Cite

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

/courses/arithmetic-hierarchy/index-sets
@misc{arithmetic-hierarchy-index-sets,
  title = {{Classifying index sets}},
  howpublished = {Lesson 3 of \emph{The arithmetic hierarchy}, Computability Theory course notes},
  url = {/courses/arithmetic-hierarchy/index-sets},
  note = {Unreleased working version}
}

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

Lesson 3 of 5

Classifying index sets

Upper bounds by counting quantifiers, lower bounds by reductions; Fin is Σ⁰₂-complete, Tot and Inf are Π⁰₂-complete, Cof is Σ⁰₃-complete.

45 min4 questions
Before this lesson:Post's theorem · Rice's theorem (Fundamentals of computability)
The arithmetic hierarchy · lesson 3 of 5
  1. Σ⁰ₙ and Π⁰ₙ sets
  2. Post's theorem
  3. Classifying index sets
  4. Arithmetic definability
  5. Codes for sets

Rice’s theorem says that no non-trivial index set is computable. The arithmetic hierarchy measures how far from computable they are. To classify a set is to find the level of the hierarchy at which it is complete. This involves two steps: an upper bound, obtained by writing a definition and counting quantifiers, and a lower bound, obtained by reducing a complete set of that level to it. This lesson classifies the basic index sets; by Post’s theorem, the classification also determines their Turing degrees.

3.1 The method

If AA is Σn0\Sigma^0_n-complete, then by Theorem 2.2, A≡m∅(n)A \equiv_m \emptyset^{(n)}, so A≡T∅(n)A \equiv_T \emptyset^{(n)}, and AA is not Πn0\Pi^0_n (Corollary 2.3). Hence to show that a set BB is Σn0\Sigma^0_n-complete, it suffices to show that BB is Σn0\Sigma^0_n and that A≤mBA \leq_m B for some Σn0\Sigma^0_n-complete set AA, or directly that every Σn0\Sigma^0_n set reduces to BB. Reductions are built with the SMN theorem, in the parameter form (Parameter form): from xx, one computes an index of a c.e. set or partial computable function whose behavior encodes the membership of xx in the given set.

3.2 The second level

Theorem 3.1.#

The set Fin={e:We is finite}\mathrm{Fin} = \{ e : W_e \text{ is finite} \} is Σ20\Sigma^0_2-complete, and Inf=N∖Fin\mathrm{Inf} = \nat \setminus \mathrm{Fin} is Π20\Pi^0_2-complete.

Proof of Theorem 3.1.

Fin\mathrm{Fin} is Σ20\Sigma^0_2 by Example 1.2. Let AA be Σ20\Sigma^0_2: x∈A  ⟺  ∃y ∀z R(x,y,z)x \in A \iff \exists y\ \forall z\ R(x, y, z) with RR computable. Consider the sets

Vx={n:∀y<n  ∃z ¬R(x,y,z)}.V_x = \{ n : \forall y < n\ \ \exists z\ \neg R(x, y, z) \}.

The relation n∈Vxn \in V_x is c.e., by closure under bounded quantification, so there is a computable ff with Wf(x)=VxW_{f(x)} = V_x for every xx. If x∉Ax \notin A, every yy has a zz with ¬R(x,y,z)\neg R(x, y, z), so Vx=NV_x = \nat is infinite. If x∈Ax \in A, let y0y_0 be such that ∀z R(x,y0,z)\forall z\ R(x, y_0, z); then no n>y0n > y_0 belongs to VxV_x, so VxV_x is finite. Hence x∈A  ⟺  f(x)∈Finx \in A \iff f(x) \in \mathrm{Fin}. The statement on Inf\mathrm{Inf} follows by complementation.

Theorem 3.2.#

The set Tot={e:Φe is total}\mathrm{Tot} = \{ e : \Phi_e \text{ is total} \} is Π20\Pi^0_2-complete.

Proof of Theorem 3.2.

Tot\mathrm{Tot} is Π20\Pi^0_2 by Example 1.2. Let AA be Π20\Pi^0_2: x∈A  ⟺  ∀y ∃z R(x,y,z)x \in A \iff \forall y\ \exists z\ R(x, y, z) with RR computable. Let ff be computable with Φf(x)(y)≃μz R(x,y,z)\Phi_{f(x)}(y) \simeq \mu z\ R(x, y, z). Then Φf(x)\Phi_{f(x)} is total if and only if every yy has a witness zz, that is, if and only if x∈Ax \in A.

So Fin\mathrm{Fin}, Inf\mathrm{Inf} and Tot\mathrm{Tot} are Turing equivalent to ∅′′\emptyset'', as announced in Proposition 3.5 of “The Turing jump”. None of them is computable in ∅′\emptyset'.

3.3 The third level

Theorem 3.3.#

The set Cof={e:We is cofinite}\mathrm{Cof} = \{ e : W_e \text{ is cofinite} \} is Σ30\Sigma^0_3-complete.

Proof of Theorem 3.3.

Cof\mathrm{Cof} is Σ30\Sigma^0_3 by Example 1.2. Let AA be Σ30\Sigma^0_3: x∈A  ⟺  ∃n P(x,n)x \in A \iff \exists n\ P(x, n) with PP a Π20\Pi^0_2 relation. By Theorem 3.1, Inf\mathrm{Inf} is Π20\Pi^0_2-complete, so there is a computable gg with P(x,n)  ⟺  Wg(x,n)P(x, n) \iff W_{g(x, n)} is infinite. We build, uniformly in xx, a c.e. set VxV_x which is cofinite if and only if some Wg(x,n)W_{g(x, n)} is infinite.

Construction. We enumerate VxV_x by stages, starting with Vx[0]=∅V_x[0] = \emptyset. At stage s+1s + 1, let nn be the least n≤sn \leq s such that Wg(x,n)[s+1]≠Wg(x,n)[s]W_{g(x, n)}[s + 1] \neq W_{g(x, n)}[s], if there is one; we say that nn acts at stage s+1s + 1. In that case, list the elements of {0,…,s−1}∖Vx[s]\{0, \dots, s - 1\} \setminus V_x[s] in increasing order, and enumerate into VxV_x all of them except the first nn. If no n≤sn \leq s acts, do nothing. The construction is computable uniformly in xx, so there is a computable hh with Wh(x)=VxW_{h(x)} = V_x.

The numbers not in VxV_x at stage ss, in increasing order, are called the markers at stage ss. A stage at which nn acts keeps the first nn markers below ss and removes the others.

If some Wg(x,n)W_{g(x, n)} is infinite, VxV_x is cofinite. Fix such an nn. Then Wg(x,n)W_{g(x, n)} receives new elements at infinitely many stages, and at each of them some n′≤nn' \leq n acts. Suppose m0<m1<⋯<mnm_0 < m_1 < \dots < m_n are n+1n + 1 numbers outside VxV_x. Take a stage s+1s + 1 with s>mns > m_n at which some n′≤nn' \leq n acts. The numbers m0,…,mnm_0, \dots, m_n are all below ss and outside Vx[s]V_x[s], and mnm_n is not among the first nn markers below ss, since m0,…,mn−1m_0, \dots, m_{n-1} precede it. So mnm_n is enumerated into VxV_x, a contradiction. Hence the complement of VxV_x has at most nn elements.

If every Wg(x,n)W_{g(x, n)} is finite, VxV_x is coinfinite. We show that for every kk, the kk-th marker (counting from 00) eventually reaches a final value. Only actions of some n≤kn \leq k remove the kk-th marker or a smaller one, since an action of nn keeps the first nn markers. Each n≤kn \leq k acts only finitely often, since Wg(x,n)W_{g(x, n)} is finite. After the last action of the numbers n≤kn \leq k, the first k+1k + 1 markers are never removed: they are final. So the complement of VxV_x contains the final values of all markers, and is infinite.

Hence x∈A  ⟺  h(x)∈Cofx \in A \iff h(x) \in \mathrm{Cof}.

The construction is a first example of a marker argument: each nn may, when it acts, destroy all the markers below the current stage except the first nn. If some nn acts infinitely often, at most nn markers survive; if every nn acts finitely often, each marker settles. Arguments of this kind, with priorities between the actors, are the subject of the course on Post’s problem.

3.4 Summary

The following table collects the classification of the basic index sets. The sets marked as admitted are classified in the references; the others are classified in this lesson or its exercises.

Set Definition Complete for
∅′\emptyset', {e:We≠∅}\{ e : W_e \neq \emptyset \} Φe(e) ⁣ ⁣↓\Phi_e(e)\converge; ∃x x∈We\exists x\ x \in W_e Σ10\Sigma^0_1
{e:We=∅}\{ e : W_e = \emptyset \} ∀x x∉We\forall x\ x \notin W_e Π10\Pi^0_1
Fin\mathrm{Fin} ∃n ∀x>n  x∉We\exists n\ \forall x > n\ \ x \notin W_e Σ20\Sigma^0_2
Inf\mathrm{Inf}, Tot\mathrm{Tot}, {e:We=N}\{ e : W_e = \nat \} ∀n ∃x>n  x∈We\forall n\ \exists x > n\ \ x \in W_e; ∀x Φe(x) ⁣ ⁣↓\forall x\ \Phi_e(x)\converge Π20\Pi^0_2
Cof\mathrm{Cof} ∃n ∀x>n  x∈We\exists n\ \forall x > n\ \ x \in W_e Σ30\Sigma^0_3
Rec={e:We computable}\mathrm{Rec} = \{ e : W_e \text{ computable} \} (admitted) see Σ⁰ₙ and Π⁰ₙ sets Σ30\Sigma^0_3

The completeness of Rec\mathrm{Rec} requires a more elaborate construction, which we admit; a proof is in Soare’s Recursively Enumerable Sets and Degrees, Chapter IV, Section 3.

3.5 Exercises

Exercise 3.4 ★.

Show that {e:We=N}\{ e : W_e = \nat \} is Π20\Pi^0_2-complete, and that {e:We≠∅}\{ e : W_e \neq \emptyset \} is Σ10\Sigma^0_1-complete.

Solution

We=N  ⟺  ∀x ∃s Φe(x)[s] ⁣ ⁣↓W_e = \nat \iff \forall x\ \exists s\ \Phi_e(x)[s]\converge, a Π20\Pi^0_2 condition, and We=dom(Φe)W_e = \mathrm{dom}(\Phi_e), so the reduction of Theorem 3.2 also reduces every Π20\Pi^0_2 set to this set. The set {e:We≠∅}\{ e : W_e \neq \emptyset \} is Σ10\Sigma^0_1, and ∅′\emptyset' reduces to it by Proposition 3.5 of “The halting problem”.

Exercise 3.5 ★.

Show that Cof\mathrm{Cof} is not Π30\Pi^0_3, and that Fin\mathrm{Fin} is not Π20\Pi^0_2.

Solution

If Cof\mathrm{Cof} were Π30\Pi^0_3, every Σ30\Sigma^0_3 set, being many-one reducible to Cof\mathrm{Cof}, would be Π30\Pi^0_3 by Proposition 1.6, contradicting Corollary 2.3. The same argument applies to Fin\mathrm{Fin} at level 22.

Exercise 3.6 ★★.

Show that the set {e:Φe is total and constant}\{ e : \Phi_e \text{ is total and constant} \} is Π20\Pi^0_2-complete.

Solution

Φe\Phi_e is total and constant if and only if Φe\Phi_e is total and ∀x ∀y ∀s (Φe(x)[s] ⁣ ⁣↓∧Φe(y)[s] ⁣ ⁣↓→Φe(x)[s]=Φe(y)[s])\forall x\ \forall y\ \forall s\ \big( \Phi_e(x)[s]\converge \wedge \Phi_e(y)[s]\converge \to \Phi_e(x)[s] = \Phi_e(y)[s] \big); this is the conjunction of a Π20\Pi^0_2 and a Π10\Pi^0_1 condition, hence Π20\Pi^0_2. For completeness, let A={x:∀y ∃z R(x,y,z)}A = \{ x : \forall y\ \exists z\ R(x, y, z) \} with RR computable, and let ff be computable with Φf(x)(y)≃0\Phi_{f(x)}(y) \simeq 0 if ∃z R(x,y,z)\exists z\ R(x, y, z), the program searching for zz and diverging if there is none. Then Φf(x)\Phi_{f(x)} is defined exactly on the yy that have a witness, and takes the value 00 there; it is total and constant if and only if x∈Ax \in A.

3.6 Notes and references

The classification of the index sets of this lesson is presented in Rogers’ and Soare’s books. Marker arguments, of which the proof for Cof\mathrm{Cof} is a simple instance, are a basic tool in the study of c.e. sets.

  • H. Rogers, Theory of Recursive Functions and Effective Computability, McGraw-Hill, 1967.
  • R. I. Soare, Recursively Enumerable Sets and Degrees, Springer, 1987.