Cite

This course has no released version yet: the text at this address may change. Cite a released version when one exists.

/courses/pi01-classes/pa-degrees
@misc{pi01-classes-pa-degrees,
  title = {{PA degrees}},
  howpublished = {Lesson 4 of \emph{Π⁰₁ classes}, Computability Theory course notes},
  url = {/courses/pi01-classes/pa-degrees},
  note = {Unreleased working version}
}

To cite a single statement, use the # link that appears next to its heading.

Lesson 4 of 5

PA degrees

Sets computing a diagonally non-computable {0,1}-valued function; equivalence with computing a member of every non-empty Π⁰₁ class and with separating disjoint c.e. sets; completions of Peano arithmetic.

45 min3 questions
Before this lesson:Basis theorems
Π⁰₁ classes · lesson 4 of 5
  1. Trees and Π⁰₁ classes
  2. Computable members
  3. Basis theorems
  4. PA degrees
  5. Finitely branching trees

Some sets compute a member of every non-empty Π10\Pi^0_1 class: they can solve, uniformly, every problem whose solutions form a Π10\Pi^0_1 class. These sets are characterized by a simple diagonal property — they compute a {0,1}\{0, 1\}-valued function that differs from every computable function on the diagonal — and, historically, by their ability to compute a complete consistent extension of Peano arithmetic. Their degrees are the PA degrees. They are not necessarily complete, and not even necessarily above any given non-computable set: by the basis theorems, there are low PA degrees.

4.1 Definition and characterizations

Definition 4.1.#

A set XX is of PA degree, or PA, if XX computes a member of DNC2\mathrm{DNC}_2, that is, a function f:N→{0,1}f : \nat \to \{0, 1\} with f(e)≠Φe(e)f(e) \neq \Phi_e(e) for every ee such that Φe(e) ⁣ ⁣↓\Phi_e(e)\converge.

The class of PA sets is closed upward under Turing reducibility. The halting set is PA: the function f(e)=1f(e) = 1 if Φe(e) ⁣ ⁣↓=0\Phi_e(e)\converge = 0, and f(e)=0f(e) = 0 otherwise, is computable in ∅′\emptyset' and belongs to DNC2\mathrm{DNC}_2.

Theorem 4.2.#

For every set XX, the following are equivalent:

  1. XX is PA;
  2. for every pair of disjoint c.e. sets AA and BB, XX computes a set CC separating them: A⊆CA \subseteq C and C∩B=∅C \cap B = \emptyset;
  3. XX computes a member of every non-empty Π10\Pi^0_1 class.
Proof of Theorem 4.2.

(1 ⇒ 2) Let f≤TXf \leq_T X be a member of DNC2\mathrm{DNC}_2, and AA, BB disjoint c.e. sets. By the parameter form of SMN, there is a computable qq such that the program q(n)q(n), on any input, enumerates AA and BB, outputs 00 if nn appears in AA, and 11 if nn appears in BB; it diverges if nn belongs to neither. So Φq(n)(q(n))=0\Phi_{q(n)}(q(n)) = 0 if n∈An \in A, =1= 1 if n∈Bn \in B, and diverges otherwise. Let C={n:f(q(n))=1}C = \{ n : f(q(n)) = 1 \}, computable in XX. If n∈An \in A, then f(q(n))≠0f(q(n)) \neq 0, so n∈Cn \in C; if n∈Bn \in B, then f(q(n))≠1f(q(n)) \neq 1, so n∉Cn \notin C.

(2 ⇒ 3) Let P=[T]\mathcal P = [T] be non-empty, with TT computable. For a string σ\sigma, non-extendibility of σ0\sigma 0 and of σ1\sigma 1 are Σ10\Sigma^0_1 properties (Proposition 1.9); fix enumerations of them by stages. Let AA be the set of strings σ\sigma such that the non-extendibility of σ0\sigma 0 is confirmed at some stage ss and that of σ1\sigma 1 is not confirmed at any stage t<st < s, and BB the set of strings σ\sigma such that the non-extendibility of σ1\sigma 1 is confirmed at some stage ss and that of σ0\sigma 0 is not confirmed at any stage t≤st \leq s. These sets are c.e. and disjoint, so XX computes a set CC separating them. Define σ0=ε\sigma_0 = \varepsilon and σn+1=σn1\sigma_{n+1} = \sigma_n 1 if σn∈C\sigma_n \in C, σn+1=σn0\sigma_{n+1} = \sigma_n 0 otherwise. We show by induction that σn\sigma_n is extendible. If σn\sigma_n is extendible, at least one of its successors is. If both are, so is σn+1\sigma_{n+1}. If only σn1\sigma_n 1 is, then the non-extendibility of σn0\sigma_n 0 is confirmed at some stage and that of σn1\sigma_n 1 never, so σn∈A⊆C\sigma_n \in A \subseteq C and σn+1=σn1\sigma_{n+1} = \sigma_n 1. If only σn0\sigma_n 0 is, then σn∈B\sigma_n \in B, so σn∉C\sigma_n \notin C and σn+1=σn0\sigma_{n+1} = \sigma_n 0. Hence the sequence ⋃nσn\bigcup_n \sigma_n, computable in CC and so in XX, is a member of P\mathcal P.

(3 ⇒ 1) DNC2\mathrm{DNC}_2 is a non-empty Π10\Pi^0_1 class (Theorem 2.6).

The proof of (2 ⇒ 3) is uniform: from an index of TT, one computes the indices of AA and BB, and the member is obtained from the separating set by a fixed procedure. Combined with (1 ⇒ 2), it shows that a single member of DNC2\mathrm{DNC}_2 computes members of all non-empty Π10\Pi^0_1 classes, uniformly in their indices. For this reason DNC2\mathrm{DNC}_2 is called a universal Π10\Pi^0_1 class.

4.2 Consequences of the basis theorems

Corollary 4.3.#
  1. There is a low PA set.
  2. There is a computably dominated PA set.
  3. For every non-computable set CC, there is a PA set that does not compute CC.
Proof of Corollary 4.3.

Apply Theorem 3.1, Theorem 3.5 and Theorem 3.7 to the non-empty Π10\Pi^0_1 class DNC2\mathrm{DNC}_2, all of whose members are PA.

So the ability to compute a member of every non-empty Π10\Pi^0_1 class does not require much computational power in other respects: a low PA set has the same jump as a computable set, and it computes a solution to every problem with a Π10\Pi^0_1 set of solutions. On the other hand, a PA set is never computable, since it computes a member of DNC2\mathrm{DNC}_2.

4.3 Completions of Peano arithmetic

The name PA comes from logic. Fix a computable coding of the sentences of arithmetic by integers. A set SS of sentences is consistent if no contradiction is provable from it, and complete if for every sentence φ\varphi, either φ∈S\varphi \in S or ¬φ∈S\neg \varphi \in S. We use two facts from logic, which we admit.

  • Proofs are checkable. For a computable set of axioms, the relation ”pp codes a proof of φ\varphi from the axioms” is computable, so the set of theorems is c.e.
  • Peano arithmetic proves true computations. If Φe(e) ⁣ ⁣↓=v\Phi_e(e)\converge = v, then Peano arithmetic PA\mathrm{PA} proves the sentence θe,v\theta_{e,v} expressing ”Φe(e)\Phi_e(e) converges with value vv”, and proves ¬θe,w\neg \theta_{e,w} for every w≠vw \neq v. Here θe,v\theta_{e,v} is the Σ1\Sigma_1 sentence obtained from Corollary 4.6 of “Arithmetic definability”, and the statement is the Σ1\Sigma_1-completeness of PA\mathrm{PA}, together with the uniqueness of the output, provable in PA\mathrm{PA}.

We also use that PA\mathrm{PA} is consistent, since its axioms are true in N\nat.

Theorem 4.4.#

A set XX is PA if and only if it computes a complete consistent set of sentences containing the axioms of PA\mathrm{PA}.

Proof of Theorem 4.4.

(⇒\Rightarrow) Identify a set of sentences with its characteristic function on codes, and let C\mathcal C be the class of sets SS such that: for every sentence φ\varphi, exactly one of φ\varphi and ¬φ\neg \varphi belongs to SS; every axiom of PA\mathrm{PA} belongs to SS; and no proof of a contradiction uses only premises from SS. Its members are exactly the complete consistent sets of sentences containing PA\mathrm{PA}. A set fails to belong to C\mathcal C if and only if one of the following finite witnesses exists: a sentence φ\varphi with both or neither of φ,¬φ\varphi, \neg \varphi in SS; an axiom of PA\mathrm{PA} not in SS; or a proof of a contradiction whose premises are in SS. Each witness involves finitely many bits of SS, and, by the first fact, can be recognized by an algorithm. So the complement of C\mathcal C is a Σ10\Sigma^0_1 class, and C\mathcal C is a Π10\Pi^0_1 class. It is non-empty, since the set of sentences true in N\nat belongs to it. By Theorem 4.2, XX computes a member of C\mathcal C.

(⇐\Leftarrow) Let S≤TXS \leq_T X be a complete consistent set containing PA\mathrm{PA}. Let f(e)=1f(e) = 1 if θe,0∈S\theta_{e,0} \in S, and f(e)=0f(e) = 0 otherwise; ff is computable in XX. If Φe(e) ⁣ ⁣↓=0\Phi_e(e)\converge = 0, then PA⊢θe,0\mathrm{PA} \vdash \theta_{e,0}; since SS is complete, contains PA\mathrm{PA} and is consistent, θe,0∈S\theta_{e,0} \in S (otherwise ¬θe,0∈S\neg \theta_{e,0} \in S and SS would be inconsistent), so f(e)=1≠0f(e) = 1 \neq 0. If Φe(e) ⁣ ⁣↓=1\Phi_e(e)\converge = 1, then PA⊢¬θe,0\mathrm{PA} \vdash \neg \theta_{e,0}, so θe,0∉S\theta_{e,0} \notin S by consistency, and f(e)=0≠1f(e) = 0 \neq 1. If Φe(e)\Phi_e(e) converges to another value or diverges, there is nothing to check. So f∈DNC2f \in \mathrm{DNC}_2.

The second direction contains Gödel’s first incompleteness theorem in Rosser’s form: a complete consistent extension of PA\mathrm{PA} is not computable, since it would compute a member of DNC2\mathrm{DNC}_2. The first direction shows that this is the only obstruction: any degree that computes a member of DNC2\mathrm{DNC}_2 computes a completion. Combined with Corollary 4.3, there are low completions of Peano arithmetic.

4.4 Exercises

Exercise 4.5 ★.

Show that XX is PA if and only if every partial computable function ψ:N⇀{0,1}\psi : \nat \rightharpoonup \{0, 1\} has a total XX-computable extension.

Solution

If XX is PA, let A={n:ψ(n) ⁣ ⁣↓=0}A = \{ n : \psi(n)\converge = 0 \} and B={n:ψ(n) ⁣ ⁣↓=1}B = \{ n : \psi(n)\converge = 1 \}, disjoint c.e. sets, and let D≤TXD \leq_T X separate BB from AA: B⊆DB \subseteq D and D∩A=∅D \cap A = \emptyset. The characteristic function of DD is a total XX-computable extension of ψ\psi.

Conversely, let ψ(e)≃1−Φe(e)\psi(e) \simeq 1 - \Phi_e(e) when Φe(e) ⁣ ⁣↓∈{0,1}\Phi_e(e)\converge \in \{0, 1\}, and ψ(e)\psi(e) undefined otherwise; ψ\psi is partial computable with values in {0,1}\{0, 1\}. Let gg be a total XX-computable extension, and let f(e)=g(e)f(e) = g(e) if g(e)∈{0,1}g(e) \in \{0, 1\} and f(e)=0f(e) = 0 otherwise. If Φe(e) ⁣ ⁣↓∈{0,1}\Phi_e(e)\converge \in \{0, 1\}, then f(e)=g(e)=1−Φe(e)≠Φe(e)f(e) = g(e) = 1 - \Phi_e(e) \neq \Phi_e(e). If Φe(e) ⁣ ⁣↓≥2\Phi_e(e)\converge \geq 2, then f(e)∈{0,1}f(e) \in \{0, 1\} differs from it. So f∈DNC2f \in \mathrm{DNC}_2, and XX is PA.

Exercise 4.6 ★★.#

A set YY is PA relative to XX if it computes a function f:N→{0,1}f : \nat \to \{0, 1\} with f(e)≠ΦeX(e)f(e) \neq \Phi^X_e(e) whenever ΦeX(e) ⁣ ⁣↓\Phi^X_e(e)\converge. Show that if YY is PA relative to XX, then X≤TYX \leq_T Y.

Solution

The proof of Theorem 4.2 relativizes: YY computes a member of every non-empty Π10(X)\Pi^0_1(X) class, that is, every class [T][T] with TT an XX-computable tree. The singleton {X}\{ X \} is such a class, with tree {X ⁣↾n:n∈N}\{ X \uh n : n \in \nat \}. So YY computes its only member, XX.

4.5 Notes and references

The degrees of completions of Peano arithmetic were studied by Scott (1962), who showed that they compute members of all non-empty Π10\Pi^0_1 classes, and by Jockusch and Soare (1972), who proved the characterization by {0,1}\{0, 1\}-valued diagonally non-computable functions, also obtained by Solovay. Jockusch and Soare also proved the existence of low PA degrees. The incompleteness theorem in the form used here is due to Rosser (1936).

  • D. Scott, Algebras of sets binumerable in complete extensions of arithmetic, in Recursive Function Theory, Proceedings of Symposia in Pure Mathematics 5, American Mathematical Society, 1962, 117–121.
  • C. G. Jockusch and R. I. Soare, Degrees of members of Π⁰₁ classes, Pacific Journal of Mathematics 40 (1972), 605–616.
  • C. G. Jockusch and R. I. Soare, Π⁰₁ classes and degrees of theories, Transactions of the American Mathematical Society 173 (1972), 33–56.
  • J. B. Rosser, Extensions of some theorems of Gödel and Church, Journal of Symbolic Logic 1 (1936), 87–91.