Cite

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

/courses/induction-conservation/induction-collection
@misc{induction-conservation-induction-collection,
  title = {{Induction, collection and least numbers}},
  howpublished = {Lesson 1 of \emph{Induction and conservation}, Computability Theory course notes},
  url = {/courses/induction-conservation/induction-collection},
  note = {Unreleased working version}
}

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

Lesson 1 of 4

Induction, collection and least numbers

The schemes of Σ⁰ₙ and Π⁰ₙ induction, collection and least number principles; closure of the classes Σ⁰ₙ under bounded quantification in the presence of collection; Σ⁰ₙ and Π⁰ₙ induction are equivalent; the hierarchy IΣ⁰ₙ₊₁ ⇒ BΣ⁰ₙ₊₁ ⇒ IΣ⁰ₙ.

40 min4 questions
Before this lesson:The base theory RCA₀ (Second-order arithmetic)
Induction and conservation · lesson 1 of 4
  1. Induction, collection and least numbers
  2. The pigeonhole principle and Σ⁰₂ collection
  3. First-order parts and conservation
  4. Harrington's conservation theorem

In an ω-model, every form of induction holds, because the first-order part is the true N\nat. In a model with non-standard numbers, induction may fail for complicated formulas: a formula may define a proper initial segment closed under successor, a cut, which contains 00 but not every number. The axioms of RCA0\mathsf{RCA}_0 only grant induction for Σ10\Sigma^0_1 formulas, and many principles of ordinary mathematics turn out to be equivalent to forms of induction or of the related collection principle. This lesson sets up the hierarchy of these schemes, following Paris and Kirby.

1.1 The schemes

All formulas are formulas of L2\mathcal L_2, with number and set parameters (Definition 1.1 of “The language and models of second-order arithmetic”); the classes Σn0\Sigma^0_n and Πn0\Pi^0_n count blocks of quantifiers of the same kind as one.

Definition 1.1.#

Let n≥0n \geq 0 and Γ\Gamma be Σn0\Sigma^0_n or Πn0\Pi^0_n. For every formula φ∈Γ\varphi \in \Gamma, the universal closures of the following formulas form respectively the schemes IΓ\mathsf I\Gamma, BΓ\mathsf B\Gamma and LΓ\mathsf L\Gamma:

  1. induction: (φ(0)∧∀x (φ(x)→φ(x+1)))→∀x φ(x)(\varphi(0) \wedge \forall x\, (\varphi(x) \rightarrow \varphi(x+1))) \rightarrow \forall x\, \varphi(x);
  2. collection: ∀x<a ∃y φ(x,y)→∃b ∀x<a ∃y<b φ(x,y)\forall x < a\ \exists y\ \varphi(x, y) \rightarrow \exists b\ \forall x < a\ \exists y < b\ \varphi(x, y);
  3. least number principle: ∃x φ(x)→∃x (φ(x)∧∀y<x ¬φ(y))\exists x\, \varphi(x) \rightarrow \exists x\, (\varphi(x) \wedge \forall y < x\ \neg \varphi(y)).

The theory IΔ00\mathsf I\Delta^0_0 consists of the basic axioms (Definition 1.2 of “The language and models of second-order arithmetic”) and the scheme IΣ00\mathsf I\Sigma^0_0 of induction for bounded formulas.

The formula φ\varphi may contain other free variables, which act as parameters. Collection says that a “finite” family of nonempty sets of witnesses has a common bound; here finite means indexed by the numbers below some aa, which may be non-standard. In the true N\nat, every instance of the three schemes holds. All theories in this course extend IΔ00\mathsf I\Delta^0_0, and RCA0\mathsf{RCA}_0 contains IΣ10\mathsf I\Sigma^0_1.

Fact 1.2.#

IΔ00\mathsf I\Delta^0_0 proves the elementary properties of addition, multiplication and order: << is a discrete linear order with least element 00, every x≠0x \neq 0 is y+1y + 1 for a unique yy, x≤yx \leq y if and only if x+z=yx + z = y for some zz, which is then unique, polynomials with non-negative coefficients are monotone, and the pairing function ⟨u,v⟩=(u+v)(u+v+1)2+v\langle u, v \rangle = \frac{(u + v)(u + v + 1)}{2} + v is a bijection, definable by a bounded formula, with u,v≤⟨u,v⟩u, v \leq \langle u, v \rangle.

These properties are proved by bounded induction, in the same way as in Peano arithmetic; see Kaye, Models of Peano Arithmetic, Chapter 2. We use them freely.

A formula obtained from Σn0\Sigma^0_n formulas by conjunction, disjunction, bounded quantification and existential quantification is not Σn0\Sigma^0_n in general, because a bounded universal quantifier cannot be moved across an unbounded existential one. In the true N\nat, ∀x<a ∃y φ\forall x < a\ \exists y\ \varphi is equivalent to ∃b ∀x<a ∃y<b φ\exists b\ \forall x < a\ \exists y < b\ \varphi; this equivalence is exactly an instance of collection.

Lemma 1.3.#

Let n≥1n \geq 1. Over IΔ00+BΣn0\mathsf I\Delta^0_0 + \mathsf B\Sigma^0_n:

  1. every formula obtained from Σn0\Sigma^0_n formulas by ∧\wedge, ∨\vee, bounded quantifiers ∃x<t\exists x < t, ∀x<t\forall x < t and existential quantifiers is provably equivalent to a Σn0\Sigma^0_n formula;
  2. every formula obtained from Πn0\Pi^0_n formulas by ∧\wedge, ∨\vee, bounded quantifiers and universal quantifiers is provably equivalent to a Πn0\Pi^0_n formula.
Proof of Lemma 1.3.

Item 2 follows from item 1 by negation, since the negation of a Σn0\Sigma^0_n formula is logically equivalent to a Πn0\Pi^0_n formula. It suffices to treat one operation at a time. By renaming bound variables, ∃y α∧∃z β\exists y\, \alpha \wedge \exists z\, \beta is equivalent to ∃y ∃z (α∧β)\exists y\, \exists z\, (\alpha \wedge \beta), and similarly for ∨\vee; ∃x<t ∃y α\exists x < t\ \exists y\, \alpha is equivalent to ∃y ∃x<t α\exists y\ \exists x < t\ \alpha; and an existential quantifier merges with the first block. We may also assume that the first block of a Σn0\Sigma^0_n formula consists of a single quantifier: ∃y1 ∃y2 α\exists y_1\, \exists y_2\, \alpha is equivalent to ∃w ∃y1<w ∃y2<w α\exists w\ \exists y_1 < w\ \exists y_2 < w\ \alpha, and ∃y1<w ∃y2<w α\exists y_1 < w\ \exists y_2 < w\ \alpha, with α∈Πn−10\alpha \in \Pi^0_{n-1}, is bounded if n=1n = 1 and equivalent to a Πn−10\Pi^0_{n-1} formula by item 2 for n−1n - 1 if n≥2n \geq 2, in the induction below. The only non-trivial case is the bounded universal quantifier. We proceed by induction on nn.

For n=1n = 1, let φ=∃y δ(x,y)\varphi = \exists y\, \delta(x, y) with δ\delta bounded. Then ∀x<t ∃y δ\forall x < t\ \exists y\ \delta implies ∃b ∀x<t ∃y<b δ\exists b\ \forall x < t\ \exists y < b\ \delta by BΣ10\mathsf B\Sigma^0_1, and the converse is logical. The second formula is Σ10\Sigma^0_1.

Let n≥1n \geq 1, assume the lemma for nn, and assume BΣn+10\mathsf B\Sigma^0_{n+1}, which contains BΣn0\mathsf B\Sigma^0_n. A Σn+10\Sigma^0_{n+1} formula has the form ∃y ψ\exists y\, \psi with ψ∈Πn0\psi \in \Pi^0_n. Conjunctions and disjunctions of Πn0\Pi^0_n formulas, and bounded quantifications of them, are equivalent to Πn0\Pi^0_n formulas by item 2 for nn; so the easy cases above stay within Σn+10\Sigma^0_{n+1}. For the bounded universal quantifier, ∀x<t ∃y ψ(x,y)\forall x < t\ \exists y\ \psi(x, y) is equivalent to ∃b ∀x<t ∃y<b ψ(x,y)\exists b\ \forall x < t\ \exists y < b\ \psi(x, y) by BΣn+10\mathsf B\Sigma^0_{n+1} applied to ψ\psi, which is Σn+10\Sigma^0_{n+1}. By item 2 for nn, ∀x<t ∃y<b ψ\forall x < t\ \exists y < b\ \psi is equivalent to a Πn0\Pi^0_n formula, so the whole formula is equivalent to a Σn+10\Sigma^0_{n+1} formula.

1.2 Σ⁰ₙ and Π⁰ₙ induction

Theorem 1.4 (Paris, Kirby).#

For every n≥1n \geq 1, IΔ00\mathsf I\Delta^0_0 proves that IΣn0\mathsf I\Sigma^0_n and IΠn0\mathsf I\Pi^0_n are equivalent.

Proof of Theorem 1.4.

(IΣn0⇒IΠn0\mathsf I\Sigma^0_n \Rightarrow \mathsf I\Pi^0_n) Let φ\varphi be Πn0\Pi^0_n with φ(0)\varphi(0) and ∀x (φ(x)→φ(x+1))\forall x\, (\varphi(x) \rightarrow \varphi(x+1)), and suppose that ¬φ(a)\neg \varphi(a). Let

θ(y)≡y≤a→∃x (x+y=a∧¬φ(x)).\theta(y) \equiv y \leq a \rightarrow \exists x\, (x + y = a \wedge \neg \varphi(x)).

The formula ¬φ\neg \varphi is equivalent to a Σn0\Sigma^0_n formula, and the conjunction with the bounded formula x+y=ax + y = a can be moved inside its innermost block; so θ\theta is equivalent to a Σn0\Sigma^0_n formula. We have θ(0)\theta(0), with x=ax = a. Suppose θ(y)\theta(y), and let y+1≤ay + 1 \leq a. Then y<ay < a, and there is xx with x+y=ax + y = a and ¬φ(x)\neg \varphi(x); x≠0x \neq 0, since y<ay < a, so x=x′+1x = x' + 1. Since φ(x′)\varphi(x') would imply φ(x)\varphi(x), we have ¬φ(x′)\neg \varphi(x'), and x′+(y+1)=ax' + (y + 1) = a: this is θ(y+1)\theta(y + 1). By IΣn0\mathsf I\Sigma^0_n, θ(a)\theta(a) holds, which gives xx with x+a=ax + a = a, so x=0x = 0, and ¬φ(0)\neg \varphi(0): a contradiction.

(IΠn0⇒IΣn0\mathsf I\Pi^0_n \Rightarrow \mathsf I\Sigma^0_n) Let φ\varphi be Σn0\Sigma^0_n with φ(0)\varphi(0) and ∀x (φ(x)→φ(x+1))\forall x\, (\varphi(x) \rightarrow \varphi(x+1)), and suppose ¬φ(a)\neg \varphi(a). Let

η(y)≡y≤a→∀x (x+y=a→¬φ(x)),\eta(y) \equiv y \leq a \rightarrow \forall x\, (x + y = a \rightarrow \neg \varphi(x)),

equivalent to a Πn0\Pi^0_n formula. Then η(0)\eta(0) holds since ¬φ(a)\neg \varphi(a). Suppose η(y)\eta(y) and y+1≤ay + 1 \leq a, and let x+(y+1)=ax + (y + 1) = a. Then (x+1)+y=a(x + 1) + y = a, so ¬φ(x+1)\neg \varphi(x + 1), hence ¬φ(x)\neg \varphi(x). So η(y+1)\eta(y+1). By IΠn0\mathsf I\Pi^0_n, η(a)\eta(a) holds, and with x=0x = 0 it gives ¬φ(0)\neg \varphi(0): a contradiction.

The proof reverses the direction of the induction: the property ”a−ya - y fails” is propagated from y=0y = 0 up to y=ay = a. Both formulas θ\theta and η\eta have the same complexity as the negation of φ\varphi, which is why the trick exchanges Σn0\Sigma^0_n and Πn0\Pi^0_n.

1.3 Induction implies collection

Theorem 1.5 (Paris, Kirby).#

For every n≥1n \geq 1, IΔ00+IΣn0\mathsf I\Delta^0_0 + \mathsf I\Sigma^0_n proves BΣn0\mathsf B\Sigma^0_n.

Proof of Theorem 1.5.

By induction on nn. Assume IΣn0\mathsf I\Sigma^0_n, and, if n≥2n \geq 2, the theorem for n−1n - 1, so that BΣn−10\mathsf B\Sigma^0_{n-1} holds and Lemma 1.3 applies to Πn−10\Pi^0_{n-1} formulas.

Let φ(x,y)≡∃z ψ(x,y,z)\varphi(x, y) \equiv \exists z\, \psi(x, y, z) with ψ∈Πn−10\psi \in \Pi^0_{n-1} (bounded if n=1n = 1), and suppose that ∀x<a ∃y φ(x,y)\forall x < a\ \exists y\ \varphi(x, y). Let

χ(x,w)≡∃y<w ∃z<w ψ(x,y,z).\chi(x, w) \equiv \exists y < w\ \exists z < w\ \psi(x, y, z).

It is equivalent to a Πn−10\Pi^0_{n-1} formula: for n=1n = 1 it is bounded, and for n≥2n \geq 2 by Lemma 1.3. Moreover ∀x<a ∃w χ(x,w)\forall x < a\ \exists w\ \chi(x, w): take ww larger than a witness yy and a witness zz. Let

θ(c)≡c≤a→∃b ∀x<c ∃w<b χ(x,w).\theta(c) \equiv c \leq a \rightarrow \exists b\ \forall x < c\ \exists w < b\ \chi(x, w).

The formula ∀x<c ∃w<b χ\forall x < c\ \exists w < b\ \chi is equivalent to a Πn−10\Pi^0_{n-1} formula, by the same argument, so θ\theta is equivalent to a Σn0\Sigma^0_n formula. Then θ(0)\theta(0) holds with b=0b = 0. Suppose θ(c)\theta(c) and c+1≤ac + 1 \leq a; let bb be given by θ(c)\theta(c), and w0w_0 with χ(c,w0)\chi(c, w_0), which exists since c<ac < a. Then b′=max⁡(b,w0+1)b' = \max(b, w_0 + 1) witnesses θ(c+1)\theta(c+1). By IΣn0\mathsf I\Sigma^0_n, θ(a)\theta(a) holds: there is bb such that for every x<ax < a some w<bw < b satisfies χ(x,w)\chi(x, w), hence some y<w<by < w < b satisfies φ(x,y)\varphi(x, y).

1.4 Least number principles

Proposition 1.6.#

For every n≥1n \geq 1, the schemes IΣn0\mathsf I\Sigma^0_n, IΠn0\mathsf I\Pi^0_n, LΣn0\mathsf L\Sigma^0_n and LΠn0\mathsf L\Pi^0_n are equivalent over IΔ00\mathsf I\Delta^0_0.

Proof of Proposition 1.6.

(LΠn0⇒IΣn0\mathsf L\Pi^0_n \Rightarrow \mathsf I\Sigma^0_n) Let φ∈Σn0\varphi \in \Sigma^0_n with φ(0)\varphi(0) and ∀x (φ(x)→φ(x+1))\forall x\, (\varphi(x) \rightarrow \varphi(x+1)), and suppose ¬φ(a)\neg \varphi(a). By LΠn0\mathsf L\Pi^0_n applied to the Πn0\Pi^0_n formula ¬φ\neg \varphi, there is a least mm with ¬φ(m)\neg \varphi(m). Then m≠0m \neq 0, so m=m′+1m = m' + 1 with φ(m′)\varphi(m'), hence φ(m)\varphi(m): a contradiction. The same argument gives LΣn0⇒IΠn0\mathsf L\Sigma^0_n \Rightarrow \mathsf I\Pi^0_n.

(IΣn0⇒LΠn0\mathsf I\Sigma^0_n \Rightarrow \mathsf L\Pi^0_n) Let φ∈Πn0\varphi \in \Pi^0_n with φ(c)\varphi(c), and suppose there is no least xx with φ(x)\varphi(x). Let θ(x)≡∀y≤x ¬φ(y)\theta(x) \equiv \forall y \leq x\ \neg \varphi(y). By Theorem 1.5, BΣn0\mathsf B\Sigma^0_n holds, so θ\theta is equivalent to a Σn0\Sigma^0_n formula by Lemma 1.3. If φ(0)\varphi(0), then 00 is least; so θ(0)\theta(0). If θ(x)\theta(x) and φ(x+1)\varphi(x+1), then x+1x + 1 is least; so θ(x)\theta(x) implies θ(x+1)\theta(x+1). By induction, θ(c)\theta(c) holds, contradicting φ(c)\varphi(c).

(IΠn0⇒LΣn0\mathsf I\Pi^0_n \Rightarrow \mathsf L\Sigma^0_n) The same argument, with φ∈Σn0\varphi \in \Sigma^0_n: the formula ∀y≤x ¬φ(y)\forall y \leq x\ \neg \varphi(y) is equivalent to a Πn0\Pi^0_n formula, since IΠn0\mathsf I\Pi^0_n implies IΣn0\mathsf I\Sigma^0_n by Theorem 1.4, hence BΣn0\mathsf B\Sigma^0_n.

Combined with Theorem 1.4, the four schemes are equivalent.

The least number principle gives the model-theoretic meaning of induction. If IΣn0\mathsf I\Sigma^0_n fails, some Σn0\Sigma^0_n formula φ\varphi holds at 00, is preserved by successor, and fails somewhere; the set of numbers below every failure is then a cut, and its complement has no least element. In a model of IΣn0\mathsf I\Sigma^0_n, no Σn0\Sigma^0_n or Πn0\Pi^0_n formula defines a cut. For instance, in a non-standard model of IΣn0\mathsf I\Sigma^0_n, the standard numbers {0,1,2,… }\{ 0, 1, 2, \dots \} form a cut, so they are not definable by such a formula: if a Σn0\Sigma^0_n or Πn0\Pi^0_n formula holds for every standard number, it holds for some non-standard number. This is the overspill principle.

1.5 Collection implies induction

Theorem 1.7 (Paris, Kirby).#

For every n≥1n \geq 1, IΔ00+BΣn+10\mathsf I\Delta^0_0 + \mathsf B\Sigma^0_{n+1} proves IΣn0\mathsf I\Sigma^0_n.

Proof of Theorem 1.7.

By induction on nn. Assume BΣn+10\mathsf B\Sigma^0_{n+1}; it contains BΣn0\mathsf B\Sigma^0_n and BΣn−10\mathsf B\Sigma^0_{n-1}. When n≥2n \geq 2, the induction hypothesis (the theorem for n−1n - 1, applied to BΣn0\mathsf B\Sigma^0_n) gives IΣn−10\mathsf I\Sigma^0_{n-1}, hence also IΠn−10\mathsf I\Pi^0_{n-1} by Theorem 1.4.

Let φ(x)≡∃y ψ(x,y)\varphi(x) \equiv \exists y\, \psi(x, y) with ψ∈Πn−10\psi \in \Pi^0_{n-1}, suppose φ(0)\varphi(0) and ∀x (φ(x)→φ(x+1))\forall x\, (\varphi(x) \rightarrow \varphi(x+1)), and fix aa. For every x≤ax \leq a there is uu with ∀y ¬ψ(x,y)∨ψ(x+1,u)\forall y\, \neg \psi(x, y) \vee \psi(x + 1, u). This formula is equivalent to a Πn0\Pi^0_n formula, hence is Σn+10\Sigma^0_{n+1}, and BΣn+10\mathsf B\Sigma^0_{n+1} gives a bound b0b_0. Let y0y_0 be such that ψ(0,y0)\psi(0, y_0), and b=max⁡(b0,y0+1)b = \max(b_0, y_0 + 1). Then for every x<ax < a,

∃y<b ψ(x,y)→∃u<b ψ(x+1,u).\exists y < b\ \psi(x, y) \rightarrow \exists u < b\ \psi(x + 1, u).

Indeed, if ψ(x,y)\psi(x, y), the first disjunct fails, so ψ(x+1,u)\psi(x + 1, u) for some u<b0≤bu < b_0 \leq b. Let

θ(x)≡x≤a→∃u<b ψ(x,u).\theta(x) \equiv x \leq a \rightarrow \exists u < b\ \psi(x, u).

For n=1n = 1, θ\theta is bounded; for n≥2n \geq 2, it is equivalent to a Πn−10\Pi^0_{n-1} formula by Lemma 1.3. We have θ(0)\theta(0), since y0<by_0 < b, and θ(x)→θ(x+1)\theta(x) \rightarrow \theta(x+1) by the display. By IΔ00\mathsf I\Delta^0_0 or IΠn−10\mathsf I\Pi^0_{n-1}, θ(a)\theta(a) holds, so φ(a)\varphi(a).

The idea is that collection bounds, uniformly for x≤ax \leq a, the witnesses needed to pass from xx to x+1x + 1. Once the witnesses are bounded, the induction becomes an induction on a formula of lower complexity.

We obtain the hierarchy, for every n≥1n \geq 1, over IΔ00\mathsf I\Delta^0_0:

IΣn+10  ⇒  BΣn+10  ⇒  IΣn0  ⇒  BΣn0.\mathsf I\Sigma^0_{n+1} \;\Rightarrow\; \mathsf B\Sigma^0_{n+1} \;\Rightarrow\; \mathsf I\Sigma^0_n \;\Rightarrow\; \mathsf B\Sigma^0_n.

Paris and Kirby showed that none of these implications reverses: for every n≥1n \geq 1, there are models of IΣn0\mathsf I\Sigma^0_n in which BΣn+10\mathsf B\Sigma^0_{n+1} fails, and models of BΣn0\mathsf B\Sigma^0_n in which IΣn0\mathsf I\Sigma^0_n fails. The constructions use non-standard models of arithmetic and are admitted here (Hájek and Pudlák, Chapter IV). Slaman proved that BΣn0\mathsf B\Sigma^0_n is equivalent to induction for Δn0\Delta^0_n formulas, over IΔ00\mathsf I\Delta^0_0 with the totality of exponentiation.

1.6 Exercises

Exercise 1.8 ★.#

Show that, over IΔ00\mathsf I\Delta^0_0, BΣn+10\mathsf B\Sigma^0_{n+1} is equivalent to BΠn0\mathsf B\Pi^0_n for every n≥1n \geq 1.

Solution

Every Πn0\Pi^0_n formula is Σn+10\Sigma^0_{n+1}, so BΣn+10\mathsf B\Sigma^0_{n+1} implies BΠn0\mathsf B\Pi^0_n. Conversely, assume BΠn0\mathsf B\Pi^0_n, and let φ(x,y)≡∃z ψ(x,y,z)\varphi(x, y) \equiv \exists z\, \psi(x, y, z) with ψ∈Πn0\psi \in \Pi^0_n, such that ∀x<a ∃y φ(x,y)\forall x < a\ \exists y\ \varphi(x, y). Using the pairing function of Fact 1.2, let

ψ′(x,w)≡∀u≤w ∀v≤w (w=⟨u,v⟩→ψ(x,u,v)).\psi'(x, w) \equiv \forall u \leq w\ \forall v \leq w\ (w = \langle u, v \rangle \rightarrow \psi(x, u, v)).

Since every ww is the pair of unique u,v≤wu, v \leq w, ψ′(x,w)\psi'(x, w) holds if and only if ψ(x,u,v)\psi(x, u, v) for the components of ww. Since exactly one pair (u,v)(u, v) satisfies w=⟨u,v⟩w = \langle u, v \rangle, the prefix ∀u≤w ∀v≤w (w=⟨u,v⟩→⋅)\forall u \leq w\ \forall v \leq w\ (w = \langle u, v \rangle \rightarrow \cdot) amounts to substituting the components of ww, and commutes with every quantifier of ψ\psi: moving it down to the bounded matrix of ψ\psi, ψ′\psi' is equivalent to a Πn0\Pi^0_n formula, without any use of collection. Now ∀x<a ∃w ψ′(x,w)\forall x < a\ \exists w\ \psi'(x, w), so BΠn0\mathsf B\Pi^0_n gives bb with ∀x<a ∃w<b ψ′(x,w)\forall x < a\ \exists w < b\ \psi'(x, w). If w=⟨u,v⟩<bw = \langle u, v \rangle < b, then u≤w<bu \leq w < b and φ(x,u)\varphi(x, u). So ∀x<a ∃y<b φ(x,y)\forall x < a\ \exists y < b\ \varphi(x, y).

Exercise 1.9 ★.

The strong induction scheme for a formula φ\varphi is ∀x ((∀y<x φ(y))→φ(x))→∀x φ(x)\forall x\, ((\forall y < x\ \varphi(y)) \rightarrow \varphi(x)) \rightarrow \forall x\, \varphi(x). Show that IΣn0\mathsf I\Sigma^0_n proves strong induction for Σn0\Sigma^0_n formulas, for every n≥1n \geq 1.

Solution

Let φ∈Σn0\varphi \in \Sigma^0_n satisfy the hypothesis, and fix zz. Let θ(x)≡x≤z→∀y<x φ(y)\theta(x) \equiv x \leq z \rightarrow \forall y < x\ \varphi(y). By Theorem 1.5 and Lemma 1.3, θ\theta is equivalent to a Σn0\Sigma^0_n formula. We have θ(0)\theta(0). If θ(x)\theta(x) and x+1≤zx + 1 \leq z, then ∀y<x φ(y)\forall y < x\ \varphi(y), so φ(x)\varphi(x) by hypothesis, and ∀y<x+1 φ(y)\forall y < x + 1\ \varphi(y): this is θ(x+1)\theta(x+1). By IΣn0\mathsf I\Sigma^0_n, θ(z)\theta(z) holds, so ∀y<z φ(y)\forall y < z\ \varphi(y), and φ(z)\varphi(z) by hypothesis.

1.7 Notes and references

The hierarchy of induction and collection schemes is due to Paris and Kirby (1978), building on Parsons (1970), who showed that IΣ1\mathrm I\Sigma_1 proves the same Π2\Pi_2 sentences as primitive recursive arithmetic. Hájek and Pudlák’s monograph is the standard reference; Kaye’s book gives a model-theoretic introduction.

  • J. B. Paris and L. A. S. Kirby, Σₙ-collection schemas in arithmetic, in Logic Colloquium ‘77, North-Holland, 1978, 199–209.
  • C. Parsons, On a number theoretic choice schema and its relation to induction, in Intuitionism and Proof Theory, North-Holland, 1970, 459–473.
  • P. Hájek and P. Pudlák, Metamathematics of First-Order Arithmetic, Springer, 1993, Chapter I.2 and Chapter IV.
  • R. Kaye, Models of Peano Arithmetic, Oxford University Press, 1991, Chapters 2 and 10.
  • T. A. Slaman, Σₙ-bounding and Δₙ-induction, Proceedings of the American Mathematical Society 132 (2004), 2449–2456.