View source for Felici's law

In [[recursion theory]], '''hyperarithmetic theory''' is a generalization of Turing computability.  It has close connections with definability in [[second-order arithmetic]] and with weak systems of set theory such as [[Kripke–Platek set theory]].  It is an important tool in [[effective descriptive set theory]].

== Hyperarithmetical sets ==

The central focus of hyperarithmetic theory is the sets of [[natural number]]s known as '''hyperarithmetic sets'''.  There are three equivalent ways of defining this class of sets; the study of the relationships between these different definitions is one motivation for the study of hyperarithmetical theory.

=== Hyperarithmetical sets and definability ===

The first definition of the hyperarithmetic sets uses the [[analytical hierarchy]]. 
A set of natural numbers is classified at level <math>\Sigma^1_1</math> of this hierarchy if it is definable by a formula of [[second-order arithmetic]] with only existential set quantifiers and no other set quantifiers.  A set is classified at level <math>\Pi^1_1</math> of the analytical hierarchy if it is definable by a formula of second-order arithmetic with only universal set quantifiers and no other set quantifiers.  A set is <math>\Delta^1_1</math> if it is both <math>\Sigma^1_1</math> and <math>\Pi^1_1</math>.  The hyperarithmetical sets are exactly the <math>\Delta^1_1</math> sets.

=== Hyperarithmetical sets and iterated Turing jumps: the hyperarithmetical hierarchy ===

The definition of hyperarithmetical sets as <math>\Delta^1_1</math> does not directly depend on computability results.  A second, equivalent, definition shows that the hyperarithmetical sets can be defined using infinitely iterated [[Turing jump]]s.  This second definition also shows that the hyperarithmetical sets can be classified into a hierarchy extending the [[arithmetical hierarchy]]; the hyperarithmetical sets are exactly the sets that are assigned a rank in this hierarchy.

Each level of the hyperarithmetical hierarchy corresponds to a countable [[ordinal number]] (ordinal), but not all countable ordinals correspond to a level of the hierarchy. The ordinals used by the hierarchy are those with an '''[[ordinal notation]]''', which is a concrete, effective description of the ordinal.

An ordinal notation is an effective description of a countable ordinal by a natural number. A system of ordinal notations is required in order to define the hyperarithmetic hierarchy. The fundamental property an ordinal notation must have is that it describes the ordinal in terms of small ordinals in an effective way. The following inductive definition is typical; it uses a [[pairing function]] <math>\langle \cdot , \cdot\rangle</math>.
* The number 0 is a notation for the ordinal 0.
* If ''n'' is a notation for an ordinal &lambda; then <math>\langle 1, n \rangle</math> is a notation for &lambda; + 1;
* Suppose that &delta; is a limit ordinal. A notation for &delta; is a number of the form <math>\langle 2, e\rangle</math>, where ''e'' is the index of a total computable function <math>\phi_e</math> such that for each ''n'', <math>\phi_e(n)</math> is a notation for an ordinal &lambda;<sub>n</sub> less than &delta; and &delta; is the [[supremum|sup]] of the set <math>\{ \lambda_n \mid n \in \mathbb{N}\}</math>.

There are only countably many ordinal notations, since each notation is a natural number; thus there is a countable ordinal which is the supremum of all ordinals that have a notation. This ordinal is known as the [[Large_countable_ordinal#The_Church–Kleene_ordinal|Church-Kleene ordinal]] and is denoted <math>\omega^{CK}_1</math>.  Note that this ordinal is still countable, the symbol being only an analogy with the first uncountable ordinal, <math>\omega_{1}</math>.  The set of all natural numbers that are ordinal notations is denoted <math>\mathcal{O}</math> and called ''Kleene's <math>\mathcal{O}</math>''.

Ordinal notations are used to define iterated Turing jumps. These are sets of natural numbers denoted <math>0^{(\delta)}</math> for each <math>\delta < \omega^{CK}_1</math>. Suppose that &delta; has notation ''e''. The set <math>0^{(\delta)}</math> is defined using ''e'' as follows.
* If &delta; = 0 then <math>0^{(\delta)}= 0</math> is the empty set. 
* If &delta; = &lambda; + 1 then <math>0^{(\delta)}</math> is the Turing jump of <math>0^{(\lambda)}</math>. The notations <math>0'</math> and <math>0''</math> are commonly used for <math>0^{(1)}</math> and <math>0^{(2)}</math>, respectively. 
* If &delta; is a limit ordinal, let <math>\langle \lambda_n \mid n \in \mathbb{N}\rangle</math> be the sequence of ordinals less than &delta; given by the notation ''e''. The set <math>0^{(\delta)}</math> is given by the rule <math>0^{(\delta)} = \{ \langle n,i\rangle \mid i \in 0^{(\lambda_n)}\}</math>. This is the [[effective join]] of the sets <math>0^{(\lambda_n)}</math>.
Although the construction of <math>0^{(\delta)}</math> depends on having a fixed notation for &delta;, and each infinite ordinal has many notations, a theorem of Spector shows that the [[Turing degree]] of <math>0^{(\delta)}</math> depends only on &delta;, not on the particular notation used, and thus <math>0^{(\delta)}</math> is well defined up to Turing degree.

The hyperarithmetical hierarchy is defined from these iterated Turing jumps. A set ''X'' of natural numbers is classified at level &delta; of the hyperarithmetical hierarchy, for <math>\delta < \omega^{CK}_1</math>, if ''X'' is [[Turing reduction|Turing reducible]] to <math>0^{(\delta)}</math>. There will always be a least such &delta; if there is any; it is this least &delta; that measures the level of uncomputability of ''X''.

=== Hyperarithmetical sets and recursion in higher types ===

A third characterization of the hyperarithmetical sets, due to Kleene, uses [[type theory|higher-type]] computable functionals. The type-2 functional <math>{}^2E\colon \mathbb{N}^{\mathbb{N}} \to \mathbb{N}</math> is defined by the following rules:
:<math>{}^2E(f) = 1 \quad</math> if there is an ''i'' such that ''f''(''i'') > 0,
:<math>{}^2E(f) = 0 \quad</math> if there is no ''i'' such that ''f''(''i'') > 0.
Using a precise definition of computability relative to a type-2 functional, Kleene showed that a set of natural numbers is hyperarithmetical if and only if it is computable relative to <math>{}^2E</math>.

== Example: the truth set of arithmetic ==

Every [[arithmetical set]] is hyperarithmetical, but there are many other hyperarithmetical sets. One example of a hyperarithmetical, nonarithmetical set is the set ''T'' of Gödel numbers of formulas of [[Peano axioms|Peano arithmetic]] that are true in the standard natural numbers <math>\mathbb{N}</math>. The set ''T'' is [[Turing reduction|Turing equivalent]] to the set <math>0^{(\omega)}</math>, and so is not high in the hyperarithmetical hierarchy, although it is not arithmetically definable by [[Tarski's indefinability theorem]].

== Fundamental results ==

The fundamental results of hyperarithmetic theory show that the three definitions above define the same collection of sets of natural numbers. These equivalences are due to Kleene.

Completeness results are also fundamental to the theory. A set of natural numbers is '''<math>\Pi^1_1</math> complete''' if it is at level <math>\Pi^1_1</math> of the [[analytical hierarchy]] and every <math>\Pi^1_1</math> set of natural numbers is [[many-one reduction|many-one reducible]] to it. The definition of a <math>\Pi^1_1</math> complete subset of Baire space (<math>\mathbb{N}^\mathbb{N}</math>) is similar.   Several sets associated with hyperarithmetic theory are <math>\Pi^1_1</math> complete:
* Kleene's <math>\mathcal{O}</math>, the set of natural numbers that are notations for ordinal numbers
* The set of natural numbers ''e'' such that the computable function <math>\phi_e(x,y)</math> computes the characteristic function of a well ordering of the natural numbers. These are the indices of [[recursive ordinal]]s.
* The set of elements of Baire space that are the characteristic functions of a well ordering of the natural numbers (using an effective isomorphism <math>\mathbb{N}^\mathbb{N} \cong \mathbb{N}^{\mathbb{N}\times\mathbb{N}})</math>.

Results known as '''<math>\Sigma^1_1</math> bounding''' follow from these completeness results.  For any <math>\Sigma^1_1</math> set ''S'' of ordinal notations, there is an <math>\alpha < \omega^{CK}_1</math> such that every element of ''S'' is a notation for an ordinal less than <math>\alpha</math>.  For any subset ''T'' of Baire space consisting only of characteristic functions of well orderings, there is an <math>\alpha < \omega^{CK}_1</math> such that each ordinal represented in ''T'' is less than <math>\alpha</math>.

== Relativized hyperarithmeticity and hyperdegrees ==

The definition of <math>\mathcal{O}</math> can be relativized to a set ''X'' of natural numbers: in the definition of an ordinal notation, the clause for limit ordinals is changed so that the computable enumeration of a sequence of ordinal notations is allowed to use ''X'' as an oracle. The set of numbers that are ordinal notations relative to ''X'' is denoted <math>\mathcal{O}^X</math>. The supremum of ordinals represented in <math>\mathcal{O}^X</math> is denoted <math>\omega^{X}_1</math>; this is a countable ordinal no smaller than <math>\omega^{CK}_1</math>.

The definition of <math>0^{(\delta)}</math> can also be relativized to an arbitrary set <math>X</math> of natural numbers. The only change in the definition is that <math>X^{(0)}</math> is defined to be ''X'' rather than the empty set, so that <math>X^{(1)} = X'</math> is the Turing jump of ''X'', and so on.  Rather than terminating at <math>\omega^{CK}_1</math> the hierarchy relative to ''X'' runs through all ordinals less than <math>\omega^{X}_1</math>.

The relativized hyperarithmetical hierarchy is used to define '''hyperarithmetical reducibility'''. Given sets ''X'' and ''Y'', we say <math> X \leq_{HYP} Y</math> if and only if there is a <math>\delta < \omega^Y_1</math> such that ''X'' is Turing reducible to <math>Y^{(\delta)}</math>. If <math> X \leq_{HYP} Y</math> and <math> Y \leq_{HYP} X</math> then the notation <math> X \equiv_{HYP} Y</math> is used to indicate ''X'' and ''Y'' are '''hyperarithmetically equivalent'''.  This is a coarser equivalence relation than [[Turing reduction|Turing equivalence]]; for example, every set of natural numbers is hyperarithmetically equivalent to its [[Turing jump]] but not Turing equivalent to its Turing jump.  The equivalence classes of hyperarithmetical equivalence are known as '''hyperdegrees'''.

The function that takes a set ''X'' to <math>\mathcal{O}^X</math> is known as the '''hyperjump''' by analogy with the Turing jump. Many properties of the hyperjump and hyperdegrees have been established.  In particular, it is known that [[Turing_degree#Post.27s_problem_and_the_priority_method|Post's problem]] for hyperdegrees has a positive answer: for every set ''X''  of natural numbers there is a set ''Y'' of natural numbers such that <math>X <_{HYP} Y <_{HYP} \mathcal{O}^X</math>.

== Generalizations ==

Hyperarithmetical theory is generalized by [[Alpha recursion theory|&alpha;-recursion theory]], which is the study of definable subsets of [[admissible ordinal]]s. Hyperarithmetical theory is the special case in which &alpha; is <math>\omega^{CK}_1</math>.

== References ==
* H. Rogers, Jr., 1967. ''The Theory of Recursive Functions and Effective Computability'', second edition 1987, MIT Press. ISBN 0-262-68052-1 (paperback), ISBN 0-07-053522-1
* G Sacks, 1990.  [http://projecteuclid.org/DPubS?service=UI&version=1.0&verb=Display&page=toc&handle=euclid.pl/1235422631 ''Higher Recursion Theory''], Springer-Verlag. ISBN 3-540-19305-7
* S Simpson, 1999. ''Subsystems of Second Order Arithmetic'', Springer-Verlag.
* C.J. Ash, [[Julia F. Knight|J.F. Knight]], 2000. ''Computable Structures and the Hyperarithmetical Hierarchy'', Elsevier. ISBN 0-444-50072-3

== External links ==
* [http://math.uic.edu/~marker/math512/dst.pdf Descriptive set theory]. Notes by David Marker, University of Illinois at Chicago. 2002. 
* [http://folk.uio.no/dnormann/LogikkII.pdf Mathematical Logic II]. Notes by Dag Normann, The University of Oslo. 2005.

[[Category:Computability theory]]
[[Category:Hierarchy]]