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/sigma-pi-sets
@misc{arithmetic-hierarchy-sigma-pi-sets,
  title = {{Σ⁰ₙ and Π⁰ₙ sets}},
  howpublished = {Lesson 1 of \emph{The arithmetic hierarchy}, Computability Theory course notes},
  url = {/courses/arithmetic-hierarchy/sigma-pi-sets},
  note = {Unreleased working version}
}

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

Lesson 1 of 5

Σ⁰ₙ and Π⁰ₙ sets

The arithmetic hierarchy defined by alternations of quantifiers over computable relations; normal forms, closure properties, relativization and many-one reductions.

35 min4 questions
Before this lesson:The Turing jump (Oracles and Turing degrees)
The arithmetic hierarchy · lesson 1 of 5
  1. Σ⁰ₙ and Π⁰ₙ sets
  2. Post's theorem
  3. Classifying index sets
  4. Arithmetic definability
  5. Codes for sets

The c.e. sets are the sets defined by an existential quantifier over a computable relation (Theorem 4.3 of “Computably enumerable sets”). Allowing more quantifiers, alternating between ∃\exists and ∀\forall, gives a classification of sets by the complexity of their definitions: the arithmetic hierarchy, due to Kleene and Mostowski. The set Tot\mathrm{Tot} of indices of total functions, for instance, is defined by ”∀x ∃s\forall x\ \exists s, the computation Φe(x)\Phi_e(x) converges at stage ss”: a universal quantifier followed by an existential one, over a computable relation. This lesson defines the hierarchy and establishes the closure properties used to place sets in it. The next lesson relates it to the Turing jump.

1.1 Definition

Definition 1.1.#

Let n≥1n \geq 1 and k≥1k \geq 1. A relation A⊆NkA \subseteq \nat^k is

  1. Σn0\Sigma^0_n if there is a computable relation R⊆Nk+nR \subseteq \nat^{k + n} such that
x⃗∈A  ⟺  ∃y1 ∀y2 ∃y3⋯Qyn  R(x⃗,y1,…,yn),\vec x \in A \iff \exists y_1\ \forall y_2\ \exists y_3 \cdots Q y_n\ \ R(\vec x, y_1, \dots, y_n),

with nn alternating quantifiers, the last one QQ being ∃\exists if nn is odd and ∀\forall if nn is even; 2. Πn0\Pi^0_n if the same holds with the quantifiers ∀y1 ∃y2⋯\forall y_1\ \exists y_2 \cdots, starting with a universal one; 3. Δn0\Delta^0_n if it is both Σn0\Sigma^0_n and Πn0\Pi^0_n.

By convention, the Σ00\Sigma^0_0, Π00\Pi^0_0 and Δ00\Delta^0_0 relations are the computable relations. A relation is arithmetic if it is Σn0\Sigma^0_n for some nn.

The superscript 00 indicates that the quantifiers range over integers, that is, first-order objects; the analytic hierarchy Σn1\Sigma^1_n, studied in the course on higher computability, quantifies over sets of integers. Some authors use Δ0\Delta_0 for the relations defined by formulas with bounded quantifiers only; these are particular computable relations.

Example 1.2.#
  1. A set is Σ10\Sigma^0_1 if it is of the form {x:∃y R(x,y)}\{ x : \exists y\ R(x, y) \} with RR computable: these are exactly the c.e. sets (Theorem 4.3 of “Computably enumerable sets”).
  2. Tot={e:∀x ∃s Φe(x)[s] ⁣ ⁣↓}\mathrm{Tot} = \{ e : \forall x\ \exists s\ \Phi_e(x)[s]\converge \} is Π20\Pi^0_2, since the relation Φe(x)[s] ⁣ ⁣↓\Phi_e(x)[s]\converge is computable (Proposition 1.4 of “Programs as numbers”).
  3. Fin={e:We is finite}={e:∃n ∀x ∀s (x>n→Φe(x)[s] ⁣ ⁣↑)}\mathrm{Fin} = \{ e : W_e \text{ is finite} \} = \{ e : \exists n\ \forall x\ \forall s\ (x > n \to \Phi_e(x)[s]\diverge) \} is Σ20\Sigma^0_2, once the two universal quantifiers are merged (see below).
  4. Cof={e:We is cofinite}={e:∃n ∀x ∃s (x≤n∨Φe(x)[s] ⁣ ⁣↓)}\mathrm{Cof} = \{ e : W_e \text{ is cofinite} \} = \{ e : \exists n\ \forall x\ \exists s\ (x \leq n \vee \Phi_e(x)[s]\converge) \} is Σ30\Sigma^0_3.

Δ01Σ01Π01Δ02Σ02Π02Δ03Σ03Π03Δ04⋯

The first levels of the arithmetic hierarchy. Each level Δn0\Delta^0_n is contained in Σn0\Sigma^0_n and in Πn0\Pi^0_n, and both are contained in Δn+10\Delta^0_{n+1}. All inclusions are strict, as shown in the next lesson.

1.2 Normal forms

Proposition 1.3.#

Let n≥1n \geq 1.

  1. A relation is Σn0\Sigma^0_n if and only if its complement is Πn0\Pi^0_n.
  2. Every Σn0\Sigma^0_n or Πn0\Pi^0_n relation is Δn+10\Delta^0_{n+1}.
  3. Contraction. If x⃗∈A  ⟺  ∃y1∃y2 P(x⃗,y1,y2)\vec x \in A \iff \exists y_1 \exists y_2\ P(\vec x, y_1, y_2) with PP a Πn−10\Pi^0_{n-1} relation, then x⃗∈A  ⟺  ∃z P(x⃗,π0(z),π1(z))\vec x \in A \iff \exists z\ P(\vec x, \pi_0(z), \pi_1(z)); dually for universal quantifiers. Consequently, a relation defined by a prefix of quantifiers over a computable relation is Σn0\Sigma^0_n if the prefix starts with ∃\exists and has nn blocks of quantifiers of the same kind, and Πn0\Pi^0_n if it starts with ∀\forall and has nn blocks.
Proof of Proposition 1.3.
  1. The negation of ∃y1 ∀y2⋯R\exists y_1\ \forall y_2 \cdots R is ∀y1 ∃y2⋯¬R\forall y_1\ \exists y_2 \cdots \neg R, and the complement of a computable relation is computable.
  2. Add a dummy quantifier: if x⃗∈A  ⟺  ∃y1⋯Qyn R\vec x \in A \iff \exists y_1 \cdots Q y_n\ R, then x⃗∈A  ⟺  ∀y0 ∃y1⋯Qyn R\vec x \in A \iff \forall y_0\ \exists y_1 \cdots Q y_n\ R and x⃗∈A  ⟺  ∃y1⋯Qyn Q′yn+1 R\vec x \in A \iff \exists y_1 \cdots Q y_n\ Q' y_{n+1}\ R, where RR does not depend on y0y_0 or yn+1y_{n+1}. The first form is Πn+10\Pi^0_{n+1}, the second Σn+10\Sigma^0_{n+1}. The case of Πn0\Pi^0_n relations is symmetric.
  3. The pairing function is a bijection, so ∃y1∃y2\exists y_1 \exists y_2 ranges over the same pairs as ∃z\exists z with (y1,y2)=(π0(z),π1(z))(y_1, y_2) = (\pi_0(z), \pi_1(z)). The relation P(x⃗,π0(z),π1(z))P(\vec x, \pi_0(z), \pi_1(z)) is Πn−10\Pi^0_{n-1}: substitute π0(z),π1(z)\pi_0(z), \pi_1(z) for y1,y2y_1, y_2 in the computable relation of a normal form of PP, which gives a computable relation, since π0\pi_0 and π1\pi_1 are computable. So merging two adjacent quantifiers of the same kind does not change the level. Applying this to the quantifiers of a prefix over a computable relation, which may stand anywhere in the prefix, merges each block into a single quantifier.

1.3 Closure properties

To place a set in the hierarchy, one writes a definition and counts alternations. This is justified by the following closure properties.

Proposition 1.4.#

Let n≥1n \geq 1. The Σn0\Sigma^0_n relations and the Πn0\Pi^0_n relations are closed under:

  1. finite unions and intersections;
  2. substitution of total computable functions: if A⊆NmA \subseteq \nat^m is Σn0\Sigma^0_n and g1,…,gm:Nk→Ng_1, \dots, g_m : \nat^k \to \nat are total computable, then {x⃗:(g1(x⃗),…,gm(x⃗))∈A}\{ \vec x : (g_1(\vec x), \dots, g_m(\vec x)) \in A \} is Σn0\Sigma^0_n, and likewise for Πn0\Pi^0_n;
  3. bounded quantifiers ∃z<y\exists z < y and ∀z<y\forall z < y.

Moreover, Σn0\Sigma^0_n relations are closed under existential quantification ∃y\exists y, and Πn0\Pi^0_n relations under universal quantification ∀y\forall y.

Proof of Proposition 1.4.

By induction on nn. For n=1n = 1, the Σ10\Sigma^0_1 relations are the c.e. relations, and their closure under finite unions and intersections, substitution (preimages under computable functions), projection and bounded universal quantification is Proposition 4.5 of “Computably enumerable sets”; bounded existential quantification ∃z<y P\exists z < y\ P is the projection ∃z (z<y∧P)\exists z\ (z < y \wedge P). The Π10\Pi^0_1 case follows by complementation, using item 1 of Proposition 1.3.

Suppose the statement holds for nn, and let AA and BB be Σn+10\Sigma^0_{n+1}: x⃗∈A  ⟺  ∃y P(x⃗,y)\vec x \in A \iff \exists y\ P(\vec x, y) and x⃗∈B  ⟺  ∃y Q(x⃗,y)\vec x \in B \iff \exists y\ Q(\vec x, y), with PP and QQ in Πn0\Pi^0_n (by definition, after grouping the quantifiers following the first one).

  1. x⃗∈A∩B  ⟺  ∃z (P(x⃗,π0(z))∧Q(x⃗,π1(z)))\vec x \in A \cap B \iff \exists z\ \big(P(\vec x, \pi_0(z)) \wedge Q(\vec x, \pi_1(z))\big), and x⃗∈A∪B  ⟺  ∃y (P(x⃗,y)∨Q(x⃗,y))\vec x \in A \cup B \iff \exists y\ (P(\vec x, y) \vee Q(\vec x, y)). The relations inside are Πn0\Pi^0_n by the induction hypothesis (substitution, then intersection or union).
  2. (g1(x⃗),… )∈A  ⟺  ∃y P(g1(x⃗),…,y)(g_1(\vec x), \dots) \in A \iff \exists y\ P(g_1(\vec x), \dots, y), and the relation inside is Πn0\Pi^0_n by the induction hypothesis.
  3. ∃z<w ∃y P(x⃗,z,y)  ⟺  ∃u (π0(u)<w∧P(x⃗,π0(u),π1(u)))\exists z < w\ \exists y\ P(\vec x, z, y) \iff \exists u\ (\pi_0(u) < w \wedge P(\vec x, \pi_0(u), \pi_1(u))), which is Σn+10\Sigma^0_{n+1}. For the bounded universal quantifier, the key equivalence is
∀z<w  ∃y P(x⃗,z,y)  ⟺  ∃v  ∀z<w  ∃y<v  P(x⃗,z,y).\forall z < w\ \ \exists y\ P(\vec x, z, y) \iff \exists v\ \ \forall z < w\ \ \exists y < v\ \ P(\vec x, z, y).

From left to right, choose for each of the finitely many z<wz < w a witness yzy_z, and let vv exceed all of them; the converse is clear. By the induction hypothesis, ∃y<v P\exists y < v\ P is Πn0\Pi^0_n (bounded quantification), then ∀z<w ∃y<v P\forall z < w\ \exists y < v\ P is Πn0\Pi^0_n (bounded quantification again), so the right-hand side is Σn+10\Sigma^0_{n+1}.

Existential quantification: ∃u ∃y P(x⃗,u,y)\exists u\ \exists y\ P(\vec x, u, y) is contracted to a single quantifier. The Πn+10\Pi^0_{n+1} case follows by complementation.

The equivalence used for the bounded universal quantifier is called collection: finitely many witnesses can be bounded by a single number. It fails for unbounded quantifiers, which is why ∀z ∃y\forall z\ \exists y cannot in general be rewritten as ∃v ∀z\exists v\ \forall z.

1.4 Relativization and reductions

Definition 1.5.#

For X∈2NX \in \cs, the classes Σn0(X)\Sigma^0_n(X), Πn0(X)\Pi^0_n(X) and Δn0(X)\Delta^0_n(X) are defined as in Definition 1.1, with XX-computable relations in place of computable ones.

All the results above relativize. In particular, the Σ10(X)\Sigma^0_1(X) sets are the XX-c.e. sets, and the Δ10(X)\Delta^0_1(X) sets are the XX-computable sets.

Proposition 1.6.#
  1. A set is Σ10\Sigma^0_1 if and only if it is c.e., and Δ10\Delta^0_1 if and only if it is computable.
  2. If A≤mBA \leq_m B and BB is Σn0(X)\Sigma^0_n(X), then AA is Σn0(X)\Sigma^0_n(X); likewise for Πn0(X)\Pi^0_n(X) and Δn0(X)\Delta^0_n(X).
Proof of Proposition 1.6.
  1. The first statement is Theorem 4.3 of “Computably enumerable sets”. A set is Δ10\Delta^0_1 if and only if it and its complement are c.e., that is, if and only if it is computable (Proposition 4.8 of “Computably enumerable sets”).
  2. If ff reduces AA to BB, then A={x:f(x)∈B}A = \{ x : f(x) \in B \}, and the result is the substitution property of Proposition 1.4, relativized to XX.

Many-one reducibility preserves the levels of the hierarchy, while Turing reducibility does not: every set is Turing equivalent to its complement, and the complement of a Σ10\Sigma^0_1 set that is not computable is not Σ10\Sigma^0_1. This is why completeness at a level of the hierarchy is defined with many-one reductions.

Definition 1.7.#

A set AA is Σn0\Sigma^0_n-complete if it is Σn0\Sigma^0_n and B≤mAB \leq_m A for every Σn0\Sigma^0_n set BB; Πn0\Pi^0_n-complete sets are defined similarly.

By Theorem 4.12 of “Computably enumerable sets”, ∅′\emptyset' is Σ10\Sigma^0_1-complete, and so its complement is Π10\Pi^0_1-complete, since A≤mBA \leq_m B if and only if N∖A≤mN∖B\nat \setminus A \leq_m \nat \setminus B. A Σn0\Sigma^0_n-complete set is not Πn0\Pi^0_n as soon as some Σn0\Sigma^0_n set is not Πn0\Pi^0_n; the next lesson shows that this is always the case.

1.5 Exercises

Exercise 1.8 ★.

Show that {e:We=∅}\{ e : W_e = \emptyset \} is Π10\Pi^0_1, that {e:∣We∣≥2}\{ e : |W_e| \geq 2 \} is Σ10\Sigma^0_1, and that {e:∣We∣=2}\{ e : |W_e| = 2 \} is Δ20\Delta^0_2.

Solution

We=∅  ⟺  ∀x ∀s Φe(x)[s] ⁣ ⁣↑W_e = \emptyset \iff \forall x\ \forall s\ \Phi_e(x)[s]\diverge. Next, ∣We∣≥2  ⟺  ∃x ∃x′ ∃s (x≠x′∧Φe(x)[s] ⁣ ⁣↓∧Φe(x′)[s] ⁣ ⁣↓)|W_e| \geq 2 \iff \exists x\ \exists x'\ \exists s\ (x \neq x' \wedge \Phi_e(x)[s]\converge \wedge \Phi_e(x')[s]\converge). Finally, ∣We∣=2|W_e| = 2 is the intersection of ∣We∣≥2|W_e| \geq 2, which is Σ10\Sigma^0_1, and of the complement of ∣We∣≥3|W_e| \geq 3, which is Π10\Pi^0_1 by the same argument. Both are Δ20\Delta^0_2, and Δ20\Delta^0_2 is closed under intersection.

Exercise 1.9 ★.

Show that for each nn there are only countably many Σn0\Sigma^0_n sets, and deduce that the Σn0\Sigma^0_n sets are not closed under arbitrary countable unions.

Solution

A Σn0\Sigma^0_n set is determined by a computable relation, which has an index; so there are countably many. Hence some set AA is not arithmetic, by Cantor. But A=⋃x∈A{x}A = \bigcup_{x \in A} \{ x \} is a countable union of finite sets, which are computable, hence Σn0\Sigma^0_n. The closure under existential quantification only applies to uniform unions, ⋃y{x:(x,y)∈B}\bigcup_y \{ x : (x, y) \in B \} with BB a single Σn0\Sigma^0_n relation.

Exercise 1.10 ★★.

Show that {e:We is computable}\{ e : W_e \text{ is computable} \} is Σ30\Sigma^0_3.

Solution

Let E(e,i,x)E(e, i, x) be the relation

(x∈We∧∃s Φi(x)[s] ⁣ ⁣↓=1)∨(x∉We∧∃s Φi(x)[s] ⁣ ⁣↓=0).\big(x \in W_e \wedge \exists s\ \Phi_i(x)[s]\converge = 1\big) \vee \big(x \notin W_e \wedge \exists s\ \Phi_i(x)[s]\converge = 0\big).

Then WeW_e is computable if and only if ∃i ∀x E(e,i,x)\exists i\ \forall x\ E(e, i, x): if ∀x E(e,i,x)\forall x\ E(e, i, x), then Φi\Phi_i is total with values in {0,1}\{0, 1\} and computes WeW_e; conversely, an index of the characteristic function of WeW_e satisfies ∀x E(e,i,x)\forall x\ E(e, i, x). The relation EE is a Boolean combination of Σ10\Sigma^0_1 relations, hence Δ20\Delta^0_2 and in particular Π20\Pi^0_2. So ∀x E\forall x\ E is Π20\Pi^0_2, and ∃i ∀x E\exists i\ \forall x\ E is Σ30\Sigma^0_3.

1.6 Notes and references

The arithmetic hierarchy was introduced independently by Kleene (1943) and Mostowski (1947). The notation Σn0\Sigma^0_n, Πn0\Pi^0_n is due to Addison, by analogy with the Borel hierarchy of descriptive set theory, of which it is the effective counterpart. Rogers’ and Soare’s textbooks treat the hierarchy in detail.

  • S. C. Kleene, Recursive predicates and quantifiers, Transactions of the American Mathematical Society 53 (1943), 41–73.
  • A. Mostowski, On definable sets of positive integers, Fundamenta Mathematicae 34 (1947), 81–112.
  • H. Rogers, Theory of Recursive Functions and Effective Computability, McGraw-Hill, 1967.
  • R. I. Soare, Recursively Enumerable Sets and Degrees, Springer, 1987.