1 ((b)) = 3(b). Also,
P2(O(a)) p2((b))
= p2 (y(b)) = a(b) = c = a(a)
= a(b). Then 6 = y by the
and similarly,
uniqueness of
so y(a) = 0(a) = y(b).
By claim 2, all values y(a) for a e K are equal; call the common value c.
00
Proof:
(x) (a)
14.
Claim 3: K = {x e Njy(x) = c}.
Proof: By claim 2, K C {x e Nly(x) = c}. If y(x) = c and x c K, then x = a(x) = p2(y(x))
p2 (c) = C *
Since y is induced by a partial recursive function, K = {x e Nly(x) = ci is recursively enumerable *.
00 00
Hence the product N x N cannot exist.
Some additional remarks might make clearer why
R,
cannot have products. The natural picture of the00 00 C0
product N x N is {<x,y>ix, y e Ni, which would allow
pairs of the form <n,o> for example (n # co). We must have <n,00> l W, since pl(<n,oo>) = n, while
pl(w) = co. In our example above, we would expect, for
a number n E
K,
to have y(u) = <1,w>; but to definesuch a y, we would need to be able to fill in the
second element of the pair, which we rould do only by
observing that n e K, an observation which cannot be
made effectively.
Proof: We show
R
has finite products. Let X, Y C N. The functions T, 7r1, 7rr2 are familiar. Let P = T"(X x Y). Then P is a product for X, Y,with projections 7 1 : P - X, r2 : P + Y. For let
f : Z -+ X, f2 : Z - Y. The map z + T(f 1 (z), f2(z))
is the required unique map Z + P (uniqueness follows
from the fact that T is'l-1 and Onto).
2.7. Notation. Let f1 : A -+ B 1
both in
R2,
i.e. inRl
with ker 0. Yf1 x f2 the unique map y : A x C -+ B
making the following diagram commute:
and f2 : C + E, Denote by x E with ker 0 A<-A x C - C + f2 f + B BK E E Pt B x E -?E
Let f1 : A B 1 f: A + C with ker 0. Denote by f @ g the unique map y:A + B x C with
ker 0 making the following commute: A p1 P2 B,,--- B x C - C 2.8. Remark. A x B = B x A; A @ B = B 4 A A x (B x C) = (A x B) x C; A $ (B $ C) = (A $ B) @ C We have seen that
R
admits finite but not infinitecoproducts nor does it admit finite products. However,
R2
does admit finite products. We see finally that:2.9. Remark.
R2
does not admit infinite products (similarly forR
1).
Proof: Let Y = N for r = 0, 1, 2,..., and
suppose P = X Y = X N, with projections p i P + N.
i i
Let f0 : N +Y = N be the function which is every-where 0. Let f: N -+ Y = N (i > 0) be given by
1'
- if $ (x) fails to converge in < i steps
O).(x)
1 if $(x) converges in < i steps.
Note each f is total, so is a map in R1. Then
IT:N+P such that p y = fi for &ll i. Note y is total,
since it too is a map in
R',
Claim 1. If x,y c X, then y(x) = y(y).
Proof. This follows from the uniqueness of y, as in the proof of (2.5). Let c be the common value
of y(x) (x C K).
Claim 2: K = {xjy(x) = c} which is recursively
enumerable * . This claim also follows as in (2.5).
Mitchell (1.17.3) proves the following useful
2.10* Theorem. The following statements are equivalent in a category
A.
(a)
A
has finite intersections and finite products (b)A
has equalizers and finite products18.
2.11. Axiom.
R
has equilizers.Proof: Let a, : A + B. Let
C = {x e Ala(x) = a(x)}, and i the natural inclusion i : C A. Clearly i serves as Equalizer (c,3).
2.12. * Corollary.
R
2 has equalizersProof: Given a, : A -+ B in R2. Let i : C4C-A
be the equalizer in R 1
.
Then i is a monomorphism,hence has ker 0, hence is an R2-map. Let f : E -+ A
in R2 such that af = $f. Then f factors through i, in
Ri,
say f = is. Since f has ker 0, so does s(cf. 2.19 (11)), i.e. s is an R2-map.
2.13. * Corollary.
R2
has pullbacks and finite intersections.We can actually do a little better than this.
2.14. Axiom. Let u : A -+ C, v : B -+ C be
R
1-maps with ker 0. Then inR,
there exists a pullback P, P2 P -- 9 B + + A- C u-L j. For x E A, y E B, u(x) 00 and v(y) W0 both $ w since p = {T(x,y)Ix e
U, v have ker 0. Let A,
00
y e B & u(x) = v(y)} {0}. Let
00
be induced by (i = 1, 2). Then the above diagram commutes.
Suppose u and v are induced by 0 n
resp., and let f1 : E - A f2 : E - B be induced by
so uf1 = vf 2. Define
e
as follows: dovetail computations $mx).
If either is e(x) = T($k W, Clearly 64 ,(x))
convergent, let when convergent.is the required map for the pullback diagram: f2
3'
-1 Plj tv - A C u Proof: are pi resp. and m given for Onok(x) and W i xILU.
2.15. * Corollary.
R,
has finite intersections. Proof: An intersection of two monomorphisms is just a pullback. Higher order finite intersections are gotten by induction.2.16. * Corollary. If f : A -+ B is an
R
1-map with ker 0, then f has inverse images;R2
has inverse images.2.17. * Corollary.
R
2 is finitely complete.Proof: As Mitchell (II 2.6) points out being-finitely complete is equivalent to having finite intersections and finite products.
We close this section with some simple remarks
about kernels and 0.
2.18. * Remarks. (i) Let
A
be an arbitrary category with 0. If u : A C-0O is a monomorphism,then u : A + 0.
u v
kernels, and A + C + B.
(ii) Let
A
have 0 and21.
has ker 0; also ker (vu) = 0 => ker (u) = 0.
(iii) Let u K C+A,
v : A4*'B, a : B + C such that vu = ker (a). Then
u = ker (av).
(iv) Let u K C!A be the kernel of d: A +E and say E
:E-+C has kernel 0.
Then u = ker (0a).
Proof: (i) Let v : 044 A be the unique
(mono)-morphism from 0 to A. Then uv : 0 - 0,
hence uv = 10. But then uvu = u, and u is a
monomorphism, so vu = 1 , hence u is an isomorphism. (ii) Consider: 0 + 0 + 0. The squares
A +C +B u v
are pullbacks, since u, v have kernel 0 (MItchell, p. 14), and 0 -+ B is a monomorphismn, so the outer rectangle is
a pullback (Mitchell, I 7.2), hence vu has ker 0.
For the second assertion, we have
ker cu) '.pker (vu) = 0 as in Mitchell, p. 15, so by part (i), ker (u) = 0.
22.
K +0
A v
B +C
(iii) We have a pullback
The claim is that K 0 is also a
u
AC-B +* C
v a
pullback. It is obviously commutative. Given a
map f : E - A such that avf = 0, we get a
unique map y : E -+ K such that vuy = vf. Since v is a monomorphism, uy = f. That such a y with the property that uy = f is unique follows from the fact that u is a monomorphism.
(iv) Consider the following diagram.
K +0 +0
uf + +
A +E +C
a S
Each square is a pullback, since 0 = ker (S) and u = ker (a) and 0 -+ C is a monomorphism, so the
3. A Question of Balance
Recall that a category is balanced iff every
mono-epimorphism is an isomorphism. It is this property we wish now to explore for Rl.
3.1. Theorem. Let u : A'(-+B be a mono-epimorphism.
u is an isomorphism iff u is induced by a partial
recursive e which is 1-1 on its domain.
Proof: => Let u be induced by n' u~ by
$m. Define 0 as follows: given x, compute $ *(x) If it converges, compute $mn x). If this is convergent, see if $ n (x) = x. If so, give output $ n(x; if
not diverge.
Claim 1: 6 is 1-1 on its domain. For let
x, y e domain (0), with $n(x) = O(x) =
6(y)
= o (y). Then x o otp (x) =4
o Y) = yo The x=
m n M n*
Claim 2: 0 induces u, since if x e A, 0(x) is convergent and equals n Cx) = u(x).
: Let , induce u, with 0 1-1 on its domain4
given x, dovetail computations for e(0), e(1), ... until some convergence to x is found; if 6 (y) = x
is the first such convergence found, let 4(x) = y.
Note that G 1-1 =>
3
at most one y such that0(y) = x. Further, if x e B, such a y exists and
00
belongs to A, hence 4 induces v : B + A. Clearly
00
v serves as inverse for u.
We shall see shortly that all infinite recursively
00
enumerable sets X are isomorphic in
R',
and that any set isomorphic to a recursively enumerable set isitself recursively enumerable; we may use this to
3.3. Remark.
3
non-isomorphic objects A, B,and a mono-epimorphism u : A B. In particular,
u is not an isomorphism, so 1 is not balanced.
Proof: It suffices to show
3
non-isomorphic sets X, Y _ N, and a mono-epimorphism u : Xc+"Y inR1.
Suppose not. Let 0 be the following (total) recursive function: O(x) = least prime factor of x. Note 0 : N-.-P = {primes}, and P is recursivelyset of integers with the properties that
U) 0 is 1-1 on Z, i.e. no two distinct elements of Z have the same least prime factor
and (ii) for each prime p,
3
z s Z such that p is the least prime factor of z.Then 6 induces ZIC*OP, so Z
=
P, i.e. Z is recursively enumerable. However, there are clearly uncountably many such Z, and only countably manyrecursively enumerable sets. *
In particular, let {p1, p2, ... } enumerate the
primes, and let {a,, a ... } be a (non-effective) enumeration of K. Let Z = {pnanln e N}. Then 0 induces ZC'P, but Z is not recursively enumerable, since Z recursively enumerable => K recursively enumerable clearly.
In fact, this result can be strengthened to show that
R
andR2
are extremely unbalanced: given any infinite set X, I a set Y and mono-epimorphismf : Y -+ X which is not an isomorphism. We need first
the following:
26.
3.4. Lemma. For every pair of infinite sets
X, Z, 3 a set Y and mono-epimorphisms
f Y4**X, g : YC-*?z.
Proof: Enumerate X = {x1, x2, .. 1 and Z {Z 2 ... } (non-effectively). Let
Y = {<Xi, z >|i E NI and take f = 1st projection,
g
= 2nd projection.3.5. Theorem. Given any infinite set X,
3
a set Y and mono-epimorphism f : Yc+)X which isnot an isomorphism.
Proof: Since the number of partial-recursive functions is countable, and the number of infinite sets is uncountable,
3
a set Z such that there is no mono-epimorphism XC+iZ. Pick Y, f, g as in the Lemma: f : YC-*-X, g : Yc-+*Z. Note f is not anisomorphism, for if so, we have gf 1 : XC+OZ *. The imbalance of
R
and R2 is disappointing, since it immediately forechoses the possibilities of normality and exactness, notions which have been of tremendous value in categorical studies of algebra.27 .
Evidently
R
1 is not an algebraic beast, but, as we shall see, there remains much of interest to say aboutit.
Theorem (3.2) yields easily as a dorollary the fact that if A is recursively enumerable and
00
u : A - B is a mono-epimorphism, then u is an isomorphism. We shall, in fact, prove this result
categorically later on; at this point it interests
us as it raises the follo*ing question: does this property categorically characterize the recursively
enumerable sets, i.e. if A is such that every
mono-epimorphism with domain A is an isomorphism, must A be recursively enumerable? Our question is
00
answered in the negative by the following
3.6. Theorem.
3
a set X such that X,are immune (in particular, Xis not recursively enumerable) with the property that any mono-epimorphism
00
u : XC4OB is an isomorphism.
Proof: We construct X by stages. At each stage
at most a finite number of integers will be placed in
cd.
respectively.
Stage 2n: Consider Wn, the nth recursively enumerable set. If it is infinite, find integers an, bn e Wn different from all integers so far marked. Mark an with a 1 and bn with a 0.
If Wn is finite, go to the next stage.
Stage 2n+l: Consider $n. Let
Y2n+1 = {integers so far marked with a 1},
Z2n+l = {integers so far marked with a 0}. See
(not effectively) if $n is 1-1 on its domain. If so, go to the next stage; if not, there are 2 cases:
case (i). x, y e dom (4 ) such that
$n(x) = 4n(y) and both of x, y e Z2n+l. Then pick
such a pair and mark x, y with a 1 (if they are not already so marked), and proceed to the next stage.
case (ii) nCx) = n (y) (both convergent)
=>
eitherx e Z2n+1 or y c Z2n+i. Then go immediately to the
29.
Let X = U Y 2n+1= {integers marked with a 1}. neN
Clearly the even numbered steps insure that X, X are both immune.
00
Now let u : XC**B be given, a mono-epimorphism, induced by say $n. We want u to be an isomorphism, and by (3.1) it suffices to show that u is induced by some partial recursive 6 which is 1-1 on its
domain. If 4n is 1-1 on its domain, then u is an isomorphism and we are done. So suppose n is not 1-1 on its domain. There are two cases.
Case (i) x, y e dom ( n) such that n Cx) n(y) and both x, y i Z 2n+l* Then for some such pair ,
x, y s X, so $n is not 1-1 on X *.
Case (ii). $n Cx) = n (y) => either x e Z2n+l or y e Z 2n+l* Now Z 2n+ is finite, hence recursive,
so the function
e
is partial recursive, where rn(x) if x i Z 2n+0(x) =. Note
divergent if x e Z2n+l
domain 6. Since X Cdom ($n) (u has kernel 0) and X
fl
Z2n+1 = $, X C dom 6, so 0 induces u. Finally 0 is ll on it5 domain, for if not, we would be in case (i). Hence n is an isomorphism.4. Image, Coimages, Reversible Monomorphisms, Quasi-Unions and Essential Domains
A natural question to ask of a category is whether arbitrary maps have images. Recall that the image of
a map a : A -+ B is a monomorphism i : I4 B such
that a factors through i, and such that for any other monomorphism
j
: JC-vB such that a factorsthrough j, i factors through j.
4.l. Remark In
R
it is not in general the case that the natural inclusion of the set theoretic image serves as the categorical image.Proof: Let f : Z'4-P be as in (3.3), with P
00
recursively enumerable but Z not recursively enumerable. 00
Just suppose 1p : P + P serves as Im (f), Since f factors through f, 1 must also factor through f.
U b
A moment's thought establishes that this forces f to be an isomorphism *. It should be pointed out that f does have an image, namely itself.
The question of general images in
R
remains unsettled. The following is known, however.4.2. Proposition. Let A be a recursively enumerable set, and let f : A + B. Then f has an
image; in fact the natural inclusion of the set theoretic image works in this case.
Proof: Let f"A be the set-theoretic image of f, and let i : f"A -+ B be the natural inclusion. The
assertion is that i = Im (f). Clearly f factors through i, say f = i7. Let j : J B be a
A A
monomorphism such that f = Jf for some f : A -+ J.
We must show i factors through j. Let $n induce f; since A is recursively enumerable we may assume
00
dom (n) C A. Define a partial recursive function 0
00
as follows: given x. begin a Stand'ard dovetailing
32.
t he first integer such that this procedure gives
n (y) = x, if any such exists. Let O(x) = f(y). Claim 1: 6 induces k : f"A -+ J,, For let
x e (f"tA).. Then y e A such that n(y) =(y) X.
Then f(y) s J.
Claim 2: jk = i. For if x c (f"A) , then k(x) = f(y) for some y c A such that n(y) = X.
A
Then jk(x) = jf(y) = f(y) = iT(y) = i4)n(y) = i(x).
Fortunately,
R
1 behaves much more nicely withrespect to coimages. In fact, if a : A -+ B is a map,
then the induced map a : A++cat"A serves as the coimage, and we have the following
4.3. Axiom. Every
R
1 map a : A + B factors- i
into Aa ++ B, with A ++ I the coimage of a.
Proof: Let : A -+a"A be induced by a, and i the natural inclusion. Clearly a = i. Let
: A + J be an epimorphism such that for some
6-) C
x F J,
3
y c A such that $(y) = x. Thenk(x) = kS(y) = a(y) c a"A, so k factors through i. We have the following diagram
A -I
I,
with iU = k. monomorphism, commutative.
Then i3= a=k so a = j , so the = ij, but i is a entire diagram is 4.4. Definition. Let u : u A - B u C i A + B be be the factorization a monomorphism, through the
coimage. Note U is a mono-epimorphism. u is reversible iff u is an isomorphism.
4.5. Remark. u : ACa;B is reversible iff u
induces an isomorphism A + u"A.
4.6. * Remark. Let a : A -+ B have coimage
factorization A DB
C
. Then v is reversible. v
Proof: Let v have coimage factorization
V C C-"-'2B
. We show v a: A + E also serves
as coimage (a), hence that v is an isomorphism by
the universal property of coimage. Clearly v a is
an epimorphism, being the composition of epimorphisms, a
and a factors through v a . Let A -+ B be given.
Then j: G -+ C such that jf = a. Then
Vjf : G + 7 and vjf = v a.
4.7. * Corollary. Equalizers (hence kernels) are
reversible.
Proof: Let a,B : A -+ B, and let i : E A be
EqU (a, f). Factor i through its coimage i =
jf,
with
j
: E + A a monomorphism. Claim : j = Equ (a,$). We have aji = ai = 0i = aJr, but is an epimorphism,so aj =j. Given f : C + A such that af f, we
have g : C E such that ig = f. Then g =C and jig ig = f.
We seek a concept of union in Fl, but it turns out to be more convenient to define a related concept,
which is simply the notion of union restricted to reversible monomorphisms. To be more precise:
4.8. Definition. Let {u i : A i-+ A} be a
family of (not necessarily reversible) monomorphisms. A quasi-union u : A' C+ A is a reversible monomorphism
such that
(i) each u factors through u
(ii) given any object B, map f : A -+ B, and reversible monomorphism v : B' C B, if each ui gets
carried into v, then so does u.
4.9 . * Theorem. Quasi-unions are unique (if they exist) up to isomorphism as subobjects.
Proof: Let ul : A C+ A and u" : A" 4 A be two quasi-unions of {u ::A 4 Al. Since u" is reversible, and 1A carries each u
into u", u' is also carried into u". We have
'35.-w
A' + A"
u U" and since lAu is a monomorphism, so
A - A 1 A
is w. Hence ul precedes u" as subobject of A; similarly u" precedes u', so the two are isomorphic
as subobjects of A. (cf. Mitchell, p. 6).
4.10. Axiom.
R
admits arbitrary quasi-unions.Proof: Let {u i: A C A} be given. Then each u i"A C A,
the natural
so A' = U ui"A C A. Let u -A' + A be
is
inclusion so u is reversible and every factors through u.
Let f : A -+ B be given and let v : B' C+. B be
a reversible monomorphism WOLOG, v is the natural
inclusion (otherwise we may replace B' by the
isomorphic subobject v"B' C B). Suppose f carried
each u. into v. Then fu "A C B' for each i, so f"A' = f"(Uu i"Ai) = U fu "A CB', so f carries
i i
37.
Since no mention shall be made of traditional unions, we shall reserve the notation U u for the
i quasi-union.
The following fact is worth noting. Since no further use is made of it, it is not presented as an axiom.
4.11. Fact. Let {ui : A iC A} be a family of reversible monomorphisms. Then u exists.
Proof: WOLOG, assume each ui is a literal inclusion. Let A' = A and v : A'(-'l A the
natural inclusion. Then v factors through each of the ui. Let w : B -+ A factor through each u.; then for each i, w"B C A , so w"B n A = A', so w factors through v.
4.12. Definition. Let a : A + B be an
Rl-map.
Let D be a singleton (i.e. an R2-null object).Let C = {u : D-+ Ajau P 0}. The essential domain of a, denoted Edom (a), is U u.
4.13. Remark. Let a : A -+ B. The natural
inclusion {a e Aja(a) convergent}w C A serves as Edom (a).
5. Singletons
Clearly singletons will play a very important role in this work. The present section is devoted to
exploring some of their basic categorical properties. (Throughout this section, let D denote a singleton).
5.1. Axiom. Every map in
R,
with D as domainand kernel 0 is a monomorphism.
5.2. Axiom. B # 0 iff
3
a monomorphismv : D C+ B.
5.3. * Remark. Let u : A + D be a monomorphism. Then A = 0 or u is an isomorphism.
Proof: If A # 0,
3
v : D + A a monomorphism. Since uv : D -+ D is a monomorphism, it has ker 0,hence uv = D m iap D + D). Then uvu
= lDu = u = ulA and u is a monomorphism, so vu = 1A* Thus u is an
39.
5.4. * Remark. If A
$
0, and a : A - D hasEdom (a) # 0, then a is an epimorphism.
Proof: Suppose we have g, h : D + E such that
ga = ha; we must show g = h. Let w : BC+ A be Edom (a).
Since B p 0,
3f:
DC4 B. Note awf : D-+D is not 0;it follows that awf has kernel 0 (if not then
A q : D ( D such that awfq = 0, but q must be 1D since it is a kernel 0 map from D to D). Further awf = 1D. Then g = glD = gawf = hawf = hlD = h.
5.5. * Remark. Let A (# 0) be such that every
map with A as (categorical) domain and kernel 0 is
a monomorphism. Then A is a singleton.
Proof: Let v : A -+ D with kernel 0 (such exists
since D is R2 null). By hypothesis, v is a
monomorphism, so by (5.3), A 0 or v : A + D. 5.6. * Remark. D x D D.
Proof: It suffices to show that D x D is R2 null.
Given object A, we have v : A -+ D with kernel 0,
U .
a, : A + D x D with kernel 0. Let p1, p2 be the projections. Then p1a, piS : A -+ D with kernel 0,
hence pla = plO. Similarly, p2a = P2r. Hence a =
The following axiom asserts that every object is made up of singletons:
5.7. Axiom. For every object A, 1A = U{u : D-+, Al Proof: Clearly 1A is reversible, and each
u factors through 1 Given f : A - B, and v : B' C+ B, reversible such that each u gets
carried into v, we may sssume that v is a literal
inclusion. Then each u"D CB', and since each u i DC* A picks out a member of A and conversely, it follows that A
C
B', so 1A gets carried into v.As we shall see subsequently, the following axiom asserts that eacy singleton is recursive.
5.8. Axiom. Let u : D C+ B. Then
I
E anda : B - E such that u = ker (a).
Proof: Let D = {O,oo}, and let x = u(O) e B. 00
Define 0(y) =
y if y d x
convergent if = x.
Then 0 induces
B - N with kernel u clearly.
Actually we shall want another formulation of this recursiveness, to wit:
5.9. Axiom. Let u : DCE*B be given. Then
3
w : A C-0B such that B = D $ A, with injections u and w.Proof: Simply take the natural inclusion
B - u"Dc4B. 00
6. Complements
Throughout this section, D denotes a fixed singleton.
6.1. Definition. Maps u : A + C and v : B + C are disjoint iffm monomorphism DeN9C factors through both of them. Clearly u and v are disjoint
iff (u"A)00 and (v"A)0 are disjoint sets.
6.2. * Remark. Let u : A -+ C ahd v : B +
42.
Then au and av are disjoint.
Proof: Suppose w : D -+ E factors through both au and av, i.e. we have commutative diagrams
u a A + CC E V a B - C'+ E and g D D w
Then auf = w = avg, and a is a monomorphism, so uf = vg. Moreover, since w has kernel 0, so does uf = vg : D -+ C. Since D is a singleton, uf = vg
must be a monomorphism (5.1) D4+ C factoring through both u and v *.
6.3. Definition. Maps u : A + C and v : B + C
are complementary iff
(i) they are disjoint and
(ii) every monomorphism D 4 C factors
through one of u or v.
43.
(u"A). and (v"B)O are relative complements in C.
00
Given a map a : A -+ B, a complement of a is a
reversible monomorphism C 4 B, complementary to a.
6.4. * Theorem. Complements (when they exist) are unique, up to isomorphism as subobjects.
Proof: Let
8
: C -+ B and8'
: C' 4 B betwo complements to a. Let v : D + C. Then av
factors through 8, hence not through a, hence through
8'. We have the following situation:
D -- -- >Ct
4?
, i.e. a carries v into8'.
C
c-
V
Since 1C = U{v : D - C1, it follows that # carries 1C into 8', hence that a precedes 8'. Similarly
8' precedes 8, so the two are isomorphic subobjects
44.
6.5. Axiom.
R,
admits complements.1Proof: Given a : A -+ B, the natural inclusion
i (B - a"A)"f4 B serves as the complement.
We shall denote the complement of a : A -+ B by
ca : K + B.
7. The ordering <c
7.1. Remark. Let X N. Then in
R
X X join $ via Xx[2x1.7.2. Lemma. Suppose in
R',
X = Y join V forsome V C N, and Y = X join W for some W CN. Then X = Y.
Proof: If X and Y are finite, it is easy to
see they must have the same cardinality, hence are isomorphic. Assume for the duration that X and Y
are infinite. Let $n induce the map X join W + Y, 1 This axiom is actually superfluous. In fact, with the aid of (III, 1.1), one can easily show categorically that
for any map a : A B U{u : D 4 Blu does not factor through a) serves as the complement.
45.
with inverse induced by $n,. Let $m induce the map
Y join V + X, with inverse induced by $m,. The maps
Y 4 Y join U and Xc+ X join W are induced by Xx[2x].
Define 01, 02 as follows:
On , o $ (2u)
e (u) = n' 2 (note divergent if 0 $~a0 4n (2u) is odd)
0m rft (2u)
0,2(u) = 4)'~ 2
Now let R be a recursively enumerable set of ordered triples of integers, gotten from the following instructions: list (a, b, c) iff
(i) a = 61(u) for some u & $ n n' '1 n (2u)
is convergent and = 4)n (2u) = b & c = 1
or (ii) b = 2 (v) for some v & 4m 0 m ' c m (2v)
is convergent and = $m (2v) = a & c = 2.
By dovetailing computations, one can effectively generate R in some order. Let
46. R = {(al, b1 , c= 1 or 2. Claim 0: similarly, Claim 1: as an aj in c 1 ), (a2, b2, c2), ... }, where u e X => v e Y => For each
61(u) = u & no n On(2u) =
%n
(2u) e Y02(V) = v & 0mO mi m(2v) = $m(2v) c X
U e X, u appears at most twice
R; one time for sure with ci = 1, and possibly another time with ci = 2. Analogously for
each V E Y.
Proof: Let u e X. Clearly the triple
(U = 01 (u), $m(2u), 1) e R. Suppose (u, b, 1) e R. Then u = 1 (t) for some t such that
On(2t) = n 0 0n' o n (2t) and b = $n (2t). Now
U n' On(2t)
u = 2 , so
2u = 0n' A n (2t), so
On (2u) = $n 0 0 n' a n(2t) = $ n(2t) = b. Thus
(u, $n(2u), 1) is the only triple in R with first coordinate u and last coordinate 1.
4'.
Suppose a triple (u, b, 2) e R. Then b = 02(v) for some v such that
m m' C m(2v) = 4m(2v) = u. Then $m,
4(2v)
b = 2 so, since u e X, $ (u) b = 2-7- C4,(u) . Note 4,m,(u) must be even $m,(u) C Y join U, and is even, hence
Claim 2: For each (a, b, c) Proof: Immediate
Notation: For x
s R, a e X <=> b s Y. from claims 0, 1, and their proofs. s X, let fi(x) = unique y
such that (x, y, 1) s R. For v e Y, let
f2(v) = the unique u such that (u, v, 2) e R.
We now list a set R' of ordered quadruples of
integers inductively as follows. Say
(a t, b1', c1t, d 11), ... (ak', bkt, ck', dk') have
been listed, subject to the following conditions: If c ' = 1, then (d.', bit, 2) E R; if c
j
= 2,to the following procedure.
(1) For u not yet appearing as some a ,
search in R for u = a with c = 1. If such is found, momentarily stop dovetailing, and define f(a ) as
follows: If b 1 b for all i < k with c ' = 2,
let f(a ) =b . Say b =b c 1' = 2. If
(i < k) and
d t b ' for all k < k such that C = 2, let f(a4) = di '. If d 1' = b 2' for some i2 < 2 k with c 2 = 2, continue the procedure
1i2
with d2' . In this way the sequence
b , d 1' , d 2
is built up. Let f(a ) be the first repetition in the sequence if such occurs, otherwise f(a ) = d i
s for least s such that d b for all k < k with c zI = 2, We shall say a is "searching for"
f(a ). Now go back over all computations that have so far converged according to this procedure,
including part (2) below (only finitely many have thus
far converged, since each such computation must arise
from a number preceding a. in the list of tuples
in R), and see if any of them are searching for a j If v is searching for aj, and is the first such,
then add the quadruple (aj, v, 2, b ) to the list. Note that it satisfies the conditions. If no
computation is presently searching for aj, continue the dovetailing computation by searching through R
for some t such that (t,f(a ), 2) e R. If one is found either in the course of this computation, or subsequently in the course of some other computation, add the tuple (aj, f(a ), 1, t) to the list. Note that it satisfies the condition.
(2) Simultaneously, perform analogous computations for all v which have already appeared as some a i
but have not yet appeared as some b '. In the course of this computation we shall say v = b is searching
for g(b ). At any given stage in the process there
will be at most one tuple appropriate for adding
to the list. If there is one such, add it and begin
recursively enumerable.
Claim 1*: If (u, v, c, d) e R', then
u C X <=> v C Y.
Proof: Say (u, v, c, d) c R' and c 1. We must be in situation (1), i.e.
(u, v, c, d) = (aj, f(a ), 1, t). Note
(a4 , b , c ) c R, so u e X <=> b e Y <=> b ' = b c Y <=> a ' 1 X <=> d ' e Y <=> b ' Y <=> 1 12 a ' c X <=> d 1 t Y <=> ... <=> v = f(a ) = d i I Y. 2 i2 s
If c = 2, the argument is similar: here the tuple
must have form (g(b ), b , 1, t).
Claim 2*: (V u C X) (3 at most one v)
(3
c)(Jd)[(u,v,c,d)c R'
analogously, ( v c Y) (
3
at most one u) (3 c)(3 d)[(u,v,c,d) e R'1Proof: Supppose (u, v, C, d) c R' was put in as
the k+1st tuple, u e X, and (u, v', C', d) c R' was
put in earlier, and this is the 1 place of such
difficulty. If c = 1, we have an immediate contradiction,
51.
computation at stage k. So say c = 2. Then v = b and u = g(b ) for some J. Since u c X, v = b e Y. Subelaim: Since v e Y, the appropriate sequence
a = d ' , d 1' , d 2 , ... contains no repetitions..
Proof: of subclaim: We have (a , b , 2) e R, aj = a a =s1, so (aj, b' 1, d 1') e R', with
b b ( by const. of type (2)), hence
(di ' b 2) e R'. Note b e Y => a E X =>b ' e Y >
d CX. Since b bi ' and both
(aj, b , 2), (d 1', b 1', 2) c R, a # d i' by claim 1'.
Similarly, the sequence consists of all distinct members. There are now two cases:
(i) c' = 1. Since the above sequence has no repetitions, u = g(b ) cannot equal a 1 for any
>2.
(ii) c' = 2.
We have sequences v = b , a = d 0, v, U 1, V2, u2 ***,vsus=t
and v' = bj,, aj, = di , v 1 , u', v2', U2 v t',s = U,
satisfying:
(i) (aj1 V, 2), (u1 , v1 , 2), ...,(u=ussvS2) eR
(a '$ v', 2), .. , (u ' = u, v t1 2) e R;
and (ii)
have already been listed in R', as have been
(a ' v ' 1 , u I),...(ut, ', 1 , i u t')9 By
claim 1 and the fact that R' has no repetitions on each of the first two coordinates so far we conclude that
Us = u = Ut; v s Vt,; U = us-l ***
There are three cases:
(i) s = t, i.e. v' = v
(ii) s < t. Then v = v ' for some Z, and would already have appeared as a second coordinate in the list,
so could not now have been considered for a type (2) calculation.
(iii) s > t. Then v' = v for some X < s. Then (u2_1, vx,1 1, u5 ) already appears in R', but R'
contains no repetitions as yet, so u = u = u . It follows that v x 1 = v5 by the "1-1 ness" of R
with respect to X, Y (Claim 1), which implies that R' does have repetitions so far *
This includes the proof of half of claim 2 The other half is proved similarly.
*.
Claim 3*: ( V u E X) (3 v, c, d)[(u, v, c, d) e R'];
(V v e Y) ( 3u, c, d) [(u, v, c, d) e R'].
Proof: Let u e X. The preliminary search for some j such that u = a with cj = 1 eventually succeeds, Suppose k tuples have been listed by such time as a convergent computation is reached. One
immediately defines f(a ) and begins searching for it. Eventually, since f(a ) E Y, a t such that
54.
(t, f(a ), 2) e R will be found; however, by this time more tuples may have been added to R', say tuples k+l through k. Suppose u has still not been added as a first coordinate in any tuple.
Subclaim: At this point in the construction,
if one defines f'(a ) to be the new sought integer, fr(a') = f(a ).
Proof of subclaim: It suffices to show f(a ) b s for any s such that k+1<s<k . Suppose f(a) =b
for some such s. If c' = 1, then since b5 I= f(a ) e Y, as' e X and we have
(as, f(a ), s , e R'. Then we have sequences
as',
vl'
, u2 Pv
2, *.., u', f(a ) andaj, v1, u22 2, ... , up, f(a ), as in the proof of
the previous claim. As was shown there,
u ' = up, u = UP-1,... Also, v1 = bm for some
m < k. There are three cases:
(i) u = aj = a * (u has not yet appeared as first coordinate)
55
(ii) as' = u. for some uj in the sequence
aj, v1 u 2 , ... ; but all such uj actually appear as
first coordinttes before the k+lst tuple. *
(iii) a = ui I for some ui in the sequence as vt l, u2 , ... ; but each u i' actually appears as
first coordinate before the sth tuple *,
Hence if f(a ) = bs for some s' such that k + 1 < s < X, then cs' = 2; but then in the
computation for bs' we would actually discover some t such that (t, bs' = f(a ), 2) e R. This would conclude our search for f(a ) in the computation for a,
and according to the instructions, the tuple (a , f(a ), 1, t) would have been added at this time. * (aj = u has not
yet appeared as first coordinate).
This establishes the subclaim. Since f'(a) =f(a
and the search for f(a ) has been concluded, the tuple (aj, f(a ), 1, t) is now added to R', establishing half the claim. The other half is established similarly.
Putting claims 1*, 2*, and 3* together, we have:
(Vu e X)(J!v)(3c)(3d)[(u,v,c,d) e R'] and that
V e
Y(V v s Y)(3!u)(3c)(Jd)[(u,v,c,d) e R'] and that u e X.
Clearly the following partial recursive function a induces an isomorphism X + Y in R2
':
a(u) = b ',where b ' is the first integer such that c, d are discovered making (u, b ', c, d) e R', The inverse is induced by (v) = aj, where a is first integer such that
3
c, d making (aj v, c, d) e R'.7.3. Remark. A = B in R, <=> A = B in
R
<=> A = B in R21'.20 0 0
7.4. Definition. GIven A, B, objects in
Rl.
A is categorically smaller than B (A <c B) iffah object C such that B A $ C. Note c is well-defined on isomorphism classes.
7.5. Theorem. -c< is a partial ordering. Proof:
(i) < is reflexive: Given A, A = A @ join $
-c 00 Go
in R2
',
so A= (A joini)
= A e 0 inRl.
00
(1i) < is transitive: Suppose B = A e E, C B $ F. -C
Then C = (A $ E) $ F = A e (E e F).
(iii) -C < is antisymmetric: Suppose A < B & B < A.
-c
-Then 3 C, E such that B = A $ C, A = B $ E. Then B = (A @ )00
=
A join C; A = B join E; so by the lemma,0600 00- 0 00
00
A B in R2 ' , hence A B. 00 00
The ordering <c has been studied extensively by Dekker and Myhill. Letting fl = the set of isomorphism classes in
R
2',
and viewing @ as a binary operation,they show (2, @) is a commutative, partially ordered groupoid with the finite refinement property. Our
work shall not take this direction, since it seems very unlikely that the < ordering defined in w-recursion
theory will tnt gewralb to a partial ordering in generalized a-recursion theory for admissible a. Aside from one application in the next chapter, in
providing a second characterization of N, we shall make no further use of c'
56.
Chapter II
Recursion Theoretic Properties Expressed Categorically
This chapter aims to translate various basic recursion theoretic notions into statements about Surprisingly, many such notions are amenable to this procedure, including recursively enumerable, recursive,
immune, cohesive, index (G8del-numbering) and
productive. Throughout this chapter, D denotes a singleton.
1. Characterizations of
N
1.1. Definition: An object A in
R
1 is finite iff it is a finite coproduct of singletons. If A is not finite or 0, A is infinite.1.2. Remark. A is finite iff A is finite.
00 In fact, A= q i=l,. . D iff A = n. o
1.3. Lemma. X C-N is infinite recursively
enumerable iff N = X in R2t
.
Proof: => : Say X -is infinite recursively enumerable. Then X is the set-theoretic image of a