This action might not be possible to undo. Are you sure you want to continue?

# Turing Oracle Machines, Online Computing, and

**Three Displacements in Computability Theory
**

Robert I. Soare

∗

January 3, 2009

Contents

1 Introduction 4

1.1 Terminology: Incompleteness and Incomputability . . . . . . 4

1.2 The Goal of Incomputability not Computability . . . . . . . 5

1.3 Computing Relative to an Oracle or Database . . . . . . . . . 5

1.4 Continuous Functions and Calculus . . . . . . . . . . . . . . . 6

2 Origins of Computability and Incomputability 7

2.1 G¨odel’s Incompleteness Theorem . . . . . . . . . . . . . . . . 7

2.2 Alonzo Church . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.3 Herbrand-G¨odel Recursive Functions . . . . . . . . . . . . . . 9

2.4 Stalemate at Princeton Over Church’s Thesis . . . . . . . . . 10

2.5 G¨odel’s Thoughts on Church’s Thesis . . . . . . . . . . . . . . 11

3 Turing Breaks the Stalemate 11

3.1 Turing Machines and Turing’s Thesis . . . . . . . . . . . . . . 11

3.2 G¨odel’s Opinion of Turing’s Work . . . . . . . . . . . . . . . . 13

3.2.1 G¨odel [193?] Notes in Nachlass [1935] . . . . . . . . . 14

3.2.2 Princeton Bicentennial [1946] . . . . . . . . . . . . . . 15

∗

Parts of this paper were delivered in an address to the conference, Computation and

Logic in the Real World, at Siena, Italy, June 18–23, 2007. Keywords: Turing ma-

chine, automatic machine, a-machine, Turing oracle machine, o-machine, Alonzo Church,

Stephen C. Kleene, Alan Turing, Kurt G¨odel, Emil Post, computability, incomputability,

undecidability, Church-Turing Thesis, Post-Turing Thesis on relative computability, com-

putable approximations, Limit Lemma, eﬀectively continuous functions, computability in

analysis, strong reducibilities. Thanks are due to C.G. Jockusch, Jr., P. Cholak, and

T. Slaman for corrections and suggestions.

1

3.2.3 The Flaw in Church’s Thesis . . . . . . . . . . . . . . 16

3.2.4 G¨odel on Church’s Thesis . . . . . . . . . . . . . . . . 17

3.2.5 G¨odel’s Letter to Kreisel [1968] . . . . . . . . . . . . . 17

3.2.6 Gibbs Lecture [1951] . . . . . . . . . . . . . . . . . . . 17

3.2.7 Godel’s Postscriptum 3 June, 1964 to G¨odel [1934] . . 18

3.3 Hao Wang Reports on G¨odel . . . . . . . . . . . . . . . . . . 19

3.4 Kleene’s Remarks About Turing . . . . . . . . . . . . . . . . 19

3.5 Church’s Remarks About Turing . . . . . . . . . . . . . . . . 20

4 Oracle Machines and Relative Computability 20

4.1 Turing’s Oracle Machines . . . . . . . . . . . . . . . . . . . . 21

4.2 Modern Deﬁnitions of Oracle Machines . . . . . . . . . . . . . 21

4.2.1 The Graph of a Partial Computable Function . . . . . 23

4.2.2 The Graph of an Oracle Computable Functional . . . 23

4.3 The Oracle Graph Theorem . . . . . . . . . . . . . . . . . . . 23

4.4 Equivalent Deﬁnitions of Relative Computability . . . . . . . 24

4.4.1 Notation for Functions and Functionals . . . . . . . . 25

5 Emil Post Expands Turing’s Ideas 25

5.1 Post’s Work in the 1930’s . . . . . . . . . . . . . . . . . . . . 26

5.2 Post Steps Into Turing’s Place During 1940–1954 . . . . . . . 26

5.3 Post’s Problem on Incomplete C.E. Sets . . . . . . . . . . . . 28

5.4 Post Began With Strong Reducibilities . . . . . . . . . . . . . 28

6 Post Highlights Turing Computability 29

6.1 Post Articulates Turing Reducibility . . . . . . . . . . . . . . 29

6.2 The Post-Turing Thesis . . . . . . . . . . . . . . . . . . . . . 30

7 Continuous and Total Functionals 31

7.1 Representations of Open and Closed Sets . . . . . . . . . . . 31

7.2 Notation for Trees . . . . . . . . . . . . . . . . . . . . . . . . 31

7.3 Dense Open Subsets of Cantor Space . . . . . . . . . . . . . . 33

7.4 Eﬀectively Open and Closed Sets . . . . . . . . . . . . . . . . 33

7.5 Continuous Functions on Cantor Space . . . . . . . . . . . . . 34

7.6 Eﬀectively Continuous Functionals . . . . . . . . . . . . . . . 35

7.7 Continuous Functions are Relatively Computable . . . . . . . 36

8 Bounded Reducibilities 36

8.1 A Matrix M

x

for Bounded Reducibilities . . . . . . . . . . . . 36

8.2 Bounded Turing Reducibility . . . . . . . . . . . . . . . . . . 37

2

8.3 Truth-Table Reductions . . . . . . . . . . . . . . . . . . . . . 37

8.4 Diﬀerence of c.e. sets, n-c.e., and ω-c.e. sets . . . . . . . . . . 39

9 Online Computing 40

9.1 Turing Machines and Online Processes . . . . . . . . . . . . . 42

9.2 Trial and Error Computing . . . . . . . . . . . . . . . . . . . 42

9.3 The Limit Lemma . . . . . . . . . . . . . . . . . . . . . . . . 43

9.4 Two Models for Computing With Error . . . . . . . . . . . . 44

9.4.1 The Limit Computable Model . . . . . . . . . . . . . . 44

9.4.2 The Online Model . . . . . . . . . . . . . . . . . . . . 44

10 Three Displacements in Computability Theory 45

11 “Computable” versus “Recursive” 45

11.1 Church Defends Church’s Thesis with “Recursive” . . . . . . 46

11.2 Church and Kleene Deﬁne “Recursive” as “Computable” . . . 46

11.3 G¨odel Rejects “Recursive Function Theory” . . . . . . . . . . 47

11.4 The Ambiguity in the Term “Recursive” . . . . . . . . . . . . 48

11.5 Changing “Recursive” Back to “Inductive” . . . . . . . . . . 48

12 Renaming it the “Computability Thesis?” 50

12.1 Kleene Called it “Thesis I” in [1943] . . . . . . . . . . . . . . 50

12.2 Kleene Named it “Church’s thesis” in [1952] . . . . . . . . . . 50

12.3 Kleene Dropped “Thesis I” for “Church’s thesis” . . . . . . . 51

12.4 Evidence for the Computability Thesis . . . . . . . . . . . . . 51

12.5 Who First Demonstrated the Computability Thesis? . . . . . 52

12.6 The Computability Thesis and the Calculus . . . . . . . . . . 54

12.7 Founders of Computability and the Calculus . . . . . . . . . . 55

13 Turing a-machines versus o-machines? 57

13.1 Turing, Post, and Kleene on Relative Computability . . . . . 57

13.2 Relative Computability Uniﬁes Incomputability . . . . . . . . 57

13.3 The Key Concept of the Subject . . . . . . . . . . . . . . . . 57

13.4 When to Introduce Relative Computability . . . . . . . . . . 58

14 Conclusions 59

Abstract

We begin with the history of the discovery of computability in the

1930’s, the roles of G¨odel, Church, and Turing, and the formalisms of

recursive functions and Turing automatic machines (a-machines). To

3

whom did G¨odel credit the deﬁnition of a computable function? We

present Turing’s notion [1939, '4] of an oracle machine (o-machine)

and Post’s development of it in [1944, '11], [1948], and ﬁnally Kleene-

Post [1954] into its present form.

A number of topics arose from Turing functionals including con-

tinuous functionals on Cantor space and online computations. Almost

all the results in theoretical computability use relative reducibility and

o-machines rather than a-machines and most computing processes in

the real world are potentially online or interactive. Therefore, we argue

that Turing o-machines, relative computability, and online computing

are the most important concepts in the subject, more so than Turing

a-machines and standard computable functions since they are special

cases of the former and are presented ﬁrst only for pedagogical clar-

ity to beginning students. At the end in '10 – '13 we consider three

displacements in computability theory, and the historical reasons they

occurred. Several brief conclusions are drawn in '14.

1 Introduction

In this paper we consider the development of Turing oracle machines and

relative computability and its relation to continuity in analysis and to online

computing in the real world. We also challenge a number of traditional views

of these subjects as often presented in the literature since the 1930’s.

1.1 Terminology: Incompleteness and Incomputability

The two principal accomplishments in computability and logic in the 1930’s

were the discovery of incompleteness by G¨odel [1931] and of incomputabil-

ity independently by Church [1936] and Turing in [1936]. We use the term

“noncomputable” or “incomputable” for individual instances, but we of-

ten use the term “incomputability” for the general concept because it is

linguistically and mathematically parallel to “incomplete.” The term “in-

computable” appeared in English as early as 1606 with the meaning, that

which “cannot be computed or reckoned; incalculable,” according to the

Oxford English Dictionary. Websters dictionary deﬁnes it as “greater than

can be computed or enumerated; very great.” Neither dictionary lists an

entry for “noncomputable” although it has often been used in the subject to

mean “not computable” for a speciﬁc function or set analogously as “non-

measurable” is used in analysis.

4

1.2 The Goal of Incomputability not Computability

For several thousand years the study of algorithms had led to new the-

oretical algorithms and sometimes new devices for computability and cal-

culation. In the 1930’s for the ﬁrst time the goal was the refutation of

Hilbert’ two programs, a ﬁnite consistency proof for Peano arithmetic, and

the Entscheidungsproblem (decision problem). For the latter, two main dis-

coverers of computability, Alonzo Church and Alan Turing, wanted to give

formal deﬁnitions of a computable function so that they could diagonalize

over all computable functions and produce an incomputable (unsolvable)

problem. The speciﬁc models of computable functions produced by 1936,

Turing a-machines, λ-deﬁnable functions, and recursive functions, would

all have deep applications to the design and programming of computing

machines, but not until after 1940. Meanwhile, the researchers spent the re-

mainder of the 1930’s investigating more of the new world of incomputability

they had created by diagonal arguments, just as Georg Cantor had spent the

last quarter of the nineteenth century exploring the world of uncountable

sets which he had created by the diagonal method. In '1 and '2 we consider

this historical development from 1931 to 1939 and we introduce quotes from

G¨odel to show convincingly that he believed “the correct deﬁnition of me-

chanical computability was established beyond any doubt by Turing” and

only by Turing.

1.3 Computing Relative to an Oracle or Database

In 1936 Turing’s a-machines and Church’s use of G¨odel’s recursive functions

solved an immediate problem by producing a deﬁnition of a computable

function, with which one could diagonalize and produce undecidable prob-

lems in mathematics. The Turing a-machine is a good model for oﬄine

computing such as a calculator or batch processor where the process is ini-

tially given a procedure and an input and continues with no further outside

information.

However, many modern computer processes are online processes in that

they consult an external data base of other resource during the computation

process. For example, a laptop computer might consult the World Wide Web

via an ethernet or wireless connection while it is computing. These processes

are sometimes called online or interactive or database processes depending

on the way they are set up.

Turing spent 1936–1938 at Princeton writing a Ph.D. thesis under Church

on ordinal logics. A tiny and obscure part of his paper [1939, '4] included

5

a description of an oracle machine (o-machine) roughly a Turing a-machine

which could interrogate an “oracle” (external database) during the computa-

tion. The one page description was very sketchy and Turing never developed

it further.

Emil Post [1944, '11] considerably expanded and developed relative com-

putability and Turing functionals. These concepts were not well understood

when Post began, but in Post [1943], [1944], [1948] and Kleene-Post [1954]

they emerged into their modern state. These are summarized in the Post-

Turing Thesis of '6.2. This remarkable role by Post has been underempha-

sized in the literature but is discussed here in '5 and '6.

The theory of relative computability developed by Turing and Post and

the o-machines provide a precise mathematical framework for database or

online computing just Turing a-machines provide one for oﬄine computing

processes such as batch processing. Oracle computing processes are those

most often used in theoretical research and also in real world computing

where a laptop computer may communicate with a database like the World

Wide Web. Often the local computer is called the “client” and the remote

device the “server.”

In '4–'6 we study Turing’s oracle machines (o-machines) and Post’s de-

velopment of them into relative computability. It is surprising that so much

attention has been paid to the Church-Turing Thesis 3.2 over the last sev-

enty years and so little to the Post-Turing Thesis 6.1 on relative reducibil-

ity, particularly in view of the importance of relative computability (Turing

o-machines) in comparison with plain computability (Turing a-machines) in

both theoretical and practical areas.

1.4 Continuous Functions and Calculus

In '7 we show that any Turing functional on Cantor space 2

ω

is an eﬀec-

tively continuous partial map. Conversely, for any continuous functional

there is a Turing functional relative to some real parameter which deﬁnes it,

thereby linking computability with analysis. This makes Turing functionals

the analogue in computability of the continuous functions in analysis as we

explain. In contrast, Turing a-machines viewed as functionals on Cantor

space produce only a constant functional. Surprisingly, there appears to

be more emphasis in calculus books on continuous functionals than there is

in introductory computability books on computably continuous functionals

and relative computability.

6

2 Origins of Computability and Incomputability

Mathematicians have studied algorithms and computation since ancient

times, but the modern study of computability and incomputability be-

gan around 1900. David Hilbert was deeply interested in the foundations

of mathematics. Hilbert [1899] gave an axiomatization of geometry and

showed [1900] that the question of the consistency of geometry reduced

to that for the real-number system, and that in turn, to arithmetic by

results of Dedekind (at least in a second order system). Hilbert [1904]

proposed proving the consistency of arithmetic by what emerged [1928]

as his ﬁnitist program. He proposed using the ﬁniteness of mathematical

proofs in order to establish that contradictions could not be derived. This

tended to reduce proofs to manipulation of ﬁnite strings of symbols de-

void of intuitive meaning which stimulated the development of mechanical

processes to accomplish this. Hilbert’s second major program concerned

the Entscheidungsproblem (decision problem). Kleene [1987b, p. 46] wrote,

“The Entscheidungsproblem for various formal systems had been posed by

Schr¨oder [1895], L¨owenheim [1915], and Hilbert [1918].” The decision prob-

lem for ﬁrst order logic emerged in the early 1920’s in lectures by Hilbert and

was stated in Hilbert and Ackermann [1928]. It was to give a decision proce-

dure (Entscheidungsverfahren) “that allows one to decide the validity of the

sentence.” Hilbert characterized this as the fundamental problem of mathe-

matical logic. Davis [1965, p. 108] wrote, “This was because it seemed clear

to Hilbert that with the solution of this problem, the Entscheidungsproblem,

it should be possible, at least in principle, to settle all mathematical ques-

tions in a purely mechanical manner.” Von Neumann (1927) doubted that

such a procedure existed but had no idea how to prove it.

2.1 G¨ odel’s Incompleteness Theorem

Hilbert retired in 1930 and was asked to give a special address in the fall of

1930 in K¨onigsberg, the city of his birth. Hilbert spoke on natural science

and logic, the importance of mathematics in science, and the importance

of logic in mathematics. He asserted that there are no unsolvable problems

and stressed,

Wir m¨ ussen wissen. (We must know.)

Wir werden wissen. (We will know.)

At a mathematical conference preceding Hilbert’s address, a quiet, obscure

young man, Kurt G¨odel, only a year a beyond his Ph.D., announced a result

7

which would forever change the foundations of mathematics. He formalized

the liar paradox, “This statement is false,” to prove roughly that for any

eﬀectively axiomatized, consistent extension T of number theory (Peano

arithmetic) there is a sentence σ which asserts its own unprovability in T.

John von Neumann in the audience immediately understood the importance

of G¨odel’s Incompleteness Theorem. He was at the conference representing

Hilbert’s proof theory program, and recognized that Hilbert’s program was

over. In the next few weeks von Neuman realized that by arithmetizing

the proof of G¨odel’s ﬁrst theorem, one could prove an even better one, that

no such formal system T could prove its own consistency. A few weeks

later he brought his proof to G¨odel who thanked him and informed him

politely that he had already submitted the Second Incompleteness Theorem

for publication. G¨odel’s Incompleteness Theorem [1931] not only refuted

Hilbert’s program on proving consistency, but it also had a profound ef-

fect on refuting Hilbert’s second theme of the Entscheidungsproblem. G¨odel

had successfully challenged the Hilbert’s ﬁrst proposal. This made it eas-

ier to challenge Hilbert on the second topic of the decision problem. Both

refutations used diagonal arguments. Of course, diagonal arguments had

been known since Cantor’s work, but G¨odel showed how to arithmetize the

syntactic elements of a formal system and diagonalize within that system.

Crucial elements in computability theory, such as the Turing universal ma-

chine, the Kleene µ-recursive functions, or the self reference in the Kleene’s

Recursion Theorem, all depend upon giving code numbers to computations

and elements within a computation, and in calling algorithms by their code

numbers (G¨odel numbers). These ideas spring from G¨odel’s [1931] incom-

pleteness proof.

2.2 Alonzo Church

By 1931 computability was a young man’s game. Hilbert had retired and no

longer had much inﬂuence on the ﬁeld. As the importance of G¨odel’s Incom-

pleteness Theorem began to sink in, and researchers began concentrating on

the Entscheidungsproblem, the major ﬁgures were all young. Alonzo Church

(born 1903), Kurt G¨odel (b. 1906), and Stephen C. Kleene (b. 1909) were

all under thirty. Turing (b. 1912), perhaps the most inﬂuential of all on

computability theory, was not even twenty. Only Emil Post (b. 1897) was

over thirty, and he was not yet thirty-ﬁve. These young men were about

to leave Hilbert’s ideas behind and open the path of computability for the

next two thirds of the twentieth century, which would solve the theoretical

problems and would show the way toward practical computing devices.

8

After completing his Ph.D. at Princeton in 1927, Alonzo Church spent

one year at Harvard and one year at G¨ottingen and Amsterdam. He returned

to Princeton as an Assistant Professor of Mathematics in 1929. In 1931

Church’s ﬁrst student, Stephen C. Kleene, arrived at Princeton. Church

had begun to develop a formal system now called the λ-calculus. In 1931

Church knew only that the successor function was λ-deﬁnable. Kleene began

proving that certain well-known functions were λ-deﬁnable. By the time

Kleene received his Ph.D. in 1934 he had shown that all the usual number

theoretic functions were λ-deﬁnable. On the basis of this evidence and

his own intuition, Church proposed to G¨odel around March, 1934 the ﬁrst

version of his thesis on functions which are eﬀectively calculable, the term

in the 1930’s for a function which is computable in the informal sense. (See

Davis [1965, p. 8–9].)

Deﬁnition 2.1. Church’s Thesis (First Version) [1934]. A function

if eﬀectively calculable if and only if it is λ-deﬁnable.

When Kleene ﬁrst heard the thesis, he tried to refute it by a diagonal

argument but since the λ-deﬁnable functions were only partial functions, the

diagonal was one of the rows. Instead of a contradiction, Kleene had proved

a beautiful new theorem, the Kleene Recursion Theorem whose proof is a

diagonal argument which fails (see Soare [1987, p. 36]). Although Kleene was

convinced by Church’s ﬁrst thesis, G¨odel was not. G¨odel rejected Church’s

ﬁrst thesis as “thoroughly unsatisfactory.”

2.3 Herbrand-G¨ odel Recursive Functions

From 1933 to 1939 G¨odel spent time both in Vienna pursuing his academic

career and at Fine Hall in Princeton which housed both the Princeton Uni-

versity faculty in mathematics and the Institute for Advanced Study, of

which he was a member. G¨odel spent the ﬁrst part of 1934 at Princeton.

The primitive recursive functions which he had used in his 1931 paper did

not constitute all computable functions. He expanded on a concept of Her-

brand and brought closer to the modern form. At Princeton in the spring of

1934 G¨odel lectured on the Herbrand-G¨odel recursive functions which came

to be known as the general recursive functions to distinguish them from

the primitive recursive functions which at that time were called “recursive

functions.” Soon the preﬁx “primitive” was added to the latter and the

preﬁx “general” was generally dropped from the former. G¨odel’s deﬁnition

gave a remarkably succinct system whose simple rules reﬂected the way a

mathematician would informally calculate a function using recursion.

9

Church and Kleene attended G¨odel’s lectures on recursive functions.

Rosser and Kleene took notes which appeared as G¨odel [1934]. After seeing

G¨odel’s lectures, Church and Kleene changed their formalism (especially for

Church’s Thesis) from “λ-deﬁnable” to “Herbrand-G¨odel general recursive.”

Kleene [1981] wrote,

“I myself, perhaps unduly inﬂuenced by rather chilly receptions

from audiences around 1933–35 to disquisitions on λ-deﬁnability,

chose, after general recursiveness had appeared, to put my work

in that format. . . . ”

Nevertheless, λ-deﬁnability is a precise calculating system and has close

connections to modern computing, such as functional programming.

2.4 Stalemate at Princeton Over Church’s Thesis

Church reformulated his thesis, with Herbrand-G¨odel recursive functions

in place of λ-deﬁnable ones. This time without consulting G¨odel, Church

presented to the American Mathematical Society on April 19, 1935, his

famous proposition described in his paper [1936].

“In this paper a deﬁnition of recursive function of positive in-

tegers which is essentially G¨odel’s is adopted. It is maintained

that the notion of an eﬀectively calculable function of positive

integers should be identiﬁed with that of a recursive function,

. . . ”

It has been known since Kleene [1952] as Church’s Thesis in the following

form.

Deﬁnition 2.2. Church’s Thesis [1936]. A function on the positive

integers is eﬀectively calculable if and only if it is recursive.

As further evidence, Church and Kleene had shown the formal equiva-

lence of the Herbrand-G¨odel recursive functions and the λ-deﬁnable func-

tions. Kleene introduced a new equivalent deﬁnition, the µ-recursive func-

tions, functions deﬁned by the ﬁve schemata for primitive recursive func-

tions, plus the least number operator µ. The µ-recursive functions had the

advantage of a short standard mathematical deﬁnition, but the disadvan-

tage that any function not primitive recursive could be calculated only by a

tedious arithmetization as in G¨odel’s Incompleteness Theorem.

10

2.5 G¨ odel’s Thoughts on Church’s Thesis

In spite of this evidence, G¨odel still did not accept Church’s Thesis by the

beginning of 1936. G¨odel had become the most prominent ﬁgure in math-

ematical logic. It was his approval that Church wanted most. Church had

solved the Entscheidungsproblem only if his characterization of eﬀectively

calculable functions was accurate. G¨odel had considered the question of

characterizing the calculable functions in [1934] when he wrote,

“[Primitive] recursive functions have the important property that,

for each given set of values for the arguments, the value of the

function can be computed by a ﬁnite procedure

3

.”

Footnote 3.

“The converse seems to be true, if, besides recursion according to

scheme (V) [primitive recursion], recursions of other forms (e.g.,

with respect to two variables simultaneously) are admitted. This

cannot be proved, since the notion of ﬁnite computation is not

deﬁned, but it serves as a heuristic principle.”

The second paragraph, G¨odel’s footnote 3, gives crucial insight into his

thinking about the computability thesis and his later pronouncements about

the achievements of Turing versus others. G¨odel says later that he was not

sure that his system of Herbrand-G¨odel recursive functions comprised all

possible recursions. Second, his ﬁnal sentence suggests that he may have

believed such a characterization “cannot be proved,” but is a “heuristic

principle.”

This suggests that G¨odel was waiting not only for a formal deﬁnition

(such as recursive functions or Turing machines which came later) but evi-

dence that these captured the informal notion of eﬀectively calculable (which

Turing later gave, but which G¨odel did not ﬁnd in Church’s work). Here

he even suggests that such a precise mathematical characterization of the

informal notion cannot be proved which makes his acceptance of Turing’s

later work even more impressive.

3 Turing Breaks the Stalemate

3.1 Turing Machines and Turing’s Thesis

At the start of 1936 those gathered at Princeton: G¨odel, Church, Kleene,

Rosser, and Post nearby at City College in New York, constituted the most

11

distinguished and powerful group in the world investigating the notion of

a computable function and Hilbert’s Entscheidungsproblem, but they could

not agree on whether recursive functions constituted all eﬀectively calculable

functions. At that moment stepped forward a twenty-two year old youth,

far removed from Princeton. Well, this was not just any youth. Alan Turing

had already proved the Central Limit Theorem in probability theory, not

knowing it had been previously proved, and as a result Turing had been

elected a Fellow of King’s College, Cambridge.

The work of Hilbert and G¨odel had become well-known around the world.

At Cambridge University topologist M.H.A. (Max) Newman gave lectures

on Hilbert’s Entscheidungsproblem in 1935. Alan Turing attended. Turing’s

mother had had a typewriter which fascinated him as a boy. He designed

his automatic machine (a-machine) as a kind of idealized typewriter with

an inﬁnite carriage over which the reading head passes with the ability to

read, write, and erase one square at a time before moving to an immediately

adjacent square, just like a typewriter.

Deﬁnition 3.1. Turing’s Thesis [1936]. A function is intuitively com-

putable (eﬀectively calculable) if and only if it is computable by a Tur-

ing machine, i.e., an automatic machine (a-machine), as deﬁned in Turing

[1936].

Turing showed his solution to the astonished Max Newman in April,

1936. The Proceedings of the London Mathematical Society was reluctant

to publish Turing’s paper because Church’s had already been submitted to

another journal on similar material. Newman persuaded them that Tur-

ing’s work was suﬃciently diﬀerent, and they published Turing’s paper in

volume 42 on November 30, 1936 and December 23, 1936. There has been

considerable misunderstanding in the literature about exactly when Turing’s

seminal paper was published. This is important because of the appearance

in 1936 of related papers by Church, Kleene, and Post, and Turing’s priority

is important here.

1

1

Many papers, Kleene [1943, p. 73], [1987], [1987b], Davis [1965, p. 72], Post [1943,

p. 20], G¨ odel Collected Works, Vol. I, p. 456, and others, mistakenly refer to this paper

as “Turing [1937],” perhaps because the volume 42 is 1936-37 covering 1936 and part

of 1937, or perhaps because of the two page minor correction [1937a]. Others, such as

Kleene [1952], [1981], [1981b], Kleene and Post [1954, p. 407], Gandy [1980], Cutland

[1980], and others, correctly refer to it as “[1936],” or sometimes “[1936-37].” The journal

states that Turing’s manuscript was “Received 28 May, 1936–Read 12 November, 1936.”

It appeared in two sections, the ﬁrst section of pages 230–240 in Volume 42, Part 3, issued

on November 30, 1936, and the second section of pages 241–265 in Volume 42, Part 4,

12

Turing’s a-machine has compelling simplicity and logic which makes it

even today the most convincing model of computability. Equally important

with the Turing machine was Turing’s analysis of the intuitive conception of

a “function produced by a mechanical procedure.” In a masterful demon-

stration, which Robin Gandy considered as precise as most mathematical

proofs, Turing analyzed the informal nature of functions computable by a

ﬁnite procedure and demonstrated that they coincide with those computable

by an a-machine. Also Turing [1936, p. 243] introduced the universal Turing

machine which has great theoretical and practical importance.

3.2 G¨ odel’s Opinion of Turing’s Work

G¨odel never accepted Church’s Thesis in the form above, even though it

was formulated with his own general recursive functions, but G¨odel and

most others accepted Turing’s Thesis. G¨odel knew of the extensional ev-

idence. Church and Kleene [1936b] had shown the formal equivalence of

λ-deﬁnable functions, general recursive functions, and Kleene had proved

the equivalence with µ-recursive functions based on G¨odel’s own arithme-

tization of 1931. G¨odel was also well aware of Turing’s proof [1937b] of

the equivalence of λ-deﬁnable functions with Turing computable ones, and

hence the conﬂuence of all the known deﬁnitions.

However, G¨odel was interested in the intensional analysis of ﬁnite pro-

cedure as given by Turing [1936]. He had not accepted the argument of

conﬂuence as suﬃcient to justify Church’s Thesis. G¨odel clearly expresses

his opinion in his three volume collected works, G¨odel [1986], G¨odel [1990],

and G¨odel [1995]. Let us examine there what G¨odel has to say. In the

following article G¨odel considers all these as equivalent formal deﬁnitions.

The question was whether they captured the informal concept of a function

speciﬁed by a ﬁnite procedure. The best source for articles of G¨odel is the

three volume Collected Works, which we have listed by year of publication:

Volume I, G¨odel [1986]; Volume II, G¨odel [1990], and Volume III, G¨odel

[1995].

issued December 23, 1936. No part of Turing’s paper appeared in 1937, but the two page

minor correction [1937a] did. Determining the correct date of publication of Turing’s work

is important to place it chronologically in comparison with Church [1936], Post [1936], and

Kleene [1936].

13

3.2.1 G¨odel [193?] Notes in Nachlass [1935]

This article has an introductory note by Martin Davis in G¨odel [1995, p. 156].

Davis wrote, “This article is taken from handwritten notes in English, ev-

idently for a lecture, found in the Nachlass in a spiral notebook.” In the

Nachlass printed in G¨odel [1995, p. 166] G¨odel wrote,

“When I ﬁrst published my paper about undecidable proposi-

tions the result could not be pronounced in this generality, be-

cause for the notions of mechanical procedure and of formal sys-

tem no mathematically satisfactory deﬁnition had been given at

that time. . . .

The essential point is to deﬁne what a procedure is.”

To formally capture this crucial informal concept, G¨odel, who was giving

an introductory lecture, began with a deﬁnition of the primitive recursive

functions (which he quickly proved inadequate by a diagonal argument)

and then his own Herbrand-G¨odel recursive functions on p. 167. (G¨odel

gave the Herbrand-G¨odel recursive function deﬁnition rather than Turing

machines because he knew they were equivalent. He intended his talk to be

as elementary as possible for a general audience, which he knew would be

more familiar with recursion.) G¨odel continued on p. 168,

“That this really is the correct deﬁnition of mechanical com-

putability was established beyond any doubt by Turing.”

The word “this” evidently refers to the recursive functions. G¨odel knew

that Turing had never proved anything about recursive functions. What

did he mean? G¨odel knew that by work of Turing, Church, and Kleene,

the formal classes of Turing computable functions, recursive functions, and

λ-deﬁnable functions all coincided. G¨odel was asserting that it was Turing

who had demonstrated that these formal classes captured the informal no-

tion of a procedure. It was Turing’s proof [1936] and the formal equivalences

which had elevated Herbrand-G¨odel recursive functions to a correct charac-

teristic of eﬀectively calculable functions, not that the Herbrand-G¨odel re-

cursive functions had elevated Turing computable functions. Indeed G¨odel

had seen Church’s Thesis 2.2 expressed in terms of Herbrand-G¨odel recur-

sive functions, and had rejected it in 1935 and 1936 because he was not sure

his own deﬁnition had captured the informal concept of procedure.

G¨odel had begun with the recursive function as more easily explained to

a general audience, but having introduced Turing, G¨odel now went forward

with Turing’s work.

14

“But Turing has shown more. He has shown that the computable

functions deﬁned in this way are exactly those for which you can

construct a machine with ﬁnite number of parts which will do

the following thing. If you write down any number n

1

, . . . n

r

,

on a slip of paper and put the slip into the machine and turn

the crank then after ﬁnite number of turns the machine will stop

and the value of the function for the argument n

1

, . . . n

r

will be

printed on the paper.”

3.2.2 Princeton Bicentennial [1946]

To fully understand this article one should be familiar with G¨odel’s

¨

Uber die

L¨ange von Beweisen” (“On the length of proofs) [1936a] in Volume I [1986,

p. 396–398]. G¨odel discussed what it means for a function to be computable

in a formal system S and remarked that given a sequence of formal systems

S

i

, S

i+1

. . . it is possible that passing from one formal system S

i

to one of

higher order S

i+1

not only allows us to prove certain propositions which were

not provable before, but also makes it possible to shorten by an extraordinary

amount proofs already available.

Now for G¨odel’s Princeton Bicentennial address [1946] refer to the Col-

lected Works Volume II, G¨ odel [1990, p. 150]. G¨odel muses on the remark-

able fact of the absoluteness of computability, that it is not necessary to

distinguish orders (diﬀerent formal systems). Once we have a suﬃciently

strong system (such as Peano arithmetic) we can prove anything about com-

putable functions which could be proved in a more powerful system. Once

again, G¨odel identiﬁes those formal systems known to be equivalent, general

recursiveness, Turing computability, and others, as a single formal system.

“Tarski has stressed in his lecture (and I think justly) the great

importance of the concept of general recursiveness (or Turing’s

computability). It seems to me that this importance is largely

due to the fact that with this concept one has for the ﬁrst time

succeeded in giving an absolute deﬁnition of an interesting episte-

mogical notion, i.e., one not depending on the formalism chosen.

2

In all other cases treated previously, such as demonstrability or

deﬁnability, one has been able to deﬁne them only relative to a

given language, and for each individual language it is clear that

the one thus obtained is not the one looked for. For the concept

2

. . . “A function of integers is computable in any formal system containing arithmetic

if and only if it is computable in arithmetic” . . . .

15

of computability, however, although it is merely a special kind of

demonstrability or decidability, the situation is diﬀerent. By a

kind of miracle it is not necessary to distinguish orders, and the

diagonal procedure does not lead outside the deﬁned notion.”

G¨odel stated,

“. . . one has for the ﬁrst time succeeded in giving an absolute

deﬁnition of an interesting epistemogical notion . . . ”

Who is the “one” who has linked the informal notion of procedure or

eﬀectively calculable function to one of the formal deﬁnitions. This becomes

irrefutably clear in '3.2.5.

3.2.3 The Flaw in Church’s Thesis

G¨odel himself was the ﬁrst to provide one of the formalisms later recognized

as a deﬁnition of computability, the general recursive functions. However,

G¨odel himself never claimed to have made this link. Church claimed it in

his announcement of Church’s Thesis in 1935 and 1936, but G¨odel did not

accept it then and gave no evidence later of believing that Church had done

this. Modern scholars found weaknesses in Church’s attempted proof [1936]

that the recursive functions constituted all eﬀectively calculable functions.

If the basic steps are stepwise recursive, then it follows easily by the

Kleene Normal Form Theorem which Kleene had proved and communicated

to G¨odel before November, 1935 (see Kleene [1987b], p. 57]), that the entire

process is recursive. The fatal weakness in Church’s argument was the core

assumption that the atomic steps were stepwise recursive, something he did

not justify. Gandy [1988, p. 79] and especially Sieg [1994, pp. 80, 87] in

their excellent analyses brought out this weakness in Church’s argument.

Sieg [p. 80] wrote, “. . . this core does not provide a convincing analysis:

steps taken in a calculus must be of a restricted character and they are

assumed, for example by Church, without argument to be recursive.” Sieg

[p. 78] wrote, “It is precisely here that we encounter the major stumbling

block for Church’s analysis, and that stumbling block was quite clearly seen

by Church,” who wrote that without this assumption it is diﬃcult to see

how the notion of a system of logic can be given any exact meaning at all.

It is exactly this stumbling block which Turing overcame by a totally new

approach.

16

3.2.4 G¨odel on Church’s Thesis

G¨odel may not have found errors in Church’s demonstration, but he never

gave any hint that he thought Church had been the ﬁrst to show that the

recursive functions coincided with the eﬀectively calculable ones. On the

contrary, G¨odel said,

“As for previous equivalent deﬁnitions of computability, which,

however, are much less suitable for our purpose, [i.e., verifying

the Computability Thesis], see A. Church 1936, pp. 256–358.”

– G¨odel, Princeton Bicentennial, [1946, p. 84], and G¨odel

Collected Works, Vol. I, pp 150–153.

3.2.5 G¨odel’s Letter to Kreisel [1968]

-G¨odel: letter to Kreisel of May 1, 1968 [Sieg, 1994, p. 88].

“ But I was completely convinced only by Turing’s paper.”

3.2.6 Gibbs Lecture [1951]

G¨odel, Collected Works Volume III, [G¨odel, 1995, p.304–305].

Martin Davis in his introduction wrote,

“On 26 December 1951, at a meeting of the American Mathe-

matical Society at Brown University, G¨odel delivered the twenty-

ﬁfth Josiah Willard Gibbs Lecture, “Some basic theorems on the

foundations of mathematics and their implications.” . . .

It is probable as Wang suggests ([1987, pages 117–18]), that the lecture

was the main project G¨odel worked on in the fall of 1951. . . . “

G¨odel [1951] in his Gibbs lecture wrote,

“Research in the foundations of mathematics during the past few

decades has produced some results of interest, not only in them-

selves, but also with regard to their implications for the tradi-

tional philosophical problems about the nature of mathematics.

The results themselves, I believe, are fairly widely known, but

nevertheless I think it will be useful to present them in outline

once again, especially in view of the fact that due to the work

17

of various mathematicians, they have taken on a much more

satisfactory form than they had had originally. The greatest im-

provement was made possible through the precise deﬁnition of

the concept of ﬁnite procedure, [“ . . . equivalent to the concept of

a ’computable function of integers’ . . . ”] which plays a decisive

role in these results. There are several diﬀerent ways of arriving

at such a deﬁnition, which, however, all lead to exactly the same

concept. The most satisfactory way, in my opinion, is that of

reducing the concept of a ﬁnite procedure to that of a machine

with a ﬁnite number of parts, as has been done by the British

mathematician Turing.”

In this one paragraph,

1. G¨odel stressed the importance of the results to mathematics and phi-

losophy.

2. G¨odel gave full credit to Turing and his “machine with a ﬁnite number

of parts” for capturing the concept of ﬁnite procedure.

3. G¨odel never mentions Church or G¨odel’s own deﬁnition of recursive

functions.

This is one of the most convincing and explicit demonstrations of G¨odel’s

opinion of Turing’s work.

3.2.7 Godel’s Postscriptum 3 June, 1964 to G¨odel [1934]

-Godel’s Postscriptum 3 June, 1964 to G¨odel [1934], see “The Un-

decidable,” M. Davis, [1965, p. 71] and G¨odel Collected Works, Volume I

[1986, p. 369–370].

“ In consequence of later advances, in particular of the fact that,

due to A. M. Turing’s work, a precise and unquestionably ade-

quate deﬁnition of the general concept of formal system can now

be given, the existence of undecidable arithmetical propositions

and the non-demonstrability of the consistency of a system in the

same system can now be proved rigorously for every consistent

formal system containing a certain amount of ﬁnitary number

theory.”

18

Turing’s work gives an analysis of the concept of “mechanical

procedure” (alias “algorithm” or “computation procedure” or

“ﬁnite combinatorial procedure”). This concept is shown to be

equivalent with that of a “Turing machine.”

3.3 Hao Wang Reports on G¨ odel

Hao Wang was a very close friend and professional colleague of G¨odel, whom

he called “G” in the following passage. Wang [1987, p. 96] wrote about

G¨odel’s opinion of Turing’s work.

“Over the years G habitually credited A.M. Turing’s paper of

1936 as the deﬁnitive work in capturing the intuitive concept [of

computability], and did not mention Church or E. Post in this

connection. He must have felt that Turing was the only one who

gave persuasive arguments to show the adequacy of the precise

concept . . . In particular, he had probably been aware of the

arguments oﬀered by Church for his “thesis” and decided that

they were inadequate. It is clear that G and Turing (1912–1954)

had great admiration for each other, . . . “

3.4 Kleene’s Remarks About Turing

“Turing’s computability is intrinsically persuasive” but “λ-de-

ﬁnability is not intrinsically persuasive” and “general recursive-

ness scarcely so (its author G¨odel being at the time not at all

persuaded).”

-Stephen Cole Kleene [1981b, p. 49]

“Turing’s machine concept arises from a direct eﬀort to ana-

lyze computation procedures as we know them intuitively into

elementary operations. Turing argued that repetitions of his el-

ementary operations would suﬃce for any possible computation.

“ For this reason, Turing computability suggests the thesis more

immediately than the other equivalent notions and so we choose

it for our exposition.”

-Stephen Cole Kleene, second book [1967, p. 233]

19

3.5 Church’s Remarks About Turing

Computability by a Turing machine, “ has the advantage of mak-

ing the identiﬁcation with eﬀectiveness in the ordinary (not ex-

plicitly deﬁned) sense evident immediately—i.e., without the ne-

cessity of proving preliminary theorems.”

-Alonzo Church [1937], Review of Turing [1936]

In modern times it is sometimes stated as follows, recognizing that

Church [1935], [1936] got it ﬁrst, but that Turing [1936] got it right, in

the opinion of G¨odel and many modern scholars.

Deﬁnition 3.2. Church-Turing Thesis. A function is intuitively com-

putable if and only if it is computable by a Turing machine, or equivalently

if it is speciﬁed by a recursive function.

We strongly believe that it should not be called any of the three (Church’s

Thesis, Turing’s Thesis, or the Church-Turing Thesis) but rather should be

called the “Computability Thesis” as we argue in '13 and '14, just as the

calculus is named for neither of its discoverers, Newton and Leibniz.

4 Oracle Machines and Relative Computability

After introducing deﬁnitions of computable functions: Turing a-machines,

recursive functions, and λ-deﬁnable functions; the originators continued dur-

ing 1936–1939 to explore incomputable phenomena, rather than computable

applications of these devices, which came only a decade or more later.

Church and Kleene [1936] as well as Church [1938] and Kleene [1938] studied

computable well-orderings and deﬁned recursive ordinals which were later

used to extend the jump operation to the arithmetic hierarchy and beyond

to the hyperarithmetic hierarchy up to the ﬁrst nonrecursive ordinal ω

CK

1

.

Turing spent 1936–1938 at Princeton working on a Ph.D. with Church.

His thesis was completed in 1938 and published in Turing [1939]. Church

and other mathematicians had found G¨odel’s Incompleteness Theorem un-

settling. By G¨odel’s proof an eﬀective extension T

1

of Peano arithmetic

cannot prove its own consistency con

T

1

. However, we can add the arithmeti-

cal statement con

T

1

to T

1

to get a strictly stronger theory T

2

. Continuing,

we can get an increasing hierarchy of theories ¦T

α

¦

α∈S

over a set S of or-

dinals. Turing’s Ph.D. thesis [1939] concerned such an increasing array of

undecidable theories.

20

4.1 Turing’s Oracle Machines

In one of the most important and most obscure parts of all of computability

theory, Turing wrote in his ordinal logics paper [1939, '4] a short statement

about oracle machines.

“Let us suppose we are supplied with some unspeciﬁed means of

solving number-theoretic problems; a kind of oracle as it were.

. . . this oracle . . . cannot be a machine.

With the help of the oracle we could form a new kind of ma-

chine (call them o-machines), having as one of its fundamental

processes that of solving a given number-theoretic problem.”

This is virtually all Turing said of oracle machines. His description was

only a page long and half of that was devoted to the unsolvability of related

problems such as whether an o-machine will output an inﬁnite number of

0’s or not.

In 1939 Turing left this topic never to return. It mostly lay dormant

for ﬁve years until it was developed in a beautiful form by Post [1944],

[1948], and other papers as we shall explain in '5 and '6. Before doing so,

we complete this section '4.1 with a modern treatment of oracle machines

and Turing functionals including some of the more important properties even

though these were mostly discovered much later, even after Post. More mod-

ern properties of Turing functionals, such as eﬀective continuity on Cantor

space, will be developed in '7.6.

4.2 Modern Deﬁnitions of Oracle Machines

There are several equivalent ways that a Turing machine with oracle may be

deﬁned. We prefer the deﬁnition in Soare’s book [1987, p. 46] of a machine

with a head which reads the work tape and oracle tape simultaneously, but

many other formulations produce the same class of functionals.

Deﬁnition 4.1. A Turing oracle machine (o-machine) is a Turing machine

with an extra “read only” tape, called the oracle tape, upon which is writ-

ten the characteristic function of some set A (called the oracle), and whose

symbols cannot be printed over. The old tape is called the work tape and

operates just as before. The reading head moves along both tapes simulta-

neously. As before, Q is a ﬁnite set of states, S

1

= ¦B, 0, 1¦ is the oracle

tape alphabet, S

2

= ¦B, 1¦ is the work tape alphabet, and ¦R, L¦ the set of

21

head moving operations right and left. A Turing oracle program

P

e

is now

simply a partial map,

δ : QS

1

S

2

−→ QS

2

¦R, L¦,

where δ(q, a, b) = (p, c, X) indicates that the machine in state q reading

symbol a on the oracle tape and symbol b on the work tape passes to state

p, prints “c” over “b” on the work tape, and moves one space right (left) on

both tapes if X = R (X = L). The other details are just as previously in

Soare [1987]. The Turing oracle program

P

e

takes some oracle A and deﬁnes

a partial A-computable functional Φ

A

e

(x) = y.

Notation 4.2. (i) We let lower case Greek letters ϕ, ψ represent partial

functions from ω to ω and lower case Latin letters f, g, and h represent

total functions.

(ii) We let upper case Greek letters represent partial functionals from 2

ω

to 2

ω

. If A ⊆ ω then Ψ

A

(x) may be deﬁned for some or all x. If B ⊆ ω we

write Ψ

A

= B if Ψ

A

(x) = B(x) for all x ∈ ω.

(iii) As in Soare [1987] we use ¦P

e

¦

e∈ω

for an eﬀective listing of Turing

programs for Turing a-machines and let ϕ

e

be the partial computable func-

tion deﬁned by P

e

. We let ¦

P

e

¦

e∈ω

be an eﬀective listing of oracle Turing

programs, and let Φ

e

be the computable partial functional deﬁned by

P

e

. If

Φ

A

e

is total and computes B we say B is computable in A and write B ≤

T

A.

We refer to ϕ

e

as a partial computable function ω to ω because its input and

output are integers. On the other hand, Φ

A

e

is called a partial computable

functional because it takes a set A to a set B and is viewed as a map on

Cantor Space 2

ω

.

(iv) Since Rogers book [1967], researchers have used ϕ

e

(x) or ¦e¦(x) for the

partial computable function with program P

e

. Since Lachlan about 1970,

researchers have used Φ

A

e

(x) for the Turing functional with oracle program

P

e

and have used ϕ

A

e

(x) for the use function, the maximum element of A

examined during the computation. Lachlan also used matched pairs Ψ, ψ;

Γ, γ and so forth for partial computable functionals and their use functions

in many papers, and this is the general usage today. There is no confusion

between the notation ϕ

e

(x) as a partial computable function and ϕ

A

e

(x) as a

use function for Φ

A

e

(x) because the former ϕ

e

(x) will never have an exponent

A and the latter use function ϕ

A

e

(x) always will.

22

4.2.1 The Graph of a Partial Computable Function

Deﬁnition 4.3. Given a partial computable (p.c) function ϕ

e

deﬁne the

graph of ϕ

e

as follows.

(1) g

e

= graph(ϕ

e

) =

dfn

¦ 'x, y` : ϕ

e

(x) = y ¦

Note that if ϕ

e

is a partial computable (p.c.) function then graph(ϕ

e

)

is a computable enumerable (c.e.) set. Likewise, given any (c.e.) set W

e

we can ﬁnd a single-valued c.e. subset V

e

⊆ W

e

which is the graph of a p.c.

function. The the notions of a Turing program to compute a p.c. function

ψ and a description of its graph are interchangeable and equally powerful in

describing ψ.

4.2.2 The Graph of an Oracle Computable Functional

Deﬁnition 4.4. For an oracle machine program

P

e

we likewise deﬁne the

oracle graph of the corresponding computable functional Φ

e

but now tak-

ing into consideration the ﬁnite strings read by the oracle head during the

computation.

(2) G

e

=

dfn

graph(Φ

e

) =

dfn

¦ 'σ, x, y` : Φ

σ

e

(x) = y ¦

where σ ranges over 2

<ω

,

Here Φ

σ

e

(x) = y denotes that the oracle program

P

e

with oracle σ on its

oracle tape, and x on its input tape, eventually halts and outputs y, and

does not read more of the oracle tape than σ during the computation. The

crucial property of the oracle graph G

e

and the one which makes a Turing

functional Φ

e

independent of any particular machine representation is the

following.

4.3 The Oracle Graph Theorem

Theorem 4.5. Oracle Graph Theorem. If

P

e

is an oracle Turing pro-

gram deﬁning a Turing functional Φ

e

, then the graph G

e

deﬁned in (2) is a

computably enumerable (c.e.) set.

Proof. From the deﬁnitions G

e

is Σ

0

1

and therefore c.e.

The converse holds for a c.e. set V which satisﬁes a singlevaluedness

condition (3) and a continuity condition (4).

23

Theorem 4.6. Let V ⊂ 2

<ω

ω ω be a computably enumerable set which

satisﬁes the following two conditions. Then there is a Turing functional Φ

e

such that G

e

= V .

(3) 'σ, x, y` ∈ V =⇒ (∃τ ⊇ σ)(∃z)[ z = y & 'τ, x, z` ∈ V ].

(4) 'σ, x, y` ∈ V =⇒ (∀τ ⊃ σ)[ 'τ, x, y` ∈ V ].

Proof. Given V satisfying (3) and (4) deﬁne a partial computable functional

Ψ

σ

(x) as follows. If 'σ, x, y` appears in V deﬁne Ψ

σ

(x) = y. This deﬁnes an

eﬀective reduction Ψ. There must be a Turing oracle machine Φ

e

such that

Φ

e

= Ψ.

Theorem 4.7. Furthermore, given any c.e. set U ⊆ 2

<ω

ω ω there is a

c.e. subset V ⊆ U satisfying (3) and (4) and having the same domain, i.e.,

¦x : (∃σ)(∃y)[ 'σ, x, y` ∈ U ]¦ = ¦x : (∃σ)(∃y)[ 'σ, x, y` ∈ V ]¦.

Proof. Similar to single valuedness theorem in Soare [1987, Chapter II].

4.4 Equivalent Deﬁnitions of Relative Computability

There are several diﬀerent formal deﬁnitions of relative computability. This

includes an oracle machine with a single reading head reading the work

tape and oracle tape, or two independent reading heads, or other varia-

tions. In addition, several authors deﬁne relative computability from oracle

A by adding the characteristic function of A either to the Herbrand-G¨odel

general recursive function deﬁnition or to the Kleene µ-recursive function

deﬁnition. Each of these formal deﬁnitions produces a c.e. graph G

e

and

these deﬁnitions are all equivalent.

Furthermore, any Turing a-machine can clearly be simulated by a Turing

o-machine as we note in the following theorem. Therefore, in presenting the

subject we can bypass a-machines altogether, present o-machines, and then

draw a-machines as special cases. This reinforces the claim that it is the

o-machine, not the a-machine, which is the central concept of the subject.

Theorem 4.8. If P

e

is a Turing program for a Turing a-machine, then there

is a Turing oracle program

P

i

which on input x and any oracle A produces

the same output y.

Proof. Let P

e

be a Turing program to compute ϕ

e

. Now P

e

consists of a

ﬁnite partial map which can be identiﬁed with a set of 5-tuples,

δ : QS

2

−→ QS

2

¦R, L¦,

24

where Q is a ﬁnite set of states, S

2

= ¦B, 1¦ is the work tape alphabet, and

¦R, L¦ the set of head moving operations right and left. Deﬁne an oracle

program

P

i

as follows with transition function

δ : QS

1

S

2

−→ QS

2

¦R, L¦,

for S

1

= ¦B, 0, 1¦ the oracle tape alphabet as follows. For each line in P

e

of the form δ(q, a) = (p, b, X) for p, q ∈ Q and a, b ∈ S

2

, we add to oracle

program

P

i

a line

δ(q, c, a) = (p, b, X) for both c = 0 and c = 1. Hence,

P

i

has exactly the same eﬀect on input x as P

e

regardless of the oracle A.

4.4.1 Notation for Functions and Functionals

The standard notation is that given above.

Remark 4.9. Note that the oracle Turing machine Φ

e

is a ﬁnite object

represented by an oracle program

P

e

or an oracle graph G

e

and has no oracle

associated with it, but it can use any oracle A which may be attached. This

is analogous to a laptop computer with no active connection to a database

which may later be connected to the World Wide Web.

Recently, some researchers have unfortunately used Φ

e

to denote ϕ

e

.

This is unwise because it blurs the distinction of types in which ϕ

e

operates

on integers and Φ

e

on sets. Furthermore, sometimes we would like to write

Φ

e

alone without its exponent A to identify it with

P

e

or its oracle graph

G

e

as a ﬁnite object, like a laptop computer whose link with the web has

temporarily been removed. Doing so under the new convention leads to con-

fusion with ϕ

e

which is given by a diﬀerent type of program. The functional

Φ

e

is deﬁned by a program

P

e

which is a ﬁnite set of 6-tuples operating on

sets, while ϕ

e

is deﬁned by P

e

a ﬁnite set of 5-tuples operating on integers.

Furthermore, there is no justiﬁcation for the necessity of Φ

e

to denote ϕ

e

since the current notation ϕ

e

is quite satisfactory. We recommend against

using Φ

e

to denote ϕ

e

the partial computable function.

5 Emil Post Expands Turing’s Ideas

The spirit of Turing’s work was taken up by the American mathematician

Emil Post, who had been appointed to a faculty position at City College of

New York in 1932.

25

5.1 Post’s Work in the 1930’s

Post [1936] independently of Turing (but not independently of the work by

Church and Kleene in Princeton) had deﬁned a “ﬁnite combinatory pro-

cess” which closely resembles a Turing machine. From this it is often and

erroneously written (Kleene [1987b, p. 56], and [1981, p. 61]) that Post’s

contribution here was “essentially the same” as Turing’s, but in fact it was

much less. Post did not attempt to prove that his formalism coincided with

any other formalism, such as general recursiveness, but merely expressed

the expectation that this would turn out to be true, while Turing [1937b]

proved the Turing computable functions equivalent to the λ-deﬁnable ones.

Post gave no hint of a universal Turing machine. Most important, Post gave

no analysis, as did Turing, of why the intuitively computable functions are

computable in his formal system. Post oﬀers only as a “working hypoth-

esis” that his contemplated “wider and wider formulations” are “logically

reducible to formulation 1.” Lastly, Post, of course, did not prove the un-

solvability of the Entscheidungsproblem because at the time Post was not

aware of Turing’s [1936], and Post believed that Church [1936] had settled

the Entscheidungsproblem. Furthermore, Post wrote [1936] that Church’s

identiﬁcation of eﬀective calculability and recursiveness was working hy-

pothesis which is in “need of continual veriﬁcation.” This irritated Church

who criticized it in his review [1937b] of Post [1936].

Post’s contributions during the 1930’s were original and insightful, cor-

responding in spirit to Turing’s more than to Church’s, but they were not

as inﬂuential as those of Church and Turing. It was only during the next

phase from 1940 to 1954 that Post’s remarkable inﬂuence was fully felt.

5.2 Post Steps Into Turing’s Place During 1940–1954

As Turing left the subject of pure computability theory in 1939, his mantle

fell on the shoulders of Post. This was the mantle of clarity and intuitive ex-

position, the mantle of exploring the most basic objects such as computably

enumerable sets, and most of all, the mantle of relative computability and

Turing reducibility. During the next decade and a half from 1940 until his

death in 1954, Post played an extraordinary role in shaping the subject.

Post [1941] and [1943] introduced a second and unrelated formalism

called a production system and (in a restricted form) a normal system,

which he explained again in [1944]. Post’s (normal) canonical system is

a generational system, rather than a computational system as in general

recursive functions or Turing computable functions, because it gives an al-

26

gorithm for generating (listing) a set of integers rather than computing a

function. This led Post to concentrate on eﬀectively enumerable sets rather

than computable functions. Post, like Church and Turing, gave a thesis

[1943, p. 201], but stated it in terms of generated sets and production sys-

tems, which asserted that “any generated set is a normal set.” That is,

any eﬀectively enumerable set in the intuitive sense could be produced as a

normal set is his formal system. Although he had used other terminology

earlier, by the 1940’s Post had adopted the Kleene-Church terminology of

“recursively enumerable set” for the formal equivalent of Post’s eﬀectively

enumerable set.

Deﬁnition 5.1. [Post’s Thesis, 1943, 1944]. A nonempty set is eﬀec-

tively enumerable (listable in the intuitive sense) iﬀ it is recursively enumer-

able (the range of a recursive function) or equivalently iﬀ it is generated by

a (normal) production system.

Post showed that every recursively enumerable set (one formally gen-

erated by a recursive function) is a normal set (one derived in his normal

canonical system) and conversely. Therefore, normal sets are formally equiv-

alent to recursively enumerable sets. Since recursively enumerable sets are

equideﬁnable with partial computable functions, this deﬁnition of normal set

gives a new formal deﬁnition of computability which is formally equivalent

to the deﬁnitions of Church or Turing. (Equideﬁnable here means that from

the deﬁnition of a partial computable function we can derive a c.e. set as its

range and from the deﬁnition of a c.e. set one can ﬁnd a single valued c.e.

subset which is the graph of a partial computable function.) Post’s Thesis

is equivalent to Turing’s Thesis.

Post used the terms “eﬀectively enumerable set” and “generated set”

almost interchangeably, particularly for sets of positive integers. Post [1944,

p. 285], like Church [1936], deﬁned a set of positive integers to be recur-

sively enumerable if it is the range of a recursive function and then stated,

“The corresponding intuitive concept is that of an eﬀectively enumerable

set of positive integers.” (This is Church’s [1936] terminology also). Post

[1944, p. 286], explained his informal concept of a “generated set” of positive

integers this way,

“Suﬃce it to say that each element of the set is at some time

written down, and earmarked as belonging to the set, as a result

of predetermined eﬀective processes. It is understood that once

an element is placed in the set, it stays there.”

Post then [1944, p. 286], restated Post’s Thesis 5.1 in the succinct form,

27

“every generated set of positive integers is recursively enumer-

able.”

He remarked that “this may be resolved into the two statements: every

generated set is eﬀectively enumerable, every eﬀectively enumerable set of

positive integers is recursively enumerable.” Post continued, “their con-

verses are immediately seen to be true.” Post’s concentration on c.e. sets

rather than partial computable functions may be even more fundamental

than the thesis of Church and Turing characterizing computable functions

because Sacks [1990] has remarked that often in higher computability the-

ory it is more convenient to take the notion of a generalized c.e. set as basic

and to derive generalized computable functions as those whose graphs are

generalized computably enumerable.

5.3 Post’s Problem on Incomplete C.E. Sets

Post’s most inﬂuential achievement during this period was the extraordinar-

ily clear and intuitive paper, Recursively enumerable sets of positive integers

and their decision problems, [1944]. Here Post introduced the terms degree

of unsolvability and the concept that one set has lower degree of unsolvability

than another. Post later expanded on these deﬁnitions in [1948].

Post’s paper [1944] revealed with intuition and great appeal the signif-

icance of the computably enumerable sets and the signiﬁcance of G¨odel’s

Incompleteness Theorem. Post called G¨odel’s diagonal set,

K = ¦e : e ∈ W

e

¦

the complete set because every c.e. set W

e

is computable in K (W

e

≤

T

K).

Moreover, Post felt that the creative property of K revealed the inherent

creativeness of the mathematical process.

5.4 Post Began With Strong Reducibilities

Post posed his famous “Post’s problem” of whether there exists a com-

putably enumerable (c.e.) set W which is not computable but which cannot

compute G¨odel’s diagonal set K, i.e., such that ∅ <

T

W <

T

K. In 1944 re-

searchers did not understand Turing reducibility, even as little as presented

above in '4. Post himself was struggling to understand it, and did not ex-

plicitly discuss it until the very end of his paper, and even then only in

general terms.

28

Post’s contributions from 1943 to 1954 concerning relative computabil-

ity are remarkable. First, Post resurrected in [1944] the concept of oracle

machines which had been buried in Turing’s [1939] paper and which other

researchers had apparently ignored for ﬁve years. Second, Post deﬁned a

sequence of strong reducibilities to better understand the concept of a set

B being reducible to a set A.

Along with these strong reducibilities, Post deﬁned families of c.e. sets

with thin complements, simple, hyper-simple, hyper-hypersimple, in an at-

tempt to ﬁnd an incomplete set for these reducibilities. These concepts have

pervaded the literature and proved useful and interesting, but they did not

lead to a solution of Post’s problem. Post was able to exhibit incomplete

incomputable c.e. sets for several of these stronger reducibilities but not for

Turing reducibility. Post’s Problem stimulated a great deal of research in

the ﬁeld and had considerable inﬂuence.

Slowly Post’s understanding deepened of the general case of one set B

being reducible (Turing reducible) to another set A. Post steadily contin-

ued gaining a deeper and deeper understanding from 1943 to 1954 until he

had brought it to full development. Our modern understanding of relative

computability and Turing functionals is due more to Post and his patient,

persistent eﬀorts over a decade and a half than it is due to the brief remark

by Turing in 1939.

6 Post Highlights Turing Computability

When Post wrote his famous paper [1944], Turing’s notion of relative com-

putability from an oracle discussed in '4.1 had been mostly ignored for ﬁve

years. It was only at the end of Post’s paper [1944] in the last section, '11

General (Turing) Reducibility, that Post deﬁned and named for the ﬁrst time

“Turing Reducibility,” denoted B ≤

T

A, and began to discuss it in intuitive

terms. Post’s four and a half page discussion there is the most revealing

introduction to eﬀective reducibility of one set from another. In the same

crisp, intuitive style as in the rest of the paper, Post described the manner

in which the decision problem for one set S

1

could be reduced to that of a

second set S

2

. Post wrote it for a c.e. set S

2

in studying Post’s problem, but

the analysis holds for any set S

2

.

6.1 Post Articulates Turing Reducibility

Post wrote in [1944, '11],

29

“Now suppose instead, says Turing [1939] in eﬀect, this situation

obtains with the following modiﬁcation. That at certain times

the otherwise machine determined process raises the question is

a certain positive integer in a given recursively enumerable set

S

2

of positive integers, and that the machine is so constructed

that were the correct answer to this question supplied on every

occasion that arises, the process would automatically continue to

its eventual correct conclusion. We could then say that the ma-

chine eﬀectively reduces the decision problem of S

1

to that of S

2

.

Intuitively, this would correspond to the most general concept

of reducibility of S

1

to S

2

. For the very concept of the decision

problem of S

2

merely involves the answering for an arbitrarily

given single positive integer m of the question is m in S

2

; and in a

ﬁnite time but a ﬁnite number of such questions can be asked. A

corresponding formulation of “Turing reducibility” should then

be the same degree of generality for eﬀective reducibility as say

general recursive function is for eﬀective calculability.”

6.2 The Post-Turing Thesis

Post’s statement may be restated in succinct modern terms and incorporates

the statement implicit in Turing [1939, '4] in the following extension of

Turing’s ﬁrst Thesis 3.1 and Post’s ﬁrst Thesis 5.1.

Deﬁnition 6.1. Post-Turing Thesis, Turing [1939 §4], Post [1944,

§11]. A set B is eﬀectively reducible to another set A iﬀ B is Turing

reducible to A by a Turing oracle machine (B ≤

T

A).

Turing’s brief introduction of oracles did not state this as a formal the-

sis, but it is largely implied by his presentation. Post makes it explicit and

claims that this is the formal equivalent of the intuitive notion of eﬀec-

tively reducible, a step as signiﬁcant as the Church-Turing characterization

of “calculable.” If we identify a Turing reduction Φ

e

with its graph G

e

both

informally and formally then the Post-Turing Thesis is equivalent to Post’s

Thesis 5.1 (because G

e

is c.e.), which is equivalent to Turing’s Thesis 3.1.

However, there has been little analysis (along the lines of the extensive

analysis of the Church-Turing Thesis 3.2 for unrelativized computations) of

what constitutes a relative computation of B from A. This is surprising

because the Post-Turing Thesis was stated clearly in Post [1944]. It is even

more surprising because relative computability is used much more often than

30

ordinary computability in the theory of computability, applications of com-

putability to other areas such as algebra, analysis, model theory, algorithmic

complexity and many more. Also interactive or online computing in the real

world is more common than batch processing or oﬄine computing, using

processes contained entirely inside the machine.

7 Continuous and Total Functionals

7.1 Representations of Open and Closed Sets

Deﬁnition 7.1. (i) Using ordinal notation identify the ordinal 2 with the

set of smaller ordinals ¦0, 1¦. Identify the sets A ⊆ ω with their character-

istic functions f : ω → ¦0, 1¦ and represent the set of such f as 2

ω

.

(ii) Let 2

<ω

denote the set of ﬁnite strings of 0

**s and 1’s, i.e., ﬁnite initial
**

segments of functions f ∈ 2

ω

. Let σ and τ represent ﬁnite strings in 2

<ω

and σ ≺ τ or σ ≺ f denote that σ is an initial segment of τ or f.

(iii) Recall the deﬁnition of a ﬁnite set D

y

with canonical index y where

D

y

= ¦x

1

< x

2

, . . . < x

k

¦ and y = 2

x

1

+ 2

x

2

. . . = 2

x

k

. Deﬁne the string σ

y

with canonical index y by σ

y

(z) = 1 if z ∈ D

y

and σ

y

(z) = 0 otherwise.

(iv) Cantor space is 2

ω

with the following topology (class of open sets). For

every σ ∈ 2

<ω

the basic open (clopen) set

^

σ

= ¦ f : f ∈ 2

ω

& σ ≺ f ¦.

The sets ^

σ

are called clopen because they are both closed and open. The

open sets of Cantor space are the countable unions of basic open sets.

(iv) Set A ⊆ 2

<ω

is an open representation of the open set ^

A

=

σ∈A

^

σ

.

We may assume A is closed upwards, i.e., σ ∈ A and σ ≺ τ implies τ ∈ A.

(v) A set ( is (topologically) closed if its complement ^

A

is open, i.e.,

( = ^

A

= (2

ω

−^

A

). In this case T =

dfn

2

<ω

−A is a closed representation

for (. Now T is closed downwards (because A is closed upwards). Hence,

we shall see that T forms a tree as in Deﬁnition 7.2 (i).

7.2 Notation for Trees

Deﬁnition 7.2. (i) A tree T ⊆ 2

<ω

is a set closed under initial segments,

i.e., σ ∈ T and τ ≺ σ imply τ ∈ T. Fix any tree T.

31

(ii) The set of inﬁnite paths through T is

(5) [ T ] = ¦ f : (∀n) [ fn ∈ T ] ¦.

Note that [ T ] is always a closed set. If ( ⊆ 2

ω

is any closed set then by

Theorem 7.3 there is a (nonunique) tree T ⊆ 2

<ω

such that ( = [ T ], which

is called a closed representation for (.

(iii) For σ ∈ T deﬁne the subtree T

σ

of nodes τ ∈ T comparable with σ,

(6) T

σ

= ¦ τ : σ _ τ or τ ≺ σ ¦.

(iv) Deﬁne the subtree of extendible nodes σ ∈ T.

(7) T

ext

= ¦ σ ∈ T : (∃f ~ σ)[ f ∈ [ T ] ] ¦

Note that if the tree T is computable then T

ext

is co-c.e. Usually for a

given tree T there are many other trees T

such that [ T ] = [ T

], i.e., many

diﬀerent representations for the same closed set [ T ]. The closed sets are

closed under ﬁnite union and countable intersection since the open sets have

the dual properties in Deﬁnition 7.1 (iii), closure under ﬁnite intersection

and countable union. The clopen sets are both open and closed, so any

countable union of them is open and any countable intersection is closed.

Theorem 7.3. If T is a tree, then [ T ] is a closed set, and for every closed

set ( there is a tree T such that ( = [ T ].

Proof. (=⇒). Given tree T let A = 2

<ω

−T. Then ^

A

is open. Therefore,

[ T ] = 2

ω

−^

A

and [ T ] is closed.

(⇐=). Given any closed ( with complement ^

A

deﬁne T by putting σ in T

if (∀τ _ σ)[ τ ∈ A]. Then T is closed downward and [ T ] = (.

Theorem 7.4. [Compactness]. The following very easy and well-known

properties hold for Cantor Space 2

ω

. The term “compactness” refers to any

of them, but particularly to (iv). These properties can be proved from one

another but here we give direct proofs of each.

(i) K¨onig’s Lemma. If T ⊆ 2

<ω

is an inﬁnite tree, then [ T ] = ∅.

(ii) If T

0

⊇ T

1

. . . is a decreasing sequence of trees with [ T

n

] = ∅ for every

n, and intersection T

ω

= ∩

n∈ω

T

n

, then [ T

ω

] = ∅.

(iii) If ¦(

i

¦

i∈ω

is a countable family of closed sets such that ∩

i∈F

(

i

= ∅ for

every ﬁnite set F ⊆ ω, then ∩

i∈ω

(

i

= ∅ also.

(iv) Finite subcover. Any open cover ¦^

σ

¦

σ∈A

⊇ 2

ω

has a ﬁnite open

subcover ¦^

σ

¦

σ∈F

⊇ 2

ω

for some ﬁnite subset F ⊆ A.

32

7.3 Dense Open Subsets of Cantor Space

Deﬁnition 7.5. Let o be Cantor space 2

ω

or Baire space ω

ω

.

(i) A set / ⊆ o is dense if (∀σ) (∃f ~ σ) [ f ∈ / ].

(ii) A set / ⊆ o is dense open if

(8) (∀τ)(∃σ _ τ)(∀f ~ σ) [ f ∈ / ].

(iii) Let T ⊆ 2

<ω

be a tree. A point f ∈ [ T ] is isolated in [ T ] if

(9) (∃σ)[ [ T

σ

] = ¦ f ¦ ].

We say that σ isolates f because ^

σ

∩[ T ] = ¦f ¦. If f is not isolated, then

f is a limit point.

(iv) A space o is separable if it has a countable base of open sets. (Both

Cantor space and Baire space are separable.)

(v) A class B ⊆ o is G

δ

, i.e., boldface Π

0

2

, if B = ∩

i

/

i

a countable

intersection of open sets /

i

.

After open and closed sets, much attention is paid in point set topology

to G

δ

sets (see Oxtoby[1971]). If the open sets /

i

are also dense open, then

they have special signiﬁcance. Banach-Mazur games can be used for ﬁnd-

ing a point f ∈ ∩

i

/

i

where the the sets /

i

are dense and open. This is

the paradigm for the ﬁnite extension constructions in computability theory

and oracle constructions as Kleene and Post, especially the Finite Extension

Paradigm of Kleene and Post, which we use to construct sets and degrees

meeting an countable sequence of “requirements.” Meeting a given require-

ment R

i

amounts to meeting the corresponding dense open set /

i

.

7.4 Eﬀectively Open and Closed Sets

Deﬁnition 7.6. (i) If A ⊆ 2

<ω

is c.e. and / = ^

A

, then / is eﬀectively

(computably) open.

(ii) If ( = 2

ω

−^

A

for A c.e., or equivalently if ( = [ T ] for a computable

tree T ⊆ 2

<ω

, then ( is eﬀectively (computably) closed.

(iii) ( ⊆ 2

ω

is a Π

0

1

class if there is a computable relation R(x) such that

(10) ( = ¦f : (∀x) R(f(x))¦.

33

We call this lightface Π

0

1

because it is deﬁned in (10) by a universal quantiﬁer

outside of a computable relation R(x).

(iv) A set ( ⊆ 2

ω

is boldface Π

1

if there is some set S ⊆ ω such that

( = ¦f : (∀x) R

S

(f(x))¦,

and we say ( is in Π

S

1

. Here R

S

denotes a relation computable in the set S

which we call the parameter determining R

S

. A set / ⊂ 2

ω

is boldface Σ

1

or Σ

S

1

if its complement 2

ω

−/ is boldface Π

1

or Π

S

1

.

Theorem 7.7. The open (closed) sets of 2

ω

are exactly the boldface Σ

1

(Π

1

) sets and any boldface set is lightface in some parameter A.

Proof. If / is open, then / = ^

A

for some countable set A. Now A de-

termines exactly which σ contribute in the countable union / = ∪

σ∈A

^

σ

.

Hence, if we ﬁx A as a parameter, then the deﬁnition and properties become

lightface Σ

A

1

, i.e., eﬀectively open relative to the oracle A. (But this entire

section is about working relative to an oracle.)

We often convert an open set / = ^

A

into the realm of computability

theory as follows. We: (1) usually ﬁx the parameter A and do a construction

which is computable relative to the parameter A; (2) often replace the open

set ^

A

by its complement the closed set ( = ^

A

; and (3) replace the closed

set by ( = [ T

A

] for a A-computable tree T. By ﬁxing the parameter A we

can apply all the methods of this chapter on A-computable constructions,

such as the Recursion Theorem, construction of an A-c.e. set B, and so

forth.

Theorem 7.8. A class ( is a Π

0

1

class iﬀ ( is eﬀectively closed, i.e., ( = [ T ]

for some computable tree T.

Proof. (=⇒). Let ( = ¦ f : (∀x) R(f(x)) ¦ for R computable. Deﬁne a

computable tree T = ¦ σ : (∀τ ⊆ σ)[R(τ)] ¦. Then [ T ] = (.

(⇐=). Let ( = [ T ] for T a computable tree. Deﬁne R(σ) iﬀ σ ∈ T. Then

¦f : (∀x) R(f(x))¦ = [ T ].

7.5 Continuous Functions on Cantor Space

In elementary calculus courses a continuous function is usually deﬁned with

δ and concepts or with limits. In advanced analysis or topology courses

34

the more general deﬁnition is presented that a function F is continuous iﬀ

the image of every basic open set is open. We state continuity now for

functionals on the Cantor space 2

ω

. For arbitrary topological spaces an

open set is an arbitrary union of basic open sets, but for Cantor space we

may take countable unions.

Deﬁnition 7.9. (i) A functional Ψ on Cantor space 2

ω

is continuous if for

every τ ∈ 2

<ω

there is a countable set X

τ

such that

(11) Ψ

−1

(^

τ

) = ∪ ¦ ^

σ

: σ ∈ X

τ

¦.

By identifying string σ

y

with code number y deﬁned in Deﬁnition 7.1 (iii)

we can think of X

τ

as a subset of ω.

(ii) A functional Ψ on Cantor space 2

ω

is total if Ψ

A

(x) is deﬁned for every

A ⊆ ω and x ∈ ω, i.e., if (∀A)(∃B)(∀x)[ Ψ

A

(x) = B(x) ].

Theorem 7.10. Let Ψ be a continuous functional on Cantor Space 2

ω

.

Then Ψ is total iﬀ for each set X

τ

of (11) there is a ﬁnite subset

X

τ

⊆ X

τ

such that,

(12) 2

ω

⊆ ∪ ¦ ^

σ

: σ ∈

X

τ

¦.

Proof. (⇐=). Given such

X

τ

for every τ we see that Ψ is total by (12).

(=⇒). Assume Ψ is total. Then for every τ there is a countable set X

τ

which covers 2

ω

in the sense of (12). Now the Compactness Theorem asserts

that any open cover of Cantor space 2

ω

has a ﬁnite subcover. Therefore, we

can replace every set X

τ

by a ﬁnite subset

X

τ

.

Note that this involves only compactness and has no computable content,

although it is usually presented in its eﬀective analogue as the theorem that

Φ

e

is total iﬀ it is a truth-table reduction.

7.6 Eﬀectively Continuous Functionals

A Turing functional Φ

e

is not only continuous but eﬀectively continuous in

the following sense.

Theorem 7.11. For Φ

e

deﬁne for every τ ∈ 2

<ω

the set of strings

(13) X

e

τ

= ¦ σ : (∃s)( ∀x < [τ[ ) [ Φ

σ

e,s

(x)↓ = τ(x) ] ¦.

Then X

e

τ

is c.e. and uniformly so in the sense that there is a computable

function h(e, τ) such that W

h(e,τ)

= X

e

τ

.

35

Proof. Use the Oracle Graph G

e

. Identify string σ with ^

σ

. Now the set of

strings ¦ X

e

τ

: τ ∈ 2

ω

¦ witnesses that the functional Φ

e

is continuous.

Deﬁnition 7.12. Since X

e

τ

is not only countable but also computably enu-

merable uniformly in e the functional Φ

e

is eﬀectively continuous, i.e., that

X

e

τ

is a computably enumerable set of strings.

7.7 Continuous Functions are Relatively Computable

We showed that any Turing functional is continuous, indeed eﬀectively con-

tinuous. Now we prove that any continuous functional on 2

ω

is a Turing

functional relative to some real parameter X ⊆ ω and therefore is eﬀectively

continuous relative to X.

Theorem 7.13. Suppose Ψ is a continuous functional on 2

ω

. Then Ψ is a

Turing functional relative to some real parameter X ⊆ ω.

Proof. Since Ψ is continuous, the inverse image of every basic open set ^

τ

,

τ ∈ 2

<ω

, is open and therefore is a countable union of basic open sets.

Hence, (identifying strings σ with their code numbers as integers) there is a

set X

τ

⊆ ω such that,

Ψ

−1

( ^

τ

) = ∪ ¦ ^

σ

: σ ∈ X

τ

¦.

Therefore, the set X = ⊕¦X

τ

: τ ∈ 2

<ω

¦ provides a complete oracle for

calculating Ψ : 2

ω

→ 2

ω

.

8 Bounded Reducibilities

8.1 A Matrix M

x

for Bounded Reducibilities

A bounded reducibility is a Turing reducibility Φ

A

e

(x) with a computable

function h(x) which bounds the use function, i.e., ϕ

A

e

(x) ≤ h(x). Given

h(x) imagine a matrix M

x

whose rows are all the strings σ of length h(x).

The action on x is entirely determined by the action of Φ

σ

e

(x) for these rows

σ ∈ M

x

. For example, if B ≤

m

A via a computable function f(x) then

this reduction is bounded by h(x) = f(x) and x ∈ B iﬀ σ(f(x)) = 1 where

σ ≺ A, i.e., where σ is an initial segment of A. In Deﬁnition 7.1 (iii) we

deﬁned σ

y

to be the string with canonical index y derived from D

y

. This

gives a method for indexing the rows σ

y

∈ M

x

.

We begin in '8.2 with the most general bounded reducibility called

bounded Turing reducibility (B ≤

bT

A) where we are given only the com-

putable bound h(x), i.e., only the matrix M

x

. In '8.3 we study the more

36

restrictive case of truth-table reducibility (B ≤

tt

A) where Φ

σ

e

(x) converges

for every x and every row σ ∈ M

x

. If Φ

σ

e

(x) diverges for even one x and

σ ∈ M

x

then the reduction is a partial bT-reduction, but if it converges for

every x and every σ ∈ M

x

then it gives a genuine truth-table as in beginning

logic courses as deﬁned in Theorem 8.3 (iii). This convergence on all strings

σ

y

∈ M

x

is the key deﬁning property for a tt-reduction as expressed in (14).

8.2 Bounded Turing Reducibility

Deﬁnition 8.1. (i) A set B is bounded Turing reducible (bT-reducible) to

a set A (B ≤

bT

A) if there is a Turing reduction Φ

A

e

= B and a computable

function h(x) such that the use ϕ

A

e

(x) ≤ h(x).

(ii) A set B is identity bounded Turing reducible to A (B ≤

ibT

A) if B ≤

bT

A

with h(x) = x.

The bT and ibT reductions occur naturally in several parts of the subject.

For example, often we are given a noncomputable c.e. set A and we construct

a simple set B ≤

T

A by simple permitting where an element x is allowed to

enter B at stage some stage only if some y ≤ x has just entered A. When

Ax has settled then Bx has settled also, and hence B ≤

ibT

A.

More recently, ibT has occurred in applications of computability to dif-

ferential geometry in Soare [2004] and Csima-Soare [ta], as described later,

and ibT reducibility has also been used in applications to algorithmic ran-

domness and Kolmogorov complexity. Barmpalias and Lewis [ta] have shown

the nondensity of the ibT-degrees of c.e. sets.

8.3 Truth-Table Reductions

Deﬁnition 8.2. A set B is truth-table reducible to set a A (B ≤

tt

A) if

B ≤

bT

A via Φ

A

e

= B and h(x) as in Deﬁnition 8.1, and also

(14) (∀x)(∀σ)[ [σ[ = h(x) =⇒ Φ

σ

e

(x)↓ ].

Theorem 8.3. [Truth-Table Theorem, Nerode]. The following deﬁ-

nitions of tt-reducible are equivalent.

(i) B ≤

tt

A as deﬁned in Deﬁnition 8.2 and (14).

(ii) Φ

A

e

= B for some total Φ

X

e

that is (∀X)(∃Y )(∀x)[ Φ

X

e

(x) = Y (x) ].

37

(iii) B ≤

tt

A via Φ

A

e

and h(x) as in Deﬁnition 8.2. In addition, there is a

computable function g(x) such that

D

g(x)

= ¦ y : [σ

y

[ = h(x) & Φ

σy

e

(x) = 1 ¦.

(iv) There exist computable functions g and h such that, for all x,

x ∈ B ⇐⇒ (∃z ∈ D

g(x

) [ A(h(x) + 1) = D

z

].

Proof. The implications (i) =⇒ (ii), (iii) =⇒ (i), and (iii) ⇐⇒ (iv) are

obvious. It remains to prove (ii) =⇒ (iii).

(ii) =⇒ (iii). Uniformly in x we can eﬀectively enumerate the set

U

x

= ¦ σ : Φ

σ

e

(x)↓ ¦.

Since Φ

e

is total apply the Compactness Theorem 7.4 (iv) to get a ﬁnite

subset V

x

⊆ U

x

such that ∪¦ ^

σ

: σ ∈ V

x

¦ ⊇ 2

ω

. Deﬁne h(x) and g(x) by

h(x) = max¦ [σ[ : σ ∈ V

x

¦.

D

g(x)

= ¦ y : [σ

y

[ = h(x) & Φ

σy

e

(x) = 1 ¦.

Remark 8.4. Post [1944, '6] introduced the ﬁrst deﬁnition of B ≤

tt

A.

For every integer n he required a eﬀective procedure to produce integers

k, and m

1

, m

2

. . . m

k

. He then drew a diagram of a matrix with columns

m

1

, . . . , m

k

and 2

k

rows with individual entries of + for x ∈ A and − for

x / ∈ A. Letting h(x) be the maximum of the ¦m

i

¦

i≤k

, ﬁlling in all columns

less than h(x), and replacing + by 1 and − by 0 we have exactly the matrix

M

x

described in '8.1. Next Post drew a vertical line to the right of this

matrix and added an extra column such that an entry v

z

in this column

following row R

z

indicated that if A satisﬁes row R

z

then x ∈ B iﬀ v

z

= +.

Post’s extra column converted the matrix M

x

into a total reduction

because A had to extend exactly one of the rows R

z

and then the value

for B(x) was completely speciﬁed by v

z

given in advance. Therefore, Post’s

original deﬁnition [1944] of B ≤

tt

A is virtually identical in intuition and

description to our Deﬁnition 8.2 above where we could attach to matrix

M

x

an extra column with the value Φ

σ

e

(x) on row σ. In both cases the

crucial point is that Φ

σ

e

(x) is deﬁned for every string of length h(x) before

examining the oracle A. Property (iv) of B ≤

tt

A has been used in the

38

past e.g., Soare [1987, p. 83] and is short and slick but is does not give the

necessary elegance and intuition for beginning students.

Turing reducibility was not well understood in 1944 and an understand-

ing of it in its modern state emerged only gradually during the 1940’s and

1950’s. Post’s tt-reducibility in 1944 was understood at once. Therefore,

in 1959 Friedberg and Rogers introduced wtt-reducibility (B ≤

wtt

A) as

a weakening of tt-reducibility which was already a strengthening of ≤

T

.

Therefore, the concept of wtt has distance two from the central concept ≤

T

.

The concept of bounded Turing reducibility (bT), which has the same deﬁni-

tion as wtt, goes immediately to the main concept of bounding the queried

information by a computable function h(x). Thus, ≤

bT

has distance one

from the main reducibility of ≤

T

. The concept of tt-reducibility is tied to

the totality of the reduction Φ

A

e

which is exactly what bT-reducibility lacks.

Also bT is more recognizable than wtt whose meaning cannot be guessed

from its name.

8.4 Diﬀerence of c.e. sets, n-c.e., and ω-c.e. sets

The Limit Lemma characterized sets A ≤

T

∅

**as those such that A = lim
**

s

A

s

for a computable sequence ¦A

s

¦

s∈ω

. Now consider special cases based upon

how many times the approximation changes on x. These notions are more

general than c.e. sets A but not the most general case of A ≤

T

∅

.

Deﬁnition 8.5. (i) A set D is the diﬀerence of c.e. sets (d.c.e.) if D = A−B

where A and B are c.e. sets.

(ii) The set A is omega-c.e. (written ω-c.e.) if there is a computable sequence

¦A

s

¦

s∈ω

with A

0

= ∅ and a computable function g(x) which bounds the

number of changes in the approximation ¦A

s

¦

s∈ω

in the following sense,

(15) A = lim

s

A

s

& [ ¦ s : A

s

(x) = A

s+1

(x) ¦ [ ≤ g(x).

(iii) For n ∈ ω the set A is n-c.e. if g(x) ≤ n.

For example, the only 0-c.e. set is ∅, the 1-c.e. sets are the usual c.e. sets,

and the 2-c.e. sets are the d.c.e. sets. The next theorem gives an elegant

characterization of ω-c.e. sets.

Theorem 8.6. The following are equivalent.

(i) A ≤

bT

∅

.

(ii) A is ω-c.e.

(iii) A ≤

tt

∅

.

39

Proof. Clearly, (iii) implies (i).

(i) =⇒ (ii). Suppose Φ

∅

e

= A with computable bound g(x) ≥ ϕ

∅

e

(x). Let

A

s

(x) = Φ

∅

e

(x)[ s ] where we may speed up the computation so the latter is

always deﬁned. Now A

s+1

(x) = A

s

(x) only if some element z ≤ g(x) enters

∅

**which can happen at most g(x) + 1 times, once for each z ∈ [0, g(x)].
**

(ii) =⇒ (iii). Assume A = lim

s

A

s

satisﬁes (15) via g(x). Deﬁne the c.e.

change set C by

C = ¦ 'k, x` : 1 < k ≤ [ ¦s : A

s

(x) = A

s+1

(x)¦ [ ¦.

(Intuitively, whenever A

s

(x) = A

s+1

(x) we put into C the least element

of the form 'k, x`, i.e., on row ω

[x]

, not already in C.) Therefore, [C

[x]

[ is

the number of changes on x during the approximation, and [C

[x]

[ ≤ g(x)

by hypothesis (ii). Furthermore, A(x) = 1 iﬀ [C

[x]

[ is odd, because the

approximation changes between 0 and 1 starting with 0.

First build a truth-table to demonstrate that A ≤

tt

C. For a given x

the truth-table has rows of width g(x) as in Deﬁnition 8.2 (ii). For each

k ≤ g(x) construct a row beginning with k many 1

s followed by all 0

s.

Now to tt-compute from C whether x ∈ A compute k = [C

x

[. Next ﬁnd the

row with k many 1’s. Now A(x) = 1 iﬀ k is odd. Hence, A ≤

tt

C. However,

C is c.e. Hence, C ≤

m

0

and therefore A ≤

tt

0

.

This result shows the diﬀerence between T-reductions and tt-reductions.

First, the assumption on h(x) ensures that A ≤

bT

D. Second, the approx-

imation always begins with A

s

(x) = 0 for s = 0 and changes between 0

and 1 because all values are in ¦0, 1¦ rather than in ω. Therefore, we can

make up a row in the truth-table which gives value for A(x) based only on

the number of changes. This ensures that A ≤

tt

D. This case is unusual

since most reductions we consider only produce A ≤

T

B for some set B and

sometimes produce A ≤

bT

B as in permitting arguments.

9 Online Computing

The original implementations of computing devices were generally oﬄine

devices such as calculators or batch processing devices. However, in recent

years the implementations have been increasingly online computing devices

which can access or interact with some external database or other device.

The Turing o-machine is a better model to study them because the Turing

a-machine lacks this online capacity.

40

Deﬁnition 9.1. (i) An online or interactive computing process is one which

interacts with its environment, for example a computer communicating with

an external data base such at the World Wide Web.

(ii) An oﬄine computing process is one which begins with a program and

input data, and proceeds internally, not interacting with any external device.

This includes a calculator, and batch processing devices where a user handed

a deck of punched IBM cards to an operator, who fed them to the computer

and produced paper output later.

There are many descriptions in the computing literature about online

and interactive processes. In [Goldin-Smolka-Wegner] a chapter by Yuri

Gurevich, Interactive Algorithms 2005 is described,

“In this chapter, Gurevich asserts that computer science is largely

about algorithms, and broadens the notion of algorithms to in-

clude interaction by allowing intra-step interaction of an algo-

rithm with its environment.”

About the chapter, A Theory of Interactive Computation by Jan van

Leeuwen and Jiri Wiedermann, the book states,

“This chapter asks what a computational theory of interactive,

evolving programs should look like. The authors point out that a

theory of interactive computation must necessarily lead beyond

the classical, ﬁnitary models of computation. A simple model of

interactive computing is presented consisting of one component

and an environment, interacting using single streams of input

and output signals.”

It appears that the Turing o-machine is a good theoretical model to

analyze an interactive process because there is usually a ﬁxed algorithm

or procedure at the core, which by Turing’s thesis we can identify with a

Turing a-machine, and there is a mechanism for the process to communicate

with its environment, which when coded into integers may be regarded as

a Turing type oracle. Under the Post-Turing Thesis 6.1 these real world

online or interactive processes can be described by a Turing oracle machine.

In real world computing the oracle may be called a database or an en-

vironment. A laptop obtaining data from the World Wide Web is a good

example. In the real world the database would not be literally inﬁnite but

may appear so (like the web) and is usually too large to be conveniently

downloaded into the local computer. Sometimes the local computer is called

the “client” and the remote device the “server.”

41

9.1 Turing Machines and Online Processes

We could continue analyzing to what extent Turing oracle machines can

model modern online processes, but let us now examine the reverse direction,

that o-machines are online while a-machines are not. The following points

appear self-evident.

• Turing oracle machines are online. An o-machine has ﬁxed program

but has the capacity to interact with its environment and receive new

data during its computation.

• The original Turing a-machines are not online. A Turing a-machine,

even a universal machine, begins with a ﬁxed program and ﬁxed input

and proceeds without further outside input until (if ever) it halts.

• A large number and rapidly increasing number of computing processes

in the real world are online or interactive. See the authors in [Goldin-

Smolka-Wegner] for only a few.

• A large number of books presenting an introduction to computability

mention Turing oracle machines and relative computability very late

in the book or not at all.

• A large number of books with articles on Turing and the Church-

Turing Thesis do not mention Turing oracle machines or relative com-

putability at all, e.g., Christof Teuscher [2004].

Discussing only Turing a-machines in modern texts, or only the Church-

Turing Thesis, and not the Post-Turing Thesis on oracle computers, is like

discussing only batch processing machines of the 1950’s long after the emer-

gence of online computing.

9.2 Trial and Error Computing

We expect Turing a-machines and o-machines to be absolutely correct. How-

ever, there are many computing processes in the real world which give a se-

quence of approximations to the ﬁnal answer. Turing considered machines

which make mistakes. In his talk to the London Mathematical Society,

February 20, 1947, quoted in Hodges, p. 360–361, Turing said,

“I would say that fair play must be given to the machine. Instead

of it sometimes giving no answer we could arrange that it gives

42

occasional wrong answers. But the human mathematician would

likewise make blunders when trying out new techniques. It is

easy for us to regard these blunders as not counting and give him

another chance, but the machine would probably be allowed no

mercy. In other words if a machine is expected to be infallible, it

cannot also be intelligent. There are several theorems which say

exactly that. But these theorems say nothing about how much

intelligence may be displayed if a machine makes no pretence at

infallibility.

Hillary Putnam [1965] described trial and error predicates as ones for

which there is a computable function which arrives at the correct answer

after a ﬁnite number of mistakes. In modern terminology this is called

a limit computable function as described in Soare [1987] or Soare [CTA]

Chapter 3. This is a model for many processes in the real world which allow

ﬁnitely many mistakes but gradually move closer to the correct answer.

9.3 The Limit Lemma

There are several diﬀerent approaches for computing with ﬁnite errors.

Deﬁnition 9.2. A computable sequence of computable sets ¦B

s

¦

s∈ω

is a

computable (∆

0

) approximating sequence for a set B if B = lim

s

B

s

. If there

is such a sequence we say that B is limit computable,

Lemma 9.3. [Limit Lemma, Shoenﬁeld, 1959]. The following are

equivalent.

(i) B ≤

T

A for some c.e. “oracle set” A.

(ii) B is limit computable.

(iii) B ∈ ∆

2

, that is, B is is both two quantiﬁer forms.

Proof. (See Soare [1987] or Soare [CTA] Lemma 3.3.6.)

Now in the real world imagine an example of a computing process with

error such as: a robot learning a maze; a ﬁnancial trader receiving informa-

tion from around the world updated every second; a meteorologist predicting

the weather a week from now given constantly updated weather conditions

today. In our idealized model we assume that the individual makes ﬁnitely

many errors during the process ¦B

t

¦

t∈T

for time periods t ∈ T but even-

tually gets the correct answer. (In practice, the ﬁnal answer may not be

43

exactly correct in these examples, but is presumably more accurate than

the ﬁrst approximation B

0

. The sequences of improving approximations,

even if not exact, are usually useful in ﬁnancial trading, meteorology, and

other approximations in real time.)

9.4 Two Models for Computing With Error

9.4.1 The Limit Computable Model

In the limit computable or approximation model we have a sequence of

Turing programs ¦P

t

: t ∈ T¦ so that P

t

computes function g

t

at time t ∈ T.

There is not necessarily any connection between diﬀerent programs and we

may have to compute all over again with a new program as we pass from

time t to t + 1.

Suppose the ﬁnancial trader in Chicago receives data every second t ∈ T

about currency prices in London, Milan, New York and Tokyo. The conﬁg-

uration at his trading desk may be described using the Limit Lemma by a

computable function where g

t

is the computable characteristic function of

B

t

, the conﬁguration of his computation at the end of time t. The com-

putable function g

t

gives an algorithm to compute the condition B

t

at time

t but it gives no relationship between B

t

and B

t+1

. It will not be possible

for the trader to write a new program every second. How will the trader

write a program to uniformly compute the index g

t

for t ∈ T?

9.4.2 The Online Model

By the Limit Lemma there is a c.e. set A (or even a ∆

0

2

set) and oracle

machine Φ

e

such that B = Φ

A

e

. Now the trader can program the algorithm

Φ

e

into his laptop once and for all at the start of the trading day. Every

second t ∈ T he receives from New York and abroad the latest quotes A

t

which enter directly into his computer by an internet connection. He does

not (and cannot) change the program Φ

e

every second. His algorithm simply

receives the “oracle” information A

t

from the internet as it is continually

updated, and computes the approximation B

t

(x) = Φ

At

e

(x). His program

then executes a trade when the algorithm determines that conditions are

favorable. It is diﬃcult to see how this trader could have carried out his

business using a batch processing, Turing a-machine model, instead of an

online model.

44

10 Three Displacements in Computability Theory

The dictionary deﬁnes “displacement” to be the moving of something from

its rightful place or position, often when replaced by something else. There

are three important issues in computability which have at one time been

displaced from their correct or proper positions as evaluated by historical

and scientiﬁc criteria. These issues are at the very heart of the subject

and they deﬁne how we think about computability. We now examine these

three issues one by one in the next three sections. These items may have

been displaced accidentally or without conscious thought or decision, simply

acting from the exigencies of the situation at the time, but are not consistent

with a careful scientiﬁc analysis later.

When computabiliy theory originated in the 1930’s it was a very small

ﬁeld attempting to consolidate its ideas. Furthermore, three of its leaders,

G¨odel, Turing, and Church eﬀectively left the ﬁeld after 1940 and had little

direct inﬂuence thereafter on its development, although Church supervised

the Ph.D. theses of many prominent researchers in the ﬁeld. After 1940

the ﬁeld was developed and promulgated primarily by Stephen C. Kleene

with some additional inﬂuence by Emil Post as described earlier. In dedicat-

ing his book Degrees of Unsolvability [1971] Shoenﬁeld recognized Kleene’s

overwhelming inﬂuence,

“Dedicated to S.C. Kleene who made recursive function theory

into a theory.”

This is entirely accurate and without Kleene’s leadership we would not have

the ﬁeld as we know it today. However, a number of changes took place after

1940, perhaps by accident, that were not consistent with the original devel-

opment of computability in the 1930’s and would not have been approved

of by G¨odel and Turing.

11 “Computable” versus “Recursive”

Starting in 1936, Church and Kleene used the term “recursive” to mean

“computable” even though Turing and G¨odel later objected. Kleene later

introduced the term “recursive function theory” for the subject although

G¨odel disagreed (see below). This was the ﬁrst time in history that the

term “recursive” which had meant roughly “inductive” acquired the addi-

tional meaning “computable” of “calculable.” After 1996 the term the term

‘recursive” was again used only to mean “inductive” not “computable” or

45

“calculable.” From 1931 to 1934 Church and Kleene had used the λ-deﬁnable

functions as the formal equivalent of eﬀectively calculable functions, and

Church had ﬁrst proposed his thesis privately to G¨odel in that form.

11.1 Church Defends Church’s Thesis with “Recursive”

Recall Church’s Thesis 2.1 (First Version) [1934] “A function is eﬀectively

calculable if and only if it is λ-deﬁnable.” When G¨odel strongly rejected to

this thesis Church turned instead to G¨odel’s own Herbrand-G¨odel general

recursive functions as a formalism and proposed in [1935] and [1936] the

well-known form of his thesis, Church’s Thesis 2.2 [1936] A function on the

positive integers is eﬀectively calculable if and only if it is recursive.

Church and Kleene knew almost immediately and published by 1936

the proof of the formal equivalence of recursive functions with λ-deﬁnable

functions. After seeing G¨odel’s lectures in 1934 Church and Kleene dropped

the λ-deﬁnable functions and adopted the recursive functions. This was

not because of the inadequacy of λ-deﬁnable functions in comparison to the

recursive functions. Indeed Church seems to have preferred the λ-deﬁnable

functions and caused Turing to write his thesis [1939] in that formalism.

Church was very eager for mathematicians to accept his thesis and he

know that the recursive functions were more familiar to a mathematical au-

dience than λ-deﬁnable ones. Church and Kleene used the second version

of Church’s Thesis above, phrased in terms of recursive functions primarily

for public relations as Kleene [1981] explained, “I myself, perhaps unduly

inﬂuenced by rather chilly receptions from audiences around 1933–35 to dis-

quisitions on λ-deﬁnability, chose, after general recursiveness had appeared,

to put my work in that format. . . . ” Church and Kleene were simply doing

what most scientists do, arrange the work in a framework which will be

understandable and appealing to as large a scientiﬁc audience as possible.

Ironically, this is exactly what caused the change in 1996 from “recursive”

back to “computable” because in 1996 the term “computable” was much

better understood by a general audience than “recursive.” The irony is that

the term “computable” was there all along and was preferred by Turing and

G¨odel.

11.2 Church and Kleene Deﬁne “Recursive” as “Computable”

By 1936 Kleene and Church had begun thinking of the word “recursive”

to mean “computable.” Church had seen his ﬁrst thesis rejected by G¨odel

and was heavily invested in the acceptance of his 1936 thesis in terms of

46

recursive functions. Without the acceptance of this thesis Church had no

unsolvable problem. Church wrote in [1936, p. 96] printed in Davis [1965]

that a “recursively enumerable set” is one which is the range of a recursive

function. This is apparently the ﬁrst appearance of the term “recursively

enumerable” in the literature and the ﬁrst appearance of “recursively” as

an adverb meaning “eﬀectively” or “computably.”

In the same year Kleene [1936, p. 238] cited in Davis [1965] [p. 238]

mentioned a “recursive enumeration” and noted that there is no recursive

enumeration of Herbrand-G¨odel systems of equations which gives only the

systems which deﬁne the (total) recursive functions. By a “recursive enu-

meration” Kleene states that he means “a recursive sequence (i.e., the suc-

cessive values of a recursive function of one variable).” Post [1944], under

the inﬂuence of Church and Kleene, adopted this terminology of “recursive”

and “recursively enumerable” over his own terminology [1943], [1944] of “ef-

fectively generated set,” “normal set,” “generated set.” Thereafter, it was

ﬁrmly established.

11.3 G¨ odel Rejects “Recursive Function Theory”

Neither Turing nor G¨odel ever used the word “recursive” to mean “com-

putable.” G¨odel never used the term “recursive function theory” to name

the subject; when others did G¨odel reacted sharply negatively, as related by

Martin Davis.

In a discussion with G¨odel at the Institute for Advanced Study in

Princeton about 1952–54, Martin Davis casually used the term

“recursive function theory” as it was used then. Davis related,

“To my surprise, G¨odel reacted sharply, saying that the term in

question should be used with reference to the kind of work Rosza

Peter did.”

(See Peter’s work on recursions in [1934] and[1951].) By 1990 the situation

had become very diﬃcult. Most people access to a personal computer on

their desks and the terms of computing were familiar to the general popula-

tion but “recursive” was limited to very small number who mainly associated

it with a ﬁrst year programming course or a deﬁnition by induction on math-

ematics and almost never with computability. So few people understood the

meaning of “recursive” that by 1990 I had to begin my papers with,

“Let f be a recursive function (that is, a computable function),”

as if I were writing in Chinese and translating back into English.

47

11.4 The Ambiguity in the Term “Recursive”

The traditional meaning of “recursive” as “inductive” led to ambiguity.

Kleene often wrote of calculation and algorithms dating back to the Baby-

lonians, the Greeks and other early civilizations. However, Kleene [1981b,

p. 44] wrote,

“I think we can say that recursive function theory was born

there ninety-two years ago with Dedekind’s Theorem 126 (‘Satz

der Deﬁnition durch Induktion’) that functions can be deﬁned

by primitive recursion.”

Did he mean that recursion and inductive deﬁnition began with Dedekind or

that computability and algorithms began there? The latter would contradict

his several other statements, such as Kleene [1988, p. 19] where he wrote,

“The recognition of algorithms goes back at least to Euclid (c. 330 B.C.).”

When one uses a term like “recursive” to also mean “computable” or “algo-

rithmic” as Kleene did, then one is never sure whether a particular instance

means “calculable” or “inductive” and our language has become indistinct.

Returning “recursive” to its original meaning of “inductive” has made its

use much clearer. We do not need another word to mean “computable.” We

already have one.

11.5 Changing “Recursive” Back to “Inductive”

By 1996 the confusion had become intolerable. I wrote an article on Com-

putability and Recursion for the Bulletin of Symbolic Logic [1996] on the

history and scientiﬁc reasons for why we should use “computable” and not

“recursive” to mean “calculable.” “Recursive” should mean “inductive” as

is had for Dedekind and Hilbert. At ﬁrst few were willing to make such a

dramatic change, overturning a sixty year old tradition of Kleene, and the

words “computability theory” and “computably enumerable (c.e.) set” did

not come tripping from the lips. However, in a few months more people

were convinced by the undeniable logic of the situation. Three years later

at the A.M.S. conference in Boulder, Colorado referenced in Soare [2000],

most researchers, especially those under forty years old, had adopted the

new terminology and conventions. Changing back from from “recursive” to

“computable” during 1996–1999 has had a number of advantages.

Historical Accuracy. The founders of the two key models of com-

putabilty, Turing and G¨odel, had never used “recursive” to mean com-

putable and indeed had objected when it was so used. The object of

48

everyone was to formally capture the informal concept of “eﬀectively

calculable” which Turing machines did to G¨odel’s satisfaction, while

at ﬁrst G¨odel’s own model of Herbrand-G¨odel recursive functions did

not. The object was never to understand the notion of recursions

or inductive deﬁnitions. In 1935 Church adopted G¨odel’s recursive

functions as a deﬁnition of eﬀectively calculable before seeing Turing

machines As Kleene relates, Church and Kleene did this as a matter

of public relations to relate to a concept mathematicians could under-

stand, not in an attempt to better understand the nature of recursion

and inductive deﬁnition.

Scientiﬁc Accuracy. The words “calculate” and “compute” are very

close in the dictionary, the former being a bit more general. The

word “recursive” means a procedure characterized by recurrence or

repetition from the Latin verb “recurrere,” to run back. The word

has nothing to do with “calculate” or “compute.” The general public

understands the ﬁrst two words in this context. To the extent that they

have any idea about “recursive” they understand it in this context.

For example, a ﬁrst year programming course speaks of deﬁnition by

iteration versus deﬁnition by recursion.

Name Recognition. Suppose a student is scanning a catalogue for a

course to take and sees the course title “Recursive Function Theory”

versus “Computability Theory.” Which will give him more information

about the content of the course? Should a fresh Ph.D. apply for jobs

under the general area of his work as the ﬁrst or second? Should a

professor write an abstract for his lecture at another university under

the ﬁrst title or second?

Names do matter. They mattered to Church and Kleene in 1936 when

they changed from the term “λ-deﬁnable function” to “recursive function”

to achieve greater name recognition among mathematicians and to make

Church’s Thesis more convincing before the appearance of Turing. Names

matter today as we try to relate our specialty of computability to the world

of computers and algorithmic procedures all around us, a world partially

created by Turing.

49

12 Renaming it the “Computability Thesis?”

12.1 Kleene Called it “Thesis I” in [1943]

In the 1930’s both Church and Turing thought they were giving deﬁnitions

of an eﬀectively calculable function, not putting forth a “thesis.” These were

not even called “theses” at all until Kleene [1943, p. 60] referred to Church’s

“deﬁnition” as “Thesis I.”

12.2 Kleene Named it “Church’s thesis” in [1952]

Later in his very inﬂuential book [1952] it is fascinating to see how Kleene’s

thinking and terminology progressed from “Thesis I” to “Church’s Thesis”

and not to “Turing’s Thesis” or the “Church-Turing Thesis.” Kleene [1952,

p. 300] took up where Kleene [1943] had left oﬀ.

“This heuristic evidence and other considerations led Church

1936 to propose the following thesis.

Thesis I. Every eﬀectively calculable function (eﬀectively decid-

able predicate) is general recursive.”

This is identical with what we previously called Church’s Thesis 2.2. Of

course, Kleene was aware of other similar “theses” advanced nearly simul-

taneously and he continued [1952, p.300],

“This thesis is also implicit in the conception of a computing

machine formulated by Turing 1936–7 and Post 1936.”

Next Kleene begins a subtle shift of terminology from “Thesis I” to

“Church’s thesis.” Apparently he did not feel it necessary to include Turing’s

name when he used the term “Church’s thesis.” Kleene wrote [1952, p.317]

began a new section '62 and called it “Church’s thesis” instead of “Thesis I”

as he had been doing. Kleene wrote,

“ '62. Church’s thesis. One of the main objectives of this and

the next chapter is to present the evidence for Church’s thesis

(Thesis I) '60.”

50

12.3 Kleene Dropped “Thesis I” for “Church’s thesis”

As Kleene progressed through [1952, '62, pp. 318–319] he dropped any ref-

erence to “Thesis I” and used only “Church’s thesis” with no mention of

Turing or Post in the thesis. He wrote [1952, p. 318–319],

“ Church’s thesis, by supplying a precise delimitation of all ef-

fectively calculable functions, . . . ”

“While we cannot prove Church’s thesis, since its role is to de-

limit precisely an hitherto vaguely conceived totality, we require

evidence that it cannot conﬂict with the intuitive notion which

it is supposed to complete; . . . ”

“The converse of Church’s thesis, i.e., that every general recur-

sive function ϕ is eﬀectively calculable, we take to be already

conﬁrmed by the intuitive notion (cf. '60).”

12.4 Evidence for the Computability Thesis

In one of the most familiar parts of the book, Kleene summarized the ev-

idence for “Church’s thesis” in [1952, '62, p. 319]. These arguments have

been cited in hundreds of computability books and papers for the last half

century. After some heuristic evidence in (A), Kleene presented perhaps

the most powerful evidence, the “(B) Equivalence of diverse formulations:”

such as the (Herbrand-G¨odel) general recursive functions, λ-deﬁnable func-

tions, Turing computable functions, and systems of Post [1943] and [1946] of

canonical and normal systems. Ironically, to clinch the evidence for Church’s

thesis Kleene began a new part (C) where he appealed to Turing’s work and

wrote,

“(C) Turing’s Concept of a Computing Machine.”

Turing’s computable functions [1936–7] are those which can be

computed by a machine of a kind which is designed, according

to his analysis, to reproduce all sorts of operations which a hu-

man computer could perform, working according to preassigned

instructions. Turing’s notion is thus the result of a direct at-

tempt to formulate mathematically the notion of eﬀective calcu-

lability, while other notions arose diﬀerently and were afterwards

identiﬁed with eﬀective calculability. Turing’s formulation hence

constitutes an independent statement of Church’s thesis.”

51

We see here the amalgamation of diﬀerent meanings into a single term

(just as for “recursive” above). Kleene here was using the term “Church’s

thesis” to include Turing’s Thesis and Turing’s justiﬁcation. Turing’s work

was to establish the connection between eﬀectively calculable functions and

Turing computable functions in Turing’s Thesis 3.1. As an intensional

claim it had nothing to do with the recursive functions of Church’s The-

sis 2.2. It was only the equivalence of Turing computable functions ﬁrst

with λ-deﬁnable functions and hence with recursive functions which links

them extensionally but not intensionally.

Remark 12.1. [The Computability Thesis Emerges]. It is exactly

here that Kleene introduces (without explicit mention) another convention

and term which has lasted until today. Kleene knows that the various for-

mal deﬁnitions all coincide extensionally, and he regards them interchange-

ably, regardless of their intensional or historical import. From now on when

Kleene uses the term “Church’s thesis” he means the following “Computabil-

ity Thesis.” Kleene omits Turing’s name from the thesis even though in part

(C) above he gave Turing credit as the only one who made a “direct attempt

to formulate mathematically the notion of eﬀective calculability.”

Deﬁnition 12.2. [The Computability Thesis]. A function on the pos-

itive integers is eﬀectively calculable if and only if it is recursive or Turing

computable, or deﬁned by any of the other formalisms.

Note that the Computability Thesis is essentially the same as what we

have called the Church-Turing Thesis 3.2. It does not refer to one single man

or to one single formalism. The Computability Thesis, used extensively in

the subject, is usually listed under the name “Church’s Thesis” as in the

title of the new book [Olszewski, 2007] with no mention of Turing. Many

people today use “Church’s Thesis” in this way.

12.5 Who First Demonstrated the Computability Thesis?

Demonstrating something like the Computability Thesis requires two steps.

The Formalism. A formal mathematical deﬁnition for eﬀectively calcula-

ble functions. Church [1936] proposed G¨odel’s general recursive func-

tions [1934]. Turing invented the new formalism of Turing machines

because this was closest to his idea of a mechanical process and be-

cause it lent itself to proving that any eﬀectively calculable function

lay in this class.

52

The Demonstration. The second and perhaps more important step is to

prove that the informal notion of a person calculating a function can be

simulated by the formal model. This step is not a purely mathematical

one but it needs to be as convincing and logical as possible. We refer

to this step as a “demonstration” rather than a formal “proof.”

Although it is not a formal proof, this second step is so crucial that it has

been referred to as a “theorem” by Gandy and others. Gandy [1988, p. 82]

observed, “Turing’s analysis does much more than provide an argument for

‘Turing’s Thesis,’ it proves a theorem.”

3

Furthermore, as Gandy [1988,

pp. 83–84] pointed out, “Turing’s analysis makes no reference whatsoever to

calculating machines. Turing machines appear as a result, a codiﬁcation, of

his analysis of calculations by humans.” Turing’s Thesis [1936, '9] stated

in Deﬁnition 3.1 is that every intuitively computable (eﬀectively calculable)

function is computable by a Turing machine.

In contrast, Church used the Herbrand-G¨odel general recursive functions

as his formal model but even their inventor, G¨odel, was not convinced as

we have seen in '2 and '3. No modern book uses the H-G general recursive

formalism to deﬁne the eﬀectively calculable functions.

A considerably more serious objection is that there was a ﬂaw in Church’s

demonstration that every eﬀectively calculable function is general recursive.

The ﬂaw in Church’s argument [1936, '7] for his thesis was this. Church

began by deﬁning an “eﬀectively calculable” function to be one for which

“there exists an algorithm for the calculation of its values.” Church analyzed

the informal notion of the calculation of a value f(n) = m according to a

step-by-step approach (so called by Gandy [1988, p. 77]) from two points of

view, ﬁrst by an application of an algorithm, and second as the derivation in

some formal system, because as he pointed out, G¨odel had shown that the

steps in his formal system P were primitive recursive. Following Davis [1958,

p. 64] or Shoenﬁeld [1967, pp. 120–121] it is reasonable to suppose that the

calculation of f proceeds by writing expressions on a sheet of paper, and

that the expressions have been given code numbers, c

0

, c

1

, . . . c

n

. Deﬁne

'c

0

, c

1

, . . . c

n

` = p

c

0

0

p

c

1

1

. . . p

cn

n

. We say that the calculation is stepwise

recursive if there is a partial recursive function ψ such that ψ('c

0

, . . . , c

i

`) =

c

i+1

for all i, 0 ≤ i < n.

3

Gandy actually wrote “Church’s thesis” not “Turing’s thesis” as written here, but

surely Gandy meant Turing’s Thesis, i.e., the computability thesis, at least intensionally,

because Turing did not prove anything in [1936] or anywhere else about general recursive

functions.

53

If the basic steps are stepwise recursive, then it follows easily by the

Kleene Normal Form Theorem which Kleene had proved and communicated

to G¨odel before November, 1935 (see Kleene [1987b, p. 57]), that the entire

process is µ-recursive. The fatal weakness in Church’s argument was the

core assumption that the atomic steps were stepwise recursive, something

he did not justify. Gandy [1988, p. 79] and especially Sieg [1994, pp. 80, 87],

in their excellent analyses, brought out this weakness in Church’s argument.

Sieg [p. 80] wrote, “. . . this core does not provide a convincing analysis:

steps taken in a calculus must be of a restricted character and they are

assumed, for example by Church, without argument to be recursive.” Sieg

[p. 78] wrote, “It is precisely here that we encounter the major stumbling

block for Church’s analysis, and that stumbling block was quite clearly seen

by Church,” who wrote that without this assumption it is diﬃcult to see

how the notion of a system of logic can be given any exact meaning at all.

It is exactly this stumbling block which Turing overcame by a totally new

approach.

12.6 The Computability Thesis and the Calculus

Why all the fuss over names? Why not simply use the term “Church’s

Thesis” invented by Kleene [1952], and let it refer to the “Computability

Thesis?” This is in fact what is widely (and incorrectly) done.

If we had to attach a single name to “the calculus” every time we men-

tioned it, whose name should it be? Isaac Newton began working on a form

of the calculus in 1666 but did not publish it until much later. Gottfried

Leibniz began work on the calculus in 1674 and published his account of the

diﬀerential calculus in 1684 and the integral calculus in 1686. Newton did

not publish it until 1687 although most believe he had been working on it

before Leibniz began in 1674. There was a great controversy about priority

up until Leibniz’s death in 1716. The British Royal Society handed down

a verdict in 1715 crediting Isaac Newton with the discovery of the calculus,

and stating that Leibniz was guilty of plagarism (although these charges

were later proved false). Newton was more established than Leibniz and

had vigorous supporters. Newton and his followers campaigned vigorously

for his position. The Wiki encylopedia wrote,

“Despite this ruling of the Royal Society, mathematics through-

out the eighteenth century was typiﬁed by an elaboration of the

diﬀerential and integral calculus in which mathematicians gen-

erally discarded Newton’s ﬂuxional calculus in favor of the new

methods presented by Leibniz.”

54

12.7 Founders of Computability and the Calculus

There is a parallel between the development of the Calculus and the demon-

stration of the Computability Thesis.

1. Church and Turing both worked independently on it and came up with

diﬀerent models.

2. Turing began slightly later than Church. Leibnitz began later than

Newton. Both Leibniz and Turing worked at a distance and indepen-

dently of Newton and Church, respectively, unaware of the work by

another researcher.

3. Turing’s model of Turing machines is overwhelmingly more appealing

and popular that Church’s model of the (Herbrand-G¨odel) general

recursive functions, a model which is rarely presented in any books

and never used for actual calculations in any courses. The general

recursive functions are used only in an historical discussion. (This

refers to the Herbrand-G¨odel general recursive functions. The Kleene

µ-recursive functions have other uses but are were not mentioned in

the original Church’s Thesis 2.2.)

4. For the calculus, despite the Royal Society ruling, “mathematicians

generally discarded Newton’s ﬂuxional calculus in favor of the new

methods presented by Liebniz.”

5. In computer science, the Turing machines (and other calculating ma-

chines like register machines) dominate. The general recursive func-

tions are never seen. Kleene’s µ-recursive functions are sometimes

used in courses and mistakenly called recursive functions but to prove

that an eﬀectively calculable function is µ-recursive requires a tedious

arithmetization as in G¨odel’s incompleteness theorem [1931] and is

virtually never done.

6. Newton was in the position of power with the Royal Society which

not only aﬃrmed his claim but denied the claim by his rival Leibniz.

Church was the senior ﬁgure in computability (after G¨odel). Church’s

claim to the Computability Thesis was aﬃrmed by his former Ph.D.

student Stephen Kleene [1952]. Kleene occasionally mentioned Tur-

ing’s Thesis and repeatedly used the Turing machine model and Tur-

ing’s demonstration of its success to demonstrate the Computability

Thesis. However, Kleene deliberately and overwhelmingly established

55

the phrase “Church’s thesis” to stand for the “Computability Thesis, ”

at a time when nobody called it a thesis and when the ﬁeld was about

to rapidly expand in the 1950’s and 1960’s and would turn to Kleene

for direction as the last representative of the original computability

researchers of the 1930’s and the most prominent and inﬂuential of

all computability theorists in the 1950’s. The Royal Society was not

successful in excluding Leibniz from the calculus, but Kleene was cer-

tainly successful in excluding Turing’s name from the Computability

Thesis.

Virtually all the papers and books including Rogers [1967] and Soare

[1987] and many others followed Kleene’s lead. Unlike the calculus, the

participants, Church, Turing, and their followers, have given credit to the

others. The problem is simply that Kleene has not given Turing credit in

his naming of the Computability Thesis. Kleene could have called it “the

Computability Thesis” analogously with “the calculus.” We never refer to

“the Newton calculus” or the “Leibniz calculus.” Why do we need to give

a person’s name to the Computability Thesis? Kleene never denied credit

to Turing and in many places such as his books [1952] and [1967] he gives

Turing credit for the most intuitive presentation of computability. Kleene

just does not include Turing in the name he chose.

Remark 12.3. Kleene thought and wrote with tokens, words that are given

arbitrary and nonstandard meaning by the author: “recursive” means “com-

putable,” “Church’s Thesis” means the “Computability Thesis.” The arbi-

trary and sometimes misleading use of words has diminished our communi-

cations among ourselves and with other scientiﬁc and scholarly colleagues.

It is ambiguous to use “recursive” with both meanings, inductive and

computable, as we have seen. Second, it is simply wrong to use “Church’s

Thesis” to refer to a proposition ﬁrst demonstrated by Turing and never

successfully demonstrated by Church. It would also be wrong to refer to

“the Newton Calculus” without mentioning Leibniz. Today we refer to “the

calculus” without any founder’s name. Why not call it simply, “the Com-

putability Thesis” and not “Church’s Thesis,” or the “Church-Turing The-

sis?” Which of the three terms is more understandable to an outsider who

has never heard about the subject?

56

13 Turing a-machines versus o-machines?

13.1 Turing, Post, and Kleene on Relative Computability

In '4 we have seen how Turing [1939, '4] brieﬂy introduced an oracle machine

(o-machine) and how and in '5–'7 how Post developed relative computabil-

ity through the inﬂuential Kleene-Post [1954] paper. Kleene [1952, p. 314]

took up the theme of relative computabilty of a function from an oracle

set. Analogous to his version of Thesis I considered above, Kleene deﬁned

Thesis I* to be the corresponding relative computability thesis which we

have called Post-Turing Thesis 6.1. He recommended this thesis but did not

give a separate justiﬁcation. Kleene [1952, p. 314] wrote, “The evidence for

Thesis I will also apply to Thesis I*,” and on [1952, p. 319] he wrote,

“We now summarize the evidence for Church’s Thesis (and The-

sis I*, end '61) under three main headings (A)–(C), and one

other (D) which may be included under (A).”

13.2 Relative Computability Uniﬁes Incomputability

The ﬁeld of computability theory deals mostly with incomputable not com-

putable objects. The objects considered in degrees of unsolvability, in com-

putable model theory, in diﬀerential geometry as in Csima-Soare [2006] or

Soare [2004] all deal with incomputable objects. However, we should not

call the subject “incomputability theory” because the underlying theme is

the notion of relative computability as absolute because of the Oracle Graph

Theorem 4.5, and because it relates and uniﬁes the myriad incomputable

objects.

13.3 The Key Concept of the Subject

The notion of an oracle machine and relative computability is the single

most important in the subject.

1. A Turing a-machine can easily be simulated by a Turing o-machine

and the latter is scarcely more complicated to explain.

2. Most of the objects considered in computability theory and applica-

tions to algebra, model theory, geometry, analysis and other ﬁelds are

incomputable not computable and relative computability uniﬁes them.

57

3. Many if not most computing processes in the real world are online

or interactive processes, better modelled by an o-machine than an

a-machine.

4. A relative computability process Φ

A

e

corresponds to a continuous func-

tional on Cantor space analogous to continuous functions in analysis.

A function on Cantor space given by an a-machine is merely a constant

function.

13.4 When to Introduce Relative Computability

In view of the importance of relative computability, and online computing in

both theoretical results and real world computing it is surprising how many

computability books introduce oracle machines and Turing functionals so

late in the book or not at all. For example, Kleene’s book [1952] was the

ﬁrst real book on computability theory and the principal reference for at

least ﬁfteen years until Rogers [1967] appeared. Kleene introduced relative

computability in Chapter 11 on page 266 by adding the characteristic func-

tion of the oracle set A to the Herbrand-G¨odel general recursive functions.

Rogers [1967] took the Turing machine approach and immediately deﬁned

computability using regular Turing machines (a-machines). Rogers quickly

became the most readable textbook on computability and remains a pop-

ular reference. Rogers introduced relative computability only in Chapter 9

on page 128 using Turing’s original deﬁnition [1939] of an ordinary Turing

machine with the additional capacity to consult an oracle A occasionally

during the computation. In another popular introduction, Computability,

Cutland [1980] introduces relative computability relatively late on page 167.

Boolos and Jeﬀrey in Computability and Logic [1974] do not discuss it at all.

The more recent and very extensive books by Odifreddi, Classical Recursion

Theory Vol. I [1989] and Vol. II [1999] introduce relative computability only

on page 175 by adding the characteristic function of oracle set A to the

Kleene µ-recursive functions. Lerman [1983] deﬁnes relative computability

from an oracle by adding the characteristic function of the oracle to the

Kleene µ-recursive functions. This occurs on page 11 but Lerman is as-

suming that the reader has already mastered a ﬁrst course in computability

using a text such as Rogers [1967]. Cooper’s new book [2004] introduces

oracle Turing machines on page 139.

Kleene’s second and more introductory book, Mathematical Logic [1967,

p. 267] has a brief discussion of reducing one predicate to another and on

degrees of unsolvability. The only genuine introduction to computability I

58

found which introduces relative computability immediately is Martin Davis

Computability and Unsolvability [1958], which deﬁnes it on page 20 of Chap-

ter 1 using oracle Turing machines. In the former book Soare [1987] and new

book [CTA] Turing a-machines come in Chapter I and Turing o-machines

at the beginning of Chapter III after which the book is based on Turing

reducibility. I thought of moving o-machines to Chapter I and doing it all

at once, but my students persuaded me that people need time to absorb the

concepts. Nevertheless, I believe that one should introduce o-machines and

relative computability as soon as possible.

14 Conclusions

Conclusion 14.1. We should use the term “recursive” to mean “deﬁned

inductively” not “calculable” or “computable.” The subject is called “Com-

putability Theory” not “Recursive Function Theory” or “Recursion Theory.”

Church and Kleene [1936] introduced the term “recursive” to mean “com-

putable” primarily for public relations reasons as Kleene [1981] explained

(see '2.3). Over the next few decades Kleene reinforced and promulgated

this convention, but by the 1990’s it had become much more useful for

communication and more accurate scientiﬁcally and historically to remove

the meaning of “computable” from the term “recursive,” particularly since

Turing and G¨odel had both rejected this usage. This change has largely

been accomplished since the papers on computability and recursion in Soare

[1996] and Soare [1999]. See these papers and the discussion in '11. This

emphasis on computability (rather than recursion) and its relation to in-

computability has been developed in many recent books and papers such as

Cooper [2004].

Conclusion 14.2. It is most accurate and informative to refer to the central

thesis of the subject as the “Computability Thesis” not “Church’s Thesis,”

“Turing’s Thesis,” or the “Church-Turing Thesis.”

It is misleading to refer to it as “Church’s Thesis” as many people do

because Church never demonstrated the thesis (at least to the satisfaction of

G¨odel and modern scholars like Gandy [1988] and Sieg [1994]), but Turing

did demonstrate his thesis in a manner convincing to essentially everyone.

Church gave a formal model (the Herbrand-G¨odel general recursive func-

tions) which was not convincing even to its author, while Turing invented

a new model, the Turing a-machine which everyone, including Church and

Kleene, agreed was the most convincing of the thesis.

59

Neither of Turing nor G¨odel thought of this as a thesis. The term

“Church’s thesis” was started arbitrarily by Kleene alone in [1952]. We

use the term “the calculus” without the name of either founder, Newton or

Leibniz, attached. Why not replace the name in computability by a descrip-

tive and informative term like “Computability Thesis?” See the discussion

in '12.

Conclusion 14.3. The subject is primarily about incomputable objects not

computable ones, and has been since the 1930’s. The single most important

concept is that of relative computability to relate incomputable objects.

See '13 and '4–'7.

Conclusion 14.4. For pedagogical reasons with beginning students it is

reasonable to ﬁrst present Turing a-machines and ordinary computability.

However, any introductory computability book should then present as soon

as possible Turing oracle machines (o-machines) and relative computabil-

ity. Parallels should be drawn with oﬄine and online computing in the real

world.

See '13.4.

References

[1] [Ambos-Spies and Fejer, ta] K. Ambos-Spies, and P. Fejer, Degrees of

unsolvability Handbook of Logic, volume 9, to appear.

[2] [Boolos and Jeﬀrey, 1974] G. Boolos and R. Jeﬀrey, Computability and

Logic, Cambridge Univ. Press, Cambridge, Engl., 1974.

[3] [Church, 1935] A. Church, An unsolvable problem of elementary num-

ber theory, Preliminary Report (abstract), Bull. Amer. Math. Soc. 41

(1935), 332-333.

[4] [Church, 1936] A. Church, An unsolvable problem of elementary num-

ber theory, American J. of Math., 58 (1936), 345-363.

[5] [Church, 1936b] A. Church, A note on the Entscheidungsproblem, J.

Symbolic Logic, 1 (1936), 40–41. Correction 101–102.

[6] [Church, 1937] A. Church, Review of Turing 1936, J. Symbolic Logic

2(1) (1937), 42–43.

60

[7] [Church, 1937b] A. Church, Review of Post 1936, J. Symbolic Logic

2(1) (1937), 43.

[8] [Church, 1938] A. Church, The constructive second number class,

Bull. A.M.S. 44 (1938), 224–232.

[9] [Church and Kleene, 1936] A. Church and S. C. Kleene, Formal deﬁ-

nitions in the theory of ordinal numbers, Fund. Math. 28 (1936) 11–21.

[10] [Cooper, 2004] S.B. Cooper, Computablility Theory, Chapman &

Hall/CRC Mathematics, London, New York, 2004.

[11] [Cooper, 2004b] S.B. Cooper, The incomparable Alan Turing, Lec-

ture at Manchester University, 5 June, 2004, published electronically,

http://www.bcs.org/server.php?show=ConWebDoc.17130

[12] [Cooper-Lowe-Sorbi, 2007] S.B. Cooper, B. L¨owe, A. Sorbi, (eds.),

Computation and Logic in the Real World, Proceedings of the Third

Conference on Computability in Europe, CiE 2007, Siena, Italy, June

18–23, 2007, Lecture Notes in Computer Science, No. 4497, S.B.

Cooper, B. L¨owe, Andrea Sorbi (Eds.), (Springer-Verlag, Berlin, Hei-

delberg, 2007.

[13] [Cooper-Lowe-Sorbi, 2008] S.B. Cooper, B. L¨owe, A. Sorbi, (eds.),

New computational paradigms: changing conceptions of what is com-

putable, Springer-Verlag, 2008.

[14] [Cooper-Odifreddi, 2003] S.B. Cooper and P. Odifreddi, Incom-

putability in nature. In: S.B. Cooper and S.S. Goncharov (Eds.)

Computability and Models, Perspectives East and West, Kluwer Aca-

demic/Plenum, Dordrecht, (2003) 137–160.

B. L¨owe, A. Sorbi, (eds.), New computational paradigms: changing

conceptions of what is computable, Springer-Verlag, 2008.

[15] [Copeland-Posy-Shagrir, ta] Jack Copeland, Carl Posy, and Oron

Shagrir, Computability: G¨odel, Church, Turing, and Beyond, MIT

Press, to appear.

[16] [Csima-Soare, 2006] B. F. Csima and R.I. Soare, Computability Re-

sults Used in Diﬀerential Geometry, J. Symbolic Logic, vol. 71 (2006),

pp. 1394–1410.

61

[17] [Cutland, 1980] Nigel Cutland, Computability: An introduction to

recursive function theory, Cambridge Univ. Press, Cambridge, Engl.,

1980, reprinted 1983.

[18] [Davis, 1958] M. Davis, Computability and Unsolvability, Mc-Graw-

Hill, New York, 1958; reprinted in 1982 by Dover Publications.

[19] [Davis, 1965] M. Davis, (ed.), The Undecidable. Basic Papers on Un-

decidable Propositions, Unsolvable Problems, and Computable Func-

tions, Raven Press, Hewlett, New York, 1965.

[20] [Davis, 1982] M. Davis, Why G¨odel did not have Church’s Thesis,

Information and Control 54 (1982), 3–24.

[21] [Davis, 1988] M. Davis, Mathematical logic and the origin of modern

computers, In: Herken, 1988, 149–174.

[22] [Davis, 2000] M. Davis, The universal computer: The road from Leib-

niz to Turing, W.W. Norton & Co., New York, London, 2000. (The

same book was also published by Norton in 2000 under the title “En-

gine of Logic.”)

[23] [Davis, 2004] M. Davis, The myth of hypercomputation. In:

Teutscher, C. (Ed), Alan Turing: Life and Legacy of a Great Thinker,

Springer-Verlag, Berlin, Heidelberg, New York, (2004) 195–211.

[24] [Dawson, 1997] J. W. Dawson, Logical dilemmas: The life and work of

Kurt G¨odel, A.K. Peters Press, Cambridge, 1997.

[25] [Epstein and Carnielli, 1989] Epstein and Carnielli, Computability,

Computable Functions, Logic, and the Foundations of Mathematics,

1989, Second Printing Wadsworth Thomson Learning, 2000.

[26] [Friedberg, 1957] R. M. Friedberg, Two recursively enumerable sets

of incomparable degrees of unsolvability, Proc. Natl. Acad. Sci. USA

43 (1957), 236–238.

[27] [Friedberg-Rogers, 1959] R. M. Friedberg and H. Rogers, Jr. Re-

ducibility and completeness for sets of integers, Z. Math. Logik Grund-

lag. Math. 5 (1959), 117–125.

[28] [Gandy, 1980] R. Gandy, Church’s thesis and principles for mecha-

nisms, In: The Kleene Symposium, North-Holland, (1980), 123–148.

62

[29] [Gandy, 1988] R. Gandy, The conﬂuence of ideas in 1936, In: Herken,

55–111.

[30] [G¨odel, 1931] K. G¨odel,

¨

Uber formal unentscheidbare s¨atze der Prin-

cipia Mathematica und verwandter systeme. I, Monatsch. Math. Phys.

38 (1931) 173-178. (English trans. in Davis [1965, 4–38], and in van

Heijenoort, 1967, 592–616.)

[31] [G¨odel, 1934] K. G¨odel, On undecidable propositions of formal math-

ematical systems, Notes by S. C. Kleene and J. B. Rosser on lectures

at the Institute for Advanced Study, Princeton, New Jersey, 1934, 30

pp. (Reprinted in Davis [1965, p. 39–74].)

[32] [G¨odel, 193?] K. G¨odel, Undecidable diophantine propositions, In:

G¨odel [1995, 156–175].

[33] [G¨odel, 1946] K. G¨odel, Remarks before the Princeton bicentennial

conference of problems in mathematics, 1946. Reprinted in: Davis

[1965, p. 84–88].

[34] [G¨odel, 1951] K. G¨odel, Some basic theorems on the foundations of

mathematics and their implications, In: G¨odel [1995, 304–323]. (This

was the Gibbs Lecture delivered by G¨odel on December 26, 1951 to the

Amer. Math. Soc.)

[35] [G¨odel, 1964] K. G¨odel, Postscriptum to G¨odel [1931], written in 1946,

printed in Davis [1965, 71–73].

[36] [G¨odel, 1972] K. G¨odel, Some remarks on the undecidability results,

(written in 1972); In: G¨odel [1990, p. 305–306].

[37] [G¨odel, 1986] K. G¨odel, Collected works Volume I: Publications 1929–

1936, S. Feferman et. al., editors, Oxford Univ. Press, Oxford, 1986.

[38] [G¨odel, 1990] K. G¨odel, Collected works Volume II: Publications 1938–

1974, S. Feferman et. al., editors, Oxford Univ. Press, Oxford, 1990.

[39] [G¨odel, 1995] K. G¨odel, Collected works Volume III: Unpublished es-

says and lectures, S. Feferman et. al., editors, Oxford Univ. Press, Ox-

ford, 1995.

[40] [Goldin-Smolka-Wegner] D. Goldin, S. Smolka, P. Wegner, Interac-

tive Computation: The new Paradigm, Springer-Verlag, 2006.

63

[41] [Herken, 1988] R. Herken (ed.), The Universal Turing Machine: A

Half-Century Survey, Oxford Univ. Press, 1988.

[42] [Hilbert, 1899] D. Hilbert, Grundlagen der Geometrie, 7th ed.,

Tuebner-Verlag, Leipzig, Berlin, 1930.

[43] [Hilbert, 1904] D. Hilbert,

¨

Uber die Grundlagen der Logik

und der Arithmetik, In:Verhandlungen des Dritten Internationalen

Mathematiker-Kongresses in Heidelberg vom 8. bis 13. August 1904,

pp. 174–185, Teubner, Leipzig, 1905. Reprinted in van Heijenoort [1967,

129–138].

[44] [Hilbert, 1926] D. Hilbert,

¨

Uber das Unendliche, Mathematische An-

nalen 95 (1926), 161–190. (English trans. in van Heijenoort [1967, 367–

392].)

[45] [Hilbert, 1927] D. Hilbert, Die Grundlagen der Mathematik, Abhand-

lungen aus dem mathematischen Seminar der Hamburgischen Univer-

sit¨at 6 (1928), 65–85. Reprinted in van Heijenoort [1967, 464–479].

[46] [Hilbert and Ackermann, 1928] D. Hilbert and W. Ackermann,

Grundz¨ uge der theoretischen Logik, Springer, Berlin, 1928 (English

translation of 1938 edition, Chelsea, New York, 1950).

[47] [Hilbert and Bernays, 1934] D. Hilbert and P. Bernays, Grundlagen

der Mathematik I (1934), II (1939), Second ed., I (1968), II (1970),

Springer, Berlin.

[48] [Hodges, 1983] A. Hodges, Alan Turing: The Enigma, Burnett Books

and Hutchinson, London, and Simon and Schuster, New York, 1983.

[49] [Hodges, 2004] A. Hodges, Alan Turing: the logical and physical basis

of computing, Lecture at Manchester University, 5 June, 2004, pub-

lished electronically, http://www.bcs.org/ewics.

[50] [Kleene, 1936] S. C. Kleene, General recursive functions of natural

numbers, Math. Ann. 112 (1936), 727–742.

[51] [Kleene, 1936b] S. C. Kleene, λ-deﬁnability and recursiveness, Duke

Math. J. 2 (1936), 340–353.

[52] [Kleene, 1936c] S. C. Kleene, A note on recursive functions, Bull.

A.M.S. 42 (1936), 544–546.

64

[53] [Kleene, 1938] S. C. Kleene, On notation for ordinal numbers, J. Sym-

bolic Logic, 3 (1938), 150–155.

[54] [Kleene, 1943] S. C. Kleene, Recursive predicates and quantiﬁers,

Trans. A.M.S. 53 (1943), 41–73.

[55] [Kleene, 1944] S. C. Kleene, On the forms of the predicates in the

theory of constructive ordinals, Amer. J. Math. 66 (1944), 41–58.

[56] [Kleene, 1952] S. C. Kleene, Introduction to Metamathematics, Van

Nostrand, New York (1952). Ninth reprint 1988, Walters-Noordhoﬀ

Publishing Co., Groning¨en and North-Holland, Amsterdam.

[57] [Kleene, 1955] S. C. Kleene, Arithmetical predicates and function

quantiﬁers, Trans. A.M.S. 79 (1955), 312–340.

[58] [Kleene, 1955b] S. C. Kleene, On the forms of the predicates in the the-

ory of constructive ordinals (second paper), Amer J. Math. 77 (1955),

405–428.

[59] [Kleene, 1955c] S. C. Kleene, Hierarchies of number-theoretical pred-

icates, Bull. A.M.S. 61 (1955), 193–213.

[60] [Kleene, 1959] S. C. Kleene, Recursive functionals and quantiﬁers of

ﬁnite type I, Trans. A.M.S. 91 (1959), 1–52.

[61] [Kleene, 1962] S. C. Kleene, Turing-machine computable functionals

of ﬁnite types I, Logic, methodology, and philosophy of science: Pro-

ceedings of the 1960 international congress, Stanford University Press,

1962, 38–45.

[62] [Kleene, 1962b] S. C. Kleene, Turing-machine computable functionals

of ﬁnite types II, Proc. of the London Math. Soc. 12 (1962), no. 3,

245–258.

[63] [Kleene, 1963] S. C. Kleene, Recursive functionals and quantiﬁers of

ﬁnite type II, Trans. A.M.S. 108 (1963), 106–142.

[64] [Kleene, 1967] S. C. Kleene, Mathematical Logic, John Wiley and

Sons, Inc., New York, London, Sydney, 1967.

[65] [Kleene, 1981] S. C. Kleene, Origins of recursive function theory, An-

nals of the History of Computing, 3 (1981), 52–67.

65

[66] [Kleene, 1981b] S. C. Kleene, The theory of recursive functions, ap-

proaching its centennial, Bull. A.M.S. (n.s.) 5, (1981), 43–61.

[67] [Kleene, 1981c] S. C. Kleene, Algorithms in various contexts, Proc.

Sympos. Algorithms in Modern Mathematics and Computer Science

(dedicated to Al-Khowarizimi) (Urgench, Khorezm Region, Uzbek,

SSSR, 1979), Springer-Verlag, Berlin, Heidelberg and New York, 1981.

[68] [Kleene, 1987] S. C. Kleene, Reﬂections on Church’s Thesis, Notre

Dame Journal of Formal Logic, 28 (1987), 490–498.

[69] [Kleene, 1987b] S. C. Kleene, G¨odel’s impression on students of logic

in the 1930’s, In: P. Weingartner and L. Schmetterer (eds.), G¨odel

Remembered, Bibliopolis, Naples, 1987, 49–64.

[70] [Kleene, 1988] S. C. Kleene, Turing’s analysis of computability, and

major applications of it, In: Herken 1988, 17–54.

[71] [Kleene-Post, 1954] S. C. Kleene and E. L. Post, The upper semi-

lattice of degrees of recursive unsolvability, Ann. of Math. 59 (1954),

379–407.

[72] [Lerman, 1983] M. Lerman, Degrees of Unsolvability: Local and Global

Theory, Springer-Verlag, Heidelberg New York Tokyo, 1983.

[73] [Muchnik, 1956] A. A. Muchnik, On the unsolvability of the problem of

reducibility in the theory of algorithms, Doklady Akademii Nauk SSR

108 (1956), 194–197, (Russian).

[74] [Odifreddi, 1989] P. Odifreddi, Classical Recursion Theory, North-

Holland, Amsterdam, Volume I 1989, Volume II 1999.

[75] [Olszewski, 2007] A. Olszewski and J. Wolenski (Eds.), Church’s The-

sis After 70 years, Ontos Verlag, 2007. (551 pages, hardbound.) (ISBN-

13: 978-3938793091.)

[76] [Peter, 1934] R. P´eter,

¨

Uber den Zussammenhang der verschiedenen

Begriﬀe der rekursiven Funktion, Mathematische Annalen 110 (1934),

612–632.

[77] [Peter, 1951] R. P´eter, Rekursive Funktionen, Akad´emaiai Kiad´o

(Akademische Verlag), Budapest, 1951, 206 pp. Recursive Functions,

Third revised edition, Academic Press, New York, 1967, 300 pp.

66

[78] [Post, 1936] E. L. Post, Finite combinatory processes–formulation, J.

Symbolic Logic 1 (1936) 103–105. Reprinted in Davis [1965, 288–291].

[79] [Post, 1941] E. L. Post, Absolutely unsolvable problems and relatively

undecidable propositions: Account of an anticipation. (Submitted for

publication in 1941.) Printed in Davis [1965, 340–433].

[80] [Post, 1943] E. L. Post, Formal reductions of the general combinatorial

decision problem, Amer. J. Math. 65 (1943), 197–215.

[81] [Post, 1944] E. L. Post, Recursively enumerable sets of positive inte-

gers and their decision problems, Bull. Amer. Math. Soc. 50 (1944),

284–316. (Reprinted in Davis [1965, 304–337].)

[82] [Post, 1946] E. L. Post, Note on a conjecture of Skolem, J. Symbolic

Logic 11 (1946), 73–74.

[83] [Post, 1947] E. L. Post, Recursive unsolvability of a problem of Thue,

J. Symbolic Logic 12 (1947), 1–11. (Reprinted in Davis [1965, 292–303].)

[84] [Post, 1948] E. L. Post, Degrees of recursive unsolvability: preliminary

report (abstract), Bull. Amer. Math. Soc. 54 (1948), 641–642.

[85] [Putnam, 1956] H. Putnam, Trial and error predicates and the solu-

tion to a problem of Mostowski, J. Symbolic Logic 30, 49–57.

[86] [Rogers, 1967] H. Rogers, Jr., Theory of Recursive Functions and

Eﬀective Computability, McGraw-Hill, New York, 1967.

[87] [Sacks, 1990] G. E. Sacks, Higher Recursion Theory, Springer-Verlag,

Heidelberg New York, 1990.

[88] [Shoenﬁeld, 1967] J. R. Shoenﬁeld, Mathematical Logic, Addison-

Wesley, Reading, Mass. (1967), 344 pp.

[89] [Shoenﬁeld, 1971] J. R. Shoenﬁeld, Degrees of Unsolvability, North-

Holland, Amsterdam, London, New York, 1971.

[90] [Shoenﬁeld, 1991] J. R. Shoenﬁeld, Recursion Theory, Lecture Notes

in Logic, Springer-Verlag, Heidelberg New York, 1991.

[91] [Shoenﬁeld, 1995] J. R. Shoenﬁeld, The mathematical work of

S. C. Kleene, Bull. A.S.L 1 (1995), 8–43.

67

[92] [Sieg, 1994] W. Sieg, Mechanical procedures and mathematical ex-

perience, In: A. George (ed.), Mathematics and Mind, Oxford Univ.

Press, 1994.

[93] [Soare, 1987] R. I. Soare, Recursively Enumerable Sets and Degrees:

A Study of Computable Functions and Computably Generated Sets,

Springer-Verlag, Heidelberg, 1987.

[94] [Soare, 1996] R. I. Soare, Computability and recursion, Bulletin of

Symbolic Logic 2 (1996), 284–321.

[95] [Soare, 1999] R. I. Soare, The history and concept of computability,

In: Handbook of Computability Theory, ed. E. Griﬀor, North-Holland,

Amsterdam, 1999, 3–36.

[96] [Soare, 2000] R. I. Soare, Extensions, Automorphisms, and Deﬁn-

ability, in: P. Cholak, S. Lempp, M. Lerman, and R. Shore, (eds.)

Computability Theory and its Applications: Current Trends and

Open Problems, American Mathematical Society, Contemporary Math.

#257, American Mathematical Society, Providence, RI, 2000. pps. 279–

307.

[97] [Soare, 2004] R. I. Soare, Computability theory and diﬀerential ge-

ometry, Bull. Symb. Logic, Vol. 10 (2004), 457–486.

[98] [Soare, 2007] R. I. Soare, Computability and Incomputability, Com-

putation and Logic in the Real World, in: Proceedings of the Third Con-

ference on Computability in Europe, CiE 2007, Siena, Italy, June 18–23,

2007, Lecture Notes in Computer Science, No. 4497, S.B. Cooper, B.

L¨owe, Andrea Sorbi (Eds.), (Springer-Verlag, Berlin, Heidelberg, 2007.

[99] [Soare, CTA] R. I. Soare, Computability Theory and Applications,

Springer-Verlag, Heidelberg, to appear.

[100] [Soare, ta] R. I. Soare, Turing-Post Oracle Computability and Re-

aligning Computability Theory, in: Jack Copeland, Carl Posy, and

Oron Shagrir (eds.), Computability: G¨odel, Church, Turing, and Be-

yond, MIT Press, to appear.

[101] [Teuscher, 2004] C. Teuscher (ed.), Alan Turing: Life and legacy of

a great thinker, Springer-Verlag, 2004.

68

[102] [Turing, 1936] A. M. Turing, On computable numbers, with an ap-

plication to the Entscheidungsproblem, Proc. London Math. Soc. ser. 2

42 (Parts 3 and 4) (1936) 230–265; [Turing, 1937a] A correction,

ibid. 43 (1937), 544–546.

[103] [Turing, 1937b] A. M. Turing, Computability and λ-deﬁnability,

J. Symbolic Logic, 2 (1937), 153–163.

[104] [Turing, 1939] A. M. Turing, Systems of logic based on ordinals,

Proc. London Math. Soc. 45 Part 3 (1939), 161–228; reprinted in Davis

[1965, 154–222].

[105] [Turing, 1948] A. M. Turing, Intelligent machinery, In: Machine In-

telligence 5, 3–23. (Written in September, 1947 and submitted to the

National Physical Laboratory in 1948.)

[106] [Turing, 1949] A. M. Turing, Text of a lecture by Turing on June 24,

1949, In: F. L. Morris and C. B. Jones, “An early program proof by

Alan Turing,” Annals of the History of Computing 6 (1984), 139-143.

[107] [Turing, 1950] A. M. Turing, Computing machinery and intelligence,

Mind 59 (1950) 433–460.

[108] [Turing, 1950b] A. M. Turing, The word problem in semi-groups with

cancellation, Ann. of Math. 52 (1950), 491–505.

[109] [Turing, 1954] A. M. Turing, Solvable and unsolvable problems, Sci-

ence News 31 (1954), 7–23.

[110] [Turing, 1986] A. M. Turing, Lecture to the London Mathematical

Society on 20 February 1947, In: B. E. Carpenter and R. W. Doran,

eds., A. M. Turing’s ACE Report of 1946 and Other Papers, Cambridge

Univ. Press, 1986, 106–124.

[111] [Wang, 1974] H. Wang, From Mathematics to Philosophy, Routledge

& Kegan Paul, London, 1974.

[112] [Wang, 1981] H. Wang, Some facts about Kurt G¨odel, J. Symbolic

Logic 46 (1981) 653–659.

[113] [Wang, 1987] H. Wang, Reﬂections on Kurt G¨odel, MIT Press, Cam-

bridge, MA.

[114] [Wang, 1993] H. Wang, On physicalism and algorithmism: can ma-

chines think?, Philosophia Mathematica, 3rd series 1 (1993), 97–138.

69

Department of Mathematics

University of Chicago

Chicago, Illinois 60637-1546

soare@uchicago.edu

URL: http://www.people.cs.uchicago.edu/∼soare/

70

3.3 3.4 3.5

3.2.3 The Flaw in Church’s Thesis . . . . . . . . 3.2.4 G¨del on Church’s Thesis . . . . . . . . . . o 3.2.5 G¨del’s Letter to Kreisel [1968] . . . . . . . o 3.2.6 Gibbs Lecture [1951] . . . . . . . . . . . . . 3.2.7 Godel’s Postscriptum 3 June, 1964 to G¨del o Hao Wang Reports on G¨del . . . . . . . . . . . . o Kleene’s Remarks About Turing . . . . . . . . . . Church’s Remarks About Turing . . . . . . . . . .

. . . . . . . . . . . . . . . . [1934] . . . . . . . . . . . .

. . . . . . . .

. . . . . . . .

16 17 17 17 18 19 19 20 20 21 21 23 23 23 24 25 25 26 26 28 28

4 Oracle Machines and Relative Computability 4.1 Turing’s Oracle Machines . . . . . . . . . . . . . . . . . 4.2 Modern Deﬁnitions of Oracle Machines . . . . . . . . . . 4.2.1 The Graph of a Partial Computable Function . . 4.2.2 The Graph of an Oracle Computable Functional 4.3 The Oracle Graph Theorem . . . . . . . . . . . . . . . . 4.4 Equivalent Deﬁnitions of Relative Computability . . . . 4.4.1 Notation for Functions and Functionals . . . . . 5 Emil Post Expands Turing’s Ideas 5.1 Post’s Work in the 1930’s . . . . . . . . . . . . . 5.2 Post Steps Into Turing’s Place During 1940–1954 5.3 Post’s Problem on Incomplete C.E. Sets . . . . . 5.4 Post Began With Strong Reducibilities . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

6 Post Highlights Turing Computability 29 6.1 Post Articulates Turing Reducibility . . . . . . . . . . . . . . 29 6.2 The Post-Turing Thesis . . . . . . . . . . . . . . . . . . . . . 30 7 Continuous and Total Functionals 7.1 Representations of Open and Closed Sets . . . . 7.2 Notation for Trees . . . . . . . . . . . . . . . . . 7.3 Dense Open Subsets of Cantor Space . . . . . . . 7.4 Eﬀectively Open and Closed Sets . . . . . . . . . 7.5 Continuous Functions on Cantor Space . . . . . . 7.6 Eﬀectively Continuous Functionals . . . . . . . . 7.7 Continuous Functions are Relatively Computable 31 31 31 33 33 34 35 36

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

8 Bounded Reducibilities 36 8.1 A Matrix Mx for Bounded Reducibilities . . . . . . . . . . . . 36 8.2 Bounded Turing Reducibility . . . . . . . . . . . . . . . . . . 37

2

8.3 8.4

Truth-Table Reductions . . . . . . . . . . . . . . . . . . . . . Diﬀerence of c.e. sets, n-c.e., and ω-c.e. sets . . . . . . . . . .

37 39 40 42 42 43 44 44 44 45 45 46 46 47 48 48 50 50 50 51 51 52 54 55 57 57 57 57 58 59

9 Online Computing 9.1 Turing Machines and Online Processes . 9.2 Trial and Error Computing . . . . . . . 9.3 The Limit Lemma . . . . . . . . . . . . 9.4 Two Models for Computing With Error 9.4.1 The Limit Computable Model . . 9.4.2 The Online Model . . . . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

10 Three Displacements in Computability Theory 11 “Computable” versus “Recursive” 11.1 Church Defends Church’s Thesis with “Recursive” . . . 11.2 Church and Kleene Deﬁne “Recursive” as “Computable” 11.3 G¨del Rejects “Recursive Function Theory” . . . . . . . o 11.4 The Ambiguity in the Term “Recursive” . . . . . . . . . 11.5 Changing “Recursive” Back to “Inductive” . . . . . . . 12 Renaming it the “Computability Thesis?” 12.1 Kleene Called it “Thesis I” in [1943] . . . . . . . . . 12.2 Kleene Named it “Church’s thesis” in [1952] . . . . . 12.3 Kleene Dropped “Thesis I” for “Church’s thesis” . . 12.4 Evidence for the Computability Thesis . . . . . . . . 12.5 Who First Demonstrated the Computability Thesis? 12.6 The Computability Thesis and the Calculus . . . . . 12.7 Founders of Computability and the Calculus . . . . . 13 Turing a-machines versus o-machines? 13.1 Turing, Post, and Kleene on Relative Computability 13.2 Relative Computability Uniﬁes Incomputability . . . 13.3 The Key Concept of the Subject . . . . . . . . . . . 13.4 When to Introduce Relative Computability . . . . . 14 Conclusions

Abstract We begin with the history of the discovery of computability in the 1930’s, the roles of G¨del, Church, and Turing, and the formalisms of o recursive functions and Turing automatic machines (a-machines). To

. . . . .

. . . . .

. . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

3

A number of topics arose from Turing functionals including continuous functionals on Cantor space and online computations. relative computability. §4] of an oracle machine (o-machine) and Post’s development of it in [1944. that which “cannot be computed or reckoned. At the end in §10 – §13 we consider three displacements in computability theory. incalculable.1 Terminology: Incompleteness and Incomputability The two principal accomplishments in computability and logic in the 1930’s were the discovery of incompleteness by G¨del [1931] and of incomputabilo ity independently by Church [1936] and Turing in [1936]. Therefore. but we often use the term “incomputability” for the general concept because it is linguistically and mathematically parallel to “incomplete.whom did G¨del credit the deﬁnition of a computable function? We o present Turing’s notion [1939. [1948]. We also challenge a number of traditional views of these subjects as often presented in the literature since the 1930’s. 1. 1 Introduction In this paper we consider the development of Turing oracle machines and relative computability and its relation to continuity in analysis and to online computing in the real world. §11]. and the historical reasons they occurred. Websters dictionary deﬁnes it as “greater than can be computed or enumerated. more so than Turing a-machines and standard computable functions since they are special cases of the former and are presented ﬁrst only for pedagogical clarity to beginning students.” The term “incomputable” appeared in English as early as 1606 with the meaning. we argue that Turing o-machines. Several brief conclusions are drawn in §14. 4 . and ﬁnally KleenePost [1954] into its present form. and online computing are the most important concepts in the subject. Almost all the results in theoretical computability use relative reducibility and o-machines rather than a-machines and most computing processes in the real world are potentially online or interactive. very great.” Neither dictionary lists an entry for “noncomputable” although it has often been used in the subject to mean “not computable” for a speciﬁc function or set analogously as “nonmeasurable” is used in analysis. We use the term “noncomputable” or “incomputable” for individual instances.” according to the Oxford English Dictionary.

the researchers spent the remainder of the 1930’s investigating more of the new world of incomputability they had created by diagonal arguments. For example. In the 1930’s for the ﬁrst time the goal was the refutation of Hilbert’ two programs. but not until after 1940. 1. These processes are sometimes called online or interactive or database processes depending on the way they are set up.3 Computing Relative to an Oracle or Database In 1936 Turing’s a-machines and Church’s use of G¨del’s recursive functions o solved an immediate problem by producing a deﬁnition of a computable function.D. many modern computer processes are online processes in that they consult an external data base of other resource during the computation process.2 The Goal of Incomputability not Computability For several thousand years the study of algorithms had led to new theoretical algorithms and sometimes new devices for computability and calculation. λ-deﬁnable functions. would all have deep applications to the design and programming of computing machines.1. just as Georg Cantor had spent the last quarter of the nineteenth century exploring the world of uncountable sets which he had created by the diagonal method. and the Entscheidungsproblem (decision problem). §4] included 5 . wanted to give formal deﬁnitions of a computable function so that they could diagonalize over all computable functions and produce an incomputable (unsolvable) problem. a ﬁnite consistency proof for Peano arithmetic. two main discoverers of computability. Alonzo Church and Alan Turing. Turing a-machines. However. For the latter. with which one could diagonalize and produce undecidable problems in mathematics. and recursive functions. In §1 and §2 we consider this historical development from 1931 to 1939 and we introduce quotes from G¨del to show convincingly that he believed “the correct deﬁnition of meo chanical computability was established beyond any doubt by Turing” and only by Turing. The Turing a-machine is a good model for oﬄine computing such as a calculator or batch processor where the process is initially given a procedure and an input and continues with no further outside information. Turing spent 1936–1938 at Princeton writing a Ph. a laptop computer might consult the World Wide Web via an ethernet or wireless connection while it is computing. Meanwhile. The speciﬁc models of computable functions produced by 1936. thesis under Church on ordinal logics. A tiny and obscure part of his paper [1939.

In contrast. It is surprising that so much attention has been paid to the Church-Turing Thesis 3. These concepts were not well understood when Post began.a description of an oracle machine (o-machine) roughly a Turing a-machine which could interrogate an “oracle” (external database) during the computation. but in Post [1943].4 Continuous Functions and Calculus In §7 we show that any Turing functional on Cantor space 2ω is an eﬀectively continuous partial map. Emil Post [1944.” In §4–§6 we study Turing’s oracle machines (o-machines) and Post’s development of them into relative computability. 1. Turing a-machines viewed as functionals on Cantor space produce only a constant functional. particularly in view of the importance of relative computability (Turing o-machines) in comparison with plain computability (Turing a-machines) in both theoretical and practical areas. This remarkable role by Post has been underemphasized in the literature but is discussed here in §5 and §6. for any continuous functional there is a Turing functional relative to some real parameter which deﬁnes it. Oracle computing processes are those most often used in theoretical research and also in real world computing where a laptop computer may communicate with a database like the World Wide Web. The one page description was very sketchy and Turing never developed it further. Often the local computer is called the “client” and the remote device the “server. thereby linking computability with analysis. This makes Turing functionals the analogue in computability of the continuous functions in analysis as we explain.1 on relative reducibility. The theory of relative computability developed by Turing and Post and the o-machines provide a precise mathematical framework for database or online computing just Turing a-machines provide one for oﬄine computing processes such as batch processing.2. These are summarized in the PostTuring Thesis of §6. there appears to be more emphasis in calculus books on continuous functionals than there is in introductory computability books on computably continuous functionals and relative computability.2 over the last seventy years and so little to the Post-Turing Thesis 6. §11] considerably expanded and developed relative computability and Turing functionals. 6 . Conversely. [1948] and Kleene-Post [1954] they emerged into their modern state. [1944]. Surprisingly.

46] wrote. “This was because it seemed clear to Hilbert that with the solution of this problem. and the importance of logic in mathematics. the city of his birth. 108] wrote.) At a mathematical conference preceding Hilbert’s address. Wir m¨ssen wissen. the Entscheidungsproblem. u Wir werden wissen.1 G¨del’s Incompleteness Theorem o Hilbert retired in 1930 and was asked to give a special address in the fall of 1930 in K¨nigsberg. to settle all mathematical questions in a purely mechanical manner.) (We will know.” The decision probo o lem for ﬁrst order logic emerged in the early 1920’s in lectures by Hilbert and was stated in Hilbert and Ackermann [1928].2 Origins of Computability and Incomputability Mathematicians have studied algorithms and computation since ancient times. p. the importance of mathematics in science. It was to give a decision procedure (Entscheidungsverfahren) “that allows one to decide the validity of the sentence. and Hilbert [1918]. but the modern study of computability and incomputability began around 1900. and that in turn. obscure young man. p. David Hilbert was deeply interested in the foundations of mathematics. L¨wenheim [1915].” Von Neumann (1927) doubted that such a procedure existed but had no idea how to prove it. a quiet.. Kleene [1987b. “The Entscheidungsproblem for various formal systems had been posed by Schr¨der [1895]. Hilbert [1904] proposed proving the consistency of arithmetic by what emerged [1928] as his ﬁnitist program.D. He proposed using the ﬁniteness of mathematical proofs in order to establish that contradictions could not be derived. Hilbert [1899] gave an axiomatization of geometry and showed [1900] that the question of the consistency of geometry reduced to that for the real-number system. only a year a beyond his Ph. Davis [1965. He asserted that there are no unsolvable problems and stressed. 2. This tended to reduce proofs to manipulation of ﬁnite strings of symbols devoid of intuitive meaning which stimulated the development of mechanical processes to accomplish this. it should be possible. Hilbert spoke on natural science o and logic. at least in principle. Hilbert’s second major program concerned the Entscheidungsproblem (decision problem).” Hilbert characterized this as the fundamental problem of mathematical logic. Kurt G¨del. announced a result o 7 . to arithmetic by results of Dedekind (at least in a second order system). (We must know.

As the importance of G¨del’s Incomo pleteness Theorem began to sink in. and researchers began concentrating on the Entscheidungsproblem. Crucial elements in computability theory. 8 . Of course. and he was not yet thirty-ﬁve. Kleene (b. 1912). and Stephen C. all depend upon giving code numbers to computations and elements within a computation. These young men were about to leave Hilbert’s ideas behind and open the path of computability for the next two thirds of the twentieth century. “This statement is false. that o no such formal system T could prove its own consistency. diagonal arguments had been known since Cantor’s work. G¨del o had successfully challenged the Hilbert’s ﬁrst proposal.” to prove roughly that for any eﬀectively axiomatized. Turing (b. Hilbert had retired and no longer had much inﬂuence on the ﬁeld. Only Emil Post (b. In the next few weeks von Neuman realized that by arithmetizing the proof of G¨del’s ﬁrst theorem. but it also had a profound effect on refuting Hilbert’s second theme of the Entscheidungsproblem. Both refutations used diagonal arguments. Kurt G¨del (b. 1897) was over thirty. These ideas spring from G¨del’s [1931] incomo o pleteness proof. the Kleene µ-recursive functions. which would solve the theoretical problems and would show the way toward practical computing devices. 1906). but G¨del showed how to arithmetize the o syntactic elements of a formal system and diagonalize within that system. Alonzo Church (born 1903). He formalized the liar paradox. consistent extension T of number theory (Peano arithmetic) there is a sentence σ which asserts its own unprovability in T . 1909) were o all under thirty.which would forever change the foundations of mathematics. one could prove an even better one. A few weeks later he brought his proof to G¨del who thanked him and informed him o politely that he had already submitted the Second Incompleteness Theorem for publication. He was at the conference representing o Hilbert’s proof theory program. 2.2 Alonzo Church By 1931 computability was a young man’s game. John von Neumann in the audience immediately understood the importance of G¨del’s Incompleteness Theorem. or the self reference in the Kleene’s Recursion Theorem. and in calling algorithms by their code numbers (G¨del numbers). perhaps the most inﬂuential of all on computability theory. the major ﬁgures were all young. was not even twenty. such as the Turing universal machine. and recognized that Hilbert’s program was over. G¨del’s Incompleteness Theorem [1931] not only refuted o Hilbert’s program on proving consistency. This made it easier to challenge Hilbert on the second topic of the decision problem.

(See Davis [1965. G¨del was not.D. Alonzo Church spent one year at Harvard and one year at G¨ttingen and Amsterdam. G¨del spent the ﬁrst part of 1934 at Princeton. Church’s Thesis (First Version) [1934].” 2. 8–9]. he tried to refute it by a diagonal argument but since the λ-deﬁnable functions were only partial functions. On the basis of this evidence and his own intuition. Stephen C. Instead of a contradiction.3 Herbrand-G¨del Recursive Functions o From 1933 to 1939 G¨del spent time both in Vienna pursuing his academic o career and at Fine Hall in Princeton which housed both the Princeton University faculty in mathematics and the Institute for Advanced Study. When Kleene ﬁrst heard the thesis. He returned o to Princeton as an Assistant Professor of Mathematics in 1929. G¨del’s deﬁnition o gave a remarkably succinct system whose simple rules reﬂected the way a mathematician would informally calculate a function using recursion. Church had begun to develop a formal system now called the λ-calculus.” Soon the preﬁx “primitive” was added to the latter and the preﬁx “general” was generally dropped from the former. By the time Kleene received his Ph. Kleene. at Princeton in 1927.) Deﬁnition 2.D. Kleene began proving that certain well-known functions were λ-deﬁnable. In 1931 Church knew only that the successor function was λ-deﬁnable. of which he was a member. p.1. Although Kleene was convinced by Church’s ﬁrst thesis. He expanded on a concept of Herbrand and brought closer to the modern form. Church proposed to G¨del around March. G¨del rejected Church’s o o ﬁrst thesis as “thoroughly unsatisfactory. the Kleene Recursion Theorem whose proof is a diagonal argument which fails (see Soare [1987.After completing his Ph. o The primitive recursive functions which he had used in his 1931 paper did not constitute all computable functions. In 1931 Church’s ﬁrst student. p. 36]). A function if eﬀectively calculable if and only if it is λ-deﬁnable. Kleene had proved a beautiful new theorem. At Princeton in the spring of 1934 G¨del lectured on the Herbrand-G¨del recursive functions which came o o to be known as the general recursive functions to distinguish them from the primitive recursive functions which at that time were called “recursive functions. 9 . the diagonal was one of the rows. in 1934 he had shown that all the usual number theoretic functions were λ-deﬁnable. the term in the 1930’s for a function which is computable in the informal sense. arrived at Princeton. 1934 the ﬁrst o version of his thesis on functions which are eﬀectively calculable.

As further evidence. Deﬁnition 2. 1935. to put my work in that format. functions deﬁned by the ﬁve schemata for primitive recursive functions. Church’s Thesis [1936]. Church and Kleene had shown the formal equivalence of the Herbrand-G¨del recursive functions and the λ-deﬁnable funco tions. λ-deﬁnability is a precise calculating system and has close connections to modern computing.” o Kleene [1981] wrote. plus the least number operator µ. . his famous proposition described in his paper [1936]. “In this paper a deﬁnition of recursive function of positive integers which is essentially G¨del’s is adopted. Church and Kleene changed their formalism (especially for o Church’s Thesis) from “λ-deﬁnable” to “Herbrand-G¨del general recursive. After seeing o G¨del’s lectures. Kleene introduced a new equivalent deﬁnition. . after general recursiveness had appeared. chose. . “I myself. Church o presented to the American Mathematical Society on April 19. ” Nevertheless. with Herbrand-G¨del recursive functions o in place of λ-deﬁnable ones. such as functional programming. A function on the positive integers is eﬀectively calculable if and only if it is recursive.. perhaps unduly inﬂuenced by rather chilly receptions from audiences around 1933–35 to disquisitions on λ-deﬁnability. o Rosser and Kleene took notes which appeared as G¨del [1934]. 2. This time without consulting G¨del. .4 Stalemate at Princeton Over Church’s Thesis Church reformulated his thesis.. the µ-recursive functions.2. It is maintained o that the notion of an eﬀectively calculable function of positive integers should be identiﬁed with that of a recursive function. The µ-recursive functions had the advantage of a short standard mathematical deﬁnition. o 10 .” It has been known since Kleene [1952] as Church’s Thesis in the following form. but the disadvantage that any function not primitive recursive could be calculated only by a tedious arithmetization as in G¨del’s Incompleteness Theorem.Church and Kleene attended G¨del’s lectures on recursive functions.

G¨del’s footnote 3. Church. constituted the most 11 .5 G¨del’s Thoughts on Church’s Thesis o In spite of this evidence. It was his approval that Church wanted most. G¨del says later that he was not o sure that his system of Herbrand-G¨del recursive functions comprised all o possible recursions.” The second paragraph. since the notion of ﬁnite computation is not deﬁned.2. the value of the function can be computed by a ﬁnite procedure3 . G¨del still did not accept Church’s Thesis by the o beginning of 1936.” Footnote 3. o Rosser.” This suggests that G¨del was waiting not only for a formal deﬁnition o (such as recursive functions or Turing machines which came later) but evidence that these captured the informal notion of eﬀectively calculable (which Turing later gave. recursions of other forms (e. “The converse seems to be true. and Post nearby at City College in New York. G¨del had considered the question of o characterizing the calculable functions in [1934] when he wrote. gives crucial insight into his o thinking about the computability thesis and his later pronouncements about the achievements of Turing versus others. but which G¨del did not ﬁnd in Church’s work). Kleene. Here o he even suggests that such a precise mathematical characterization of the informal notion cannot be proved which makes his acceptance of Turing’s later work even more impressive. “[Primitive] recursive functions have the important property that. Second. but it serves as a heuristic principle. 3 3.1 Turing Breaks the Stalemate Turing Machines and Turing’s Thesis At the start of 1936 those gathered at Princeton: G¨del.” but is a “heuristic principle.. This cannot be proved. with respect to two variables simultaneously) are admitted.g. for each given set of values for the arguments. Church had solved the Entscheidungsproblem only if his characterization of eﬀectively calculable functions was accurate. besides recursion according to scheme (V) [primitive recursion]. G¨del had become the most prominent ﬁgure in matho ematical logic. his ﬁnal sentence suggests that he may have believed such a characterization “cannot be proved. if.

This is important because of the appearance in 1936 of related papers by Church. The work of Hilbert and G¨del had become well-known around the world. Gandy [1980]. Davis [1965. Turing’s mother had had a typewriter which fascinated him as a boy. or perhaps because of the two page minor correction [1937a]. 1936. 1936. but they could not agree on whether recursive functions constituted all eﬀectively calculable functions. not knowing it had been previously proved. Kleene. i. and Turing’s priority is important here. 73]. and Post. 72]. 1936–Read 12 November. and as a result Turing had been elected a Fellow of King’s College. [1981]. At that moment stepped forward a twenty-two year old youth. Well. far removed from Princeton. and the second section of pages 241–265 in Volume 42. write. Part 3. Alan Turing attended.” or sometimes “[1936-37]. and erase one square at a time before moving to an immediately adjacent square. p.distinguished and powerful group in the world investigating the notion of a computable function and Hilbert’s Entscheidungsproblem. Turing’s Thesis [1936]. the ﬁrst section of pages 230–240 in Volume 42.A. There has been considerable misunderstanding in the literature about exactly when Turing’s seminal paper was published. Post [1943. [1981b]. [1987]. this was not just any youth. (Max) Newman gave lectures on Hilbert’s Entscheidungsproblem in 1935. Turing showed his solution to the astonished Max Newman in April. p. Cambridge. Part 4. Alan Turing had already proved the Central Limit Theorem in probability theory.” perhaps because the volume 42 is 1936-37 covering 1936 and part of 1937.1 Many papers. and they published Turing’s paper in volume 42 on November 30. Cutland [1980]. and others. 1 12 . p. Vol. 1936 and December 23. such as Kleene [1952]. G¨del Collected Works. 456.e. Others. and others. Deﬁnition 3. correctly refer to it as “[1936]. [1987b]. p. He designed his automatic machine (a-machine) as a kind of idealized typewriter with an inﬁnite carriage over which the reading head passes with the ability to read. mistakenly refer to this paper o as “Turing [1937]. issued on November 30.” It appeared in two sections. p. Kleene [1943.. 1936. Newman persuaded them that Turing’s work was suﬃciently diﬀerent.H. A function is intuitively computable (eﬀectively calculable) if and only if it is computable by a Turing machine. o At Cambridge University topologist M. I.” The journal states that Turing’s manuscript was “Received 28 May. as deﬁned in Turing [1936].1. an automatic machine (a-machine). 1936. The Proceedings of the London Mathematical Society was reluctant to publish Turing’s paper because Church’s had already been submitted to another journal on similar material. just like a typewriter. Kleene and Post [1954. 20]. 407].

G¨del was also well aware of Turing’s proof [1937b] of o the equivalence of λ-deﬁnable functions with Turing computable ones. Volume II. issued December 23. Let us examine there what G¨del has to say. G¨del [1990]. 1936.Turing’s a-machine has compelling simplicity and logic which makes it even today the most convincing model of computability. even though it o was formulated with his own general recursive functions. G¨del knew of the extensional evo idence. and Kleene [1936]. Equally important with the Turing machine was Turing’s analysis of the intuitive conception of a “function produced by a mechanical procedure. G¨del o o o [1995]. general recursive functions. No part of Turing’s paper appeared in 1937. In the o o following article G¨del considers all these as equivalent formal deﬁnitions. Also Turing [1936. p. and hence the conﬂuence of all the known deﬁnitions. Church and Kleene [1936b] had shown the formal equivalence of λ-deﬁnable functions. 13 . The best source for articles of G¨del is the o three volume Collected Works. G¨del [1986]. Determining the correct date of publication of Turing’s work is important to place it chronologically in comparison with Church [1936]. He had not accepted the argument of conﬂuence as suﬃcient to justify Church’s Thesis. G¨del clearly expresses o his opinion in his three volume collected works. 3. G¨del [1990].2 G¨del’s Opinion of Turing’s Work o G¨del never accepted Church’s Thesis in the form above.” In a masterful demonstration. which we have listed by year of publication: Volume I. G¨del [1986]. and Volume III. o The question was whether they captured the informal concept of a function speciﬁed by a ﬁnite procedure. but the two page minor correction [1937a] did. Post [1936]. which Robin Gandy considered as precise as most mathematical proofs. Turing analyzed the informal nature of functions computable by a ﬁnite procedure and demonstrated that they coincide with those computable by an a-machine. G¨del was interested in the intensional analysis of ﬁnite proo cedure as given by Turing [1936]. and Kleene had proved the equivalence with µ-recursive functions based on G¨del’s own arithmeo tization of 1931. but G¨del and o most others accepted Turing’s Thesis. 243] introduced the universal Turing machine which has great theoretical and practical importance. o o and G¨del [1995]. However.

2. o the formal classes of Turing computable functions. G¨del. not that the Herbrand-G¨del reo cursive functions had elevated Turing computable functions. and had rejected it in 1935 and 1936 because he was not sure his own deﬁnition had captured the informal concept of procedure. o “That this really is the correct deﬁnition of mechanical computability was established beyond any doubt by Turing. because for the notions of mechanical procedure and of formal system no mathematically satisfactory deﬁnition had been given at that time. G¨del knew o that Turing had never proved anything about recursive functions. What did he mean? G¨del knew that by work of Turing. . Indeed G¨del o had seen Church’s Thesis 2.” The word “this” evidently refers to the recursive functions.” In the Nachlass printed in G¨del [1995. G¨del now went forward o with Turing’s work.3. G¨del was asserting that it was Turing o who had demonstrated that these formal classes captured the informal notion of a procedure. It was Turing’s proof [1936] and the formal equivalences which had elevated Herbrand-G¨del recursive functions to a correct characo teristic of eﬀectively calculable functions. Church. found in the Nachlass in a spiral notebook. and Kleene. which he knew would be more familiar with recursion. but having introduced Turing. 168. p.1 G¨del [193?] Notes in Nachlass [1935] o This article has an introductory note by Martin Davis in G¨del [1995. 166] G¨del wrote. recursive functions. G¨del had begun with the recursive function as more easily explained to o a general audience. He intended his talk to be as elementary as possible for a general audience.” To formally capture this crucial informal concept. . o o “When I ﬁrst published my paper about undecidable propositions the result could not be pronounced in this generality. began with a deﬁnition of the primitive recursive functions (which he quickly proved inadequate by a diagonal argument) and then his own Herbrand-G¨del recursive functions on p. (G¨del o o gave the Herbrand-G¨del recursive function deﬁnition rather than Turing o machines because he knew they were equivalent. “This article is taken from handwritten notes in English. 14 .) G¨del continued on p. evidently for a lecture. 156]. o Davis wrote. p. The essential point is to deﬁne what a procedure is. and λ-deﬁnable functions all coincided. .2 expressed in terms of Herbrand-G¨del recuro sive functions. 167. who was giving o an introductory lecture.

Once we have a suﬃciently strong system (such as Peano arithmetic) we can prove anything about computable functions which could be proved in a more powerful system. i. nr . p. . He has shown that the computable functions deﬁned in this way are exactly those for which you can construct a machine with ﬁnite number of parts which will do the following thing. . G¨del [1990. . . . 2 15 .e.” 3. a p. For the concept . on a slip of paper and put the slip into the machine and turn the crank then after ﬁnite number of turns the machine will stop and the value of the function for the argument n1 . Turing computability.. nr will be printed on the paper. . G¨del muses on the remarko o able fact of the absoluteness of computability. it is possible that passing from one formal system Si to one of higher order Si+1 not only allows us to prove certain propositions which were not provable before. It seems to me that this importance is largely due to the fact that with this concept one has for the ﬁrst time succeeded in giving an absolute deﬁnition of an interesting epistemogical notion. “A function of integers is computable in any formal system containing arithmetic if and only if it is computable in arithmetic” . . but also makes it possible to shorten by an extraordinary amount proofs already available.2. . and others.2 In all other cases treated previously. as a single formal system. “Tarski has stressed in his lecture (and I think justly) the great importance of the concept of general recursiveness (or Turing’s computability). . general o recursiveness. G¨del discussed what it means for a function to be computable o in a formal system S and remarked that given a sequence of formal systems Si . such as demonstrability or deﬁnability. Now for G¨del’s Princeton Bicentennial address [1946] refer to the Colo lected Works Volume II. Si+1 . 396–398]. . G¨del identiﬁes those formal systems known to be equivalent. . one has been able to deﬁne them only relative to a given language. If you write down any number n1 . that it is not necessary to distinguish orders (diﬀerent formal systems). one not depending on the formalism chosen. . 150]. Once again.2 Princeton Bicentennial [1946] ¨ To fully understand this article one should be familiar with G¨del’s Uber die o L¨nge von Beweisen” (“On the length of proofs) [1936a] in Volume I [1986. .“But Turing has shown more. and for each individual language it is clear that the one thus obtained is not the one looked for.

. Gandy [1988. It is exactly this stumbling block which Turing overcame by a totally new approach. 1935 (see Kleene [1987b]. but G¨del did not o accept it then and gave no evidence later of believing that Church had done this. “.2. G¨del himself never claimed to have made this link. 80] wrote. for example by Church. If the basic steps are stepwise recursive. 80.” Sieg [p.2. Church claimed it in o his announcement of Church’s Thesis in 1935 and 1936.3 The Flaw in Church’s Thesis G¨del himself was the ﬁrst to provide one of the formalisms later recognized o as a deﬁnition of computability. without argument to be recursive. although it is merely a special kind of demonstrability or decidability. pp. This becomes irrefutably clear in §3. . Sieg [p. By a kind of miracle it is not necessary to distinguish orders. and that stumbling block was quite clearly seen by Church. o “. and the diagonal procedure does not lead outside the deﬁned notion. ” Who is the “one” who has linked the informal notion of procedure or eﬀectively calculable function to one of the formal deﬁnitions. 3. “It is precisely here that we encounter the major stumbling block for Church’s analysis. 78] wrote.of computability. something he did not justify. . 87] in their excellent analyses brought out this weakness in Church’s argument. . the general recursive functions. However. . the situation is diﬀerent. then it follows easily by the Kleene Normal Form Theorem which Kleene had proved and communicated to G¨del before November. 57]). p.” who wrote that without this assumption it is diﬃcult to see how the notion of a system of logic can be given any exact meaning at all. p. .5. this core does not provide a convincing analysis: steps taken in a calculus must be of a restricted character and they are assumed. that the entire o process is recursive.” G¨del stated. The fatal weakness in Church’s argument was the core assumption that the atomic steps were stepwise recursive. 79] and especially Sieg [1994. 16 . however. one has for the ﬁrst time succeeded in giving an absolute deﬁnition of an interesting epistemogical notion . Modern scholars found weaknesses in Church’s attempted proof [1936] that the recursive functions constituted all eﬀectively calculable functions.

not only in themselves. “Some basic theorems on the foundations of mathematics and their implications.5 G¨del’s Letter to Kreisel [1968] o -G¨del: letter to Kreisel of May 1. o “ But I was completely convinced only by Turing’s paper.” 3. but also with regard to their implications for the traditional philosophical problems about the nature of mathematics.. are fairly widely known. “ o G¨del [1951] in his Gibbs lecture wrote.2. . 1995. however. . It is probable as Wang suggests ([1987. verifying the Computability Thesis].” – G¨del. but he never o gave any hint that he thought Church had been the ﬁrst to show that the recursive functions coincided with the eﬀectively calculable ones. Church 1936. see A.3. . p. which. 1994.2.2. p. Princeton Bicentennial.4 G¨del on Church’s Thesis o G¨del may not have found errors in Church’s demonstration. . o “As for previous equivalent deﬁnitions of computability. Vol. at a meeting of the American Mathematical Society at Brown University. 88]. p. that the lecture was the main project G¨del worked on in the fall of 1951. I. I believe. [G¨del. The results themselves. 256–358.e. [i.6 Gibbs Lecture [1951] G¨del. pages 117–18]). [1946. pp 150–153. pp. are much less suitable for our purpose.” . 84]. Collected Works Volume III. G¨del said. On the contrary. 3. 1968 [Sieg. G¨del delivered the twentyo ﬁfth Josiah Willard Gibbs Lecture. o o Martin Davis in his introduction wrote. .304–305]. o “Research in the foundations of mathematics during the past few decades has produced some results of interest. “On 26 December 1951. but nevertheless I think it will be useful to present them in outline once again. and G¨del o o Collected Works. especially in view of the fact that due to the work 17 .

of various mathematicians, they have taken on a much more satisfactory form than they had had originally. The greatest improvement was made possible through the precise deﬁnition of the concept of ﬁnite procedure, [“ . . . equivalent to the concept of a ’computable function of integers’ . . . ”] which plays a decisive role in these results. There are several diﬀerent ways of arriving at such a deﬁnition, which, however, all lead to exactly the same concept. The most satisfactory way, in my opinion, is that of reducing the concept of a ﬁnite procedure to that of a machine with a ﬁnite number of parts, as has been done by the British mathematician Turing.” In this one paragraph, 1. G¨del stressed the importance of the results to mathematics and phio losophy. 2. G¨del gave full credit to Turing and his “machine with a ﬁnite number o of parts” for capturing the concept of ﬁnite procedure. 3. G¨del never mentions Church or G¨del’s own deﬁnition of recursive o o functions. This is one of the most convincing and explicit demonstrations of G¨del’s o opinion of Turing’s work. 3.2.7 Godel’s Postscriptum 3 June, 1964 to G¨del [1934] o

-Godel’s Postscriptum 3 June, 1964 to G¨del [1934], see “The Uno decidable,” M. Davis, [1965, p. 71] and G¨del Collected Works, Volume I o [1986, p. 369–370]. “ In consequence of later advances, in particular of the fact that, due to A. M. Turing’s work, a precise and unquestionably adequate deﬁnition of the general concept of formal system can now be given, the existence of undecidable arithmetical propositions and the non-demonstrability of the consistency of a system in the same system can now be proved rigorously for every consistent formal system containing a certain amount of ﬁnitary number theory.”

18

Turing’s work gives an analysis of the concept of “mechanical procedure” (alias “algorithm” or “computation procedure” or “ﬁnite combinatorial procedure”). This concept is shown to be equivalent with that of a “Turing machine.”

3.3

Hao Wang Reports on G¨del o

Hao Wang was a very close friend and professional colleague of G¨del, whom o he called “G” in the following passage. Wang [1987, p. 96] wrote about G¨del’s opinion of Turing’s work. o “Over the years G habitually credited A.M. Turing’s paper of 1936 as the deﬁnitive work in capturing the intuitive concept [of computability], and did not mention Church or E. Post in this connection. He must have felt that Turing was the only one who gave persuasive arguments to show the adequacy of the precise concept . . . In particular, he had probably been aware of the arguments oﬀered by Church for his “thesis” and decided that they were inadequate. It is clear that G and Turing (1912–1954) had great admiration for each other, . . . “

3.4

**Kleene’s Remarks About Turing
**

“Turing’s computability is intrinsically persuasive” but “λ-deﬁnability is not intrinsically persuasive” and “general recursiveness scarcely so (its author G¨del being at the time not at all o persuaded).” -Stephen Cole Kleene [1981b, p. 49] “Turing’s machine concept arises from a direct eﬀort to analyze computation procedures as we know them intuitively into elementary operations. Turing argued that repetitions of his elementary operations would suﬃce for any possible computation. “ For this reason, Turing computability suggests the thesis more immediately than the other equivalent notions and so we choose it for our exposition.” -Stephen Cole Kleene, second book [1967, p. 233]

19

3.5

**Church’s Remarks About Turing
**

Computability by a Turing machine, “ has the advantage of making the identiﬁcation with eﬀectiveness in the ordinary (not explicitly deﬁned) sense evident immediately—i.e., without the necessity of proving preliminary theorems.” -Alonzo Church [1937], Review of Turing [1936]

In modern times it is sometimes stated as follows, recognizing that Church [1935], [1936] got it ﬁrst, but that Turing [1936] got it right, in the opinion of G¨del and many modern scholars. o Deﬁnition 3.2. Church-Turing Thesis. A function is intuitively computable if and only if it is computable by a Turing machine, or equivalently if it is speciﬁed by a recursive function. We strongly believe that it should not be called any of the three (Church’s Thesis, Turing’s Thesis, or the Church-Turing Thesis) but rather should be called the “Computability Thesis” as we argue in §13 and §14, just as the calculus is named for neither of its discoverers, Newton and Leibniz.

4

Oracle Machines and Relative Computability

After introducing deﬁnitions of computable functions: Turing a-machines, recursive functions, and λ-deﬁnable functions; the originators continued during 1936–1939 to explore incomputable phenomena, rather than computable applications of these devices, which came only a decade or more later. Church and Kleene [1936] as well as Church [1938] and Kleene [1938] studied computable well-orderings and deﬁned recursive ordinals which were later used to extend the jump operation to the arithmetic hierarchy and beyond CK to the hyperarithmetic hierarchy up to the ﬁrst nonrecursive ordinal ω1 . Turing spent 1936–1938 at Princeton working on a Ph.D. with Church. His thesis was completed in 1938 and published in Turing [1939]. Church and other mathematicians had found G¨del’s Incompleteness Theorem uno settling. By G¨del’s proof an eﬀective extension T1 of Peano arithmetic o cannot prove its own consistency conT1 . However, we can add the arithmetical statement conT1 to T1 to get a strictly stronger theory T2 . Continuing, we can get an increasing hierarchy of theories {Tα }α∈S over a set S of ordinals. Turing’s Ph.D. thesis [1939] concerned such an increasing array of undecidable theories.

20

As before.4. having as one of its fundamental processes that of solving a given number-theoretic problem. S2 = {B. The old tape is called the work tape and operates just as before. a kind of oracle as it were. even after Post.1 with a modern treatment of oracle machines and Turing functionals including some of the more important properties even though these were mostly discovered much later. . §4] a short statement about oracle machines. In 1939 Turing left this topic never to return. Before doing so. L} the set of 21 . called the oracle tape. A Turing oracle machine (o-machine) is a Turing machine with an extra “read only” tape. such as eﬀective continuity on Cantor space. cannot be a machine. [1948]. upon which is written the characteristic function of some set A (called the oracle). 1} is the work tape alphabet. . More modern properties of Turing functionals. and whose symbols cannot be printed over. and other papers as we shall explain in §5 and §6. this oracle . . 1} is the oracle tape alphabet. Q is a ﬁnite set of states. Deﬁnition 4. It mostly lay dormant for ﬁve years until it was developed in a beautiful form by Post [1944].” This is virtually all Turing said of oracle machines. p.1. His description was only a page long and half of that was devoted to the unsolvability of related problems such as whether an o-machine will output an inﬁnite number of 0’s or not. 4. The reading head moves along both tapes simultaneously. . . 46] of a machine with a head which reads the work tape and oracle tape simultaneously. S1 = {B. With the help of the oracle we could form a new kind of machine (call them o-machines). Turing wrote in his ordinal logics paper [1939. we complete this section §4.2 Modern Deﬁnitions of Oracle Machines There are several equivalent ways that a Turing machine with oracle may be deﬁned. 0. “Let us suppose we are supplied with some unspeciﬁed means of solving number-theoretic problems. and {R.1 Turing’s Oracle Machines In one of the most important and most obscure parts of all of computability theory. will be developed in §7.6. We prefer the deﬁnition in Soare’s book [1987. but many other formulations produce the same class of functionals.

L}. (i) We let lower case Greek letters ϕ. δ : Q × S1 × S2 −→ Q × S2 × {R. ΦA is called a partial computable e functional because it takes a set A to a set B and is viewed as a map on Cantor Space 2ω . X) indicates that the machine in state q reading symbol a on the oracle tape and symbol b on the work tape passes to state p. and moves one space right (left) on both tapes if X = R (X = L). Lachlan also used matched pairs Ψ. If A ⊆ ω then ΨA (x) may be deﬁned for some or all x. the maximum element of A e examined during the computation. Γ. a. where δ(q. (iv) Since Rogers book [1967]. e Notation 4. e 22 . There is no confusion between the notation ϕe (x) as a partial computable function and ϕA (x) as a e use function for ΦA (x) because the former ϕe (x) will never have an exponent e A and the latter use function ϕA (x) always will. researchers have used ΦA (x) for the Turing functional with oracle program e Pe and have used ϕA (x) for the use function. b) = (p. (iii) As in Soare [1987] we use {Pe }e∈ω for an eﬀective listing of Turing programs for Turing a-machines and let ϕe be the partial computable function deﬁned by Pe . Since Lachlan about 1970. ψ represent partial functions from ω to ω and lower case Latin letters f .2. ψ. We let {Pe }e∈ω be an eﬀective listing of oracle Turing programs. On the other hand. (ii) We let upper case Greek letters represent partial functionals from 2ω to 2ω . prints “c” over “b” on the work tape. If ΦA is total and computes B we say B is computable in A and write B ≤T A. γ and so forth for partial computable functionals and their use functions in many papers. A Turing oracle program Pe is now simply a partial map. c. and h represent total functions.head moving operations right and left. researchers have used ϕe (x) or {e}(x) for the partial computable function with program Pe . The Turing oracle program Pe takes some oracle A and deﬁnes a partial A-computable functional ΦA (x) = y. The other details are just as previously in Soare [1987]. g. If B ⊆ ω we write ΨA = B if ΨA (x) = B(x) for all x ∈ ω. e We refer to ϕe as a partial computable function ω to ω because its input and output are integers. and let Φe be the computable partial functional deﬁned by Pe . and this is the general usage today.

function ψ and a description of its graph are interchangeable and equally powerful in describing ψ. Here Φσ (x) = y denotes that the oracle program Pe with oracle σ on its e oracle tape.2 The Graph of an Oracle Computable Functional Deﬁnition 4.4.) function then graph(ϕe ) is a computable enumerable (c.e.1 The Graph of a Partial Computable Function Deﬁnition 4. The crucial property of the oracle graph Ge and the one which makes a Turing functional Φe independent of any particular machine representation is the following. subset Ve ⊆ We which is the graph of a p.3. The the notions of a Turing program to compute a p.5. then the graph Ge deﬁned in (2) is a computably enumerable (c.c) function ϕe deﬁne the graph of ϕe as follows. From the deﬁnitions Ge is Σ0 and therefore c. given any (c.e. set V which satisﬁes a singlevaluedness condition (3) and a continuity condition (4). (2) Ge =dfn graph(Φe ) =dfn { σ. x. 23 . y : ϕe (x) = y } Note that if ϕe is a partial computable (p. Likewise.4. and does not read more of the oracle tape than σ during the computation. 1 The converse holds for a c.3 The Oracle Graph Theorem Theorem 4. y : Φσ (x) = y } e where σ ranges over 2<ω .) set We we can ﬁnd a single-valued c. For an oracle machine program Pe we likewise deﬁne the oracle graph of the corresponding computable functional Φe but now taking into consideration the ﬁnite strings read by the oracle head during the computation.e.) set. eventually halts and outputs y. Proof.c.e.) set. function. 4.c. and x on its input tape.c.2.e.e. Given a partial computable (p. 4. If Pe is an oracle Turing program deﬁning a Turing functional Φe . Oracle Graph Theorem.2. (1) ge = graph(ϕe ) =dfn { x.

In addition. Furthermore. Theorem 4. then there is a Turing oracle program Pi which on input x and any oracle A produces the same output y.6. Let Pe be a Turing program to compute ϕe . y appears in V deﬁne Ψσ (x) = y. x. If Pe is a Turing program for a Turing a-machine.e. y ∈ U ]} = {x : (∃σ)(∃y)[ σ.e. Theorem 4. If σ. Proof. Similar to single valuedness theorem in Soare [1987. x. This reinforces the claim that it is the o-machine. τ. This includes an oracle machine with a single reading head reading the work tape and oracle tape. present o-machines. δ : Q × S2 −→ Q × S2 × {R.e. Therefore. y ∈ V =⇒ =⇒ ¬(∃τ ⊇ σ)(∃z)[ z = y & (∀τ ⊃ σ)[ τ. which is the central concept of the subject. or two independent reading heads. subset V ⊆ U satisfying (3) and (4) and having the same domain. Given V satisfying (3) and (4) deﬁne a partial computable functional Ψσ (x) as follows. Each of these formal deﬁnitions produces a c. Now Pe consists of a ﬁnite partial map which can be identiﬁed with a set of 5-tuples. {x : (∃σ)(∃y)[ σ. set U ⊆ 2<ω × ω × ω there is a c. x.7. Let V ⊂ 2<ω × ω × ω be a computably enumerable set which satisﬁes the following two conditions. given any c. x.4 Equivalent Deﬁnitions of Relative Computability There are several diﬀerent formal deﬁnitions of relative computability. x. Proof. x. and then draw a-machines as special cases.e. graph Ge and these deﬁnitions are all equivalent.Theorem 4. y ∈ V ].. Then there is a Turing functional Φe such that Ge = V .8. z ∈ V ]. Chapter II]. several authors deﬁne relative computability from oracle A by adding the characteristic function of A either to the Herbrand-G¨del o general recursive function deﬁnition or to the Kleene µ-recursive function deﬁnition. Furthermore. y ∈ V ]}. Proof. any Turing a-machine can clearly be simulated by a Turing o-machine as we note in the following theorem. or other variations. This deﬁnes an eﬀective reduction Ψ. There must be a Turing oracle machine Φe such that Φe = Ψ. L}. y ∈ V σ. in presenting the subject we can bypass a-machines altogether. 4. i. (3) (4) σ. not the a-machine. x. 24 .

X) for both c = 0 and c = 1. q ∈ Q and a. a) = (p. b. there is no justiﬁcation for the necessity of Φe to denote ϕe since the current notation ϕe is quite satisfactory. Hence. 4. 5 Emil Post Expands Turing’s Ideas The spirit of Turing’s work was taken up by the American mathematician Emil Post. sometimes we would like to write Φe alone without its exponent A to identify it with Pe or its oracle graph Ge as a ﬁnite object.1 Notation for Functions and Functionals The standard notation is that given above. a) = (p. S2 = {B. Deﬁne an oracle program Pi as follows with transition function δ : Q × S1 × S2 −→ Q × S2 × {R. This is unwise because it blurs the distinction of types in which ϕe operates on integers and Φe on sets. 1} the oracle tape alphabet as follows.9. while ϕe is deﬁned by Pe a ﬁnite set of 5-tuples operating on integers. The functional Φe is deﬁned by a program Pe which is a ﬁnite set of 6-tuples operating on sets. b ∈ S2 .4. We recommend against using Φe to denote ϕe the partial computable function. Recently. we add to oracle program Pi a line δ(q. Furthermore. This is analogous to a laptop computer with no active connection to a database which may later be connected to the World Wide Web. like a laptop computer whose link with the web has temporarily been removed. L} the set of head moving operations right and left. some researchers have unfortunately used Φe to denote ϕe . who had been appointed to a faculty position at City College of New York in 1932. 0. 1} is the work tape alphabet. Furthermore. c. Pi has exactly the same eﬀect on input x as Pe regardless of the oracle A. b. for S1 = {B.where Q is a ﬁnite set of states. Note that the oracle Turing machine Φe is a ﬁnite object represented by an oracle program Pe or an oracle graph Ge and has no oracle associated with it. For each line in Pe of the form δ(q. X) for p. Remark 4. but it can use any oracle A which may be attached. 25 . and {R. L}. Doing so under the new convention leads to confusion with ϕe which is given by a diﬀerent type of program.

p. Post oﬀers only as a “working hypothesis” that his contemplated “wider and wider formulations” are “logically reducible to formulation 1. corresponding in spirit to Turing’s more than to Church’s. This was the mantle of clarity and intuitive exposition. and most of all.2 Post Steps Into Turing’s Place During 1940–1954 As Turing left the subject of pure computability theory in 1939.” Lastly. Post’s contributions during the 1930’s were original and insightful. the mantle of relative computability and Turing reducibility. Most important. p. but they were not as inﬂuential as those of Church and Turing. of why the intuitively computable functions are computable in his formal system. did not prove the unsolvability of the Entscheidungsproblem because at the time Post was not aware of Turing’s [1936]. Post wrote [1936] that Church’s identiﬁcation of eﬀective calculability and recursiveness was working hypothesis which is in “need of continual veriﬁcation.5. such as general recursiveness. Post’s (normal) canonical system is a generational system. Post did not attempt to prove that his formalism coincided with any other formalism. which he explained again in [1944]. as did Turing. Post [1941] and [1943] introduced a second and unrelated formalism called a production system and (in a restricted form) a normal system. the mantle of exploring the most basic objects such as computably enumerable sets. his mantle fell on the shoulders of Post. Post gave no analysis. From this it is often and erroneously written (Kleene [1987b. During the next decade and a half from 1940 until his death in 1954. 56]. Post played an extraordinary role in shaping the subject. because it gives an al- 26 . 5. of course. while Turing [1937b] proved the Turing computable functions equivalent to the λ-deﬁnable ones. Post gave no hint of a universal Turing machine. Furthermore.” This irritated Church who criticized it in his review [1937b] of Post [1936]. and Post believed that Church [1936] had settled the Entscheidungsproblem. Post. but in fact it was much less.1 Post’s Work in the 1930’s Post [1936] independently of Turing (but not independently of the work by Church and Kleene in Princeton) had deﬁned a “ﬁnite combinatory process” which closely resembles a Turing machine. It was only during the next phase from 1940 to 1954 that Post’s remarkable inﬂuence was fully felt. but merely expressed the expectation that this would turn out to be true. rather than a computational system as in general recursive functions or Turing computable functions. 61]) that Post’s contribution here was “essentially the same” as Turing’s. and [1981.

but stated it in terms of generated sets and production systems.” (This is Church’s [1936] terminology also). Since recursively enumerable sets are equideﬁnable with partial computable functions.1 in the succinct form. which asserted that “any generated set is a normal set. “Suﬃce it to say that each element of the set is at some time written down. 285]. explained his informal concept of a “generated set” of positive integers this way. 286]. It is understood that once an element is placed in the set. Post.1. 1943. and earmarked as belonging to the set. 27 . A nonempty set is eﬀectively enumerable (listable in the intuitive sense) iﬀ it is recursively enumerable (the range of a recursive function) or equivalently iﬀ it is generated by a (normal) production system. gave a thesis [1943. Therefore. deﬁned a set of positive integers to be recursively enumerable if it is the range of a recursive function and then stated. Post [1944. as a result of predetermined eﬀective processes.e.gorithm for generating (listing) a set of integers rather than computing a function. p. “The corresponding intuitive concept is that of an eﬀectively enumerable set of positive integers. it stays there. like Church [1936]. p. Although he had used other terminology earlier. set as its range and from the deﬁnition of a c. any eﬀectively enumerable set in the intuitive sense could be produced as a normal set is his formal system. set one can ﬁnd a single valued c.” Post then [1944. subset which is the graph of a partial computable function. 201]. (Equideﬁnable here means that from the deﬁnition of a partial computable function we can derive a c. restated Post’s Thesis 5.” That is. like Church and Turing. 1944]. 286].) Post’s Thesis is equivalent to Turing’s Thesis. Post showed that every recursively enumerable set (one formally generated by a recursive function) is a normal set (one derived in his normal canonical system) and conversely. [Post’s Thesis. Deﬁnition 5. this deﬁnition of normal set gives a new formal deﬁnition of computability which is formally equivalent to the deﬁnitions of Church or Turing. particularly for sets of positive integers. by the 1940’s Post had adopted the Kleene-Church terminology of “recursively enumerable set” for the formal equivalent of Post’s eﬀectively enumerable set. p. p.e. Post [1944. normal sets are formally equivalent to recursively enumerable sets.e. Post used the terms “eﬀectively enumerable set” and “generated set” almost interchangeably. This led Post to concentrate on eﬀectively enumerable sets rather than computable functions.

. Post later expanded on these deﬁnitions in [1948]. i. Post himself was struggling to understand it.“every generated set of positive integers is recursively enumerable. set as basic and to derive generalized computable functions as those whose graphs are generalized computably enumerable. Post called G¨del’s diagonal set. [1944]. Moreover.e.” He remarked that “this may be resolved into the two statements: every generated set is eﬀectively enumerable.) set W which is not computable but which cannot compute G¨del’s diagonal set K.e. every eﬀectively enumerable set of positive integers is recursively enumerable. Recursively enumerable sets of positive integers and their decision problems. set We is computable in K (We ≤T K). even as little as presented above in §4. sets rather than partial computable functions may be even more fundamental than the thesis of Church and Turing characterizing computable functions because Sacks [1990] has remarked that often in higher computability theory it is more convenient to take the notion of a generalized c.e. Post felt that the creative property of K revealed the inherent creativeness of the mathematical process. Sets Post’s most inﬂuential achievement during this period was the extraordinarily clear and intuitive paper.” Post’s concentration on c.E. 5. and even then only in general terms. o K = {e : e ∈ We } the complete set because every c. 28 . Post’s paper [1944] revealed with intuition and great appeal the significance of the computably enumerable sets and the signiﬁcance of G¨del’s o Incompleteness Theorem. and did not explicitly discuss it until the very end of his paper. such that ∅ <T W <T K.4 Post Began With Strong Reducibilities Post posed his famous “Post’s problem” of whether there exists a computably enumerable (c. In 1944 reo searchers did not understand Turing reducibility. “their converses are immediately seen to be true.” Post continued.3 Post’s Problem on Incomplete C.e. Here Post introduced the terms degree of unsolvability and the concept that one set has lower degree of unsolvability than another. 5.e.

set S2 in studying Post’s problem. Our modern understanding of relative computability and Turing functionals is due more to Post and his patient.e. 29 . Post described the manner in which the decision problem for one set S1 could be reduced to that of a second set S2 . that Post deﬁned and named for the ﬁrst time “Turing Reducibility. and began to discuss it in intuitive terms. sets with thin complements. hyper-simple.” denoted B ≤T A. Post deﬁned a sequence of strong reducibilities to better understand the concept of a set B being reducible to a set A.1 had been mostly ignored for ﬁve years. Post resurrected in [1944] the concept of oracle machines which had been buried in Turing’s [1939] paper and which other researchers had apparently ignored for ﬁve years. §11]. Post wrote it for a c. Post deﬁned families of c. but the analysis holds for any set S2 . Post was able to exhibit incomplete incomputable c. simple. Second. hyper-hypersimple. Slowly Post’s understanding deepened of the general case of one set B being reducible (Turing reducible) to another set A.e. These concepts have pervaded the literature and proved useful and interesting. 6. §11 General (Turing) Reducibility. Post’s four and a half page discussion there is the most revealing introduction to eﬀective reducibility of one set from another. Post’s Problem stimulated a great deal of research in the ﬁeld and had considerable inﬂuence. but they did not lead to a solution of Post’s problem. It was only at the end of Post’s paper [1944] in the last section. intuitive style as in the rest of the paper. Turing’s notion of relative computability from an oracle discussed in §4. persistent eﬀorts over a decade and a half than it is due to the brief remark by Turing in 1939. Post steadily continued gaining a deeper and deeper understanding from 1943 to 1954 until he had brought it to full development. in an attempt to ﬁnd an incomplete set for these reducibilities.1 Post Articulates Turing Reducibility Post wrote in [1944. First.e. Along with these strong reducibilities. In the same crisp. 6 Post Highlights Turing Computability When Post wrote his famous paper [1944]. sets for several of these stronger reducibilities but not for Turing reducibility.Post’s contributions from 1943 to 1954 concerning relative computability are remarkable.

It is even more surprising because relative computability is used much more often than 30 . this situation obtains with the following modiﬁcation.“Now suppose instead. which is equivalent to Turing’s Thesis 3. and in a ﬁnite time but a ﬁnite number of such questions can be asked.” If we identify a Turing reduction Φe with its graph Ge both informally and formally then the Post-Turing Thesis is equivalent to Post’s Thesis 5. §11]. A corresponding formulation of “Turing reducibility” should then be the same degree of generality for eﬀective reducibility as say general recursive function is for eﬀective calculability. the process would automatically continue to its eventual correct conclusion. Intuitively. a step as signiﬁcant as the Church-Turing characterization of “calculable. but it is largely implied by his presentation.2 for unrelativized computations) of what constitutes a relative computation of B from A.). this would correspond to the most general concept of reducibility of S1 to S2 . Turing [1939 §4]. For the very concept of the decision problem of S2 merely involves the answering for an arbitrarily given single positive integer m of the question is m in S2 . §4] in the following extension of Turing’s ﬁrst Thesis 3. says Turing [1939] in eﬀect.” 6.1 and Post’s ﬁrst Thesis 5.1.1. Post [1944. Turing’s brief introduction of oracles did not state this as a formal thesis.2 The Post-Turing Thesis Post’s statement may be restated in succinct modern terms and incorporates the statement implicit in Turing [1939. This is surprising because the Post-Turing Thesis was stated clearly in Post [1944]. Post-Turing Thesis.1 (because Ge is c.1. We could then say that the machine eﬀectively reduces the decision problem of S1 to that of S2 . That at certain times the otherwise machine determined process raises the question is a certain positive integer in a given recursively enumerable set S2 of positive integers. A set B is eﬀectively reducible to another set A iﬀ B is Turing reducible to A by a Turing oracle machine (B ≤T A). and that the machine is so constructed that were the correct answer to this question supplied on every occasion that arises. there has been little analysis (along the lines of the extensive analysis of the Church-Turing Thesis 3. Deﬁnition 6. Post makes it explicit and claims that this is the formal equivalent of the intuitive notion of eﬀectively reducible. However.e.

C = N A = (2ω − NA ). Also interactive or online computing in the real world is more common than batch processing or oﬄine computing. (iii) Recall the deﬁnition of a ﬁnite set Dy with canonical index y where Dy = {x1 < x2 .1 Continuous and Total Functionals Representations of Open and Closed Sets Deﬁnition 7.ordinary computability in the theory of computability. Deﬁne the string σy with canonical index y by σy (z) = 1 if z ∈ Dy and σy (z) = 0 otherwise.. = 2xk . i. Identify the sets A ⊆ ω with their characteristic functions f : ω → {0.e.2.. . 1}.2 Notation for Trees Deﬁnition 7. i. Now T is closed downwards (because A is closed upwards).e. . algorithmic complexity and many more. 7.. (iv) Cantor space is 2ω with the following topology (class of open sets).. The open sets of Cantor space are the countable unions of basic open sets. analysis. model theory. i. applications of computability to other areas such as algebra. we shall see that T forms a tree as in Deﬁnition 7. Fix any tree T . (iv) Set A ⊆ 2<ω is an open representation of the open set NA = σ∈A Nσ . 7 7. (v) A set C is (topologically) closed if its complement NA is open. Let σ and τ represent ﬁnite strings in 2<ω and σ τ or σ f denote that σ is an initial segment of τ or f . σ ∈ T and τ σ imply τ ∈ T .e.1. For every σ ∈ 2<ω the basic open (clopen) set Nσ = { f : f ∈ 2 ω & σ f }. 1} and represent the set of such f as 2ω . (i) Using ordinal notation identify the ordinal 2 with the set of smaller ordinals {0. We may assume A is closed upwards. . (ii) Let 2<ω denote the set of ﬁnite strings of 0 s and 1’s. . < xk } and y = 2x1 + 2x2 .2 (i). The sets Nσ are called clopen because they are both closed and open. 31 . ﬁnite initial segments of functions f ∈ 2ω . In this case T =dfn 2<ω − A is a closed representation for C. using processes contained entirely inside the machine. σ ∈ A and σ τ implies τ ∈ A. (i) A tree T ⊆ 2<ω is a set closed under initial segments. . Hence. i.e.

32 . (iii) If {Ci }i∈ω is a countable family of closed sets such that ∩i∈F Ci = ∅ for every ﬁnite set F ⊆ ω. Given tree T let A = 2<ω − T . Theorem 7. . Then NA is open. The closed sets are closed under ﬁnite union and countable intersection since the open sets have the dual properties in Deﬁnition 7. Proof. (6) Tσ = { τ : σ τ or τ σ }. Any open cover {Nσ }σ∈A ⊇ 2ω has a ﬁnite open subcover {Nσ }σ∈F ⊇ 2ω for some ﬁnite subset F ⊆ A. Given any closed C with complement NA deﬁne T by putting σ in T if (∀τ σ)[ τ ∈ A ]. (iv) Deﬁne the subtree of extendible nodes σ ∈ T . Therefore. (i) K¨nig’s Lemma.e. The term “compactness” refers to any of them. [Compactness]. so any countable union of them is open and any countable intersection is closed. which is called a closed representation for C. These properties can be proved from one another but here we give direct proofs of each.(ii) The set of inﬁnite paths through T is (5) [ T ] = { f : (∀n) [ f n ∈ T ] }. (iv) Finite subcover. Note that [ T ] is always a closed set.4.. many diﬀerent representations for the same closed set [ T ]. If T ⊆ 2<ω is an inﬁnite tree. closure under ﬁnite intersection and countable union. i. and for every closed set C there is a tree T such that C = [ T ]. (⇐=).1 (iii). and intersection Tω = ∩n∈ω Tn . Usually for a given tree T there are many other trees T such that [ T ] = [ T ]. The clopen sets are both open and closed.3. [ T ] = 2ω − NA and [ T ] is closed. then [ T ] is a closed set. but particularly to (iv). (7) T ext = { σ ∈ T : (∃f σ)[ f ∈ [ T ] ] } Note that if the tree T is computable then T ext is co-c. Then T is closed downward and [ T ] = C. (iii) For σ ∈ T deﬁne the subtree Tσ of nodes τ ∈ T comparable with σ. then ∩i∈ω Ci = ∅ also. o (ii) If T0 ⊇ T1 . is a decreasing sequence of trees with [ Tn ] = ∅ for every n. then [ Tω ] = ∅. . (=⇒). The following very easy and well-known properties hold for Cantor Space 2ω .e. Theorem 7. If T is a tree. then [ T ] = ∅.3 there is a (nonunique) tree T ⊆ 2<ω such that C = [ T ]. If C ⊆ 2ω is any closed set then by Theorem 7.

If the open sets Ai are also dense open. if B = ∩i Ai a countable 2 intersection of open sets Ai . and A = NA . then they have special signiﬁcance.5.3 Dense Open Subsets of Cantor Space Deﬁnition 7. especially the Finite Extension Paradigm of Kleene and Post. boldface Π0 . (iii) Let T ⊆ 2<ω be a tree. After open and closed sets.e.” Meeting a given requirement Ri amounts to meeting the corresponding dense open set Ai ..4 Eﬀectively Open and Closed Sets If A ⊆ 2<ω is c. much attention is paid in point set topology to Gδ sets (see Oxtoby[1971]). (i) A set A ⊆ S is dense if (∀σ) (∃f (ii) A set A ⊆ S is dense open if (8) (∀τ )(∃σ τ )(∀f σ) [ f ∈ A ]. (ii) If C = 2ω − NA for A c. then f is a limit point.. If f is not isolated. (iii) (10) C ⊆ 2ω is a Π0 class if there is a computable relation R(x) such that 1 C = {f : (∀x) R(f (x))}.7. Banach-Mazur games can be used for ﬁnding a point f ∈ ∩i Ai where the the sets Ai are dense and open. then A is eﬀectively Deﬁnition 7. which we use to construct sets and degrees meeting an countable sequence of “requirements. (iv) A space S is separable if it has a countable base of open sets. or equivalently if C = [ T ] for a computable tree T ⊆ 2<ω . then C is eﬀectively (computably) closed . We say that σ isolates f because Nσ ∩ [ T ] = { f }.) (v) A class B ⊆ S is Gδ . Let S be Cantor space 2ω or Baire space ω ω .6. This is the paradigm for the ﬁnite extension constructions in computability theory and oracle constructions as Kleene and Post. A point f ∈ [ T ] is isolated in [ T ] if (9) (∃σ)[ [ Tσ ] = { f } ].e. (i) (computably) open.e. 33 . 7. i. σ) [ f ∈ A ]. (Both Cantor space and Baire space are separable.

. Deﬁne R(σ) iﬀ σ ∈ T . A class C is a Π0 class iﬀ C is eﬀectively closed. Proof. Hence.8. A set A ⊂ 2ω is boldface Σ1 or ΣS if its complement 2ω − A is boldface Π1 or ΠS .We call this lightface Π0 because it is deﬁned in (10) by a universal quantiﬁer 1 outside of a computable relation R(x).e. i. Theorem 7. In advanced analysis or topology courses 34 .. We: (1) usually ﬁx the parameter A and do a construction which is computable relative to the parameter A. (=⇒). and so forth. If A is open. and we say C is in ΠS . if we ﬁx A as a parameter. (⇐=). and (3) replace the closed set by C = [ T A ] for a A-computable tree T . Let C = { f : (∀x) R(f (x)) } for R computable. C = [ T ] 1 for some computable tree T . Deﬁne a computable tree T = { σ : (∀τ ⊆ σ)[R(τ )] }. By ﬁxing the parameter A we can apply all the methods of this chapter on A-computable constructions.e. The open (closed) sets of 2ω are exactly the boldface Σ1 (Π1 ) sets and any boldface set is lightface in some parameter A. then A = NA for some countable set A.e. i. set B. Here RS denotes a relation computable in the set S 1 which we call the parameter determining RS . construction of an A-c. such as the Recursion Theorem.) We often convert an open set A = NA into the realm of computability theory as follows.7.5 Continuous Functions on Cantor Space In elementary calculus courses a continuous function is usually deﬁned with δ and concepts or with limits. 7. 1 1 Theorem 7. Let C = [ T ] for T a computable tree. Now A determines exactly which σ contribute in the countable union A = ∪σ∈A Nσ . Then {f : (∀x) R(f (x))} = [ T ]. (But this entire 1 section is about working relative to an oracle. Proof. eﬀectively open relative to the oracle A. Then [ T ] = C. then the deﬁnition and properties become lightface ΣA . (iv) A set C ⊆ 2ω is boldface Π1 if there is some set S ⊆ ω such that C = {f : (∀x) RS (f (x))}. (2) often replace the open set NA by its complement the closed set C = N A .

Theorem 7. although it is usually presented in its eﬀective analogue as the theorem that Φe is total iﬀ it is a truth-table reduction. 7. Deﬁnition 7. (12) 2ω ⊆ ∪ { Nσ : σ ∈ Xτ }.e. Note that this involves only compactness and has no computable content.τ ) = Xτ .. Therefore. if (∀A)(∃B)(∀x)[ ΨA (x) = B(x) ]. Assume Ψ is total. Now the Compactness Theorem asserts that any open cover of Cantor space 2ω has a ﬁnite subcover. (⇐=). 35 . By identifying string σy with code number y deﬁned in Deﬁnition 7. τ ) such that Wh(e. i. (=⇒). we can replace every set Xτ by a ﬁnite subset Xτ . Then Ψ is total iﬀ for each set Xτ of (11) there is a ﬁnite subset Xτ ⊆ Xτ such that. For arbitrary topological spaces an open set is an arbitrary union of basic open sets. For Φe deﬁne for every τ ∈ 2<ω the set of strings e (13) Xτ = { σ : (∃s)( ∀x < |τ | ) [ Φσ (x) ↓ = τ (x) ] }. (i) A functional Ψ on Cantor space 2ω is continuous if for every τ ∈ 2<ω there is a countable set Xτ such that (11) Ψ −1 (Nτ ) = ∪ { Nσ : σ ∈ Xτ }. and uniformly so in the sense that there is a computable e function h(e. but for Cantor space we may take countable unions. Theorem 7.e. Given such Xτ for every τ we see that Ψ is total by (12). Let Ψ be a continuous functional on Cantor Space 2ω . (ii) A functional Ψ on Cantor space 2ω is total if ΨA (x) is deﬁned for every A ⊆ ω and x ∈ ω.the more general deﬁnition is presented that a function F is continuous iﬀ the image of every basic open set is open. We state continuity now for functionals on the Cantor space 2ω . e.11.s e Then Xτ is c. Then for every τ there is a countable set Xτ which covers 2ω in the sense of (12). Proof.9.6 Eﬀectively Continuous Functionals A Turing functional Φe is not only continuous but eﬀectively continuous in the following sense.10.1 (iii) we can think of Xτ as a subset of ω.

8 8. if B ≤m A via a computable function f (x) then this reduction is bounded by h(x) = f (x) and x ∈ B iﬀ σ(f (x)) = 1 where σ A. that e Xτ is a computably enumerable set of strings.. Since Ψ is continuous.3 we study the more 36 . the set X = ⊕{Xτ : τ ∈ 2<ω } provides a complete oracle for calculating Ψ : 2ω → 2ω . 7. the inverse image of every basic open set Nτ . Given e h(x) imagine a matrix Mx whose rows are all the strings σ of length h(x). Now the set of e strings { Xτ : τ ∈ 2ω } witnesses that the functional Φe is continuous. ϕA (x) ≤ h(x). i. where σ is an initial segment of A.Proof. Proof. Now we prove that any continuous functional on 2ω is a Turing functional relative to some real parameter X ⊆ ω and therefore is eﬀectively continuous relative to X. The action on x is entirely determined by the action of Φσ (x) for these rows e σ ∈ Mx . We begin in §8..1 (iii) we deﬁned σy to be the string with canonical index y derived from Dy . Since Xτ is not only countable but also computably enumerable uniformly in e the functional Φe is eﬀectively continuous. only the matrix Mx . i.e. (identifying strings σ with their code numbers as integers) there is a set Xτ ⊆ ω such that. i.1 Bounded Reducibilities A Matrix Mx for Bounded Reducibilities A bounded reducibility is a Turing reducibility ΦA (x) with a computable e function h(x) which bounds the use function.13. i.e. Then Ψ is a Turing functional relative to some real parameter X ⊆ ω.e. Suppose Ψ is a continuous functional on 2ω . Theorem 7.. e Deﬁnition 7.12.2 with the most general bounded reducibility called bounded Turing reducibility (B ≤bT A) where we are given only the computable bound h(x).e. This gives a method for indexing the rows σy ∈ Mx . Therefore. Ψ−1 ( Nτ ) = ∪ { Nσ : σ ∈ Xτ }. is open and therefore is a countable union of basic open sets. indeed eﬀectively continuous. In §8. For example. Identify string σ with Nσ . Use the Oracle Graph Ge . τ ∈ 2<ω .7 Continuous Functions are Relatively Computable We showed that any Turing functional is continuous.. Hence. In Deﬁnition 7.

When A x has settled then B x has settled also. (ii) ΦA = B for some total ΦX that is (∀X)(∃Y )(∀x)[ ΦX (x) = Y (x) ]. More recently.e. (i) B ≤tt A as deﬁned in Deﬁnition 8. The bT and ibT reductions occur naturally in several parts of the subject.1.3 Truth-Table Reductions Deﬁnition 8. often we are given a noncomputable c.2. ibT has occurred in applications of computability to differential geometry in Soare [2004] and Csima-Soare [ta].2 and (14). e (ii) A set B is identity bounded Turing reducible to A (B ≤ibT A) if B ≤bT A with h(x) = x. and also e (14) (∀x)(∀σ)[ |σ| = h(x) =⇒ Φσ (x) ↓ e ].3 (iii).2 Bounded Turing Reducibility Deﬁnition 8. e e e 37 . 8. nitions of tt-reducible are equivalent.restrictive case of truth-table reducibility (B ≤tt A) where Φσ (x) converges e for every x and every row σ ∈ Mx . (i) A set B is bounded Turing reducible (bT-reducible) to a set A (B ≤bT A) if there is a Turing reduction ΦA = B and a computable e function h(x) such that the use ϕA (x) ≤ h(x). This convergence on all strings σy ∈ Mx is the key deﬁning property for a tt-reduction as expressed in (14). Barmpalias and Lewis [ta] have shown the nondensity of the ibT -degrees of c. For example. Nerode]. as described later. 8. set A and we construct a simple set B ≤T A by simple permitting where an element x is allowed to enter B at stage some stage only if some y ≤ x has just entered A. The following deﬁ- Theorem 8. and ibT reducibility has also been used in applications to algorithmic randomness and Kolmogorov complexity.1. A set B is truth-table reducible to set a A (B ≤tt A) if B ≤bT A via ΦA = B and h(x) as in Deﬁnition 8.3. If Φσ (x) diverges for even one x and e σ ∈ Mx then the reduction is a partial bT-reduction. [Truth-Table Theorem. but if it converges for every x and every σ ∈ Mx then it gives a genuine truth-table as in beginning logic courses as deﬁned in Theorem 8.e. sets. and hence B ≤ibT A.

and (iii) ⇐⇒ (iv) are obvious. §6] introduced the ﬁrst deﬁnition of B ≤tt A. x ∈ B ⇐⇒ (∃z ∈ Dg(x ) [ A (h(x) + 1) = Dz ]. It remains to prove (ii) =⇒ (iii). Proof. The implications (i) =⇒ (ii). ﬁlling in all columns / less than h(x).2.(iii) B ≤tt A via ΦA and h(x) as in Deﬁnition 8. e Since Φe is total apply the Compactness Theorem 7. .1. (iii) =⇒ (i). there is a e computable function g(x) such that Dg(x) = { y : |σy | = h(x) & Φe y (x) = 1 }. . In addition. Uniformly in x we can eﬀectively enumerate the set Ux = { σ : Φσ (x) ↓ }. Deﬁne h(x) and g(x) by h(x) = max{ |σ| : σ ∈ Vx }. σ Remark 8. and replacing + by 1 and − by 0 we have exactly the matrix Mx described in §8.2 above where we could attach to matrix Mx an extra column with the value Φσ (x) on row σ.4 (iv) to get a ﬁnite subset Vx ⊆ Ux such that ∪{ Nσ : σ ∈ Vx } ⊇ 2ω . Post’s extra column converted the matrix Mx into a total reduction because A had to extend exactly one of the rows Rz and then the value for B(x) was completely speciﬁed by vz given in advance. (ii) =⇒ (iii). Letting h(x) be the maximum of the {mi }i≤k . Dg(x) = { y : |σy | = h(x) & Φe y (x) = 1 }. and m1 . Post [1944. Post’s original deﬁnition [1944] of B ≤tt A is virtually identical in intuition and description to our Deﬁnition 8. He then drew a diagram of a matrix with columns m1 . Property (iv) of B ≤tt A has been used in the 38 . . mk . . In both cases the e crucial point is that Φσ (x) is deﬁned for every string of length h(x) before e examining the oracle A. Next Post drew a vertical line to the right of this matrix and added an extra column such that an entry vz in this column following row Rz indicated that if A satisﬁes row Rz then x ∈ B iﬀ vz = +. m2 . Therefore. σ (iv) There exist computable functions g and h such that. For every integer n he required a eﬀective procedure to produce integers k. . for all x.4. . mk and 2k rows with individual entries of + for x ∈ A and − for x ∈ A.

The concept of tt-reducibility is tied to the totality of the reduction ΦA which is exactly what bT-reducibility lacks.e. e Also bT is more recognizable than wtt whose meaning cannot be guessed from its name. in 1959 Friedberg and Rogers introduced wtt-reducibility (B ≤wtt A) as a weakening of tt-reducibility which was already a strengthening of ≤T .e. Now consider special cases based upon how many times the approximation changes on x. Theorem 8.. sets are the usual c.e.5. which has the same deﬁnition as wtt. n-c. and the 2-c.past e. sets. (15) A = lims As & | { s : As (x) = As+1 (x) } | ≤ g(x).e. Thus. The next theorem gives an elegant characterization of ω-c.. The concept of bounded Turing reducibility (bT).e. sets are the d.) if there is a computable sequence {As }s∈ω with A0 = ∅ and a computable function g(x) which bounds the number of changes in the approximation {As }s∈ω in the following sense.e. (iii) For n ∈ ω the set A is n-c.e. (iii) A ≤tt ∅ . sets The Limit Lemma characterized sets A ≤T ∅ as those such that A = lims As for a computable sequence {As }s∈ω . (i) A ≤bT ∅ . ≤bT has distance one from the main reducibility of ≤T . Therefore. sets. sets. and ω-c.e.e. (written ω-c. the 1-c. set is ∅. Deﬁnition 8.e. sets A but not the most general case of A ≤T ∅ . The following are equivalent.e. sets (d.e.g. 83] and is short and slick but is does not give the necessary elegance and intuition for beginning students. Soare [1987.e. sets. sets.e.e.6. if g(x) ≤ n. the concept of wtt has distance two from the central concept ≤T . 8. These notions are more general than c. Turing reducibility was not well understood in 1944 and an understanding of it in its modern state emerged only gradually during the 1940’s and 1950’s. (ii) The set A is omega-c. (i) A set D is the diﬀerence of c. Post’s tt-reducibility in 1944 was understood at once.c. (ii) A is ω-c. goes immediately to the main concept of bounding the queried information by a computable function h(x). p. the only 0-c.e. 39 . For example.) if D = A − B where A and B are c.c. Therefore.e.4 Diﬀerence of c.

For a given x the truth-table has rows of width g(x) as in Deﬁnition 8. Now As+1 (x) = As (x) only if some element z ≤ g(x) enters ∅ which can happen at most g(x) + 1 times.e.Proof. The Turing o-machine is a better model to study them because the Turing a-machine lacks this online capacity. (i) =⇒ (ii). because the approximation changes between 0 and 1 starting with 0. First build a truth-table to demonstrate that A ≤tt C. we can make up a row in the truth-table which gives value for A(x) based only on the number of changes. Next ﬁnd the row with k many 1’s. Now A(x) = 1 iﬀ k is odd. g(x)]. Clearly. A ≤tt C. 1} rather than in ω. the assumption on h(x) ensures that A ≤bT D. However. This ensures that A ≤tt D. (iii) implies (i). Now to tt-compute from C whether x ∈ A compute k = |C x |. This result shows the diﬀerence between T -reductions and tt-reductions. 9 Online Computing The original implementations of computing devices were generally oﬄine devices such as calculators or batch processing devices. (Intuitively. First. on row ω [x] . C ≤m 0 and therefore A ≤tt 0 . whenever As (x) = As+1 (x) we put into C the least element of the form k. (ii) =⇒ (iii). Deﬁne the c. Hence. |C [x] | is the number of changes on x during the approximation. x .. i. Hence. Let e e As (x) = Φ∅ (x)[ s ] where we may speed up the computation so the latter is e always deﬁned. Assume A = lims As satisﬁes (15) via g(x). once for each z ∈ [0. the approximation always begins with As (x) = 0 for s = 0 and changes between 0 and 1 because all values are in {0. Therefore. Furthermore. in recent years the implementations have been increasingly online computing devices which can access or interact with some external database or other device. Suppose Φ∅ = A with computable bound g(x) ≥ ϕ∅ (x). 40 . not already in C. x : 1 < k ≤ | {s : As (x) = As+1 (x)} | }. Second. However. For each k ≤ g(x) construct a row beginning with k many 1 s followed by all 0 s. A(x) = 1 iﬀ |C [x] | is odd.2 (ii).) Therefore. This case is unusual since most reductions we consider only produce A ≤T B for some set B and sometimes produce A ≤bT B as in permitting arguments. change set C by C={ k.e. and |C [x] | ≤ g(x) by hypothesis (ii).e. C is c.

” About the chapter. A Theory of Interactive Computation by Jan van Leeuwen and Jiri Wiedermann. ﬁnitary models of computation.” 41 . who fed them to the computer and produced paper output later.” It appears that the Turing o-machine is a good theoretical model to analyze an interactive process because there is usually a ﬁxed algorithm or procedure at the core. which when coded into integers may be regarded as a Turing type oracle. “In this chapter. evolving programs should look like.1. not interacting with any external device. A laptop obtaining data from the World Wide Web is a good example. In the real world the database would not be literally inﬁnite but may appear so (like the web) and is usually too large to be conveniently downloaded into the local computer. and proceeds internally. (ii) An oﬄine computing process is one which begins with a program and input data. There are many descriptions in the computing literature about online and interactive processes. In real world computing the oracle may be called a database or an environment. and batch processing devices where a user handed a deck of punched IBM cards to an operator. The authors point out that a theory of interactive computation must necessarily lead beyond the classical. A simple model of interactive computing is presented consisting of one component and an environment. Sometimes the local computer is called the “client” and the remote device the “server. and broadens the notion of algorithms to include interaction by allowing intra-step interaction of an algorithm with its environment. This includes a calculator. which by Turing’s thesis we can identify with a Turing a-machine. for example a computer communicating with an external data base such at the World Wide Web. (i) An online or interactive computing process is one which interacts with its environment. Under the Post-Turing Thesis 6. Interactive Algorithms 2005 is described.1 these real world online or interactive processes can be described by a Turing oracle machine.Deﬁnition 9. interacting using single streams of input and output signals. and there is a mechanism for the process to communicate with its environment. In [Goldin-Smolka-Wegner] a chapter by Yuri Gurevich. Gurevich asserts that computer science is largely about algorithms. “This chapter asks what a computational theory of interactive. the book states.

9. An o-machine has ﬁxed program but has the capacity to interact with its environment and receive new data during its computation. p. 1947.2 Trial and Error Computing We expect Turing a-machines and o-machines to be absolutely correct. but let us now examine the reverse direction. • A large number and rapidly increasing number of computing processes in the real world are online or interactive. • A large number of books presenting an introduction to computability mention Turing oracle machines and relative computability very late in the book or not at all.g. there are many computing processes in the real world which give a sequence of approximations to the ﬁnal answer. or only the ChurchTuring Thesis. • A large number of books with articles on Turing and the ChurchTuring Thesis do not mention Turing oracle machines or relative computability at all. However. • The original Turing a-machines are not online. In his talk to the London Mathematical Society. 9. Instead of it sometimes giving no answer we could arrange that it gives 42 . that o-machines are online while a-machines are not. February 20. Discussing only Turing a-machines in modern texts. See the authors in [GoldinSmolka-Wegner] for only a few. and not the Post-Turing Thesis on oracle computers. “I would say that fair play must be given to the machine. The following points appear self-evident. is like discussing only batch processing machines of the 1950’s long after the emergence of online computing. A Turing a-machine. Christof Teuscher [2004]. e. even a universal machine. • Turing oracle machines are online. begins with a ﬁxed program and ﬁxed input and proceeds without further outside input until (if ever) it halts. quoted in Hodges.. 360–361. Turing considered machines which make mistakes.1 Turing Machines and Online Processes We could continue analyzing to what extent Turing oracle machines can model modern online processes. Turing said.

3 The Limit Lemma There are several diﬀerent approaches for computing with ﬁnite errors. equivalent. 1959]. a ﬁnancial trader receiving information from around the world updated every second.2. A computable sequence of computable sets {Bs }s∈ω is a computable (∆0 ) approximating sequence for a set B if B = lims Bs . This is a model for many processes in the real world which allow ﬁnitely many mistakes but gradually move closer to the correct answer. (In practice. In other words if a machine is expected to be infallible. Deﬁnition 9. Proof. It is easy for us to regard these blunders as not counting and give him another chance. (ii) B is limit computable. Lemma 9. (See Soare [1987] or Soare [CTA] Lemma 3. B is is both two quantiﬁer forms. 9. that is. [Limit Lemma. But the human mathematician would likewise make blunders when trying out new techniques. a meteorologist predicting the weather a week from now given constantly updated weather conditions today. “oracle set” A. but the machine would probably be allowed no mercy.) Now in the real world imagine an example of a computing process with error such as: a robot learning a maze.occasional wrong answers. But these theorems say nothing about how much intelligence may be displayed if a machine makes no pretence at infallibility.6. (i) B ≤T A for some c. There are several theorems which say exactly that.e. Shoenﬁeld. Hillary Putnam [1965] described trial and error predicates as ones for which there is a computable function which arrives at the correct answer after a ﬁnite number of mistakes. In modern terminology this is called a limit computable function as described in Soare [1987] or Soare [CTA] Chapter 3.3. If there is such a sequence we say that B is limit computable. In our idealized model we assume that the individual makes ﬁnitely many errors during the process {Bt }t∈T for time periods t ∈ T but eventually gets the correct answer.3. it cannot also be intelligent. (iii) B ∈ ∆2 . the ﬁnal answer may not be 43 The following are .

) 9.1 Two Models for Computing With Error The Limit Computable Model In the limit computable or approximation model we have a sequence of Turing programs {Pt : t ∈ T } so that Pt computes function gt at time t ∈ T . set A (or even a ∆0 set) and oracle 2 machine Φe such that B = ΦA .4 9.2 The Online Model By the Limit Lemma there is a c.4. The conﬁguration at his trading desk may be described using the Limit Lemma by a computable function where gt is the computable characteristic function of Bt .exactly correct in these examples. and other approximations in real time. the conﬁguration of his computation at the end of time t. meteorology. and computes the approximation Bt (x) = ΦAt (x). Now the trader can program the algorithm e Φe into his laptop once and for all at the start of the trading day. How will the trader write a program to uniformly compute the index gt for t ∈ T ? 9.4. He does not (and cannot) change the program Φe every second.e. There is not necessarily any connection between diﬀerent programs and we may have to compute all over again with a new program as we pass from time t to t + 1. even if not exact. The computable function gt gives an algorithm to compute the condition Bt at time t but it gives no relationship between Bt and Bt+1 . are usually useful in ﬁnancial trading. Milan. Suppose the ﬁnancial trader in Chicago receives data every second t ∈ T about currency prices in London. The sequences of improving approximations. It is diﬃcult to see how this trader could have carried out his business using a batch processing. 44 . instead of an online model. His algorithm simply receives the “oracle” information At from the internet as it is continually updated. Turing a-machine model. His program e then executes a trade when the algorithm determines that conditions are favorable. New York and Tokyo. It will not be possible for the trader to write a new program every second. Every second t ∈ T he receives from New York and abroad the latest quotes At which enter directly into his computer by an internet connection. but is presumably more accurate than the ﬁrst approximation B0 .

Kleene later o introduced the term “recursive function theory” for the subject although G¨del disagreed (see below). often when replaced by something else. theses of many prominent researchers in the ﬁeld. Kleene with some additional inﬂuence by Emil Post as described earlier. three of its leaders. perhaps by accident. “Dedicated to S. Turing.” After 1996 the term the term ‘recursive” was again used only to mean “inductive” not “computable” or 45 . When computabiliy theory originated in the 1930’s it was a very small ﬁeld attempting to consolidate its ideas. In dedicating his book Degrees of Unsolvability [1971] Shoenﬁeld recognized Kleene’s overwhelming inﬂuence. and Church eﬀectively left the ﬁeld after 1940 and had little o direct inﬂuence thereafter on its development. simply acting from the exigencies of the situation at the time. There are three important issues in computability which have at one time been displaced from their correct or proper positions as evaluated by historical and scientiﬁc criteria. These issues are at the very heart of the subject and they deﬁne how we think about computability. that were not consistent with the original development of computability in the 1930’s and would not have been approved of by G¨del and Turing.D. However. G¨del. We now examine these three issues one by one in the next three sections. o 11 “Computable” versus “Recursive” Starting in 1936. but are not consistent with a careful scientiﬁc analysis later. although Church supervised the Ph. Furthermore. These items may have been displaced accidentally or without conscious thought or decision. a number of changes took place after 1940. After 1940 the ﬁeld was developed and promulgated primarily by Stephen C. Church and Kleene used the term “recursive” to mean “computable” even though Turing and G¨del later objected.C.10 Three Displacements in Computability Theory The dictionary deﬁnes “displacement” to be the moving of something from its rightful place or position.” This is entirely accurate and without Kleene’s leadership we would not have the ﬁeld as we know it today. This was the ﬁrst time in history that the o term “recursive” which had meant roughly “inductive” acquired the additional meaning “computable” of “calculable. Kleene who made recursive function theory into a theory.

Church and Kleene knew almost immediately and published by 1936 the proof of the formal equivalence of recursive functions with λ-deﬁnable functions.” Church had seen his ﬁrst thesis rejected by G¨del o and was heavily invested in the acceptance of his 1936 thesis in terms of 46 .” The irony is that the term “computable” was there all along and was preferred by Turing and G¨del. Church’s Thesis 2. perhaps unduly inﬂuenced by rather chilly receptions from audiences around 1933–35 to disquisitions on λ-deﬁnability. Indeed Church seems to have preferred the λ-deﬁnable functions and caused Turing to write his thesis [1939] in that formalism. . Church and Kleene used the second version of Church’s Thesis above. Ironically. after general recursiveness had appeared. ” Church and Kleene were simply doing what most scientists do. . Church was very eager for mathematicians to accept his thesis and he know that the recursive functions were more familiar to a mathematical audience than λ-deﬁnable ones.” From 1931 to 1934 Church and Kleene had used the λ-deﬁnable functions as the formal equivalent of eﬀectively calculable functions. o 11. .1 (First Version) [1934] “A function is eﬀectively calculable if and only if it is λ-deﬁnable. arrange the work in a framework which will be understandable and appealing to as large a scientiﬁc audience as possible. this is exactly what caused the change in 1996 from “recursive” back to “computable” because in 1996 the term “computable” was much better understood by a general audience than “recursive. phrased in terms of recursive functions primarily for public relations as Kleene [1981] explained. o 11. and Church had ﬁrst proposed his thesis privately to G¨del in that form. After seeing G¨del’s lectures in 1934 Church and Kleene dropped o the λ-deﬁnable functions and adopted the recursive functions. chose.2 Church and Kleene Deﬁne “Recursive” as “Computable” By 1936 Kleene and Church had begun thinking of the word “recursive” to mean “computable. to put my work in that format.” When G¨del strongly rejected to o this thesis Church turned instead to G¨del’s own Herbrand-G¨del general o o recursive functions as a formalism and proposed in [1935] and [1936] the well-known form of his thesis.2 [1936] A function on the positive integers is eﬀectively calculable if and only if it is recursive.1 Church Defends Church’s Thesis with “Recursive” Recall Church’s Thesis 2.“calculable. “I myself. This was not because of the inadequacy of λ-deﬁnable functions in comparison to the recursive functions.

Martin Davis casually used the term “recursive function theory” as it was used then. under the inﬂuence of Church and Kleene. In a discussion with G¨del at the Institute for Advanced Study in o Princeton about 1952–54.recursive functions. p. Church wrote in [1936. a computable function). This is apparently the ﬁrst appearance of the term “recursively enumerable” in the literature and the ﬁrst appearance of “recursively” as an adverb meaning “eﬀectively” or “computably.” (See Peter’s work on recursions in [1934] and[1951].” In the same year Kleene [1936.” Thereafter.” “normal set. 11.” as if I were writing in Chinese and translating back into English. as related by o Martin Davis.3 G¨del Rejects “Recursive Function Theory” o Neither Turing nor G¨del ever used the word “recursive” to mean “como putable. “To my surprise. adopted this terminology of “recursive” and “recursively enumerable” over his own terminology [1943]. Most people access to a personal computer on their desks and the terms of computing were familiar to the general population but “recursive” was limited to very small number who mainly associated it with a ﬁrst year programming course or a deﬁnition by induction on mathematics and almost never with computability. saying that the term in o question should be used with reference to the kind of work Rosza Peter did. it was ﬁrmly established. Without the acceptance of this thesis Church had no unsolvable problem. “Let f be a recursive function (that is. By a “recursive enumeration” Kleene states that he means “a recursive sequence (i. 238] cited in Davis [1965] [p. So few people understood the meaning of “recursive” that by 1990 I had to begin my papers with.” “generated set. Davis related.” Post [1944].” G¨del never used the term “recursive function theory” to name o the subject.. the successive values of a recursive function of one variable). p. 96] printed in Davis [1965] that a “recursively enumerable set” is one which is the range of a recursive function. [1944] of “effectively generated set.e. 47 . 238] mentioned a “recursive enumeration” and noted that there is no recursive enumeration of Herbrand-G¨del systems of equations which gives only the o systems which deﬁne the (total) recursive functions. when others did G¨del reacted sharply negatively. G¨del reacted sharply.) By 1990 the situation had become very diﬃcult.

11.S.” When one uses a term like “recursive” to also mean “computable” or “algorithmic” as Kleene did. Turing and G¨del. most researchers.” Did he mean that recursion and inductive deﬁnition began with Dedekind or that computability and algorithms began there? The latter would contradict his several other statements.” We already have one. Colorado referenced in Soare [2000]. Three years later at the A. p. Changing back from from “recursive” to “computable” during 1996–1999 has had a number of advantages. 44] wrote.11. had adopted the new terminology and conventions. Kleene [1981b.C. overturning a sixty year old tradition of Kleene. the Greeks and other early civilizations. in a few months more people were convinced by the undeniable logic of the situation. “I think we can say that recursive function theory was born there ninety-two years ago with Dedekind’s Theorem 126 (‘Satz der Deﬁnition durch Induktion’) that functions can be deﬁned by primitive recursion. The founders of the two key models of computabilty. I wrote an article on Computability and Recursion for the Bulletin of Symbolic Logic [1996] on the history and scientiﬁc reasons for why we should use “computable” and not “recursive” to mean “calculable.). We do not need another word to mean “computable.M. p. especially those under forty years old. then one is never sure whether a particular instance means “calculable” or “inductive” and our language has become indistinct.) set” did not come tripping from the lips. The object of 48 . such as Kleene [1988.5 Changing “Recursive” Back to “Inductive” By 1996 the confusion had become intolerable. had never used “recursive” to mean como putable and indeed had objected when it was so used. conference in Boulder.4 The Ambiguity in the Term “Recursive” The traditional meaning of “recursive” as “inductive” led to ambiguity.e. However. Kleene often wrote of calculation and algorithms dating back to the Babylonians. Historical Accuracy. 19] where he wrote. and the words “computability theory” and “computably enumerable (c. At ﬁrst few were willing to make such a dramatic change. “The recognition of algorithms goes back at least to Euclid (c. However. 330 B. Returning “recursive” to its original meaning of “inductive” has made its use much clearer.” “Recursive” should mean “inductive” as is had for Dedekind and Hilbert.

Names matter today as we try to relate our specialty of computability to the world of computers and algorithmic procedures all around us. while o at ﬁrst G¨del’s own model of Herbrand-G¨del recursive functions did o o not. not in an attempt to better understand the nature of recursion and inductive deﬁnition. Scientiﬁc Accuracy. The object was never to understand the notion of recursions or inductive deﬁnitions. Suppose a student is scanning a catalogue for a course to take and sees the course title “Recursive Function Theory” versus “Computability Theory. the former being a bit more general.” to run back. Name Recognition. To the extent that they have any idea about “recursive” they understand it in this context. In 1935 Church adopted G¨del’s recursive o functions as a deﬁnition of eﬀectively calculable before seeing Turing machines As Kleene relates. apply for jobs under the general area of his work as the ﬁrst or second? Should a professor write an abstract for his lecture at another university under the ﬁrst title or second? Names do matter.everyone was to formally capture the informal concept of “eﬀectively calculable” which Turing machines did to G¨del’s satisfaction. a ﬁrst year programming course speaks of deﬁnition by iteration versus deﬁnition by recursion. The word has nothing to do with “calculate” or “compute. 49 . They mattered to Church and Kleene in 1936 when they changed from the term “λ-deﬁnable function” to “recursive function” to achieve greater name recognition among mathematicians and to make Church’s Thesis more convincing before the appearance of Turing.” The general public understands the ﬁrst two words in this context. The word “recursive” means a procedure characterized by recurrence or repetition from the Latin verb “recurrere. a world partially created by Turing. For example.D.” Which will give him more information about the content of the course? Should a fresh Ph. The words “calculate” and “compute” are very close in the dictionary. Church and Kleene did this as a matter of public relations to relate to a concept mathematicians could understand.

317] began a new section §62 and called it “Church’s thesis” instead of “Thesis I” as he had been doing. “This thesis is also implicit in the conception of a computing machine formulated by Turing 1936–7 and Post 1936.” Apparently he did not feel it necessary to include Turing’s name when he used the term “Church’s thesis. p. Kleene was aware of other similar “theses” advanced nearly simultaneously and he continued [1952. “This heuristic evidence and other considerations led Church 1936 to propose the following thesis. 300] took up where Kleene [1943] had left oﬀ.” Next Kleene begins a subtle shift of terminology from “Thesis I” to “Church’s thesis.” Kleene [1952. not putting forth a “thesis. p. Thesis I.” Kleene wrote [1952.” This is identical with what we previously called Church’s Thesis 2.2 Kleene Named it “Church’s thesis” in [1952] Later in his very inﬂuential book [1952] it is fascinating to see how Kleene’s thinking and terminology progressed from “Thesis I” to “Church’s Thesis” and not to “Turing’s Thesis” or the “Church-Turing Thesis. p.” These were not even called “theses” at all until Kleene [1943. Kleene wrote. Church’s thesis. “ §62. One of the main objectives of this and the next chapter is to present the evidence for Church’s thesis (Thesis I) §60.300].” 12. 60] referred to Church’s “deﬁnition” as “Thesis I. Every eﬀectively calculable function (eﬀectively decidable predicate) is general recursive.” 50 .2. p.12 12. Of course.1 Renaming it the “Computability Thesis?” Kleene Called it “Thesis I” in [1943] In the 1930’s both Church and Turing thought they were giving deﬁnitions of an eﬀectively calculable function.

These arguments have been cited in hundreds of computability books and papers for the last half century. the “(B) Equivalence of diverse formulations:” such as the (Herbrand-G¨del) general recursive functions. .” 12. working according to preassigned instructions. .3 Kleene Dropped “Thesis I” for “Church’s thesis” As Kleene progressed through [1952. Ironically. pp. .” Turing’s computable functions [1936–7] are those which can be computed by a machine of a kind which is designed. p. we take to be already conﬁrmed by the intuitive notion (cf. §62..4 Evidence for the Computability Thesis In one of the most familiar parts of the book. After some heuristic evidence in (A). ” “The converse of Church’s thesis. 318–319]. ” “While we cannot prove Church’s thesis. “ Church’s thesis. to clinch the evidence for Church’s thesis Kleene began a new part (C) where he appealed to Turing’s work and wrote. . 319]. that every general recursive function ϕ is eﬀectively calculable. Kleene summarized the evidence for “Church’s thesis” in [1952. Turing’s formulation hence constitutes an independent statement of Church’s thesis. and systems of Post [1943] and [1946] of canonical and normal systems. Turing’s notion is thus the result of a direct attempt to formulate mathematically the notion of eﬀective calculability. . §60). while other notions arose diﬀerently and were afterwards identiﬁed with eﬀective calculability. since its role is to delimit precisely an hitherto vaguely conceived totality. 318–319] he dropped any reference to “Thesis I” and used only “Church’s thesis” with no mention of Turing or Post in the thesis. by supplying a precise delimitation of all effectively calculable functions.” 51 . i. to reproduce all sorts of operations which a human computer could perform. Turing computable functions. §62.e. we require evidence that it cannot conﬂict with the intuitive notion which it is supposed to complete. He wrote [1952. according to his analysis.12. λ-deﬁnable funco tions. p. Kleene presented perhaps the most powerful evidence. “(C) Turing’s Concept of a Computing Machine. .

” Deﬁnition 12. or deﬁned by any of the other formalisms. 12.” Kleene omits Turing’s name from the thesis even though in part (C) above he gave Turing credit as the only one who made a “direct attempt to formulate mathematically the notion of eﬀective calculability.2. It does not refer to one single man or to one single formalism. A formal mathematical deﬁnition for eﬀectively calculable functions. is usually listed under the name “Church’s Thesis” as in the title of the new book [Olszewski. As an intensional claim it had nothing to do with the recursive functions of Church’s Thesis 2. From now on when Kleene uses the term “Church’s thesis” he means the following “Computability Thesis.2. It was only the equivalence of Turing computable functions ﬁrst with λ-deﬁnable functions and hence with recursive functions which links them extensionally but not intensionally. Many people today use “Church’s Thesis” in this way.1.We see here the amalgamation of diﬀerent meanings into a single term (just as for “recursive” above). The Formalism. A function on the positive integers is eﬀectively calculable if and only if it is recursive or Turing computable.5 Who First Demonstrated the Computability Thesis? Demonstrating something like the Computability Thesis requires two steps. Turing’s work was to establish the connection between eﬀectively calculable functions and Turing computable functions in Turing’s Thesis 3. It is exactly here that Kleene introduces (without explicit mention) another convention and term which has lasted until today. and he regards them interchangeably. Kleene here was using the term “Church’s thesis” to include Turing’s Thesis and Turing’s justiﬁcation.2. [The Computability Thesis Emerges]. 2007] with no mention of Turing. Note that the Computability Thesis is essentially the same as what we have called the Church-Turing Thesis 3.1. Turing invented the new formalism of Turing machines because this was closest to his idea of a mechanical process and because it lent itself to proving that any eﬀectively calculable function lay in this class. 52 . The Computability Thesis. [The Computability Thesis]. Church [1936] proposed G¨del’s general recursive funco tions [1934]. regardless of their intensional or historical import. Remark 12. Kleene knows that the various formal deﬁnitions all coincide extensionally. used extensively in the subject.

at least intensionally.” Turing’s Thesis [1936. §7] for his thesis was this. was not convinced as o we have seen in §2 and §3. “Turing’s analysis makes no reference whatsoever to calculating machines. The ﬂaw in Church’s argument [1936. 82] observed. but surely Gandy meant Turing’s Thesis. . Turing machines appear as a result. . G¨del had shown that the o steps in his formal system P were primitive recursive. because Turing did not prove anything in [1936] or anywhere else about general recursive functions. Deﬁne c0 . .. 53 .1 is that every intuitively computable (eﬀectively calculable) function is computable by a Turing machine.’ it proves a theorem. . 64] or Shoenﬁeld [1967. . because as he pointed out. No modern book uses the H-G general recursive formalism to deﬁne the eﬀectively calculable functions. Gandy [1988. c0 . . pcn .” Church analyzed the informal notion of the calculation of a value f (n) = m according to a step-by-step approach (so called by Gandy [1988. pp. p. . cn = pc0 · pc1 . 0 ≤ i < n. 83–84] pointed out. 3 Gandy actually wrote “Church’s thesis” not “Turing’s thesis” as written here. We say that the calculation is stepwise n 0 1 recursive if there is a partial recursive function ψ such that ψ( c0 . of his analysis of calculations by humans. The second and perhaps more important step is to prove that the informal notion of a person calculating a function can be simulated by the formal model. and that the expressions have been given code numbers. 77]) from two points of view. the computability thesis. ci ) = ci+1 for all i. . Church used the Herbrand-G¨del general recursive functions o as his formal model but even their inventor. this second step is so crucial that it has been referred to as a “theorem” by Gandy and others. a codiﬁcation. cn . p. . and second as the derivation in some formal system.” Although it is not a formal proof. ﬁrst by an application of an algorithm. p. . Church began by deﬁning an “eﬀectively calculable” function to be one for which “there exists an algorithm for the calculation of its values. pp. c1 . as Gandy [1988. Following Davis [1958. “Turing’s analysis does much more than provide an argument for ‘Turing’s Thesis.”3 Furthermore. A considerably more serious objection is that there was a ﬂaw in Church’s demonstration that every eﬀectively calculable function is general recursive.The Demonstration. G¨del. §9] stated in Deﬁnition 3. We refer to this step as a “demonstration” rather than a formal “proof. This step is not a purely mathematical one but it needs to be as convincing and logical as possible. In contrast. 120–121] it is reasonable to suppose that the calculation of f proceeds by writing expressions on a sheet of paper. i. . .e. c1 .

There was a great controversy about priority up until Leibniz’s death in 1716. that the entire o process is µ-recursive.If the basic steps are stepwise recursive. Gandy [1988. for example by Church. 87]. The fatal weakness in Church’s argument was the core assumption that the atomic steps were stepwise recursive. Newton did not publish it until 1687 although most believe he had been working on it before Leibniz began in 1674. 79] and especially Sieg [1994. Sieg [p. . The British Royal Society handed down a verdict in 1715 crediting Isaac Newton with the discovery of the calculus. . Gottfried Leibniz began work on the calculus in 1674 and published his account of the diﬀerential calculus in 1684 and the integral calculus in 1686. in their excellent analyses. 57]).” who wrote that without this assumption it is diﬃcult to see how the notion of a system of logic can be given any exact meaning at all. 78] wrote. brought out this weakness in Church’s argument. 12.” Sieg [p. It is exactly this stumbling block which Turing overcame by a totally new approach.6 The Computability Thesis and the Calculus Why all the fuss over names? Why not simply use the term “Church’s Thesis” invented by Kleene [1952]. mathematics throughout the eighteenth century was typiﬁed by an elaboration of the diﬀerential and integral calculus in which mathematicians generally discarded Newton’s ﬂuxional calculus in favor of the new methods presented by Leibniz. 80] wrote. “Despite this ruling of the Royal Society. whose name should it be? Isaac Newton began working on a form of the calculus in 1666 but did not publish it until much later.” 54 . something he did not justify. 1935 (see Kleene [1987b. and let it refer to the “Computability Thesis?” This is in fact what is widely (and incorrectly) done. Newton was more established than Leibniz and had vigorous supporters. The Wiki encylopedia wrote. Newton and his followers campaigned vigorously for his position. “. and that stumbling block was quite clearly seen by Church. this core does not provide a convincing analysis: steps taken in a calculus must be of a restricted character and they are assumed. p. If we had to attach a single name to “the calculus” every time we mentioned it. without argument to be recursive. 80. and stating that Leibniz was guilty of plagarism (although these charges were later proved false). “It is precisely here that we encounter the major stumbling block for Church’s analysis. then it follows easily by the Kleene Normal Form Theorem which Kleene had proved and communicated to G¨del before November. pp. p.

Leibnitz began later than Newton. “mathematicians generally discarded Newton’s ﬂuxional calculus in favor of the new methods presented by Liebniz.2. The general recursive functions are used only in an historical discussion. The Kleene o µ-recursive functions have other uses but are were not mentioned in the original Church’s Thesis 2.12.” 5.7 Founders of Computability and the Calculus There is a parallel between the development of the Calculus and the demonstration of the Computability Thesis. (This refers to the Herbrand-G¨del general recursive functions. Both Leibniz and Turing worked at a distance and independently of Newton and Church.D. Church’s o claim to the Computability Thesis was aﬃrmed by his former Ph. The general recursive functions are never seen. despite the Royal Society ruling. For the calculus. 3. Newton was in the position of power with the Royal Society which not only aﬃrmed his claim but denied the claim by his rival Leibniz. 1.) 4. Turing began slightly later than Church. Church and Turing both worked independently on it and came up with diﬀerent models. respectively. Kleene deliberately and overwhelmingly established 55 . the Turing machines (and other calculating machines like register machines) dominate. However. Kleene’s µ-recursive functions are sometimes used in courses and mistakenly called recursive functions but to prove that an eﬀectively calculable function is µ-recursive requires a tedious arithmetization as in G¨del’s incompleteness theorem [1931] and is o virtually never done. 2. student Stephen Kleene [1952]. unaware of the work by another researcher. Church was the senior ﬁgure in computability (after G¨del). Kleene occasionally mentioned Turing’s Thesis and repeatedly used the Turing machine model and Turing’s demonstration of its success to demonstrate the Computability Thesis. In computer science. Turing’s model of Turing machines is overwhelmingly more appealing and popular that Church’s model of the (Herbrand-G¨del) general o recursive functions. 6. a model which is rarely presented in any books and never used for actual calculations in any courses.

” “Church’s Thesis” means the “Computability Thesis. words that are given arbitrary and nonstandard meaning by the author: “recursive” means “computable. It would also be wrong to refer to “the Newton Calculus” without mentioning Leibniz. It is ambiguous to use “recursive” with both meanings. Remark 12. Kleene just does not include Turing in the name he chose. but Kleene was certainly successful in excluding Turing’s name from the Computability Thesis. The problem is simply that Kleene has not given Turing credit in his naming of the Computability Thesis. “the Computability Thesis” and not “Church’s Thesis. Unlike the calculus. ” at a time when nobody called it a thesis and when the ﬁeld was about to rapidly expand in the 1950’s and 1960’s and would turn to Kleene for direction as the last representative of the original computability researchers of the 1930’s and the most prominent and inﬂuential of all computability theorists in the 1950’s. Kleene thought and wrote with tokens. Why not call it simply.” Why do we need to give a person’s name to the Computability Thesis? Kleene never denied credit to Turing and in many places such as his books [1952] and [1967] he gives Turing credit for the most intuitive presentation of computability.3. Kleene could have called it “the Computability Thesis” analogously with “the calculus. inductive and computable. it is simply wrong to use “Church’s Thesis” to refer to a proposition ﬁrst demonstrated by Turing and never successfully demonstrated by Church. and their followers. Virtually all the papers and books including Rogers [1967] and Soare [1987] and many others followed Kleene’s lead. have given credit to the others. The Royal Society was not successful in excluding Leibniz from the calculus.” We never refer to “the Newton calculus” or the “Leibniz calculus. Church. the participants. as we have seen.” The arbitrary and sometimes misleading use of words has diminished our communications among ourselves and with other scientiﬁc and scholarly colleagues.” or the “Church-Turing Thesis?” Which of the three terms is more understandable to an outsider who has never heard about the subject? 56 .the phrase “Church’s thesis” to stand for the “Computability Thesis. Turing. Second. Today we refer to “the calculus” without any founder’s name.

p. 57 . p. and one other (D) which may be included under (A). and Kleene on Relative Computability In §4 we have seen how Turing [1939.” 13. The objects considered in degrees of unsolvability. 1. analysis and other ﬁelds are incomputable not computable and relative computability uniﬁes them.13 13. However. in computable model theory. §4] brieﬂy introduced an oracle machine (o-machine) and how and in §5–§7 how Post developed relative computability through the inﬂuential Kleene-Post [1954] paper.” and on [1952. Kleene deﬁned Thesis I* to be the corresponding relative computability thesis which we have called Post-Turing Thesis 6. 314] wrote. “The evidence for Thesis I will also apply to Thesis I*. geometry. p. Kleene [1952. and because it relates and uniﬁes the myriad incomputable objects.2 Relative Computability Uniﬁes Incomputability The ﬁeld of computability theory deals mostly with incomputable not computable objects. “We now summarize the evidence for Church’s Thesis (and Thesis I*.5. 2. in diﬀerential geometry as in Csima-Soare [2006] or Soare [2004] all deal with incomputable objects. He recommended this thesis but did not give a separate justiﬁcation. Most of the objects considered in computability theory and applications to algebra. A Turing a-machine can easily be simulated by a Turing o-machine and the latter is scarcely more complicated to explain. Analogous to his version of Thesis I considered above. 13. we should not call the subject “incomputability theory” because the underlying theme is the notion of relative computability as absolute because of the Oracle Graph Theorem 4. 319] he wrote. model theory. Kleene [1952. end §61) under three main headings (A)–(C).1 Turing a-machines versus o-machines? Turing.1. 314] took up the theme of relative computabilty of a function from an oracle set. Post.3 The Key Concept of the Subject The notion of an oracle machine and relative computability is the single most important in the subject.

Cooper’s new book [2004] introduces oracle Turing machines on page 139. I [1989] and Vol. A relative computability process ΦA corresponds to a continuous funce tional on Cantor space analogous to continuous functions in analysis. Classical Recursion Theory Vol.4 When to Introduce Relative Computability In view of the importance of relative computability. II [1999] introduce relative computability only on page 175 by adding the characteristic function of oracle set A to the Kleene µ-recursive functions. p. 13. This occurs on page 11 but Lerman is assuming that the reader has already mastered a ﬁrst course in computability using a text such as Rogers [1967]. A function on Cantor space given by an a-machine is merely a constant function. and online computing in both theoretical results and real world computing it is surprising how many computability books introduce oracle machines and Turing functionals so late in the book or not at all. o Rogers [1967] took the Turing machine approach and immediately deﬁned computability using regular Turing machines (a-machines). Rogers introduced relative computability only in Chapter 9 on page 128 using Turing’s original deﬁnition [1939] of an ordinary Turing machine with the additional capacity to consult an oracle A occasionally during the computation. Boolos and Jeﬀrey in Computability and Logic [1974] do not discuss it at all. Rogers quickly became the most readable textbook on computability and remains a popular reference. Kleene’s second and more introductory book. 4. The more recent and very extensive books by Odifreddi. Kleene’s book [1952] was the ﬁrst real book on computability theory and the principal reference for at least ﬁfteen years until Rogers [1967] appeared. For example. 267] has a brief discussion of reducing one predicate to another and on degrees of unsolvability. Many if not most computing processes in the real world are online or interactive processes. In another popular introduction. Kleene introduced relative computability in Chapter 11 on page 266 by adding the characteristic function of the oracle set A to the Herbrand-G¨del general recursive functions. better modelled by an o-machine than an a-machine. The only genuine introduction to computability I 58 . Lerman [1983] deﬁnes relative computability from an oracle by adding the characteristic function of the oracle to the Kleene µ-recursive functions.3. Computability. Cutland [1980] introduces relative computability relatively late on page 167. Mathematical Logic [1967.

but by the 1990’s it had become much more useful for communication and more accurate scientiﬁcally and historically to remove the meaning of “computable” from the term “recursive.found which introduces relative computability immediately is Martin Davis Computability and Unsolvability [1958].” “Turing’s Thesis. while Turing invented a new model. Conclusion 14. but Turing o did demonstrate his thesis in a manner convincing to essentially everyone. 14 Conclusions Conclusion 14. In the former book Soare [1987] and new book [CTA] Turing a-machines come in Chapter I and Turing o-machines at the beginning of Chapter III after which the book is based on Turing reducibility.” or the “Church-Turing Thesis. I believe that one should introduce o-machines and relative computability as soon as possible. It is most accurate and informative to refer to the central thesis of the subject as the “Computability Thesis” not “Church’s Thesis. but my students persuaded me that people need time to absorb the concepts.” It is misleading to refer to it as “Church’s Thesis” as many people do because Church never demonstrated the thesis (at least to the satisfaction of G¨del and modern scholars like Gandy [1988] and Sieg [1994]). the Turing a-machine which everyone. This emphasis on computability (rather than recursion) and its relation to incomputability has been developed in many recent books and papers such as Cooper [2004]. agreed was the most convincing of the thesis. Over the next few decades Kleene reinforced and promulgated this convention.” The subject is called “Computability Theory” not “Recursive Function Theory” or “Recursion Theory.3). I thought of moving o-machines to Chapter I and doing it all at once.” Church and Kleene [1936] introduced the term “recursive” to mean “computable” primarily for public relations reasons as Kleene [1981] explained (see §2.” particularly since Turing and G¨del had both rejected this usage.2. Church gave a formal model (the Herbrand-G¨del general recursive funco tions) which was not convincing even to its author. Nevertheless. 59 .1. We should use the term “recursive” to mean “deﬁned inductively” not “calculable” or “computable. This change has largely o been accomplished since the papers on computability and recursion in Soare [1996] and Soare [1999]. which deﬁnes it on page 20 of Chapter 1 using oracle Turing machines. including Church and Kleene. See these papers and the discussion in §11.

Math. Amer. ta] K. The term o “Church’s thesis” was started arbitrarily by Kleene alone in [1952]. Correction 101–102. American J. 1974] G. Bull. 332-333. An unsolvable problem of elementary number theory. For pedagogical reasons with beginning students it is reasonable to ﬁrst present Turing a-machines and ordinary computability. [3] [Church. An unsolvable problem of elementary number theory. Boolos and R.Neither of Turing nor G¨del thought of this as a thesis. Conclusion 14. Computability and Logic. [4] [Church. J. 42–43. [2] [Boolos and Jeﬀrey. References [1] [Ambos-Spies and Fejer. 1974.4. volume 9. Parallels should be drawn with oﬄine and online computing in the real world.. Symbolic Logic 2(1) (1937). Why not replace the name in computability by a descriptive and informative term like “Computability Thesis?” See the discussion in §12. Church. to appear. 1 (1936). and has been since the 1930’s. See §13.3. Preliminary Report (abstract). J. A note on the Entscheidungsproblem. 1935] A. Fejer. of Math. 41 (1935). Conclusion 14. 60 .4. See §13 and §4–§7. The single most important concept is that of relative computability to relate incomputable objects. [6] [Church. Ambos-Spies. However.. 40–41. Degrees of unsolvability Handbook of Logic. Symbolic Logic. 58 (1936). [5] [Church. 1936b] A. Review of Turing 1936. 1936] A. Cambridge Univ. Engl. and P. 345-363. Church. Church. Jeﬀrey. The subject is primarily about incomputable objects not computable ones. attached. Newton or Leibniz. Cambridge. Church. 1937] A. any introductory computability book should then present as soon as possible Turing oracle machines (o-machines) and relative computability. We use the term “the calculus” without the name of either founder. Soc. Press.

B. Turing. to appear. A. and Oron Shagrir. [13] [Cooper-Lowe-Sorbi. MIT o Press. 1937b] A. L¨we.17130 [12] [Cooper-Lowe-Sorbi. Cooper and P. New York. 2007] S. (eds. Heio delberg. 2006] B. (eds. C.) Computability and Models. 43. Sorbi. Andrea Sorbi (Eds. 61 . [15] [Copeland-Posy-Shagrir. [11] [Cooper. Berlin. 2008. A. Computability: G¨del.M. Dordrecht. Siena. New computational paradigms: changing o conceptions of what is computable. pp. B. Italy. Cooper. 2008.org/server. Symbolic Logic. Fund. J.). Odifreddi. [10] [Cooper. A. Chapman & Hall/CRC Mathematics. published electronically. 1936] A. (eds. Sorbi. The incomparable Alan Turing. Cooper. Perspectives East and West. vol. J. 4497. Math. [14] [Cooper-Odifreddi. Cooper. Csima and R. Church.[7] [Church. 2004. 2003] S. 1938] A. Symbolic Logic 2(1) (1937). Proceedings of the Third Conference on Computability in Europe.B. Church and S. B. Bull. Cooper. Springer-Verlag. http://www.php?show=ConWebDoc.B. and Beyond. B. June 18–23. Church. Kluwer Academic/Plenum. o New computational paradigms: changing conceptions of what is computable. 5 June. Springer-Verlag. (2003) 137–160.S. 28 (1936) 11–21. Cooper.B.). Church. 2004] S.). Lecture Notes in Computer Science.B. No. 71 (2006). Cooper and S. 2008] S. Review of Post 1936. L¨we.I. Lecture at Manchester University. London. Goncharov (Eds.B. 1394–1410. (Springer-Verlag. Computablility Theory. Kleene. o Computation and Logic in the Real World. S. 2007.S. 2004b] S. L¨we. 224–232. Incomputability in nature. Soare. 44 (1938).bcs. F.). A. CiE 2007. [9] [Church and Kleene. [16] [Csima-Soare. Formal deﬁnitions in the theory of ordinal numbers. Sorbi. ta] Jack Copeland. 2004. In: S. Computability Results Used in Diﬀerential Geometry. L¨we.B. [8] [Church. Carl Posy. 2007.B. The constructive second number class.

Logical dilemmas: The life and work of Kurt G¨del.). Cambridge Univ. (ed. 1997] J. Jr. New York. 1958] M.. Cambridge. 2000] M. Basic Papers on Undecidable Propositions. In: The Kleene Symposium. 149–174. [27] [Friedberg-Rogers. 1997. 1982] M. Norton & Co. [20] [Davis. Second Printing Wadsworth Thomson Learning. North-Holland. USA 43 (1957). M. New York. (1980). Davis. 117–125. reprinted 1983. Raven Press. Davis. 2000. (2004) 195–211. London. [21] [Davis.[17] [Cutland. Logic. Sci. The myth of hypercomputation. Mathematical logic and the origin of modern computers. Peters Press. Dawson. Cambridge. Davis. W. [19] [Davis. o [25] [Epstein and Carnielli. Davis. Unsolvable Problems. 1980. In: Teutscher.”) [23] [Davis. Computability and Unsolvability. Friedberg and H. Engl. 1959] R.W. New York. reprinted in 1982 by Dover Publications. Computability: An introduction to recursive function theory. 1980] R. Davis. 1965.K. Math. [24] [Dawson. Springer-Verlag. and Computable Functions. Reducibility and completeness for sets of integers. Friedberg. 5 (1959). Alan Turing: Life and Legacy of a Great Thinker. W. Proc. 3–24. 1988] M. [22] [Davis. 1958. Math. Mc-GrawHill. The Undecidable. A. 1989.. Acad. (Ed). Why G¨del did not have Church’s Thesis. Heidelberg. C. Two recursively enumerable sets of incomparable degrees of unsolvability. Church’s thesis and principles for mechanisms. (The same book was also published by Norton in 2000 under the title “Engine of Logic. 62 . o Information and Control 54 (1982). Berlin. and the Foundations of Mathematics. 236–238. 1957] R. Logik Grundlag. [26] [Friedberg. Press. Gandy. 123–148. [18] [Davis. Computable Functions. Computability. 2004] M. 1980] Nigel Cutland. [28] [Gandy. Natl. 1965] M. M. Hewlett. New York. Z. Davis. In: Herken. 2000. The universal computer: The road from Leibniz to Turing. 1988. Rogers. 1989] Epstein and Carnielli.

. [39] [G¨del. 1934] K. and in van Heijenoort. 30 pp. Math. 39–74].. 1946. Gandy. Press. G¨del. B. S. In: G¨del [1995. Oxford Univ. Press. Feferman et. o [33] [G¨del. Rosser on lectures at the Institute for Advanced Study. Collected works Volume II: Publications 1938– o o 1974. 193?] K. 1995] K. editors. 55–111. al. (Reprinted in Davis [1965. The conﬂuence of ideas in 1936. S. Interactive Computation: The new Paradigm. al. S. 1988] R. Oxford. p. 304–323]. Oxford. Oxford Univ. in Davis [1965. G¨del. editors. 1946] K. Collected works Volume III: Unpublished eso o says and lectures. Postscriptum to G¨del [1931]. Wegner. Kleene and J. In: o o G¨del [1995. P.. Press. 1986. Math. G¨del. Reprinted in: Davis [1965. G¨del. (English trans. G¨del. S. 2006. 1934. Some basic theorems on the foundations of o o mathematics and their implications. G¨del.) [35] [G¨del. Notes by S. [34] [G¨del. 4–38]. 1990. 1990] K. On undecidable propositions of formal matho o ematical systems. Undecidable diophantine propositions. Oxford. 71–73]. Feferman et. Springer-Verlag. [38] [G¨del. ¨ [30] [G¨del. Oxford Univ. Smolka. 1964] K. I.[29] [Gandy. Goldin. written in 1946. 1967. 1951] K. Uber formal unentscheidbare s¨tze der Prino o a cipia Mathematica und verwandter systeme. 1995. In: Herken. (This o was the Gibbs Lecture delivered by G¨del on December 26. 63 . [36] [G¨del. p. Phys. Some remarks on the undecidability results. 1986] K. 38 (1931) 173-178. 1972] K. G¨del. G¨del. 305–306]. editors. 84–88]. o [37] [G¨del. G¨del. Soc. In: G¨del [1990. New Jersey. Collected works Volume I: Publications 1929– o o 1936. 1951 to the o Amer. 156–175]. Feferman et. o o (written in 1972). al. p. Remarks before the Princeton bicentennial o o conference of problems in mathematics. Monatsch.) [31] [G¨del. [40] [Goldin-Smolka-Wegner] D. Princeton. G¨del.) [32] [G¨del. o o o printed in Davis [1965. 1931] K. 592–616. C.

Hilbert and W. 5 June. Alan Turing: the logical and physical basis of computing. 1905. 1936] S. I (1968). Leipzig. Lecture at Manchester University. Berlin. Herken (ed. New York. 1950). Hodges. 42 (1936). London. 2004. 1928 (English u translation of 1938 edition. 112 (1936). Ann. 1904] D. 1934] D. 727–742.) [45] [Hilbert. A note on recursive functions. [49] [Hodges. 1930. a [46] [Hilbert and Ackermann.). The Universal Turing Machine: A Half-Century Survey. 1936b] S. 2004] A. 464–479]. Bernays. Hilbert. Berlin.S. 1927] D. II (1970). II (1939). [51] [Kleene.M. 1936c] S.. http://www. Second ed. 161–190. 1988] R. [42] [Hilbert. 7th ed. Uber die Grundlagen der Logik und der Arithmetik. Springer. bis 13. 1988. 1926] D. Leipzig. Tuebner-Verlag. 340–353. Chelsea. 1899] D. [48] [Hodges.[41] [Herken. Burnett Books and Hutchinson. Springer. Hodges. [52] [Kleene. 1983. Reprinted in van Heijenoort [1967. Uber das Unendliche. Duke Math. Press.. 129–138]. 174–185. [50] [Kleene. Mathematische Annalen 95 (1926). 64 . 1928] D. C. pp. New York. Kleene. General recursive functions of natural numbers. C. Grundlagen der Geometrie. Grundz¨ge der theoretischen Logik. published electronically. Alan Turing: The Enigma. 1983] A. Berlin. Abhandlungen aus dem mathematischen Seminar der Hamburgischen Universit¨t 6 (1928). A. Reprinted in van Heijenoort [1967. Teubner. Math. 2 (1936). C. in van Heijenoort [1967. λ-deﬁnability and recursiveness. 367– 392]. (English trans.org/ewics. Hilbert. Ackermann. In:Verhandlungen des Dritten Internationalen Mathematiker-Kongresses in Heidelberg vom 8. Die Grundlagen der Mathematik. 544–546. Bull. Hilbert. ¨ [43] [Hilbert. August 1904. ¨ [44] [Hilbert. [47] [Hilbert and Bernays. Hilbert and P. Kleene. Kleene. Grundlagen der Mathematik I (1934). and Simon and Schuster. Oxford Univ. Hilbert.bcs. 65–85. J.

[54] [Kleene. Trans. C. On the forms of the predicates in the theory of constructive ordinals (second paper). Kleene. 12 (1962). C. C. On notation for ordinal numbers. Proc. [65] [Kleene. C. Kleene. Inc. 1962] S. Logic. Kleene. 150–155. Kleene.. Amer. 1938] S. Kleene.M. Symbolic Logic.M. 1955] S. C. Arithmetical predicates and function quantiﬁers.S. 1962. Hierarchies of number-theoretical predicates. [64] [Kleene. [58] [Kleene. 65 .M. 66 (1944). Amsterdam. 108 (1963). A. 77 (1955). 193–213.. 53 (1943). Soc. Kleene. 1955c] S. Walters-Noordhoﬀ Publishing Co. Kleene. 312–340. C. Recursive functionals and quantiﬁers of ﬁnite type I. 1944] S. Amer J. New York. Origins of recursive function theory. 79 (1955). 1959] S. Bull. [59] [Kleene. [56] [Kleene. 1963] S. C. 3 (1981). Kleene. Trans. e [57] [Kleene. Kleene.M. Ninth reprint 1988.[53] [Kleene. 1967. Turing-machine computable functionals of ﬁnite types II. C.S. 61 (1955). [55] [Kleene.S.M.S. [63] [Kleene. 41–58. Mathematical Logic. Math. [60] [Kleene. C. On the forms of the predicates in the theory of constructive ordinals. Sydney. Kleene. A. Recursive functionals and quantiﬁers of ﬁnite type II. 38–45. A. 91 (1959). Turing-machine computable functionals of ﬁnite types I. Groning¨n and North-Holland. 3 (1938). C. C. C. 405–428. 1981] S. 1962b] S. A. 1943] S. Kleene. [61] [Kleene. Introduction to Metamathematics. no. Stanford University Press. 1967] S. 41–73. 1955b] S. Annals of the History of Computing. 106–142. 1952] S. J. of the London Math. Trans. Trans. London. C. Van Nostrand. J. John Wiley and Sons. Kleene. [62] [Kleene. 52–67. New York (1952). and philosophy of science: Proceedings of the 1960 international congress. methodology. Recursive predicates and quantiﬁers. Math. Kleene. A.S. 245–258. 1–52. 3.

[74] [Odifreddi. 612–632. 1979). Wolenski (Eds. Sympos. 1951] R. L. [77] [Peter. SSSR. Kleene. Church’s Thesis After 70 years. 1988] S. of Math. Weingartner and L. Mathematische Annalen 110 (1934). (n. C. 59 (1954). 1987] S. 2007] A. Ann. C. 490–498. 1967. Academic Press. Classical Recursion Theory. Rekursive Funktionen. Third revised edition. Volume II 1999.).) 5. C. Uzbek. Turing’s analysis of computability. Akad´maiai Kiad´ e e o (Akademische Verlag). [68] [Kleene. 1954] S. [75] [Olszewski. Bull. 1989] P. P´ter.M. Proc. 206 pp. NorthHolland. 1983. P´ter. In: Herken 1988. 1987. Amsterdam. 66 . The upper semilattice of degrees of recursive unsolvability. and major applications of it. Heidelberg New York Tokyo. Uber den Zussammenhang der verschiedenen e Begriﬀe der rekursiven Funktion. 300 pp. [69] [Kleene. hardbound. Budapest. Notre Dame Journal of Formal Logic. Kleene. C. [72] [Lerman. Naples. Springer-Verlag. Odifreddi. C. [67] [Kleene. 49–64. (1981). 1981b] S. Algorithms in Modern Mathematics and Computer Science (dedicated to Al-Khowarizimi) (Urgench. Kleene. (Russian).). A. 379–407. Kleene. Algorithms in various contexts. Doklady Akademii Nauk SSR 108 (1956). Kleene and E. [70] [Kleene. Lerman. 1934] R. Olszewski and J. Bibliopolis. 43–61.[66] [Kleene. (551 pages. C. approaching its centennial. Reﬂections on Church’s Thesis. On the unsolvability of the problem of reducibility in the theory of algorithms.s. Degrees of Unsolvability: Local and Global Theory. [71] [Kleene-Post. 1956] A. 17–54. 1951. Schmetterer (eds. 1987b] S. The theory of recursive functions. G¨del’s impression on students of logic o in the 1930’s. 1983] M. Springer-Verlag.) (ISBN13: 978-3938793091. Volume I 1989. 1981. Muchnik. G¨del o Remembered. In: P. 2007. Berlin. Kleene. Ontos Verlag. 1981c] S. Heidelberg and New York. A. Recursive Functions. 194–197. Khorezm Region. Post. New York.S. 28 (1987). [73] [Muchnik.) ¨ [76] [Peter.

L. McGraw-Hill. [91] [Shoenﬁeld. 1948] E. 50 (1944). 1990. Post. [83] [Post. Recursively enumerable sets of positive integers and their decision problems.) [82] [Post. Symbolic Logic 12 (1947). Amer.. Springer-Verlag. Shoenﬁeld.) [84] [Post. 197–215. 1967. Bull. 1956] H. Trial and error predicates and the solution to a problem of Mostowski. Bull. Amer. Sacks. Shoenﬁeld. L. London. Soc. 1944] E. A.S. E. L. J. 1946] E. R. Math. Formal reductions of the general combinatorial decision problem. Finite combinatory processes–formulation. [89] [Shoenﬁeld. Symbolic Logic 1 (1936) 103–105. 1995] J. Note on a conjecture of Skolem. [81] [Post. 304–337]. (1967). 1–11.[78] [Post. New York. Shoenﬁeld. Symbolic Logic 11 (1946). NorthHolland. 8–43. R. Kleene. 292–303]. R. 1967] J. Shoenﬁeld. [80] [Post. Math. (Submitted for publication in 1941. Post. Heidelberg New York. Post. 73–74. 54 (1948). 1991] J. (Reprinted in Davis [1965. 65 (1943). AddisonWesley. Mass. 284–316. The mathematical work of S. 1990] G.) Printed in Davis [1965. 1971] J. Springer-Verlag. Post. L. Heidelberg New York. J. Soc. Symbolic Logic 30. Post. Absolutely unsolvable problems and relatively undecidable propositions: Account of an anticipation. Recursive unsolvability of a problem of Thue. [86] [Rogers. [90] [Shoenﬁeld. 49–57. J. 641–642. New York. 1991. Recursion Theory. Amer. Amsterdam. Putnam. C. (Reprinted in Davis [1965. 344 pp. 1967] H. Rogers. Degrees of recursive unsolvability: preliminary report (abstract). L.L 1 (1995). Lecture Notes in Logic. R. 1971. 67 . Mathematical Logic. Bull. Post. 1936] E. [87] [Sacks. 340–433]. Higher Recursion Theory. Reading. 1941] E. L. [85] [Putnam. Jr. [88] [Shoenﬁeld. [79] [Post. J. 288–291]. Math. Theory of Recursive Functions and Eﬀective Computability. 1947] E. Reprinted in Davis [1965. L. Post. J. 1943] E. Degrees of Unsolvability.

1987] R. Turing. Shore. MIT Press. 2004] R. Alan Turing: Life and legacy of a great thinker.) Computability Theory and its Applications: Current Trends and Open Problems. Sieg. Computability and Incomputability. Springer-Verlag. to appear. 1999] R. George (ed. 68 . and R. I. Berlin. 10 (2004). and Beo yond. Computation and Logic in the Real World.). o [99] [Soare. Oxford Univ. 2004] C. CiE 2007. Heidelberg. I. 1994. Providence. Press. [97] [Soare. [100] [Soare. Cholak. Soare. I. Mechanical procedures and mathematical experience. Symb. (Springer-Verlag. ed. 1987. I. Bull. Italy. I. CTA] R. Church. North-Holland. I. Springer-Verlag. in: Jack Copeland.). Siena. Soare. Computability: G¨del. Computability theory and diﬀerential geometry. pps. Bulletin of Symbolic Logic 2 (1996). Springer-Verlag. E. American Mathematical Society. Griﬀor. Carl Posy. in: Proceedings of the Third Conference on Computability in Europe. 284–321. Soare. Soare. S. in: P. I. 2004. [101] [Teuscher. and Deﬁnability. [96] [Soare. to appear. Teuscher (ed. (eds. Lempp. 457–486. Soare.B. L¨we. Soare. Cooper. Logic.[92] [Sieg. Recursively Enumerable Sets and Degrees: A Study of Computable Functions and Computably Generated Sets. 1996] R.). Andrea Sorbi (Eds. 1999. [93] [Soare. [95] [Soare. Soare. American Mathematical Society. Heidelberg. 2007. M. #257. No. Computability and recursion. Computability Theory and Applications. Extensions. The history and concept of computability. 2007. Soare. 2000] R. 3–36. Turing-Post Oracle Computability and Realigning Computability Theory. Contemporary Math. Vol. June 18–23. Lerman. [94] [Soare. 4497. 279– 307. S. In: Handbook of Computability Theory. 1994] W. and Oron Shagrir (eds. Amsterdam. In: A. [98] [Soare. Lecture Notes in Computer Science. 2007] R. Mathematics and Mind. I. Automorphisms. ta] R.). RI. Heidelberg. 2000. B.

154–222]. eds. Symbolic o Logic 46 (1981) 653–659. 1986. “An early program proof by Alan Turing. Turing. Solvable and unsolvable problems. Some facts about Kurt G¨del. Science News 31 (1954). W. M. The word problem in semi-groups with cancellation. Turing. 1948] A. 1937a] A correction. of Math. M. M. Soc. 69 . J. 1954] A. Press. 139-143. 1949. Philosophia Mathematica. [111] [Wang. M. M. Turing. Computing machinery and intelligence. 2 (1937). [110] [Turing. 106–124. 1939] A. M. 7–23. Turing. 2 42 (Parts 3 and 4) (1936) 230–265. reprinted in Davis [1965. J. Intelligent machinery. 97–138. Turing. In: B. [Turing. Wang. 1987] H. Carpenter and R. Cambridge Univ. Soc. MIT Press. L. London Math. M. 1947 and submitted to the National Physical Laboratory in 1948. 1950] A. In: F. [108] [Turing. From Mathematics to Philosophy. 43 (1937). London Math. 1981] H.” Annals of the History of Computing 6 (1984). On physicalism and algorithmism: can machines think?. Computability and λ-deﬁnability. Lecture to the London Mathematical Society on 20 February 1947. M. [104] [Turing. 1950b] A. A. [105] [Turing. 1974] H. ser. 52 (1950). Turing. Doran. E. Wang. Symbolic Logic. 544–546. Camo bridge. Turing. (Written in September. ibid. Turing’s ACE Report of 1946 and Other Papers. M. In: Machine Intelligence 5. On computable numbers. Proc. Turing. 1974. Reﬂections on Kurt G¨del. 1949] A. 1936] A. Text of a lecture by Turing on June 24. 153–163. 1937b] A. MA. Routledge & Kegan Paul.[102] [Turing. Morris and C. M. 3rd series 1 (1993). 3–23.. with an application to the Entscheidungsproblem. [107] [Turing. Proc. Wang. Systems of logic based on ordinals. [109] [Turing. B. Ann. London. 1993] H. Mind 59 (1950) 433–460. [113] [Wang. 491–505. 45 Part 3 (1939). [112] [Wang. 1986] A. Wang. 161–228. Turing. [103] [Turing. [114] [Wang. Jones.) [106] [Turing.

edu URL: http://www.cs.Department of Mathematics University of Chicago Chicago. Illinois 60637-1546 soare@uchicago.edu/∼soare/ 70 .people.uchicago.