Hostname: page-component-848d4c4894-4rdrl Total loading time: 0 Render date: 2024-06-16T07:11:02.635Z Has data issue: false hasContentIssue false

PRENEX NORMAL FORM THEOREMS IN SEMI-CLASSICAL ARITHMETIC

Published online by Cambridge University Press:  11 June 2021

MAKOTO FUJIWARA
Affiliation:
SCHOOL OF SCIENCE AND TECHNOLOGY MEIJI UNIVERSITY, 1-1-1 HIGASHI-MITA, TAMA-KU, KAWASAKI-SHI214-8571KANAGAWA, JAPANE-mail:makotofujiwara@meiji.ac.jp
TAISHI KURAHASHI
Affiliation:
GRADUATE SCHOOL OF SYSTEM INFORMATICS KOBE UNIVERSITY, 1-1 ROKKODAI, NADA 657-8501 KOBE, JAPANE-mail:kurahashi@people.kobe-u.ac.jp
Rights & Permissions [Opens in a new window]

Abstract

Akama et al. [1] systematically studied an arithmetical hierarchy of the law of excluded middle and related principles in the context of first-order arithmetic. In that paper, they first provide a prenex normal form theorem as a justification of their semi-classical principles restricted to prenex formulas. However, there are some errors in their proof. In this paper, we provide a simple counterexample of their prenex normal form theorem [1, Theorem 2.7], then modify it in an appropriate way which still serves to largely justify the arithmetical hierarchy. In addition, we characterize a variety of prenex normal form theorems by logical principles in the arithmetical hierarchy. The characterization results reveal that our prenex normal form theorems are optimal. For the characterization results, we establish a new conservation theorem on semi-classical arithmetic. The theorem generalizes a well-known fact that classical arithmetic is $\Pi _2$ -conservative over intuitionistic arithmetic.

Type
Article
Copyright
© Association for Symbolic Logic 2021

1 Introduction

Prenex normal form theorem is one of the most basic theorems on theories based on classical first-order predicate logic. In contrast, it does not hold for intuitionistic theories in general. Therefore it does not make sense to consider an arithmetical hierarchy in an intuitionistic theory. On the other hand, if one reasons in some semi-classical arithmetic which lies in-between classical arithmetic and intuitionistic arithmetic, one can take an equivalent formula in prenex normal form for any formula with low complexity. Akama et al. [Reference Akama, Berardi, Hayashi and Kohlenbach1] introduces the classes of formulas $\mathrm{E}_k$ and $\mathrm{U}_k$ which correspond to the classes of classical $\Sigma _k$ and $\Pi _k$ formulas respectively, and showed that the former is equivalent to the class of formulas of $\Sigma _k$ form and the latter is so for $\Pi _k$ over some semi-classical arithmetic respectively. This prenex normal form theorem justifies their investigation on the arithmetical hierarchy in the context of intuitionistic first-order arithmetic. Unfortunately, however, there are some crucial errors in their proof of the prenex normal form theorem [Reference Akama, Berardi, Hayashi and Kohlenbach1, Theorem 2.7]. In this paper, we revisit their formulation and modify their prenex normal form theorem in an appropriate way.

In §2, we recall the definitions and basic properties. In §3, we provide a simple counterexample of [Reference Akama, Berardi, Hayashi and Kohlenbach1, Theorem 2.7]. In §5, we show the corrected version of the prenex normal form theorem (see Theorem 5.3). In addition, we present a simplified version of the prenex normal form theorem for formulas which do not contain the disjunction (see Theorem 5.7). In §6, we establish a new conservation theorem on semi-classical arithmetic. The theorem generalizes a well-known fact that classical arithmetic is $\Pi _2$ -conservative over intuitionistic arithmetic. In §7, using the generalized conservation theorem in §6, we characterize several prenex normal form theorems with respect to semi-classical arithmetic. In particular, among other things, we show that for any theory ${T}$ in-between intuitionistic arithmetic and classical arithmetic, ${T}$ proves a semi-classical principle ${(\Pi _k \lor \Pi _k)}\text {-}\mathrm{DNE}$ if and only if ${T}$ satisfies the prenex normal form theorem for $ \mathrm{U}_{k'}$ and $\Pi _{k'}$ for all $k' \leq k$ (see Theorem 7.3).

Throughout this paper, we work basically over intuitionistic arithmetic. When we use some principle (including induction hypothesis [I.H.]) which is not available in intuitionistic arithmetic, it will be exhibited explicitly. As regards basic reasoning over intuitionistic first-order logic, we refer the reader to [Reference van Dalen10, §6.2].

2 Preparation

Throughout this paper, we work with a standard formulation of intuitionistic arithmetic $\mathsf {HA}$ described e.g., in [Reference Troelstra8, §1.3], which has function symbols for all primitive recursive functions. We work in the language containing all the logical constants $\forall , \exists , \to , \land , \lor , \perp $ . Let ${T}$ denote a theory (e.g., $\mathsf {HA}$ ), and $\mathrm{P}$ and $\text {Q}$ denote schemata (e.g., logical principles). Then ${T} + \mathrm{P}$ denotes the theory obtained from ${T}$ by adding $\mathrm{P}$ into the axioms. In particular, the classical variant $\mathsf {PA}$ is defined as $\mathsf {HA} + \mathrm{LEM}$ , where $\mathrm{LEM}$ is the axiom scheme of the law of excluded middle. We write ${T} \vdash \text {Q}$ (or ${T} $ proves $\text {Q}$ ) if any instance of $\text {Q}$ is provable in ${T}$ . We write ${T} \vdash \mathrm{P}+\text {Q}$ if ${T} \vdash \mathrm{P}$ and ${T} \vdash \text {Q}$ .

Notation 1. For a formula $\varphi $ , $\text { FV} \left ({\varphi }\right )$ denotes the set of free variables in $\varphi $ . Quantifier-free formulas are denoted with subscript “qf” as ${\varphi }_{\text {qf}}$ . In addition, a list of variables is denoted with an over-line as $\overline {x}$ . In particular, a list of quantifiers of the same kind is denoted as $\exists \overline {x}$ and $\forall \overline {x}$ respectively.

Definition 2.1. The classes $\Sigma _k$ and $\Pi _k$ of formulas are defined as follows:

  • $\Sigma _0$ , as well as $\Pi _0$ , is the class of all quantifier-free formulas;

  • $\Pi _{k+1}$ is the class of all formulas of form $Q_1 \overline {x_1} \cdots Q_{k+1} \overline {x_{k+1}} \, {\varphi }_{\text {qf}}$ ;

  • $\Sigma _{k+1}$ is the class of all formulas of form $Q^{\prime }_1 \overline {x_1} \cdots Q^{\prime }_{k+1} \overline {x_{k+1}} \, {\varphi }_{\text {qf}}$ ;

where $Q_i$ represents $\forall $ for odd i and $\exists $ for even i and $Q^{\prime }_i$ represents $\exists $ for odd i and $\forall $ for even i. Following [Reference Akama, Berardi, Hayashi and Kohlenbach1], we define the classes $\Sigma _k$ and $\Pi _k$ in the non-cumulative manner (namely, each $Q_i \overline {x_i} $ and $Q^{\prime }_i \overline {x_i}$ must not be empty). A formula $\varphi $ is of prenex normal form if $\varphi \in \Sigma _k \cup \Pi _k$ for some k.

Remark 2.2. Since the list of variables can be contracted into one variable in $\mathsf {HA}$ by using a fixed primitive recursive pairing function (see e.g., [Reference Troelstra8, §1.3.9]), one may assume that for each natural number $k>0$ , a formula in $\Sigma _k$ is of form $\exists x \varphi (x)$ with some $\varphi (x) \in \Pi _{k-1}$ and a formula in $\Pi _k$ is of form $\forall x \psi (x)$ with some $\psi (x) \in \Sigma _{k-1}$ without loss of generality.

Lemma 2.3. Let k be a natural number. Let $\varphi $ be in $\Pi _k$ and $\psi $ be in $\Sigma _k$ . Then, for all natural numbers $i\ \mathrm{and}\ j$ , there exist $\varphi ', \psi '\in \Pi _{k+i}$ and $\varphi '', \psi '' \in \Sigma _{k+j}$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )=\mathrm{FV} \left ({\varphi ''}\right )$ , $\mathrm{FV} \left ({\psi }\right )=\mathrm{FV} \left ({\psi '}\right )=\mathrm{FV} \left ({\psi ''}\right )$ , $ \mathsf {HA} \vdash \varphi \leftrightarrow \varphi ' \leftrightarrow \varphi '' $ and $\mathsf {HA} \vdash \psi \leftrightarrow \psi ' \leftrightarrow \psi ''$ .

Proof Straightforward by the fact that

(1) $$ \begin{align} \mathsf{HA} \vdash \xi \leftrightarrow \forall z \xi \leftrightarrow \exists z \xi \end{align} $$

for any $ z\notin \mathrm{FV} \left ({\xi }\right )$ .⊣

Definition 2.4. For a class $\Gamma $ of formulas, $\Gamma (\overline {x})$ denotes the class of formulas $\varphi $ in $\Gamma $ such that $\mathrm{FV} \left ({\varphi }\right ) \subseteq \{ \overline {x}\}$ .

Remark 2.5. In the light of Lemma 2.3, throughout this paper, we identify the classes $\Sigma _k$ and $\Pi _k$ with the classes defined as in Definition 2.1 with allowing the quantifiers $Q_i$ and $Q^{\prime }_i$ to be empty. Under this identification, for all k and $k'$ such that $k< k' $ , $\Pi _k (\overline {x}) $ and $\Sigma _k (\overline {x})$ are considered to be sub-classes of $ \Sigma _{k'} (\overline {x}) \cap \Pi _{k'} (\overline {x})$ . We frequently use this property in what follows.

Recall the logical principles from [Reference Akama, Berardi, Hayashi and Kohlenbach1] and related principles:

Definition 2.6. Let $\Gamma $ and $\Gamma '$ be classes of formulas.

  • ${\Gamma }\text {-}\mathrm{LEM}:\, \forall x \left ( \varphi (x) \lor \neg \varphi (x) \right )$ where $\varphi (x) \in \Gamma (x)$ .

  • ${\Gamma }\text {-}\text {DML}:\, \forall x \left ( \neg (\varphi (x) \land \psi (x)) \kern1.2pt{\to}\kern1.2pt \neg \varphi (x) \lor \neg \psi (x) \right )$ where $\varphi (x), \psi (x) \kern1.2pt{\in}\kern1.2pt \Gamma (x)$ .

  • ${\Gamma }\text {-}\mathrm{DNE}:\, \forall x \left ( \neg \neg \varphi (x) \to \varphi (x) \right ) $ where $\varphi (x) \in \Gamma (x)$ .

  • ${(\Gamma \lor \Gamma ')}\text {-}\mathrm{DNE}:\, \forall x \left ( \neg \neg (\varphi (x) \lor \psi (x)) \to \varphi (x) \lor \psi (x) \right )$ where $\varphi (x) \in \Gamma (x)$ and $\psi (x) \in \Gamma '(x)$ .

  • ${\Gamma }\text {-}\mathrm{DNS}: \, \forall x \left (\forall y \neg \neg \varphi (x,y ) \to \neg \neg \forall y \varphi (x,y) \right )$ where $\varphi (x,y ) \in \Gamma (x,y)$ .

  • Let $\mathrm{P} \in \{{\Gamma }\text {-}\mathrm{LEM}, {\Gamma }\text {-}\text {DML}, {\Gamma }\text {-}\mathrm{DNE}, {(\Gamma \lor \Gamma ')}\text {-}\mathrm{DNE} , {\Gamma }\text {-}\mathrm{DNS}\}$ . $\neg \neg {\mathrm{P}}:\, \neg \neg \xi $ where $\xi $ is an instance of $\mathrm{P} $ .

Note that our logical principles are equivalent also to those defined with lists of quantifiers of the same kind (cf. Remark 2.2).

Remark 2.7. One has to care about the formulation of the double negated variants. That is, one has to take the double negations of the universal closure of the original logical principles as in Definition 2.6. The double negated variants defined as such are not provable in $\mathsf {HA}$ , which has been overlooked in the proof of [Reference Akama, Berardi, Hayashi and Kohlenbach1, Theorem 2.7] (see also §3). In fact, one may think of the double negated versions as variants of the double negation shift principle (see [Reference Fujiwara and Kohlenbach3]). In addition, our double negated versions are equivalent to (the universal closures of) those with allowing free variables (cf. [Reference Fujiwara and Kohlenbach3, Remark 2.5]).

Remark 2.8. For any class $\Gamma $ of formulas, ${\Gamma }\text {-}\mathrm{DNS} $ is intuitionistically equivalent to $\neg \neg {{\Gamma }\text {-}\mathrm{DNS} }$ since

$$ \begin{align*}\begin{array}{cl} & \forall x \left( \forall y \neg \neg \varphi \to \neg \neg \forall y \varphi \right) \\ \longleftrightarrow & \forall x \neg \neg \left( \forall y \neg \neg \varphi \to \neg \neg \forall y \varphi \right) \\ \longleftrightarrow & \neg \neg \forall x \neg \neg \left( \forall y \neg \neg \varphi \to \neg \neg \forall y \varphi \right) \\ \longleftrightarrow &\neg \neg \forall x \left( \forall y \neg \neg \varphi \to \neg \neg \forall y \varphi \right).\\ \end{array} \end{align*} $$

Akama et al. [Reference Akama, Berardi, Hayashi and Kohlenbach1] introduces the classes $\text {F}_k, \mathrm{U}_k$ and $\mathrm{E}_k$ of formulas. In the following, we reformulate them and introduce two additional classes $\mathrm{U}_k^+$ and $\mathrm{E}_k^+$ in a formal manner.

Definition 2.9. An alternation path is a finite sequence of $+$ and $-$ in which $+$ and $-$ appear alternatively. For an alternation path s, let $i(s)$ denote the first symbol of s if $s \not \equiv {\langle \, \rangle } $ (empty sequence); $ \times $ if $s \equiv {\langle \, \rangle }$ . Let $s^{\perp }$ denote the alternation path which is obtained by switching $+$ and $-$ in s, and let $l(s) $ denote the length of s.

Definition 2.10. For a formula $\varphi $ , the set of alternation paths $\mathit {Alt}(\varphi )$ of $\varphi $ is defined as follows:

  • If $\varphi $ is quantifier-free, then $\mathit {Alt}(\varphi ) := \{ {\langle \, \rangle } \}$ ;

  • Otherwise, $\mathit {Alt}(\varphi )$ is defined inductively by the following rule:

    • If $\varphi \equiv \varphi _1 \land \varphi _2$ or $\varphi \equiv \varphi _1 \lor \varphi _2$ , then $\mathit {Alt}(\varphi ) := \mathit {Alt}(\varphi _1) \cup \mathit {Alt}(\varphi _2)$ ;

    • If $\varphi \equiv \varphi _1 \to \varphi _2$ , then $\mathit {Alt}(\varphi ) := \{ s^{\perp } \mid s \in \mathit {Alt}(\varphi _1)\} \cup \mathit {Alt}(\varphi _2)$ ;

    • If $\varphi \equiv \forall x \varphi _1 $ , then $\mathit {Alt}(\varphi ) :=\{s \mid s\in \mathit {Alt}(\varphi _1) \text { and } i(s)\equiv\,\,\, -\} \cup \{{-}s \mid s\in \mathit {Alt}(\varphi _1) \text { and } i(s)\not \equiv \ - \} $ ;

    • If $\varphi \equiv \exists x \varphi _1 $ , then $\mathit {Alt}(\varphi ) :=\{s \mid s\in \mathit {Alt}(\varphi _1) \text { and } i(s)\equiv + \} \cup \{+s \mid s\in \mathit {Alt}(\varphi _1) \text { and } i(s)\not \equiv + \} $ .

In addition, for a formula $\varphi $ , the degree $\mathit {deg}(\varphi )$ of $\varphi $ is defined as

$$ \begin{align*} \mathit{deg}(\varphi) := \max \{l(s) \mid s \in \mathit{Alt}(\varphi) \}. \end{align*} $$

Definition 2.11. The classes $\text {F}_k, \mathrm{U}_k, \mathrm{E}_k $ (from [Reference Akama, Berardi, Hayashi and Kohlenbach1, Definition 2.4]), $\mathrm{U}_k^+ $ and $ \mathrm{E}_k^+ $ of formulas are defined as follows:

  • $\text {F}_k := \{ \varphi \mid \mathit {deg}(\varphi )=k \} $ ;

  • $\mathrm{U}_0:=\mathrm{E}_0:=\text {F}_0$ ;

  • $\mathrm{U}_{k+1} := \{ \varphi \in \text {F}_{k+1} \mid i(s) \,{\equiv}\ \kern1pt{-}\, \text { for all }s\in \mathit {Alt}(\varphi ) \text { such that }l(s) =k+1 \}$ ;

  • $\mathrm{E}_{k+1} := \{ \varphi \in \text {F}_{k+1} \mid i(s) \equiv + \text { for all }s\in \mathit {Alt}(\varphi ) \text { such that }l(s) =k+1 \}$ ;

  • $\displaystyle \mathrm{U}_k^+ := \mathrm{U}_k \cup \bigcup _{i<k} \text {F}_i$ ;

  • $\displaystyle \mathrm{E}_k^+ := \mathrm{E}_k \cup \bigcup _{i<k} \text {F}_i$ .

Remark 2.12. From the perspective of Proposition 4.6 below, the introduction of $\mathrm{U}_k^+$ and $\mathrm{E}_k^+$ in addition to $\mathrm{U}_k$ and $\mathrm{E}_k$ is mathematically superfluous. However, we introduce these auxiliary classes for facilitating our arguments below.

Note $\text {F}_0 =\Sigma _0 =\Pi _0$ . For each formula $\varphi \in \mathrm{E}_k$ (resp. $\psi \in \mathrm{U}_k$ ) of $\mathsf {PA}$ , one can take a formula $\varphi ' \in \Sigma _k$ (resp. $\psi ' \in \Pi _k$ ) of $\mathsf {PA}$ which is equivalent to $\varphi $ (resp. $\psi $ ) over $\mathsf {PA}$ . On the other hand, this is not the case for $\mathsf {HA}$ . In what follows, we study what kind of semi-classical arithmetic in-between $\mathsf {PA}$ and $\mathsf {HA}$ captures this property for each k. In fact, Akama et al. [Reference Akama, Berardi, Hayashi and Kohlenbach1] has already undertaken this. In particular, [Reference Akama, Berardi, Hayashi and Kohlenbach1, Theorem 2.7] asserts the following:

  1. 1. For any $\varphi \in \mathrm{E}_k$ , there exists $\varphi ' \in \Sigma _k$ such that

    $$ \begin{align*} \mathsf{HA} + {\Sigma_k}\text{-}\mathrm{DNE} \vdash \varphi \leftrightarrow \varphi'. \end{align*} $$
  2. 2. For any $\varphi \in \mathrm{U}_k$ , there exists $\varphi ' \in \Pi _k$ such that

    $$ \begin{align*}\mathsf{HA} + {(\Pi_k \lor \Pi_k)}\text{-}\mathrm{DNE} \vdash \varphi \leftrightarrow \varphi'. \end{align*} $$

However, the first assertion is wrong as we show in §3. In fact, a weak variant ${\mathrm{U}_k}\text {-}\mathrm{DNS}$ of the double negation shift principle is missing in the verification theory, which will be revealed by our modified version of the prenex normal form theorem (Theorem 5.3) below. On the other hand, the second assertion is correct. This will be revealed also by Theorem 5.3.

3 A counter example

Recall that [Reference Akama, Berardi, Hayashi and Kohlenbach1, Theorem 2.7] asserts that for any $\varphi \in \mathrm{E}_k$ , there exists $\varphi ' \in \Sigma _k$ such that

$$ \begin{align*} \mathsf{HA} + {\Sigma_k}\text{-}\mathrm{DNE} \vdash \varphi \leftrightarrow \varphi'. \end{align*} $$

However, there are some errors in the proof. In particular, in [Reference Akama, Berardi, Hayashi and Kohlenbach1, p. 5, lines 15–17], it is written that “Since the double negations of $\mathrm{DNE}$ is intuitionistically provable, $\vdash _{\mathsf {HA}} \neg \neg {A_0} \leftrightarrow \neg \neg {\exists x_0. C_0 }$ (which means $\mathsf {HA} \vdash \neg \neg {A_0} \leftrightarrow \neg \neg {\exists x_0 C_0 }$ in our notation)”. As studied in [Reference Fujiwara and Kohlenbach3], however, the double negations of (the universal closure of) $\mathrm{DNE}$ is not provable in $\mathsf {HA}$ , and hence, their proof actually uses some double negated logical principles in the sense of Definition 2.6. Our counterexample below shows that such a use of some additional principle is unavoidable.

Recall the arithmetical form of Church’s thesis from [Reference Troelstra8, §3.2.14]:

$$ \begin{align*} \text{CT}_{0}: \forall x \exists y \,\varphi (x,y) \to \exists e \forall x \exists v \left(\mathrm{T}(e,x,v) \land \varphi (x, \mathrm{U}(v)) \right), \end{align*} $$

where $\text {T}$ and $\mathrm{U}$ are the standard primitive recursive predicate and function from the Kleene normal form theorem. Note that $\text {CT}_{0}$ is a sort of combination of so-called Church’s thesis stating that every function is recursive and the countable choice principle (see [Reference Troelstra and van Dalen9, §4.3.2]).

Proposition 3.1. The following sentence

$$ \begin{align*} \varphi_0:\equiv \neg \forall x \left( \neg \exists u \left( \mathrm{T}(x,x,u) \land \mathrm{U}(u)=0 \right) \lor \neg \exists u \left( \mathrm{T}(x,x,u) \land \mathrm{U}(u)\neq 0 \right) \right) \end{align*} $$

is not equivalent to any sentence $\varphi _0 ' \in \Sigma _1 $ over $\mathsf {HA} +{\Sigma _1}\text {-}\mathrm{DNE}$ .

Proof We first claim that $\mathsf {HA} + \text {CT}_{0}$ proves $\varphi _0$ . For the sake of contradiction, assume

(2) $$ \begin{align} \forall x \left( \neg \exists u \left( \mathrm{T}(x,x,u) \land \mathrm{U}(u)=0 \right) \lor \neg \exists u \left( \mathrm{T}(x,x,u) \land \text{ U}(u)\neq 0 \right) \right) \end{align} $$

and reason in $\mathsf {HA} + \text {CT}_{0}$ . Since $\varphi _1\lor \varphi _2 \leftrightarrow \exists k\left ( (k=0 \to \varphi _1) \land (k\neq 0 \to \varphi _2) \right )$ (see [Reference Troelstra8, §1.3.7]), by $\text {CT}_{0}$ , there exists e such that

$$ \begin{align*} \forall x \exists v \left( \begin{array}{l} \mathrm{T}(e,x,v) \\ \land \left(\mathrm{U}(v)=0 \to \neg \exists u \left( \mathrm{T}(x,x,u) \land \mathrm{U}(u)=0 \right) \right) \\ \land \left(\mathrm{U}(v)\neq 0 \to \neg \exists u \left( \mathrm{T}(x,x,u) \land \mathrm{U}(u) \neq 0 \right) \right) \end{array} \right). \end{align*} $$

In particular, for that e, there exists $v_e$ such that $\text {T}(e,e,v_e)$ ,

$$ \begin{align*}\mathrm{U}(v_e)=0 \to \neg \exists u \left( \mathrm{T}(e,e,u) \land \mathrm{U}(u)=0 \right) \end{align*} $$

and

$$ \begin{align*}\mathrm{U}(v_e)\neq 0 \to \neg \exists u \left( \mathrm{T}(e,e,u) \land \mathrm{U}(u) \neq 0 \right). \end{align*} $$

Since $\mathrm{U}(v_e)=0 \lor \mathrm{U}(v_e) \neq 0$ , we obtain a contradiction straightforwardly.

If $\varphi _0 $ is equivalent to some sentence $\varphi _0 ' \in \Sigma _1 $ over $\mathsf {HA} +{\Sigma _1}\text {-}\mathrm{DNE}$ , we have $\mathsf {HA} + {\Sigma _1}\text {-}\mathrm{DNE} +\text {CT}_{0} \vdash \varphi _0 '$ from the above claim. Since $\varphi _0' \in \Sigma _1$ , by the soundness of Kleene realizability (see [Reference Troelstra8, §3.2.22]), we have that

$$ \begin{align*} \mathsf{HA} + {\Sigma_1}\text{-}\mathrm{DNE} \vdash \varphi_0 ', \end{align*} $$

and hence, $\mathsf {HA} + {\Sigma _1}\text {-}\mathrm{DNE} \vdash \varphi _0 $ . On the other hand, since

$$ \begin{align*}\forall x \neg \left( \exists u \left( \mathrm{T}(x,x,u) \land \mathrm{U}(u)=0 \right) \land \exists u \left( \mathrm{T}(x,x,u) \land \mathrm{U}(u)\neq 0 \right) \right) \end{align*} $$

is provable in $\mathsf {HA}$ , we have $\mathsf {HA} +{\Sigma _1}\text {-}\text {DML} \vdash (2)$ . Therefore we have

$$ \begin{align*} \mathsf{HA} + {\Sigma_1}\text{-}\mathrm{DNE} + {\Sigma_1}\text{-}\text{DML} \vdash \perp, \end{align*} $$

and hence, $\mathsf {PA} \vdash \perp $ , which is a contradiction.⊣

Remark 3.2. One can easily see that $\varphi _0 $ in Proposition 3.1 is in $\mathrm{E}_1$ . Thus Proposition 3.1 shows that $\varphi _0 $ is a counterexample of [Reference Akama, Berardi, Hayashi and Kohlenbach1, Theorem 2.7] for $k=1$ .

4 Basic lemmata

In this section, we show several lemmata which we use in the proofs of our prenex normal form theorems.

Lemma 4.1. For any logical principle $\mathrm{P}$ in Definition 2.6 and any formula $\varphi $ (possibly containing free variables), if $\mathsf {HA} + \mathrm{P} \vdash \varphi $ , then $\mathsf {HA} + \neg \neg \mathrm{P} \vdash \neg \neg {\varphi }$ .

Proof Assume $\mathsf {HA} + \mathrm{P} \vdash \varphi $ . Then there exists finite instances $\psi _1, \dots , \psi _k$ of $\mathrm{P}$ such that $\mathsf {HA} + \psi _1 + \dots + \psi _k \vdash \varphi $ . Since $\mathsf {HA}$ satisfies the deduction theorem, we have that $\mathsf {HA}$ proves $\psi _1 \land \dots \land \psi _k \to \varphi $ , and hence, $\neg \neg {\left (\psi _1 \land \dots \land \psi _k \to \varphi \right )}$ , which is equivalent to $\neg \neg {\psi _1} \land \dots \land \neg \neg {\psi _k} \to \neg \neg {\varphi }$ . Then we have $\mathsf {HA} + \neg \neg \mathrm{P} \vdash \neg \neg {\varphi }$ .⊣

Corollary 4.2. For any logical principle $\mathrm{P}$ in Definition 2.6 and any formulas $\varphi _1$ and $\varphi _2$ (possibly containing free variables), if $\mathsf {HA} + \mathrm{P} \vdash \varphi _1 \leftrightarrow \varphi _2$ , then $\mathsf {HA} + \neg \neg \mathrm{P} \vdash \neg \neg {\varphi _1} \leftrightarrow \neg \neg {\varphi _2}$ .

Proof Immediate from Lemma 4.1 and the fact that $\neg \neg {\left (\varphi _1 \leftrightarrow \varphi _2\right )}$ is intuitionistically equivalent to $\neg \neg {\varphi _1} \leftrightarrow \neg \neg {\varphi _2}$ .⊣

Lemma 4.3. Let k be a natural number. Let $\varphi _1$ and $\varphi _2$ be formulas in $\Sigma _k$ , and let $\varphi _3$ and $\varphi _4$ be formulas in $\Pi _k$ . Then the following hold:

  1. 1. There exists a formula $\varphi \in \Sigma _k$ such that $\mathrm{FV} \left ({\varphi }\right )= \mathrm{FV} \left ({\varphi _1 \land \varphi _2}\right ) $ and $\mathsf {HA} \vdash \varphi \leftrightarrow \varphi _1 \land \varphi _2$ .

  2. 2. There exists a formula $\varphi ' \in \Pi _k$ such that $\mathrm{FV} \left ({\varphi '}\right )= \mathrm{FV} \left ({\varphi _3 \land \varphi _4}\right ) $ and $\mathsf {HA} \vdash \varphi ' \leftrightarrow \varphi _3 \land \varphi _4$ .

Proof Straightforward by simultaneous induction on k.⊣

Lemma 4.4. For any formulas $\varphi _1$ and $\varphi _2$ in $\Sigma _k$ , there exists a formula $\varphi \in \Sigma _k $ such that $\mathrm{FV} \left ({\varphi }\right )= \mathrm{FV} \left ({\varphi _1 \lor \varphi _2}\right ) $ and $\mathsf {HA} \vdash \varphi \leftrightarrow \varphi _1 \lor \varphi _2$ .

Proof Note that $\varphi _1 \lor \varphi _2$ is equivalent to

$$ \begin{align*} \exists k \left(\left(k=0 \to \varphi_1 \right) \land \left( k\neq 0 \to \varphi_2 \right) \right) \end{align*} $$

over $\mathsf {HA}$ (see [Reference Troelstra8, §1.3.7]). Since ${\varphi }_{\text {qf}} \to \exists x \psi (x)$ and ${\varphi }_{\text {qf}} \to \forall x \psi (x)$ are equivalent to $\exists x\left ({\varphi }_{\text {qf}} \to \psi (x)\right )$ and $\forall x \left ( {\varphi }_{\text {qf}} \to \psi (x) \right )$ respectively over $\mathsf {HA}$ when $x \notin \mathrm{FV} \left ({{\varphi }_{\text {qf}}}\right )$ , our assertion follows from Lemma 4.3 straightforwardly.⊣

Lemma 4.5. Let k be a natural number greater than $0$ .

  1. 1. $\varphi _1 \land \varphi _2$ is in $\mathrm{U}_k^+$ (resp. $\mathrm{E}_k^+$ ) if and only if both of $\varphi _1 $ and $\varphi _2$ are in $\mathrm{U}_k^+$ (resp. $\mathrm{E}_k^+$ ).

  2. 2. $\varphi _1 \lor \varphi _2$ is in $\mathrm{U}_k^+$ (resp. $\mathrm{E}_k^+$ ) if and only if both of $\varphi _1 $ and $\varphi _2$ are in $\mathrm{U}_k^+$ (resp. $\mathrm{E}_k^+$ ).

  3. 3. $\varphi _1 \to \varphi _2$ is in $\mathrm{U}_k^+$ (resp. $\mathrm{E}_k^+$ ) if and only if $\varphi _1 $ is in $\mathrm{E}_k^+$ (resp. $\mathrm{U}_k^+$ ) and $\varphi _2$ is in $\mathrm{U}_k^+$ (resp. $\mathrm{E}_k^+$ ).

  4. 4. $\forall x \varphi _1$ is in $\mathrm{U}_k^+$ if and only if $\varphi _1$ is in $\mathrm{U}_k^+$ .

  5. 5. $\exists x \varphi _1$ is in $\mathrm{E}_k^+$ if and only if $\varphi _1$ is in $\mathrm{E}_k^+$ .

  6. 6. $\forall x \varphi _1$ is in $\mathrm{E}_{k+1}^+$ if and only if it is in $\mathrm{U}_k^+$ .

  7. 7. $\exists x \varphi _1$ is in $\mathrm{U}_{k+1}^+$ if and only if it is in $\mathrm{E}_k^+$ .

Proof (1): Assume $\varphi _1 \land \varphi _2 \in \mathrm{U}_k^+$ . Then $l(s) \leq k$ for all $s \in \mathit {Alt}( \varphi _1 \land \varphi _2) = \mathit {Alt}( \varphi _1) \cup \mathit {Alt}( \varphi _2)$ .

  • If $l(s) < k$ for all $s \in \mathit {Alt}( \varphi _1)$ , then $\varphi _1$ is in $\displaystyle \bigcup _{i<k} \text {F}_i \subseteq \mathrm{U}_k^+$ .

  • Otherwise, there is $s_0 \in \mathit {Alt}( \varphi _1)$ such that $l(s_0)=k$ . Then, since $\varphi _1 \land \varphi _2 \notin \displaystyle \bigcup _{i<k} \text {F}_i$ , we have $\varphi _1 \land \varphi _2 \in \mathrm{U}_k$ . Then, for each $s \in \mathit {Alt}(\varphi _1)$ such that $l(s)=k$ , we have $i(s)\,{\equiv}\ \kern1pt{-}\, $ since $s \in \mathit {Alt}(\varphi _1 \land \varphi _2 )$ . Thus $\varphi _1 \in \mathrm{U}_k \subseteq \mathrm{U}_k^+$ .

We also have $\varphi _2 \in \mathrm{U}_k^+$ in the same manner.

For the converse direction, assume that $\varphi _1 $ and $\varphi _2$ are in $\mathrm{U}_k^+$ . Then, for all $s\in \mathit {Alt}(\varphi _1 \land \varphi _2)$ , since $s\in \mathit {Alt}(\varphi _1)$ or $s\in \mathit {Alt}(\varphi _2)$ , we have $l(s)\leq k$ , in particular, $i(s) \,{\equiv}\ \kern1pt{-}$ if $l(s)=k$ . Thus $\varphi _1 \land \varphi _2 $ is in $\mathrm{U}_k^+$ .

As for the case of $\mathrm{E}_k^+$ , an analogous proof works.

(2): Analogous to (1).

(3): Assume $\varphi _1 \to \varphi _2 \in \mathrm{U}_k^+$ . Let s be in $\mathit {Alt}(\varphi _1 )$ . By the definition of $\mathit {Alt}(\varphi _1 \to \varphi _2)$ , we have $s^{\perp } \in \mathit {Alt}(\varphi _1 \to \varphi _2)$ and $l(s) \leq k$ .

  • If $l(s) < k$ for all $s \in \mathit {Alt}( \varphi _1)$ , then $\varphi _1$ is in $\displaystyle \bigcup _{i<k} \text {F}_i \subseteq \mathrm{E}_k^+$ .

  • Otherwise, there is $s_0 \in \mathit {Alt}( \varphi _1)$ such that $l(s_0)=k$ . Since ${s_0}^{\perp } \in \mathit {Alt}(\varphi _1 \to \varphi _2)$ , we have $\varphi _1 \to \varphi _2 \in \mathrm{U}_k$ . Then, for each $s \in \mathit {Alt}(\varphi _1)$ such that $l(s)=k$ , we have $i(s^{\perp })\,{\equiv}\ \kern1pt{-}\,$ , and hence, $i(s)\equiv +$ . Thus $\varphi _1 \in \mathrm{E}_k \subseteq \mathrm{E}_k^+$ .

We also have $\varphi _2 \in \mathrm{U}_k^+$ in the same manner.

For the converse direction, assume $\varphi _1 \in \mathrm{E}_k^+$ and $\varphi _2 \in \mathrm{U}_k^+$ . Since $\mathit {deg}(\varphi _1)\leq k$ and $\mathit {deg}(\varphi _2)\leq k$ , we have $\mathit {deg}(\varphi _1 \to \varphi _2)\leq k$ .

  • If $\mathit {deg}(\varphi _1 \to \varphi _2)<k$ , then $\varphi _1 \to \varphi _2\in \displaystyle \bigcup _{i<k} \text {F}_i \subseteq \mathrm{U}_k^+$ .

  • If $\mathit {deg}(\varphi _1 \to \varphi _2)=k$ , for all $s\in \mathit {Alt}(\varphi _1 \to \varphi _2)$ such that $l(s)=k$ , we have $s\in \mathit {Alt}(\varphi _2)$ or $s\equiv {s_0}^{\perp }$ for some $s_0\in \mathit {Alt}(\varphi _1)$ . In the former case, we have $i(s) \,{\equiv}\ \kern1pt{-}\,$ by $\varphi _2 \in \mathrm{U}_k^+$ . In the latter case, we have $i(s_0) \equiv +$ by $\varphi _1 \in \mathrm{E}_k^+$ , and hence, $i(s)\,{\equiv}\ \kern1pt{-}\,$ .

One can also show that $\varphi _1 \to \varphi _2$ is in $\mathrm{E}_k^+$ if and only if $\varphi _1 $ is in $\mathrm{U}_k^+$ and $\varphi _2$ is in $\mathrm{E}_k^+$ analogously.

(4): Assume $\forall x \varphi _1 \in \mathrm{U}_k^+$ .

  • If $\forall x \varphi _1 \notin \mathrm{U}_k$ , then $\forall x \varphi _1 \in \displaystyle \bigcup _{i<k} \text {F}_i$ . Since $\mathit {deg}(\varphi _1)\leq \mathit {deg}(\forall x \varphi _1) <k$ , we have $\varphi _1 \in \displaystyle \bigcup _{i<k} \text {F}_i \subseteq \mathrm{U}_k^+$ .

  • Otherwise, $\mathit {deg}(\varphi _1)\leq \mathit {deg}(\forall x \varphi _1) =k$ . If $\mathit {deg}(\varphi _1)< k$ , then we have $\varphi _1 \in \displaystyle \bigcup _{i<k} \text {F}_i \subseteq \mathrm{U}_k^+$ . Assume $\mathit {deg}(\varphi _1)= k$ . Let s be an alternation path of $\varphi _1$ such that $l(s)=k$ . If $i(s) \not\equiv\ -\,$ , by the definition of $\mathit {Alt}(\forall x \varphi _1)$ , we have ${-}s\in \mathit {Alt}(\forall x \varphi _1)$ , which contradicts $\mathit {deg}(\forall x \varphi _1) =k$ since $l({-}s)=k+1$ . Then we have $i(s)\,{\equiv}\ \kern1pt{-}\,$ . Thus we have $\varphi _1 \in \mathrm{U}_k \subseteq \mathrm{U}_k^+$ .

For the converse direction, assume $\varphi _1 \in \mathrm{U}_k^+$ .

  • If $\varphi _1 \notin \mathrm{U}_k$ , then $\varphi _1 \in \displaystyle \bigcup _{i<k} \text {F}_i$ . Thus $\mathit {deg}(\varphi _1)<k$ , and hence, $\mathit {deg}(\forall x \varphi _1)\leq k$ . If $\mathit {deg}(\forall x \varphi _1)< k$ , then $\forall x \varphi _1 \in \displaystyle \bigcup _{i<k} \text {F}_i \subseteq \mathrm{U}_k^+$ . If $\mathit {deg}(\forall x \varphi _1)= k$ , since $i(s)\,{\equiv}\ \kern1pt{-}\,$ for all $s\in \mathit {Alt}(\forall x \varphi _1)$ , we have $\forall x \varphi _1 \in \mathrm{U}_k \subseteq \mathrm{U}_k^+$ .

  • Otherwise, $\mathit {deg}(\varphi _1)=k$ and $i(s)\,{\equiv}\ \kern1pt{-}\,$ for all $s \in \mathit {Alt}(\varphi _1)$ such that $l(s) =k$ . By the definition of $\mathit {Alt}(\forall x\varphi _1)$ , for all $s \in \mathit {Alt}( \forall x \varphi _1)$ , we have $l(s)\leq k$ , and hence, $\mathit {deg}(\forall x\varphi _1)= k$ . In addition, again by the definition of $\mathit {Alt}(\forall x\varphi _1)$ , we have $i(s) \equiv\ \,-$ for all $s\in \mathit {Alt}(\forall x \varphi _1)$ such that $l(s) =k$ . Thus $\forall x \varphi _1 \in \mathrm{U}_k \subseteq \mathrm{U}_k^+$ .

(5): Analogous to (4).

(6): Assume $\forall x \varphi _1 \in \mathrm{E}_{k+1}^+$ . Since $i(s)\,{\equiv}\ {-}\,$ for all $s\in \mathit {Alt}(\forall x \varphi _1)$ , $\forall x \varphi _1$ is not in $\mathrm{E}_{k+1}$ . Then $\forall x \varphi _1 \in \displaystyle \bigcup _{i\leq k} \text {F}_i $ , and hence, $\mathit {deg}(\forall x \varphi _1) \leq k$ .

  • If $\mathit {deg}(\forall x \varphi _1) < k$ , then $\forall x \varphi _1 \in \displaystyle \bigcup _{i< k} \text {F}_i \subseteq \mathrm{U}_k^+$ .

  • If $\mathit {deg}(\forall x \varphi _1) = k$ , since $i(s) \,{\equiv}\ \kern1pt{-}\,$ for all $s\in \mathit {Alt}(\forall x \varphi _1)$ , we have $\forall x \varphi _1 \in \mathrm{U}_k\subseteq \mathrm{U}_k^+$ .

The converse direction is trivial since $ \mathrm{U}_k^+ \subseteq \displaystyle \bigcup _{i< k+1} \text {F}_i \subseteq \mathrm{E}_{k+1}^+$ .

(7): Analogous to (6).⊣

As mentioned in Lemma 2.3, our non-cumulative definition of the classes $\Sigma _k$ and $\Pi _k$ does not cause any trouble. In a similar sense, the following proposition allows us to think of $\mathrm{E}_k$ and $\mathrm{U}_k$ in the cumulative manner:

Proposition 4.6. Let k be a natural number. Then the following hold:

  1. 1. If $\varphi \in \mathrm{U}_k^+$ , then there exist $\varphi '\in \mathrm{U}_k$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and $\mathsf {HA} \vdash \varphi \leftrightarrow \varphi '.$

  2. 2. If $\varphi \in \mathrm{E}_k^+$ , then there exist $\varphi '\in \mathrm{E}_k$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and $\mathsf {HA} \vdash \varphi \leftrightarrow \varphi '$ .

Proof By simultaneous induction on k. The base case is trivial.

For the induction step, assume that the assertion holds for k. We show the assertion for $k+1$ by induction on the structure of formulas.

The case of that $\varphi $ is prime: Suppose $\varphi \in \mathrm{U}_{k+1}^+ \cup \mathrm{E}_{k+1}^+$ . Since $\varphi $ is prime, the assertion for k ensures that there exist $\varphi ' \in \mathrm{U}_k$ and $\varphi ''\in \mathrm{E}_k$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )=\mathrm{FV} \left ({\varphi ''}\right )$ and $\mathsf {HA} \vdash \varphi \leftrightarrow \varphi '\leftrightarrow \varphi ''$ . Put $\psi ' :\equiv \exists x \varphi '$ and $\psi '' :\equiv \forall x \varphi ''$ where $x\notin \mathrm{FV} \left ({\varphi '}\right ) \cup \mathrm{FV} \left ({\varphi ''}\right )$ . By the definition, it is straightforward to show that $\psi ' \in \mathrm{E}_{k+1}$ , $\psi '' \in \mathrm{U}_{k+1}$ and $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\psi '}\right )=\mathrm{FV} \left ({\psi ''}\right )$ . In addition, by (1) in the proof of Lemma 2.3, we have $\mathsf {HA} \vdash \varphi \leftrightarrow \psi ' \leftrightarrow \psi ''$ .

The case of $\varphi : \equiv \varphi _1 \land \varphi _2$ : Suppose $\varphi \in \mathrm{U}_{k+1}^+$ . By Lemma 4.5.(1), we have $\varphi _1 \in \mathrm{U}_{k+1}^+$ and $\varphi _2 \in \mathrm{U}_{k+1}^+$ . By the induction hypothesis, there exist $\varphi _1' ,\varphi _2' \in \mathrm{U}_k$ such that $\mathrm{FV} \left ({\varphi _1}\right )=\mathrm{FV} \left ({\varphi _1'}\right )$ , $\mathrm{FV} \left ({\varphi _2}\right )=\mathrm{FV} \left ({\varphi _2'}\right )$ , $\mathsf {HA} \vdash \varphi _1 \leftrightarrow \varphi _1'$ and $\mathsf {HA} \vdash \varphi _2 \leftrightarrow \varphi _2'$ . Then it is straightforward to show that $\varphi _1' \land \varphi _2'\in \mathrm{U}_k$ , $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi _1' \land \varphi _2'}\right )$ and $\mathsf {HA} \vdash \varphi \leftrightarrow \varphi _1'\land \varphi _2'$ . In the same manner, if $\varphi \in \mathrm{E}_{k+1}^+$ , there exists $\varphi '\in \mathrm{E}_{k+1}$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and $\mathsf {HA} \vdash \varphi \leftrightarrow \varphi '$ .

The case of $\varphi : \equiv \varphi _1 \lor \varphi _2$ is similar to the case of $\varphi : \equiv \varphi _1 \land \varphi _2$ (use Lemma 4.5.(2) instead of Lemma 4.5.(1)).

The case of $\varphi : \equiv \varphi _1 \to \varphi _2$ : Suppose $\varphi \in \mathrm{U}_{k+1}^+$ . By Lemma 4.5.(3), we have $\varphi _1\in \mathrm{E}_{k+1}^+ $ and $\varphi _2\in \mathrm{U}_{k+1}^+$ . By the induction hypothesis, there exist $\varphi _1'\in \mathrm{E}_{k+1}$ and $\varphi _2'\in \mathrm{U}_{k+1}$ such that $\mathrm{FV} \left ({\varphi _1}\right )=\mathrm{FV} \left ({\varphi _1'}\right )$ , $\mathrm{FV} \left ({\varphi _2}\right )=\mathrm{FV} \left ({\varphi _2'}\right )$ , $\mathsf {HA} \vdash \varphi _1 \leftrightarrow \varphi _1'$ and $\mathsf {HA} \vdash \varphi _2 \leftrightarrow \varphi _2'$ . Then it is straightforward to show that $\varphi _1'\to \varphi _2'\in \mathrm{U}_{k+1}$ , $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi _1' \to \varphi _2'}\right )$ and $\mathsf {HA} \vdash \varphi \leftrightarrow (\varphi _1'\to \varphi _2')$ . In the same manner, if $\varphi \in \mathrm{E}_{k+1}^+$ , there exists $\varphi '\in \mathrm{E}_{k+1}$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and $\mathsf {HA} \vdash \varphi \leftrightarrow \varphi '$ .

The case of $\varphi :\equiv \forall x \varphi _1$ : First, suppose $\varphi \in \mathrm{U}_{k+1}^+$ . By Lemma 4.5.(4), we have $\varphi _1\in \mathrm{U}_{k+1}^+$ . By the induction hypothesis, there exists $\varphi _1'\in \mathrm{U}_{k+1}$ such that $\mathrm{FV} \left ({\varphi _1}\right )=\mathrm{FV} \left ({\varphi _1'}\right )$ and $\mathsf {HA} \vdash \varphi _1 \leftrightarrow \varphi _1'$ . Then it is straightforward to show that $\forall x \varphi _1' \in \mathrm{U}_{k+1}$ , $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\forall x \varphi _1'}\right )$ and $\mathsf {HA} \vdash \varphi \leftrightarrow \forall x \varphi _1'$ . Next, suppose $\varphi \in \mathrm{E}_{k+1}^+$ . By Lemma 4.5.(6), we have $\varphi \in \mathrm{U}_{k}^+$ . The assertion for k ensures that there exists $\varphi '\in \mathrm{U}_k$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and $\mathsf {HA} \vdash \varphi \leftrightarrow \varphi '$ . Put $\psi :\equiv \exists y\varphi '$ where $y\notin \mathrm{FV} \left ({\varphi '}\right )$ . By the definition, it is straightforward to show that $\psi \in \mathrm{E}_{k+1}$ , $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\psi }\right )$ and $\mathsf {HA} \vdash \varphi \leftrightarrow \psi $ .

The case of $\varphi :\equiv \exists x \varphi _1$ is similar to the case of $\varphi :\equiv \forall x \varphi _1$ (use Lemma 4.5.(7) and Lemma 4.5.(5)).⊣

In what follows, based on Proposition 4.6, we often identify ${\mathrm{U}_k}\text {-}\mathrm{DNS}$ with ${\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ (especially in the assertions of our theorems). The latter will play a crucial role in the arguments below.

Lemma 4.7. Let k be a natural number. For all $\varphi _1$ and $\varphi _2$ in $\Pi _k$ , there exists $\varphi \in \Pi _k$ such that $\mathrm{FV} \left ({\varphi }\right ) = \mathrm{FV} \left ({\varphi _1 \lor \varphi _2}\right )$ and $\mathsf {HA} + \neg \neg {{\Sigma _{k-1}}\text {-}\mathrm{DNE}} \, (\mathsf {HA}$ if $k=0)$ proves $\neg \neg (\varphi _1 \lor \varphi _2) \leftrightarrow \neg \neg \varphi $ .

Proof Without loss of generality, assume $k>0$ , $\varphi _1 :\equiv \forall x \rho _1(x)$ and $\varphi _2 :\equiv \forall y \rho _2(y)$ where $\rho _1(x), \rho _2(y) \in \Sigma _{k-1}$ (see Remark 2.2). By Lemma 4.4, it suffices to show

$$ \begin{align*} \mathsf{HA} + \neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}} \vdash \neg \neg \left( \forall x \rho_1(x) \lor \forall y \rho_2(y)\right) \leftrightarrow \neg \neg \forall x,y \left( \rho_1(x) \lor \rho_2 (y) \right). \end{align*} $$

The implication from the left to the right is straightforward. The converse implication is shown as follows:

$$ \begin{equation*}\begin{array}{cl} & \neg \neg \forall x,y \left( \rho_1(x) \lor \rho_2 (y) \right) \\ \underset{ \neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}}}{\longleftrightarrow}&\neg \neg \forall x,y \left( \neg \neg \rho_1(x) \lor \neg \neg \rho_2 (y) \right) \\ \longrightarrow & \forall x,y \neg \neg \left( \neg \neg \rho_1(x) \lor \neg \neg \rho_2 (y) \right) \\ \longleftrightarrow & \neg \exists x,y \neg \left( \neg \neg \rho_1(x) \lor \neg \neg \rho_2 (y) \right) \\ \longleftrightarrow & \neg \exists x,y \left( \neg \rho_1(x) \land \neg \rho_2 (y) \right)\end{array} \end{equation*} $$
$$ \begin{equation*}\begin{array}{cl}\longleftrightarrow & \neg \left(\neg \neg \exists x \neg \rho_1(x) \land \neg \neg \exists y \neg \rho_2 (y) \right) \\ \longleftrightarrow & \neg \left(\neg \forall x \neg \neg \rho_1(x) \land \neg \forall y \neg \neg \rho_2 (y) \right) \\ \longleftrightarrow& \neg \neg \left(\forall x \neg \neg \rho_1(x) \lor \forall y \neg \neg \rho_2 (y) \right)\\ \underset{ \neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}}}{\longleftrightarrow}& \neg \neg \left(\forall x \rho_1(x) \lor \forall y \rho_2 (y) \right). \end{array} \end{equation*} $$

Lemma 4.8. Let k be a natural number.

  1. 1. For all $\varphi \in \Pi _k$ , there exists $\psi \in \Sigma _k$ such that $\mathrm{FV} \left ({\varphi }\right ) = \mathrm{FV} \left ({\psi }\right )$ and $\mathsf {HA} + {\Sigma _k}\text {-}\mathrm{DNE} $ proves $ \neg \varphi \leftrightarrow \psi $ .

  2. 2. For all $\varphi \in \Sigma _k$ , there exists $\psi \in \Pi _k$ such that $\mathrm{FV} \left ({\varphi }\right ) = \mathrm{FV} \left ({\psi }\right )$ and $\mathsf {HA} + {\Sigma _{k-1}}\text {-}\mathrm{DNE} \, (\mathsf {HA}$ if $k=0) $ proves $\neg \varphi \leftrightarrow \psi $ .

Proof By simultaneous induction on k. The base case is trivial. In what follows, we show the induction step for $k+1$ .

Let $\varphi :\equiv \forall x \rho (x)$ where $\rho (x) \in \Sigma _{k}$ . By induction hypothesis, there exists $\rho '(x) \in \Pi _{k}$ such that $\mathrm{FV} \left ({\rho (x)}\right ) =\mathrm{FV} \left ({\rho '(x)}\right )$ and

$$ \begin{align*}\mathsf{HA} + {\Sigma_{k-1}}\text{-}\mathrm{DNE} \vdash \neg \rho(x) \leftrightarrow \rho'(x). \end{align*} $$

Then $\mathsf {HA} + {\Sigma _{k+1}}\text {-}\mathrm{DNE}$ proves

$$ \begin{align*}\begin{array}{cl} & \neg \forall x \rho(x) \\ \underset{{\Sigma_{k}}\text{-}\mathrm{DNE}}{\longleftrightarrow} & \neg \forall x \neg \neg \rho(x)\\ \longleftrightarrow & \neg \neg \exists x \neg \rho(x)\\ \underset{\text{[I.H.] } {\Sigma_{k-1}}\text{-}\mathrm{DNE}}{\longleftrightarrow}& \neg \neg \exists x \rho'(x)\\ \underset{{\Sigma_{k+1}}\text{-}\mathrm{DNE}}{\longleftrightarrow} & \exists x \rho'(x),\\ \end{array} \end{align*} $$

which is in $\Sigma _{k+1}$ .

Next, let $\varphi :\equiv \exists x \rho (x)$ where $\rho (x) \in \Pi _{k}$ . By induction hypothesis, there exists $\rho '(x) \in \Sigma _{k}$ such that $\mathrm{FV} \left ({\rho (x)}\right ) =\mathrm{FV} \left ({\rho '(x)}\right )$ and

$$ \begin{align*}\mathsf{HA} + {\Sigma_{k}}\text{-}\mathrm{DNE} \vdash \neg \rho(x) \leftrightarrow \rho'(x). \end{align*} $$

Then $\mathsf {HA} + {\Sigma _{k}}\text {-}\mathrm{DNE}$ proves

$$ \begin{align*}\neg \exists x \rho(x) \leftrightarrow \forall x \neg \rho(x) \underset{\text{[I.H.] } {\Sigma_{k}}\text{-}\mathrm{DNE}}{\longleftrightarrow} \forall x \rho'(x), \end{align*} $$

which is in $\Pi _{k+1}$ .⊣

Lemma 4.9. Let k be a natural number. For all $\varphi \in \Pi _k$ , there exists $\psi \in \Sigma _k$ such that $\mathrm{FV} \left ({\varphi }\right ) = \mathrm{FV} \left ({\psi }\right )$ and $\mathsf {HA} + \neg \neg {{\Sigma _k}\text {-}\mathrm{DNE}} \vdash \neg \varphi \leftrightarrow \neg \neg \psi $ .

Proof Let $\varphi \in \Pi _k$ . By Lemma 4.8.(1), there exists $\psi \in \Sigma _k$ such that $\mathrm{FV} \left ({\varphi }\right ) = \mathrm{FV} \left ({\psi }\right )$ and $\mathsf {HA} + {\Sigma _k}\text {-}\mathrm{DNE} \vdash \neg \varphi \leftrightarrow \psi $ . By Corollary 4.2, we have $\mathsf {HA} + \neg \neg {{\Sigma _k}\text {-}\mathrm{DNE}} \vdash \neg \varphi \leftrightarrow \neg \neg \psi $ .⊣

Lemma 4.10. $\mathsf {HA} + {\mathrm{U}_k}\text {-}\mathrm{DNS} \vdash \neg \neg {{\Sigma _{k-1}}\text {-}\mathrm{LEM}}$ for each natural number $k>0$ .

Proof Fix an instance of $\neg \neg {{\Sigma _{k-1}}\text {-}\mathrm{LEM}}$

$$ \begin{align*}\varphi :\equiv \neg \neg \forall x \left( \varphi_1(x) \lor \neg \varphi_1(x) \right), \end{align*} $$

where $\varphi _1(x) \in \Sigma _{k-1}$ . Note $\left ( \varphi _1(x) \lor \neg \varphi _1(x) \right ) \in \text {F}_{k-1} \subseteq \mathrm{U}_k^+$ . Since $\mathsf {HA} $ proves $\forall x \neg \neg \left ( \varphi _1(x) \kern1pt{\lor}\kern1pt \neg \varphi _1(x)\kern-0.5pt \right )$ , we have that $\mathsf {HA} \kern1.2pt{+}\kern1.2pt {\mathrm{U}_k}\text {-}\mathrm{DNS}$ proves $ \neg \neg \forall x\! \left ( \varphi _1(x)\kern1pt{\lor}\kern1pt \neg \varphi _1\kern-1pt(x) \kern-0.5pt\right )\kern-0.2pt$ , namely, $\varphi $ .⊣

Corollary 4.11. $\mathsf {HA} + {\mathrm{U}_k}\text {-}\mathrm{DNS} \vdash \neg \neg {{\Sigma _{k-1}}\text {-}\mathrm{DNE}}$ for each natural number $k>0$ .

Proof Immediate from Lemma 4.10 and the fact that ${\Sigma _{k-1}}\text {-}\mathrm{LEM}$ implies ${\Sigma _{k-1}}\text {-}\mathrm{DNE}$ .⊣

5 Prenex normal form theorems

In this section, we show the modified version of [Reference Akama, Berardi, Hayashi and Kohlenbach1, Theorem 2.7]. Prior to that, we first show a variant of the prenex normal form theorem:

Lemma 5.1. For each natural number k and a formula $\varphi $ , if $\varphi \in \mathrm{U}_k^+$ , then there exists $\varphi ' \in \Pi _k$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and

$$ \begin{align*}\mathsf{HA} + {\mathrm{U}_k}\text{-}\mathrm{DNS} \vdash \neg \neg \varphi \leftrightarrow \neg \neg \varphi'.\end{align*} $$

Proof By simultaneous induction on k, we show the following two statements (which are in fact equivalent):

  1. 1. If $\varphi \in \mathrm{E}_k^+$ , then there exists $\varphi ' \in \Pi _k$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and

    $$ \begin{align*}\mathsf{HA} + {\mathrm{U}_k^+}\text{-}\mathrm{DNS} \vdash \neg \varphi \leftrightarrow \neg \neg \varphi'. \end{align*} $$
  2. 2. If $\varphi \in \mathrm{U}_k^+$ , then there exists $\varphi ' \in \Pi _k$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and

    $$ \begin{align*}\mathsf{HA} + {\mathrm{U}_k^+}\text{-}\mathrm{DNS} \vdash \neg \neg \varphi \leftrightarrow \neg \neg \varphi'.\end{align*} $$

The base case is trivial (one can take $\varphi '$ as $\varphi $ itself). In what follows, we show the induction step.

For the induction step, assume the items 1 and 2 for $k-1$ . We show the items 1 and 2 for k simultaneously by induction on the structure of formulas. When $\varphi $ is a prime formula, by Lemma 2.3, we have $\varphi '$ which satisfies the requirement. For the induction step, assume that the items 1 and 2 hold for $\varphi _1$ and $\varphi _2$ . When it is clear from the context, we suppress the argument on free variables.

The case o $\varphi _1 \land \varphi _2$ : First, assume $\varphi _1 \land \varphi _2 \in \mathrm{E}_k^+$ . By Lemma 4.5, we have $\varphi _1,\varphi _2 \in \mathrm{E}_k^+$ . By induction hypothesis, there exist $\varphi _1' ,\varphi _2' \in \Pi _k$ such that $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ proves $\neg \varphi _1 \leftrightarrow \neg \neg \varphi _1'$ and $ \neg \varphi _2 \leftrightarrow \neg \neg \varphi _2'$ . By Lemma 4.7, there exists $\varphi ' \in \Pi _k$ such that

$$ \begin{align*}\mathsf{HA} +\neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}} \vdash \neg \neg (\varphi_1' \lor \varphi_2') \leftrightarrow \neg \neg \varphi'.\end{align*} $$

By Corollary 4.11, $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ proves

$$ \begin{align*}\begin{array}{cl} & \neg (\varphi_1 \land \varphi_2) \\ \longleftrightarrow & \neg (\neg \neg \varphi_1 \land \neg \neg \varphi_2) \\ \underset{\text{[I.H.] } {\mathrm{U}_k^+}\text{-}\mathrm{DNS}}{\longleftrightarrow} & \neg (\neg \varphi_1' \land \neg \varphi_2') \\ \longleftrightarrow & \neg \neg (\varphi_1' \lor \neg \varphi_2') \\ \underset{\neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}}}{\longleftrightarrow} & \neg \neg \varphi'. \end{array} \end{align*} $$

Next, assume $\varphi _1 \land \varphi _2 \in \mathrm{U}_k^+$ . By Lemma 4.5, we have $\varphi _1,\varphi _2 \in \mathrm{U}_k^+$ . By induction hypothesis, there exist $\varphi _1' ,\varphi _2' \in \Pi _k$ such that $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ proves $\neg \neg \varphi _1 \leftrightarrow \neg \neg \varphi _1' $ and $\neg \neg \varphi _2 \leftrightarrow \neg \neg \varphi _2'$ . By Lemma 4.3, there exists $\varphi ' \in \Pi _k$ such that $\mathsf {HA} \vdash \varphi ' \leftrightarrow \varphi _1' \land \varphi _2'$ . Then $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ proves

$$ \begin{align*}\neg \neg (\varphi_1 \land \varphi_2) \leftrightarrow \neg \neg \varphi_1 \land \neg \neg \varphi_2 \underset{\text{[I.H.] }{\mathrm{U}_k^+}\text{-}\mathrm{DNS} }{\longleftrightarrow} \neg \neg \varphi_1' \land \neg \neg \varphi_2' \leftrightarrow \neg \neg (\varphi_1' \land \varphi_2') \leftrightarrow \neg \neg \varphi'. \end{align*} $$

The case of $\varphi _1 \lor \varphi _2$ : First, assume $\varphi _1 \lor \varphi _2 \in \mathrm{E}_k^+$ . By Lemma 4.5, we have $\varphi _1,\varphi _2 \in \mathrm{E}_k^+$ . By induction hypothesis, there exist $\varphi _1' ,\varphi _2' \in \Pi _k$ such that $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS} $ proves $\neg \varphi _1 \leftrightarrow \neg \neg \varphi _1' $ and $\neg \varphi _2 \leftrightarrow \neg \neg \varphi _2'$ . By Lemma 4.3, there exists $\varphi ' \in \Pi _k$ such that $\mathsf {HA} \vdash \varphi ' \leftrightarrow \varphi _1' \land \varphi _2'$ . Then $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ proves

$$ \begin{align*}\begin{array}{cl} & \neg (\varphi_1 \lor \varphi_2)\\ \longleftrightarrow & \neg \varphi_1 \land \neg \varphi_2\\ \underset{\text{[I.H.] }{\mathrm{U}_k^+}\text{-}\mathrm{DNS} }{\longleftrightarrow} & \neg \neg \varphi_1' \land \neg \neg \varphi_2'\\ \longleftrightarrow & \neg \neg (\varphi_1' \land \varphi_2')\\ \longleftrightarrow & \neg \neg \varphi'. \end{array} \end{align*} $$

Next, assume $\varphi _1 \lor \varphi _2 \in \mathrm{U}_k^+$ . By Lemma 4.5, we have $\varphi _1,\varphi _2 \in \mathrm{U}_k^+$ . By induction hypothesis, there exist $\varphi _1' ,\varphi _2' \in \Pi _k$ such that $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS} $ proves $\neg \neg \varphi _1 \leftrightarrow \neg \neg \varphi _1' $ and $\neg \neg \varphi _2 \leftrightarrow \neg \neg \varphi _2'$ . By Lemma 4.7, there exists $\varphi ' \in \Pi _k$ such that

$$ \begin{align*}\mathsf{HA} +\neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}} \vdash \neg \neg (\varphi_1' \lor \varphi_2') \leftrightarrow \neg \neg \varphi'.\end{align*} $$

By Corollary 4.11, $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ proves

$$ \begin{align*}\begin{array}{cl} & \neg \neg (\varphi_1 \lor \varphi_2)\\ \longleftrightarrow& \neg \neg ( \neg \neg \varphi_1 \lor \neg \neg \varphi_2)\\ \underset{\text{[I.H.] }{\mathrm{U}_k^+}\text{-}\mathrm{DNS} }{\longleftrightarrow} & \neg \neg ( \neg \neg \varphi_1' \lor \neg \neg \varphi_2')\\ \longleftrightarrow &\neg \neg ( \varphi_1' \lor \varphi_2')\\ \underset{\neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}}}{\longleftrightarrow}& \neg \neg \varphi'. \end{array} \end{align*} $$

The case of $\varphi _1 \to \varphi _2$ : First, assume $\varphi _1 \to \varphi _2 \in \mathrm{E}_k^+$ . By Lemma 4.5, we have $\varphi _1 \in \mathrm{U}_k^+ $ and $\varphi _2 \in \mathrm{E}_k^+$ . By induction hypothesis, there exist $\varphi _1' ,\varphi _2' \in \Pi _k$ such that $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS} $ proves $\neg \neg \varphi _1 \leftrightarrow \neg \neg \varphi _1' $ and $ \neg \varphi _2 \leftrightarrow \neg \neg \varphi _2'$ . By Lemma 4.3, there exists $\varphi ' \in \Pi _k$ such that $\mathsf {HA} \vdash \varphi ' \leftrightarrow \varphi _1' \land \varphi _2'$ . Then $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ proves

$$ \begin{align*}\begin{array}{cl} & \neg (\varphi_1 \to \varphi_2)\\ \longleftrightarrow & \neg \neg \varphi_1 \land \neg \varphi_2\\ \underset{\text{[I.H.] }{\mathrm{U}_k^+}\text{-}\mathrm{DNS} }{\longleftrightarrow} & \neg \neg \varphi_1' \land \neg \neg \varphi_2'\\ \longleftrightarrow & \neg \neg (\varphi_1' \land \varphi_2')\\ \longleftrightarrow & \neg \neg \varphi'. \end{array} \end{align*} $$

Next, assume $\varphi _1 \to \varphi _2 \in \mathrm{U}_k^+$ . By Lemma 4.5, we have $\varphi _1 \in \mathrm{E}_k^+ $ and $\varphi _2 \in \mathrm{U}_k^+$ . By induction hypothesis, there exist $\varphi _1' ,\varphi _2' \in \Pi _k$ such that $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ proves $\neg \varphi _1 \leftrightarrow \neg \neg \varphi _1' $ and $ \neg \neg \varphi _2 \leftrightarrow \neg \neg \varphi _2'$ . By Lemma 4.7, there exists $\varphi ' \in \Pi _k$ such that

$$ \begin{align*}\mathsf{HA} +\neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}} \vdash \neg \neg (\varphi_1' \lor \varphi_2') \leftrightarrow \neg \neg \varphi'.\end{align*} $$

By Corollary 4.11, $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ proves

$$ \begin{align*}\begin{array}{cl} & \neg \neg (\varphi_1 \to \varphi_2)\\ \longleftrightarrow & \neg (\neg \neg \varphi_1 \land \neg \varphi_2)\\ \underset{\text{[I.H.] }{\mathrm{U}_k^+}\text{-}\mathrm{DNS} }{\longleftrightarrow} & \neg( \neg \varphi_1' \land \neg \varphi_2')\\ \longleftrightarrow & \neg \neg (\varphi_1' \lor \varphi_2')\\ \underset{\neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}} }{\longleftrightarrow} & \neg \neg \varphi'. \end{array} \end{align*} $$

The case of $\forall x \varphi _1(x)$ : First, assume $\forall x \varphi _1(x) \in \mathrm{E}_k^+$ . By Lemma 4.5, we have $\forall x \varphi _1(x) \in \mathrm{U}_{k-1}^+$ . By the item 2 for $k-1$ , there exists $\varphi ' \in \Pi _{k-1}$ such that

$$ \begin{align*}\mathsf{HA} + {\mathrm{U}_{k-1}^+}\text{-}\mathrm{DNS} \vdash \neg \neg \forall x \varphi_1(x) \leftrightarrow \neg \neg \varphi'. \end{align*} $$

By Lemma 4.9, there exists $\varphi ''\in \Sigma _{k-1} \subseteq \Pi _k$ (see Remark 2.5) such that

$$ \begin{align*}\mathsf{HA} +\neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}} \vdash \neg \varphi' \leftrightarrow \neg \neg \varphi''. \end{align*} $$

By Corollary 4.11, $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ proves

$$ \begin{align*}\neg \forall x \varphi_1(x) \underset{\text{[I.H.] }{\mathrm{U}_{k-1}^+}\text{-}\mathrm{DNS} }{\longleftrightarrow} \neg \varphi' \underset{\neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}} }{\longleftrightarrow} \neg \neg \varphi''. \end{align*} $$

Next, assume $\forall x \varphi _1(x) \in \mathrm{U}_k^+$ . By Lemma 4.5, we have $\varphi _1(x) \in \mathrm{U}_{k}^+$ . By induction hypothesis, there exists $\varphi _1'(x) \in \Pi _{k}$ such that

$$ \begin{align*}\mathsf{HA} + {\mathrm{U}_{k}^+}\text{-}\mathrm{DNS} \vdash \neg \neg \varphi_1(x) \leftrightarrow \neg \neg \varphi_1' (x). \end{align*} $$

Then $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ proves

$$ \begin{align*}\neg \neg \forall x \varphi_1(x) \underset{{\mathrm{U}_{k}^+}\text{-}\mathrm{DNS}}{\longleftrightarrow} \forall x \neg \neg \varphi_1(x) \underset{\text{[I.H.] }{\mathrm{U}_{k}^+}\text{-}\mathrm{DNS} }{\longleftrightarrow} \forall x \neg \neg \varphi_1'(x) \underset{{\mathrm{U}_{k}^+}\text{-}\mathrm{DNS}}{\longleftrightarrow} \neg \neg \forall x \varphi_1'(x). \end{align*} $$

The case of $\exists x \varphi _1(x)$ : First, assume $\exists x \varphi _1(x) \in \mathrm{E}_k^+$ . By Lemma 4.5, we have $\varphi _1(x) \in \mathrm{E}_{k}^+$ . By induction hypothesis, there exists $\varphi _1'(x) \in \Pi _{k}$ such that

$$ \begin{align*}\mathsf{HA} + {\mathrm{U}_{k}^+}\text{-}\mathrm{DNS} \vdash \neg \varphi_1(x) \leftrightarrow \neg \neg \varphi_1'(x). \end{align*} $$

Then $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ proves

$$ \begin{align*}\neg \exists x \varphi_1(x) \leftrightarrow \forall x \neg \varphi_1(x) \underset{\text{[I.H.] }{\mathrm{U}_{k}^+}\text{-}\mathrm{DNS} }{\longleftrightarrow} \forall x \neg \neg \varphi_1'(x) \underset{{\mathrm{U}_{k}^+}\text{-}\mathrm{DNS}}{\longleftrightarrow} \neg \neg \forall x \varphi_1'(x). \end{align*} $$

Next, assume that $\exists x \varphi _1(x) \in \mathrm{U}_k^+$ . By Lemma 4.5, we have $\exists x \varphi _1(x) \in \mathrm{E}_{k-1}^+$ . By the item 1 for $k-1$ , there exists $\varphi ' \in \Pi _{k-1}$ such that

$$ \begin{align*}\mathsf{HA} + {\mathrm{U}_{k-1}^+}\text{-}\mathrm{DNS} \vdash \neg \exists x \varphi_1(x) \leftrightarrow \neg \neg \varphi'. \end{align*} $$

By Lemma 4.9, there exists $\varphi ''\in \Sigma _{k-1} \subseteq \Pi _k$ (see Remark 2.5) such that

$$ \begin{align*}\mathsf{HA} +\neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}} \vdash \neg \varphi' \leftrightarrow \neg \neg \varphi''. \end{align*} $$

By Corollary 4.11, $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ proves

$$ \begin{align*}\neg \neg \exists x \varphi_1(x) \underset{\text{[I.H.] }{\mathrm{U}_{k-1}^+}\text{-}\mathrm{DNS} }{\longleftrightarrow} \neg \varphi'\underset{\neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}} }{\longleftrightarrow} \neg \neg \varphi''. \end{align*} $$

The following lemma is used a lot of times implicitly in the proof of our prenex normal form theorem (Theorem 5.3).

Lemma 5.2 (cf. Fact 2.2 in [Reference Akama, Berardi, Hayashi and Kohlenbach1])

Let k be a natural number.

  1. 1. $\mathsf {HA} + {\Sigma _{k+1}}\text {-}\mathrm{DNE} \vdash {(\Pi _{k} \lor \Pi _{k})}\text {-}\mathrm{DNE}$ .

  2. 2. $\mathsf {HA} +{(\Pi _{k+1} \lor \Pi _{k+1})}\text {-}\mathrm{DNE} \vdash {\Sigma _{k}}\text {-}\mathrm{DNE}$ .

  3. 3. $\mathsf {HA} +\neg \neg {{(\Pi _{k+1} \lor \Pi _{k+1})}\text {-}\mathrm{DNE}} \vdash \neg \neg {{\Sigma _{k}}\text {-}\mathrm{DNE}}$ .

  4. 4. $ \mathsf {HA} + {\Sigma _{k}}\text {-}\mathrm{DNE} \vdash {\Pi _{k+1}}\text {-}\mathrm{DNE} $ .

Proof (1): For formulas $\varphi _1$ and $\varphi _2$ in $\Pi _{k}$ , $\varphi _1 \lor \varphi _2$ is equivalent (over $\mathsf {HA}$ ) to

$$ \begin{align*} \exists k \left(\left(k=0 \to \varphi_1 \right) \land \left( k\neq 0 \to \varphi_2 \right) \right), \end{align*} $$

which is equivalent to some $\varphi \in \Sigma _{k+1}$ such that $\mathrm{FV} \left ({\varphi }\right )= \mathrm{FV} \left ({\varphi _1 \lor \varphi _2}\right )$ over $\mathsf {HA}$ by Lemma 4.3.(2). Therefore any instance of ${(\Pi _{k} \lor \Pi _{k})}\text {-}\mathrm{DNE}$ is derived from some instance of ${\Sigma _{k+1}}\text {-}\mathrm{DNE}$ .

(2): Any instance of ${\Sigma _{k}}\text {-}\mathrm{DNE}$ is derived from some instance of ${(\Pi _{k+1} \lor \Pi _{k+1})}\text {-}\mathrm{DNE}$ since $\varphi \in \Sigma _k$ is equivalent to $\forall y \varphi \in \Pi _{k+1}$ with a variable y not occurring freely in $\varphi $ (cf. Lemma 2.3).

(3): Immediate from (2) and Corollary 4.2.

(4): Note that $\neg \neg \forall x \varphi (x)$ implies $\neg \neg \forall x \neg \neg \varphi (x)$ , which is intuitionistically equivalent to $\forall x \neg \neg \varphi (x)$ . Then any instance of ${\Pi _{k+1}}\text {-}\mathrm{DNE}$ is derived from some instance of ${\Sigma _{k}}\text {-}\mathrm{DNE}$ .⊣

We are now ready to show the modified version of [Reference Akama, Berardi, Hayashi and Kohlenbach1, Theorem 2.7].

Theorem 5.3. For each natural number k and a formula $\varphi $ , the following hold:

  1. 1. If $\varphi \in \mathrm{E}_k^+$ , then there exists $\varphi ' \in \Sigma _k$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and

    $$ \begin{align*}\mathsf{HA} + {\Sigma_k}\text{-}\mathrm{DNE} +{\mathrm{U}_k}\text{-}\mathrm{DNS} \vdash \varphi \leftrightarrow \varphi'; \end{align*} $$
  2. 2. If $\varphi \in \mathrm{U}_k^+$ , then there exists $\varphi ' \in \Pi _k$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and

    $$ \begin{align*}\mathsf{HA} + {(\Pi_k\lor \Pi_k)}\text{-}\mathrm{DNE} \vdash \varphi \leftrightarrow \varphi'. \end{align*} $$

Proof For the proof, we prepare the following auxiliary assertion (which is in fact a consequence of the item 2):

  1. 3. If $\varphi \in \mathrm{E}_k^+$ , then there exists $\varphi ' \in \Pi _k$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and

    $$ \begin{align*}\mathsf{HA} + \neg \neg {{(\Pi_k\lor \Pi_k)}\text{-}\mathrm{DNE}} \vdash \neg \varphi \leftrightarrow \neg \neg \varphi'. \end{align*} $$

We show the items 13 by induction on k simultaneously. The base case is trivial (one can take $\varphi '$ as $\varphi $ itself). In what follows, we show the induction step.

Assume the items 13 for $k-1$ . Since $\mathsf {HA} + {\Pi _{k-1}}\text {-}\mathrm{DNE} \vdash {\Pi _{k-1}}\text {-}\mathrm{DNS}$ , by the item 2 for $k-1$ , we have

(3) $$ \begin{align} \mathsf{HA} + {\left(\Pi_{k-1}\lor \Pi_{k-1}\right)}\text{-}\mathrm{DNE} \vdash {\mathrm{U}_{k-1}^+}\text{-}\mathrm{DNS}. \end{align} $$

We show the items 13 simultaneously by induction on the structure of formulas. When $\varphi $ is a prime formula, by Lemma 2.3, we have $\varphi '$ which satisfies the requirement. For the induction step, assume that the items 13 hold for $\varphi _1$ and $\varphi _2$ . When it is clear from the context, we suppress the argument on free variables.

The case of $\varphi _1 \land \varphi _2$ : For the second item, assume $\varphi _1 \land \varphi _2 \in \mathrm{U}_k^+$ . By Lemma 4.5, we have $\varphi _1,\varphi _2 \in \mathrm{U}_k^+$ . By induction hypothesis, there exist $\varphi _1' ,\varphi _2' \in \Pi _k$ such that $\mathsf {HA} +{(\Pi _k\lor \Pi _k)}\text {-}\mathrm{DNE} $ proves $\varphi _1 \leftrightarrow \varphi _1'$ and $\varphi _2 \leftrightarrow \varphi _2'$ . By Lemma 4.3, there exists $\varphi ' \in \Pi _k$ such that $\mathsf {HA} \vdash \varphi ' \leftrightarrow \varphi _1'\land \varphi _2'$ . Then $\mathsf {HA} +{(\Pi _k\lor \Pi _k)}\text {-}\mathrm{DNE}$ proves

$$ \begin{align*}\varphi_1 \land \varphi_2 \underset{\text{[I.H.] }{(\Pi_k\lor \Pi_k)}\text{-}\mathrm{DNE} }{\longleftrightarrow} \varphi_1' \land \varphi_2' \leftrightarrow \varphi'.\end{align*} $$

For the first and third items, assume $\varphi _1 \land \varphi _2 \in \mathrm{E}_k^+$ . By Lemma 4.5, we have $\varphi _1,\varphi _2 \in \mathrm{E}_k^+$ . Then we have $\varphi '\in \Sigma _k$ such that $ \mathsf {HA} +{\Sigma _k}\text {-}\mathrm{DNE} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS} \vdash \varphi _1 \land \varphi _2 \leftrightarrow \varphi '$ as in the second item. On the other hand, by induction hypothesis, there exist $\varphi _1'' ,\varphi _2'' \in \Pi _k$ such that $\mathsf {HA} + \neg \neg {{(\Pi _k\lor \Pi _k)}\text {-}\mathrm{DNE}} $ proves $\neg \varphi _1 \leftrightarrow \neg \neg \varphi _1'' $ and $ \neg \varphi _2 \leftrightarrow \neg \neg \varphi _2''$ . In addition, by Lemma 4.7, there exists $\varphi '' \in \Pi _k$ such that $\mathsf {HA} + \neg \neg {{\Sigma _{k-1}}\text {-}\mathrm{DNE}} \vdash \neg \neg \varphi '' \leftrightarrow \neg \neg (\varphi _1'' \lor \varphi _2'')$ . Then, by Lemma 5.2, we have that $\mathsf {HA} + \neg \neg {{(\Pi _k\lor \Pi _k)}\text {-}\mathrm{DNE}}$ proves

$$ \begin{align*}\begin{array}{cl} &\neg (\varphi_1 \land \varphi_2) \\ \longleftrightarrow& \neg (\neg \neg \varphi_1 \land \neg \neg \varphi_2)\\ \longleftrightarrow& \neg \neg (\neg \varphi_1 \lor \neg \varphi_2)\\ \underset{\text{[I.H.] }\neg \neg {{(\Pi_{k}\lor \Pi_{k})}\text{-}\mathrm{DNE}}}{\longleftrightarrow} & \neg \neg (\neg \neg \varphi_1'' \lor \neg \neg \varphi_2'' )\\ \longleftrightarrow& \neg (\neg \varphi_1'' \land \neg \varphi_2'')\\ \longleftrightarrow&\neg \neg (\varphi_1'' \lor \varphi_2'')\\ \underset{\neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}}}{\longleftrightarrow}&\neg \neg \varphi''. \end{array} \end{align*} $$

The case of $\varphi _1 \lor \varphi _2$ : For the second item, assume $\varphi _1 \lor \varphi _2 \in \mathrm{U}_k^+$ . By Lemma 4.5, we have $\varphi _1,\varphi _2 \in \mathrm{U}_k^+$ . Then, by induction hypothesis, there exist $\rho _1(x_1), \rho _2(x_2) \in \Sigma _{k-1}$ such that $\mathsf {HA} + {(\Pi _k\lor \Pi _k)}\text {-}\mathrm{DNE} $ proves $\varphi _1 \leftrightarrow \forall x_1 \rho _1(x_1) $ and $\varphi _2 \leftrightarrow \forall x_2 \rho _2(x_2) $ . By Lemma 5.2, $\mathsf {HA} + {(\Pi _k\lor \Pi _k)}\text {-}\mathrm{DNE} $ proves

$$ \begin{align*}\begin{array}{cl} & \forall x_1 \rho_1(x_1) \lor \forall x_2 \rho_2(x_2)\\ \longrightarrow& \forall x_1, x_2 \left( \rho_1(x_1) \lor \rho_2(x_2)\right)\\ \longrightarrow& \neg \left( \exists x_1 \neg \rho_1 (x_1) \land \exists x_2 \neg \rho_2 (x_2) \right) \\ \longleftrightarrow& \neg \left( \neg \neg \exists x_1 \neg \rho_1 (x_1) \land \neg \neg \exists x_2 \neg \rho_2 (x_2) \right) \\ \underset{{\Sigma_{k-1}}\text{-}\mathrm{DNE}}{\longleftrightarrow} &\neg \left( \neg \forall x_1 \rho_1 (x_1) \land \neg \forall x_2 \rho_2 (x_2) \right) \\ \longleftrightarrow & \neg \neg \left( \forall x_1 \rho_1 (x_1) \lor \forall x_2 \rho_2 (x_2) \right) \\ \underset{{(\Pi_k\lor \Pi_k)}\text{-}\mathrm{DNE}}{\longrightarrow} &\forall x_1 \rho_1 (x_1) \lor \forall x_2 \rho_2 (x_2). \end{array} \end{align*} $$

By Lemma 4.4, there exists $\xi (x_1, x_2) \in \Sigma _{k-1}$ such that $\mathsf {HA} \vdash \xi (x_1, x_2) \leftrightarrow \rho _1(x_1) \lor \rho _2(x_2)$ . Then we have that $\mathsf {HA} + {(\Pi _k\lor \Pi _k)}\text {-}\mathrm{DNE} $ proves

$$ \begin{align*}\begin{array}{cl} &\varphi_1 \lor \varphi_2\\ \underset{\text{[I.H.] }{(\Pi_k\lor \Pi_k)}\text{-}\mathrm{DNE}}{\longleftrightarrow}& \forall x_1 \rho_1 (x_1) \lor \forall x_2 \rho_2 (x_2)\\ \underset{{(\Pi_k\lor \Pi_k)}\text{-}\mathrm{DNE}}{\longleftrightarrow} & \forall x_1, x_2 \left( \rho_1(x_1) \lor \rho_2(x_2)\right)\\ \longleftrightarrow& \forall x_1, x_2\, \xi(x_1, x_2) \in \Pi_k. \end{array}\end{align*} $$

For the first and third items, assume $\varphi _1 \lor \varphi _2 \in \mathrm{E}_k^+$ . By Lemma 4.5, we have $\varphi _1,\varphi _2 \in \mathrm{E}_k^+$ . By induction hypothesis, there exist $\rho _1(x_1), \rho _2(x_2) \in \Pi _{k-1}$ such that $\mathsf {HA} + {\Sigma _k}\text {-}\mathrm{DNE} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS} $ proves $\varphi _1 \leftrightarrow \exists x_1 \rho _1(x_1) $ and $ \varphi _2 \leftrightarrow \exists x_2 \rho _2(x_2) $ . By the item 2 for $k-1$ , there exists $\xi (x_1, x_2) \in \Pi _{k-1}$ such that

$$ \begin{align*}\mathsf{HA} + {(\Pi_{k-1}\lor \Pi_{k-1})}\text{-}\mathrm{DNE} \vdash \xi(x_1,x_2) \leftrightarrow \rho_1(x_1) \lor \rho_2(x_2).\end{align*} $$

By Lemma 5.2, we have that $\mathsf {HA} + {\Sigma _k}\text {-}\mathrm{DNE} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS} $ proves

$$ \begin{align*}\begin{array}{cl} &\varphi_1 \lor \varphi_2 \\ \underset{\text{[I.H.] }{\Sigma_k}\text{-}\mathrm{DNE},\, {\mathrm{U}_k^+}\text{-}\mathrm{DNS}}{\longleftrightarrow} & \exists x_1 \rho_1 (x_1) \lor \exists x_2 \rho_2 (x_2)\\ \longleftrightarrow &\exists x_1, x_2 (\rho_1(x_1) \lor \rho_2(x_2)) \\ \underset{\text{[I.H.] }{(\Pi_{k-1}\lor \Pi_{k-1})}\text{-}\mathrm{DNE}}{\longleftrightarrow} &\exists x_1, x_2\, \xi(x_1, x_2). \end{array} \end{align*} $$

Thus we are done for the first item. For the third item, by induction hypothesis, there exist $\varphi _1'' ,\varphi _2'' \in \Pi _k$ such that $\mathsf {HA} + \neg \neg {{(\Pi _k\lor \Pi _k)}\text {-}\mathrm{DNE}}$ proves $\neg \varphi _1 \leftrightarrow \neg \neg \varphi _1''$ and $ \neg \varphi _2 \leftrightarrow \neg \neg \varphi _2''$ . In addition, by Lemma 4.3, there exists $\varphi '' \in \Pi _k$ such that $\mathsf {HA} \vdash \varphi '' \leftrightarrow \varphi _1'' \land \varphi _2''$ . Then $\mathsf {HA} + \neg \neg {{(\Pi _k\lor \Pi _k)}\text {-}\mathrm{DNE}}$ proves

$$ \begin{align*}\begin{array}{r}\neg (\varphi_1 \lor \varphi_2 ) \leftrightarrow \neg \varphi_1 \land \neg \varphi_2 \underset{\text{[I.H.] }\neg \neg {{(\Pi_{k}\lor \Pi_{k})}\text{-}\mathrm{DNE}}}{\longleftrightarrow} \neg \neg \varphi_1'' \land \neg \neg \varphi_2''\\ \leftrightarrow \neg \neg (\varphi_1'' \land \varphi_2'') \leftrightarrow \neg \neg \varphi''.\end{array} \end{align*} $$

The case of $\varphi _1 \to \varphi _2$ : For the second item, assume $\varphi _1 \to \varphi _2 \in \mathrm{U}_k^+$ . By Lemma 4.5, we have $\varphi _1 \in \mathrm{E}_k^+$ and $\varphi _2 \in \mathrm{U}_k^+$ . By induction hypothesis, there exist $ \rho _1(x_1), \rho _2(x_2)\in \Sigma _{k-1}$ such that $\mathsf {HA} + \neg \neg {{(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE}} \vdash \neg \varphi _1 \leftrightarrow \neg \neg \forall x_1 \rho _1(x_1)$ and $\mathsf {HA} + {(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE} \vdash \varphi _2 \leftrightarrow \forall x_2 \rho _2(x_2)$ . By Lemma 4.5, we have that $\neg \rho _1(x_1) \to \rho _2(x_2)$ is in $\mathrm{E}_{k-1}^+$ . Then, by the item 1 for $k-1$ , there exists $\xi (x_1, x_2) \in \Sigma _{k-1}$ such that

$$ \begin{align*}\mathsf{HA} + {\Sigma_{k-1}}\text{-}\mathrm{DNE} + {\mathrm{U}_{k-1}^+}\text{-}\mathrm{DNS} \vdash \xi(x_1, x_2) \leftrightarrow \left( \neg \rho_1(x_1) \to \rho_2(x_2) \right).\end{align*} $$

Then, using Lemma 5.2 and (3), we have that $\mathsf {HA} + {(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE} $ proves

$$ \begin{align*}\begin{array}{cl} &\varphi_1 \to \varphi_2\\ \underset{\text{[I.H.] }{(\Pi_{k}\lor \Pi_{k})}\text{-}\mathrm{DNE}}{\longleftrightarrow} & \varphi_1 \to \forall x_2 \rho_2 (x_2)\\ \underset{{\Pi_k}\text{-}\mathrm{DNE}}{\longleftrightarrow} & \varphi_1 \to \neg \neg \forall x_2 \rho_2 (x_2)\\ \longleftrightarrow & \neg \neg \varphi_1 \to \neg \neg \forall x_2 \rho_2 (x_2) \\ \underset{\text{[I.H.] }\neg \neg {{(\Pi_{k}\lor \Pi_{k})}\text{-}\mathrm{DNE}}}{\longleftrightarrow} & \neg \forall x_1 \rho_1(x_1) \to \neg \neg \forall x_2 \rho_2 (x_2)\\ \underset{{\Sigma_{k-1}}\text{-}\mathrm{DNE} }{\longleftrightarrow} & \neg \neg \exists x_1 \neg \rho_1(x_1) \to \neg \neg \forall x_2 \rho_2 (x_2)\\ \longleftrightarrow & \neg \neg \forall x_1, x_2 \left( \neg \rho_1(x_1) \to \rho_2 (x_2) \right)\\ \underset{\text{[I.H.] }{\Sigma_{k-1}}\text{-}\mathrm{DNE} , \, {\mathrm{U}_{k-1}^+}\text{-}\mathrm{DNS} }{\longleftrightarrow} &\neg \neg \forall x_1, x_2 \, \xi( x_1, x_2)\\ \underset{{\Pi_k}\text{-}\mathrm{DNE}}{\longleftrightarrow} &\forall x_1, x_2 \, \xi( x_1, x_2) \in \Pi_k.\\ \end{array} \end{align*} $$

For the first and third items, assume $\varphi _1 \to \varphi _2 \in \mathrm{E}_k^+$ . By Lemma 4.5, we have $\varphi _1 \in \mathrm{U}_k^+$ and $\varphi _2 \in \mathrm{E}_k^+$ . By induction hypothesis, there exists $ \rho _2(x_2) \in \Pi _{k-1}$ such that $\mathsf {HA} + {\Sigma _{k}}\text {-}\mathrm{DNE} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS} \vdash \varphi _2 \leftrightarrow \exists x_2 \rho _2(x_2) $ . In addition, by Lemma 5.1, there exists $ \rho _1(x_1)\in \Sigma _{k-1}$ such that $\mathsf {HA} + {\mathrm{U}_k^+}\text {-}\mathrm{DNS} \vdash \neg \neg \varphi _1 \leftrightarrow \neg \neg \forall x_1 \rho _1(x_1)$ . By Lemma 4.5, we have $\neg \rho _2(x_2) \to \neg \rho _1(x_1)$ is in $\mathrm{U}_{k-1}^+$ . Then, by the item 2 for $k-1$ , there exists $\xi (x_1, x_2) \in \Pi _{k-1}$ such that

$$ \begin{align*}\mathsf{HA} + {(\Pi_{k-1}\lor \Pi_{k-1})}\text{-}\mathrm{DNE} \vdash \xi(x_1, x_2) \leftrightarrow \left(\neg \rho_2(x_2) \to \neg \rho_1(x_1) \right).\end{align*} $$

Then, using Lemma 5.2, we have that $\mathsf {HA} + {\Sigma _{k}}\text {-}\mathrm{DNE}+ {\mathrm{U}_k^+}\text {-}\mathrm{DNS}$ proves

$$ \begin{align*}\begin{array}{cl} & \varphi_1 \to \varphi_2 \\ \underset{\text{[I.H.] }{\Sigma_{k}}\text{-}\mathrm{DNE}, \, {\mathrm{U}_k^+}\text{-}\mathrm{DNS}}{\longleftrightarrow}& \varphi_1 \to \exists x_2 \rho_2(x_2)\\ \underset{{\Sigma_k}\text{-}\mathrm{DNE}}{\longleftrightarrow}& \varphi_1 \to \neg \neg \exists x_2 \rho_2(x_2)\\ \longleftrightarrow& \neg \neg \varphi_1 \to \neg \neg \exists x_2 \rho_2(x_2)\\ \underset{ {\mathrm{U}_k^+}\text{-}\mathrm{DNS}}{\longleftrightarrow}&\neg \neg \forall x_1 \rho_1(x_1) \to \neg \neg \exists x_2 \rho_2(x_2)\\ \longleftrightarrow&\neg \neg \exists x_2 \left( \forall x_1 \rho_1(x_1) \to \rho_2(x_2) \right) \\ \underset{{\Pi_{k-1}}\text{-}\mathrm{DNE}}{\longleftrightarrow}&\neg \neg \exists x_2 \left(\neg \rho_2(x_2) \to \neg \forall x_1 \rho_1(x_1) \right) \\ \underset{{\Sigma_{k-1}}\text{-}\mathrm{DNE}}{\longleftrightarrow}&\neg \neg \exists x_2 \left(\neg \rho_2(x_2) \to \neg \neg \exists x_1 \neg \rho_1(x_1) \right) \\ \longleftrightarrow &\neg \neg \exists x_2 \neg \neg \exists x_1 \left(\neg \rho_2(x_2) \to \neg \rho_1(x_1) \right) \\ \underset{\text{[I.H.] } {(\Pi_{k-1}\lor \Pi_{k-1})}\text{-}\mathrm{DNE}}{\longleftrightarrow}&\neg \neg \exists x_1, x_2 \, \xi (x_1, x_2)\\ \underset{ {\Sigma_k}\text{-}\mathrm{DNE}}{\longleftrightarrow}& \exists x_1, x_2 \, \xi (x_1, x_2) \in \Sigma_k .\\ \end{array} \end{align*} $$

Thus we are done for the first item. For the third item, by induction hypothesis, there exist $\varphi _1', \varphi _2' \in \Pi _k$ such that $\mathsf {HA} + {(\Pi _k \lor \Pi _k)}\text {-}\mathrm{DNE} \vdash \varphi _1 \leftrightarrow \varphi _1'$ and $\mathsf {HA} + \neg \neg {{(\Pi _k \lor \Pi _k)}\text {-}\mathrm{DNE}} \vdash \neg \varphi _2 \leftrightarrow \neg \neg \varphi _2'$ . On the other hand, by Lemma 4.3, there exists $\xi ' \in \Pi _k$ such that $\mathsf {HA} \vdash \varphi _1' \land \varphi _2' \leftrightarrow \xi '$ . Then $\mathsf {HA} +\neg \neg {{(\Pi _k \lor \Pi _k)}\text {-}\mathrm{DNE}} $ proves

$$ \begin{align*}\begin{array}{r} \neg( \varphi_1 \to \varphi_2) \leftrightarrow (\neg \neg \varphi_1 \land \neg \varphi_2)\underset{\text{[I.H.] } \neg \neg {{(\Pi_{k}\lor \Pi_{k})}\text{-}\mathrm{DNE}}}{\longleftrightarrow} \neg \neg \varphi_1' \land \neg \neg \varphi_2'\\ \leftrightarrow \neg \neg (\varphi_1' \land \varphi_2') \leftrightarrow \neg \neg \xi'. \end{array}\end{align*} $$

The case of $\forall x \varphi _1 (x )$ : For the second item, assume $\forall x \varphi _1 (x ) \in \mathrm{U}_k^+$ . By Lemma 4.5, we have $\varphi _1 (x ) \in \mathrm{U}_k^+$ . By induction hypothesis, there exists $\varphi _1'(x ) \in \Pi _{k}$ such that $\mathsf {HA} + {(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE} \vdash \varphi _1(x ) \leftrightarrow \varphi _1'(x )$ . Then $\forall x \varphi _1 (x )$ is equivalent to $\forall x \varphi _1'(x ) \in \Pi _k$ over $\mathsf {HA} + {(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE} $ .

For the first and third items, assume $\forall x \varphi _1 (x ) \in \mathrm{E}_k^+$ . By Lemma 4.5, we have $\forall x \varphi _1 (x ) \in \mathrm{U}_{k-1}^+$ . Then, by the item 2 for $k-1$ , there exists $\xi \in \Pi _{k-1} \subseteq \Sigma _k$ (see Remark 2.5) such that

(4) $$ \begin{align} \mathsf{HA} + {(\Pi_{k-1}\lor \Pi_{k-1})}\text{-}\mathrm{DNE} \vdash \forall x \varphi_1(x ) \leftrightarrow \xi. \end{align} $$

By Lemma 5.2, we are done for the first item. For the third item, by Lemma 4.8, there exists $\xi '\in \Sigma _{k-1} \subseteq \Pi _k$ (see Remark 2.5) such that $\mathsf {HA} + {\Sigma _{k-1}}\text {-}\mathrm{DNE} \vdash \neg \xi \leftrightarrow \xi '$ . By Corollary 4.2, we have $\mathsf {HA} + \neg \neg {{\Sigma _{k-1}}\text {-}\mathrm{DNE}} \vdash \neg \xi \leftrightarrow \neg \neg \xi '$ . In addition,

$$ \begin{align*}\mathsf{HA} + \neg \neg {{(\Pi_{k-1}\lor \Pi_{k-1})}\text{-}\mathrm{DNE}} \vdash \neg \neg \forall x \varphi_1(x ) \leftrightarrow \neg \neg \xi\end{align*} $$

follows from (4). Then, by Lemma 5.2, we have that $\mathsf {HA} + \neg \neg {{(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE}} $ proves

$$ \begin{align*}\neg \forall x \varphi_1(x ) \underset{\text{[I.H.] } \neg \neg {{(\Pi_{k-1}\lor \Pi_{k-1})}\text{-}\mathrm{DNE}}}{\longleftrightarrow} \neg \xi \underset{ \neg \neg {{\Sigma_{k-1}}\text{-}\mathrm{DNE}}}{\longleftrightarrow} \neg \neg \xi'. \end{align*} $$

Thus we have shown the third item.

The case of $\exists x \varphi _1 (x )$ : For the second item, assume $\exists x \varphi _1 (x ) \in \mathrm{U}_k^+$ . By Lemma 4.5, we have $ \exists x \varphi _1 (x ) \in \mathrm{E}_{k-1}^+$ . Then, by the item 1 for $k-1$ , there exists $\xi \in \Sigma _{k-1}\subseteq \Pi _k$ (see Remark 2.5) such that $\mathsf {HA} +{\Sigma _{k-1}}\text {-}\mathrm{DNE} + {\mathrm{U}_{k-1}^+}\text {-}\mathrm{DNS} \vdash \exists x \varphi _1(x ) \leftrightarrow \xi $ . By Lemma 5.2 and (3), we have $\mathsf {HA} + {(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE} \vdash \exists x \varphi _1(x ) \leftrightarrow \xi $ .

For the first and third items, assume $\exists x \varphi _1 (x ) \in \mathrm{E}_k^+$ . By Lemma 4.5, we have $\varphi _1 (x ) \in \mathrm{E}_k^+$ . By induction hypothesis, there exist $\varphi _1'(x ) \in \Sigma _{k}$ and $\varphi _1''(x ) \in \Pi _k$ such that $\mathsf {HA} +{\Sigma _{k}}\text {-}\mathrm{DNE} +{\mathrm{U}_{k}^+}\text {-}\mathrm{DNS} \vdash \varphi _1(x ) \leftrightarrow \varphi _1'(x )$ and $\mathsf {HA} + \neg \neg {{(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE}} \vdash \neg \varphi _1 (x) \leftrightarrow \neg \neg \varphi _1'' (x)$ . Then $\exists x \varphi _1 (x )$ is equivalent to $\exists x \varphi _1'(x ) \in \Sigma _k$ over $\mathsf {HA} +{\Sigma _{k}}\text {-}\mathrm{DNE} +{\mathrm{U}_{k}^+}\text {-}\mathrm{DNS} $ . Thus we are done for the first item. For the third item, since $\mathsf {HA} + \neg \neg {{(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE}}$ proves

$$ \begin{align*}\neg \exists x \varphi_1(x )\leftrightarrow \forall x \neg \varphi_1(x ) \underset{ \text{[I.H.] }\neg \neg {{(\Pi_{k}\lor \Pi_{k})}\text{-}\mathrm{DNE}}}{\longleftrightarrow} \forall x \neg \neg \varphi_1''(x ) , \end{align*} $$

we have that $\mathsf {HA} + \neg \neg {{(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE}}$ proves

$$ \begin{align*}\neg \exists x \varphi_1(x )\leftrightarrow \neg \neg \forall x \neg \neg \varphi_1''(x ). \end{align*} $$

On the other hand, the latter is equivalent to $ \neg \neg \forall x \varphi _1''(x ) $ in the presence of $ \neg \neg {{\Pi _{k}}\text {-}\mathrm{DNE}}$ . Thus we have $\mathsf {HA} + \neg \neg {{(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE}} \vdash \neg \exists x \varphi _1(x )\leftrightarrow \neg \neg \forall x \varphi _1''(x )$ .⊣

Corollary 5.4. $\mathsf {HA} + \neg \neg {{(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE}} \vdash {\mathrm{U}_k}\text {-}\mathrm{DNS} .$

Proof Since ${\mathrm{U}_k}\text {-}\mathrm{DNS} $ is intuitionistically equivalent to $\neg \neg {{\mathrm{U}_k}\text {-}\mathrm{DNS} }$ (see Remark 2.8), it suffices to show $\mathsf {HA} + {(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE} \vdash {\mathrm{U}_k}\text {-}\mathrm{DNS} $ . By Theorem 5.3.(2), any formula $\varphi \in \mathrm{U}_k$ is equivalent to some $\varphi ' \in \Pi _k$ such that $\mathrm{FV} \left ({\varphi }\right ) = \mathrm{FV} \left ({\varphi '}\right )$ over $\mathsf {HA} + {(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE}$ . Since $\mathsf {HA} + {\Pi _k}\text {-}\mathrm{DNE} \vdash {\Pi _k}\text {-}\mathrm{DNS}$ , we have $\mathsf {HA} + {(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE} \vdash {\mathrm{U}_k}\text {-}\mathrm{DNS} $ .⊣

Remark 5.5. Corollary 5.4 shows that Lemma 5.1 (equivalent to the item 1 in the proof of Lemma 5.1) is a stronger statement of the item 3 in the proof of Theorem 5.3. On the other hand, it is still open whether $\mathsf {HA} + {\mathrm{U}_k}\text {-}\mathrm{DNS}$ is a proper sub-theory of $\mathsf {HA} + \neg \neg {{(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE}} $ . In fact, in the proof of Theorem 5.3, Lemma 5.1 is only used for obtaining $\psi _1\in \Pi _k$ such that $\neg \neg \psi _1 \leftrightarrow \neg \neg \varphi _1$ (where $\varphi _1\in \mathrm{U}_k^+$ ) in the verification theory of the first item. Since this argument is available in $\mathsf {HA} + \neg \neg {{(\Pi _k \lor \Pi _k)}\text {-}\mathrm{DNE}}$ with assuming the second item for $\varphi _1$ , one can show the alternative assertions of Theorem 5.3 where ${\mathrm{U}_k}\text {-}\mathrm{DNS}$ is replaced by $\neg \neg {{(\Pi _k \lor \Pi _k)}\text {-}\mathrm{DNE}}$ without using Lemma 5.1. Thus, if $\mathsf {HA} + {\mathrm{U}_k}\text {-}\mathrm{DNS}$ proves $\neg \neg {{(\Pi _k \lor \Pi _k)}\text {-}\mathrm{DNE}}$ , we can simplify the proof of Theorem 5.3.

Remark 5.6. It follows from Theorem 5.3 and the results in §3 that $\mathsf {HA} + {\Sigma _1}\text {-}\mathrm{DNE}$ does not prove ${\mathrm{U}_1}\text {-}\mathrm{DNS}$ .

At the end of this section, we study the prenex normal form theorem for formulas which do not contain the disjunction $\lor $ . In fact, the proof of Theorem 5.3 suggests that the unusual form ${(\Pi _k \lor \Pi _k)}\text {-}\mathrm{DNE}$ of the double negation elimination is caused by the argument especially in the case of $\varphi _1 \lor \varphi _2$ . On the other hand, if a formula $\varphi $ does not contain $\lor $ , one can intuitionistically derive the original formula $\varphi $ from a formula in prenex normal form which is classically equivalent to $\varphi $ (cf. [Reference van Dalen10, Lemma 6.2.1]). Then the proof of the prenex normal form theorem for those formulas becomes to be fairly simple.

Theorem 5.7. For each natural number k and a formula $\varphi $ which does not contain $\lor $ , the following hold:

  1. 1. If $\varphi \in \mathrm{E}_k^+$ , then there exists $\varphi ' \in \Sigma _k$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and

    $$ \begin{align*}\mathsf{HA} + {\Sigma_k}\text{-}\mathrm{DNE} \vdash \varphi \leftrightarrow \varphi'. \end{align*} $$
  2. 2. If $\varphi \in \mathrm{U}_k^+$ , then there exists $\varphi ' \in \Pi _k$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and

    $$ \begin{align*}\mathsf{HA} + {\Sigma_{k-1}}\text{-}\mathrm{DNE} \,\, (\mathsf{HA} \text{ if }k=0) \vdash \varphi \leftrightarrow \varphi'. \end{align*} $$

Proof We mimic the proof of Theorem 5.3. Thus we first prepare the following auxiliary assertion (which is in fact a consequence of the item 2):

  1. 3. If $\varphi \in \mathrm{E}_k^+$ , then there exists $\varphi ' \in \Pi _k$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and

    $$ \begin{align*}\mathsf{HA} + \neg \neg {\Sigma_{k-1}}\text{-}\mathrm{DNE} \vdash \neg \varphi \leftrightarrow \neg \neg \varphi'. \end{align*} $$

Then we show the items 13 by induction on k simultaneously. The base case is trivial. Most of the parts for the induction step is the same as those for Theorem 5.3. The same proof works since each of the logical principles in the items 1 and 2 implies both of them for $k-1$ and the logical principle in the item 3 is the double negation of the logical principle in the item 2 as in Theorem 5.3.

Only the difference with the proof of Theorem 5.3 is in proving the item 1 for $\varphi : \equiv \varphi _1 \to \varphi _2\in \mathrm{E}_k^+$ , where we use Lemma 5.1. Here one can use the item 2 instead of Lemma 5.1. This is because ${\Sigma _k}\text {-}\mathrm{DNE}$ includes ${\Sigma _{k-1}}\text {-}\mathrm{DNE}$ while the verification theory of the item 1 in Theorem 5.3 contains the verification theory of Lemma 5.1. To be absolutely clear, we present the proof of this part: Let $ \varphi _1 \to \varphi _2\in \mathrm{E}_k^+$ . By Lemma 4.5, we have $ \varphi _1 \in \mathrm{U}_k^+$ and $\varphi _2\in \mathrm{E}_k^+$ . By induction hypothesis, there exists $\rho _1(x_1) \in \Sigma _{k-1}$ and $ \rho _2(x_2) \in \Pi _{k-1}$ such that $\mathsf {HA} + {\Sigma _{k-1}}\text {-}\mathrm{DNE} \vdash \varphi _1 \leftrightarrow \forall x_1 \rho _1(x_1)$ and $\mathsf {HA} + {\Sigma _{k}}\text {-}\mathrm{DNE} \vdash \varphi _2 \leftrightarrow \exists x_2 \rho _2(x_2) $ . By Lemma 4.5, we have $\neg \rho _2(x_2) \to \neg \rho _1(x_1)$ is in $\mathrm{U}_{k-1}^+$ . Then, by the item 2 for $k-1$ , there exists $\xi (x_1, x_2) \in \Pi _{k-1}$ such that

$$ \begin{align*}\mathsf{HA} + {\Sigma_{k-2}}\text{-}\mathrm{DNE} \vdash \xi(x_1, x_2) \leftrightarrow \left(\neg \rho_2(x_2) \to \neg \rho_1(x_1) \right).\end{align*} $$

Then $\mathsf {HA} + {\Sigma _{k}}\text {-}\mathrm{DNE}$ proves

$$ \begin{align*}\begin{array}{cl} & \varphi_1 \to \varphi_2 \\ \underset{\text{[I.H.] }{\Sigma_{k}}\text{-}\mathrm{DNE}}{\longleftrightarrow}& \varphi_1 \to \exists x_2 \rho_2(x_2)\\ \underset{{\Sigma_k}\text{-}\mathrm{DNE}}{\longleftrightarrow}& \varphi_1 \to \neg \neg \exists x_2 \rho_2(x_2)\\ \underset{\text{[I.H.] }{\Sigma_{k-1}}\text{-}\mathrm{DNE}}{\longleftrightarrow}& \forall x_1 \rho_1(x_1) \to \neg \neg \exists x_2 \rho_2(x_2)\\ \longleftrightarrow&\neg \neg \exists x_2 \left( \forall x_1 \rho_1(x_1) \to \rho_2(x_2) \right) \\ \underset{{\Pi_{k-1}}\text{-}\mathrm{DNE}}{\longleftrightarrow}&\neg \neg \exists x_2 \left(\neg \rho_2(x_2) \to \neg \forall x_1 \rho_1(x_1) \right) \\ \underset{{\Sigma_{k-1}}\text{-}\mathrm{DNE}}{\longleftrightarrow}&\neg \neg \exists x_2 \left(\neg \rho_2(x_2) \to \neg \neg \exists x_1 \neg \rho_1(x_1) \right) \\ \longleftrightarrow &\neg \neg \exists x_2 \neg \neg \exists x_1 \left(\neg \rho_2(x_2) \to \neg \rho_1(x_1) \right) \\ \underset{\text{[I.H.] }{\Sigma_{k-2}}\text{-}\mathrm{DNE}}{\longleftrightarrow}&\neg \neg \exists x_1, x_2 \, \xi (x_1, x_2)\\ \underset{ {\Sigma_k}\text{-}\mathrm{DNE}}{\longleftrightarrow}& \exists x_1, x_2 \, \xi (x_1, x_2) \in \Sigma_k .\\ \end{array} \end{align*} $$

Remark 5.8. It follows from Theorem 5.3 and Corollary 5.4 that ${\mathrm{E}_k}\text {-}\mathrm{LEM}$ , ${\mathrm{U}_k}\text {-}\mathrm{LEM}$ and ${\mathrm{U}_k}\text {-}\mathrm{DNE}$ are equivalent to ${\Sigma _k}\text {-}\mathrm{LEM}$ , ${\Pi _k}\text {-}\mathrm{LEM}$ and ${(\Pi _k \lor \Pi _k)}\text {-}\mathrm{DNE}$ respectively over $\mathsf {HA}$ (cf. [Reference Akama, Berardi, Hayashi and Kohlenbach1, Corollary 2.9]). This may not be the case for ${\mathrm{E}_k}\text {-}\mathrm{DNE}$ and ${\Sigma _k}\text {-}\mathrm{DNE}$ . On the other hand, Theorem 5.7 implies that $\mathsf {HA} + {\Sigma _k}\text {-}\mathrm{DNE}$ proves the double negation elimination for all formulas in $\mathrm{E}_k$ which do not contain $\lor .$

6 A conservation theorem

In this section, we generalize a well-known fact that $\mathsf {PA} $ is $\Pi _2$ -conservative over $\mathsf {HA}$ in the context of semi-classical arithmetic (see Theorem 6.14). The fact is normally shown by applying the negative translation followed by the Friedman A-translation (see e.g., [Reference Kohlenbach6, Chapter 14]). As for the negative translation, there are several equivalent forms (see [Reference Troelstra8, §1.10.1]). Here we employ Kuroda’s negative translation among them.

Definition 6.1 (cf. [Reference Kohlenbach6, Definition 10.1])

Let $\varphi $ be a $\mathsf {HA}$ -formula. Then its negative translation $\varphi ^N$ is defined as $\varphi ^N :\equiv \neg \neg {\varphi }_{N}$ , where ${\varphi }_{N}$ is defined inductively as follows:

  • ${\left ( \varphi _{\text {p}} \right )}_{N} :\equiv \varphi _{\text {p}}$ if $\varphi _{\text {p}}$ is a prime formula;

  • ${(\varphi _1 \circ \varphi _2)}_{N} :\equiv {\left ( \varphi _1 \right )}_{N} \circ {\left ( \varphi _2 \right )}_{N}$ , where $\circ \in \{ \land , \lor , \to \}$ ;

  • ${(\exists x \varphi _1)}_{N} :\equiv \exists x {\left ( \varphi _1 \right )}_{N}$ ;

  • ${(\forall x \varphi _1)}_{N} :\equiv \forall x \neg \neg {\left ( \varphi _1 \right )}_{N}$ .

Remark 6.2. By induction on the structure of formulas, one can show $\mathrm{FV} \left ({\varphi }\right ) = \mathrm{FV} \left ({{\varphi }_{N}}\right ) =\mathrm{FV} \left ({\varphi ^N}\right )$ for all formulas $\varphi $ . When it is clear from the context, we suppress the argument on free variables.

Lemma 6.3. For any $\mathsf {HA}$ -formula $\varphi $ in prenex normal form, $\mathsf {HA} $ proves $\varphi \to {\varphi }_{N}$ .

Proof By induction on the structure of formulas in prenex normal form.⊣

Proposition 6.4. For any $\mathsf {HA}$ -formula $\varphi $ , if $\mathsf {PA} \vdash \varphi $ , then $\mathsf {HA} \vdash \varphi ^N$ .

Proof By induction on the length of the derivations (see the proof of [Reference Kohlenbach6, Proposition 10.3]).⊣

Lemma 6.5. Let k be a natural number.

  1. 1. For any $\mathsf {HA}$ -formula $\varphi \in \Sigma _k$ , $\mathsf {HA} + {\Sigma _{k}}\text {-}\mathrm{DNE}$ proves $\varphi ^N \leftrightarrow \varphi $ .

  2. 2. For any $\mathsf {HA}$ -formula $\varphi \in \Pi _k$ , $\mathsf {HA} + {\Sigma _{k-1}}\text {-}\mathrm{DNE}\ (\mathsf {HA}$ if $k=0)$ proves $\varphi ^N \leftrightarrow \varphi $ .

Proof By simultaneous induction on k. The base case is trivial. For the induction step, assume the items 1 and 2 for k to show those for $k+1$ . For the first item, let $\exists x \varphi _1 \in \Sigma _{k+1}$ where $\varphi _1 \in \Pi _k$ . We have that $\mathsf {HA} + {\Sigma _{k+1}}\text {-}\mathrm{DNE}$ proves

$$ \begin{align*}\left( \exists x \varphi_1 \right)^N\equiv \neg \neg \exists x {\left( \varphi_1 \right)}_{N} \leftrightarrow \neg \neg \exists x \neg \neg {\left( \varphi_1 \right)}_{N} \underset{\text{[I.H.] }{\Sigma_{k-1}}\text{-}\mathrm{DNE}}{\longleftrightarrow} \neg \neg \exists x \varphi_1 \underset{{\Sigma_{k+1}}\text{-}\mathrm{DNE}}{\longleftrightarrow} \exists x \varphi_1. \end{align*} $$

For the second item, let $\forall x \varphi _1 \in \Pi _{k+1}$ where $\varphi _1 \in \Sigma _k$ . Since ${\Pi _{k+1}}\text {-}\mathrm{DNE}$ is derived from ${\Sigma _k}\text {-}\mathrm{DNE}$ (see Lemma 5.2.(4)), we have that $\mathsf {HA} + {\Sigma _k}\text {-}\mathrm{DNE}$ proves

$$ \begin{align*}\left( \forall x \varphi_1 \right)^N\equiv \neg \neg \forall x \neg \neg {\left( \varphi_1 \right)}_{N} \underset{\text{[I.H.] }{\Sigma_k}\text{-}\mathrm{DNE}}{\longleftrightarrow} \neg \neg \forall x \varphi_1 \underset{{\Pi_{k+1}}\text{-}\mathrm{DNE}}{\longleftrightarrow} \forall x \varphi_1. \end{align*} $$

Let $\mathsf {HA}^{*} $ denote $\mathsf {HA}$ in the extended language where a predicate symbol ${*} $ of arity $0$ , which behaves as a “place holder”, is added. In particular, $\mathsf {HA}^{*}$ has $\perp \to *$ as an axiom. To make our arguments absolutely clear, we prefer to add the distinguished new predicate ${*} $ rather than discussing about A-translation inside the original language as done in [Reference Friedman, Müller and Scott2Reference Kohlenbach6].

Definition 6.6 (A-translation [Reference Friedman, Müller and Scott2])

For a $\mathsf {HA}$ -formula $\varphi $ , we define $\varphi ^{*} $ as a formula obtained from $\varphi $ by replacing all the prime formulas $\varphi _{\text {p}}$ in $\varphi $ with $\varphi _{\text {p}} \lor {*} $ . Officially, $\varphi ^{*} $ is defined inductively as in Definition 6.1. In particular, $\perp ^{*} :\equiv \left (\perp \lor \, {*} \right )$ , which is equivalent to ${*} $ over $\mathsf {HA}^{*} $ . In what follows, $\neg _{*}\, \varphi $ denotes $\varphi \to {*} $ .

Remark 6.7. By induction on the structure of formulas, one can show $\mathrm{FV} \left ({\varphi }\right ) =\mathrm{FV} \left ({\varphi ^{*}}\right )$ for all $\mathsf {HA}$ -formulas $\varphi $ .

Proposition 6.8 (cf. [Reference Friedman, Müller and Scott2, Lemma 2])

For any $\mathsf {HA}$ -formula $\varphi $ , if $\mathsf {HA} \vdash \varphi $ , then $\mathsf {HA}^{*} \vdash \varphi ^{*} $ .

Proof By induction on the length of the derivations.⊣

Remark 6.9. An analogous assertion of Proposition 6.8 holds for $\mathsf {HA} +{\Sigma _1}\text {-}\mathrm{LEM}$ and $\mathsf {HA}^{*} +{\Sigma _1}\text {-}\mathrm{LEM}$ instead of $\mathsf {HA}$ and $\mathsf {HA}^{*}$ respectively (see [Reference Kohlenbach and Safarik7, Lemma 3.1]).

The following substitution result is important in the application of the A-translation:

Lemma 6.10 (cf. [Reference van Dalen10, Theorem 6.2.4])

Let X be a set of $\mathsf {HA}$ -sentences and $\varphi $ be a $\mathsf {HA}^{*} $ -formula. If $\mathsf {HA}^{*} + X \vdash \varphi $ , then $\mathsf {HA} + X \vdash \varphi [\psi /{*}]$ for any $\mathsf {HA}$ -formula $\psi $ such that the free variables of $\psi $ are not bounded in $\varphi $ , where $\varphi [\psi /{*}]$ is the $\mathsf {HA}$ -formula obtained from $\varphi $ by replacing all the occurrences of ${*}$ in $\varphi $ with $\psi $ .

Proof Fix a set X of $\mathsf {HA}$ -sentences. By induction on k, one can show straightforwardly that for any k and any $\mathsf {HA}^{*} $ -formula $\varphi $ , if $\mathsf {HA}^{*} + X \vdash \varphi $ with the proof of length k, then $\mathsf {HA} + X \vdash \varphi [\psi /{*} ]$ for any $\mathsf {HA}$ -formula $\psi $ such that the free variables of $\psi $ is not bounded in $\varphi $ . The variable condition is used to verify the case of the axioms and rules for quantifiers.⊣

The following lemma is a key for our generalized conservation result.

Lemma 6.11. Let k be a natural number.

  1. 1. For any $\mathsf {HA}$ -formula $\varphi \in \Sigma _k$ , $\mathsf {HA}^{*} + {\Sigma _{k-1}}\text {-}\mathrm{LEM}\ (\mathsf {HA}^{*} $ if $k=0)$ proves $\left ( {\varphi }_{N}\right )^{*} \leftrightarrow {\varphi }_{N} \lor {*} $ .

  2. 2. For any $\mathsf {HA}$ -formula $\varphi \in \Pi _k$ , $\mathsf {HA}^{*} + {\Sigma _{k}}\text {-}\mathrm{LEM}$ proves $\left ( {\varphi }_{N}\right )^{*} \leftrightarrow {\varphi }_{N} \lor {*} $ .

Proof We show the items 1 and 2 simultaneously by induction on k.

The base case: Since every quantifier-free formula ${\varphi }_{\text {qf}}$ such that $\mathrm{FV} \left ({{\varphi }_{\text {qf}}}\right )=\{\overline {x} \} $ is equivalent to a prime formula $t(\overline {x})=0$ for some closed term t (see e.g., [Reference Kohlenbach6, Proposition 3.8]), by Proposition 6.8, it suffices to show the assertions only for prime formulas. Since $\left ({\left ( \varphi _{\text {p}} \right )}_{N}\right )^{*} \equiv {\varphi _{\text {p}}}^{*} \equiv \varphi _{\text {p}} \lor {*} \equiv {\left ( \varphi _{\text {p}} \right )}_{N} \lor {*}$ , we are done.

The induction step: Assume that the items 1 and 2 hold for k. We first show the item 1 for $k+1$ . Let $\varphi _1\in \Pi _k$ . Since

$$ \begin{align*}\left({\left( \exists x \varphi_1 \right)}_{N}\right)^{*} \equiv \left( \exists x\, {\left( \varphi_1 \right)}_{N}\right)^{*} \equiv \exists x \left({\left( \varphi_1 \right)}_{N}\right)^{*},\end{align*} $$

by induction hypothesis, we have

$$ \begin{align*}\mathsf{HA}^{*} + {\Sigma_k}\text{-}\text{LEM} \vdash\left({\left( \exists x \varphi_1 \right)}_{N}\right)^{*} \leftrightarrow \exists x \left({\left( \varphi_1 \right)}_{N} \lor {*} \right).\end{align*} $$

Since $\mathsf {HA}^{*} $ proves $\exists x \left ({\left ( \varphi _1 \right )}_{N} \lor {*} \right ) \leftrightarrow \left (\exists x \, {\left ( \varphi _1 \right )}_{N} \lor {*} \right ) \equiv \left ( {\left ( \exists x \varphi _1 \right )}_{N} \lor {*} \right )$ , we have

$$ \begin{align*}\mathsf{HA}^{*} + {\Sigma_k}\text{-}\text{LEM} \vdash \left({\left( \exists x \varphi_1 \right)}_{N}\right)^{*} \leftrightarrow \left( {\left( \exists x \varphi_1 \right)}_{N} \lor {*} \right).\end{align*} $$

Thus we have shown the item 1 for $k+1$ .

Next, we show the item 2 for $k+1$ . Let $\varphi _2\in \Sigma _k$ . We shall show that $\mathsf {HA}^{*} + {\Sigma _k}\text {-}\mathrm{DNE}$ (and hence, $\mathsf {HA}^{*} + {\Sigma _{k+1}}\text {-}\mathrm{LEM}$ ) proves ${\left ( \forall x \varphi _2 \right )}_{N} \lor {*} \to \left ( {\left ( \forall x \varphi _2 \right )}_{N} \right )^{*} $ . By Lemma 6.5.(1), we have

(5) $$ \begin{align} \mathsf{HA}+{\Sigma_k}\text{-}\mathrm{DNE} \vdash \varphi_2 \leftrightarrow \left(\varphi_2 \right)^N \equiv \neg \neg {\left( \varphi_2 \right)}_{N}. \end{align} $$

Then we have that $\mathsf {HA}^{*} + {\Sigma _k}\text {-}\mathrm{DNE}$ proves

$$ \begin{align*}{\left( \forall x \varphi_2 \right)}_{N}\lor {*} \equiv \left( \forall x \neg \neg {\left( \varphi_2 \right)}_{N} \lor {*} \right) \leftrightarrow \forall x \varphi_2 \lor {*}. \end{align*} $$

By Lemma 6.3, $\mathsf {HA}$ proves $\varphi _2 \to {\left ( \varphi _2 \right )}_{N}$ . Then, using induction hypothesis and the fact that ${\Sigma _{k}}\text {-}\mathrm{DNE}$ derives ${\Sigma _{k-1}}\text {-}\mathrm{LEM}$ , we have that $\mathsf {HA}^{*} + {\Sigma _k}\text {-}\mathrm{DNE}$ proves

$$ \begin{align*}\begin{array}{rcl} {\left( \forall x \varphi_2 \right)}_{N} \lor {*} &\underset{{\Sigma_k}\text{-}\mathrm{DNE}}{\longleftrightarrow}& \forall x \varphi_2 \lor {*}\\ &\longrightarrow & \forall x {\left( \varphi_2 \right)}_{N} \lor {*}\\[2pt] & \longrightarrow & \forall x ({\left( \varphi_2 \right)}_{N} \lor {*}) \\[2pt] & \underset{\text{[I.H.] }{\Sigma_{k-1}}\text{-}\text{LEM}}{\longleftrightarrow} &\forall x \left({\left( \varphi_2 \right)}_{N} \right)^{*}\\ & \longrightarrow & \forall x \left(\left( \left({\left( \varphi_2 \right)}_{N} \right)^{*} \to {*} \right) \to {*} \right)\\ &\longleftrightarrow & \left( {\left( \forall x \varphi_2 \right)}_{N}\right)^{*}. \end{array} \end{align*} $$

In the following, we show the converse direction:

(6) $$ \begin{align} \mathsf{HA}^{*} + {\Sigma_{k+1}}\text{-}\text{LEM} \vdash \left( {\left( \forall x \varphi_2 \right)}_{N} \right)^{*} \to {\left( \forall x \varphi_2 \right)}_{N} \lor {*}. \end{align} $$

Reason in $\mathsf {HA}^{*} + {\Sigma _{k+1}}\text {-}\mathrm{LEM} $ . Suppose $\left ( {\left ( \forall x \varphi _2 \right )}_{N} \right )^{*} $ , equivalently,

(7) $$ \begin{align} \forall x \left(\left( \left({\left( \varphi_2 \right)}_{N} \right)^{*} \to {*} \right) \to {*} \right). \end{align} $$

By induction hypothesis, (7) is equivalent to $\forall x \left (\left ({\left ( \varphi _2 \right )}_{N} \lor {*} \to {*} \right ) \to {*} \right )$ , which is intuitionistically equivalent to

$$ \begin{align*} \forall x \left(\left({\left( \varphi_2 \right)}_{N} \to {*} \right)\to {*} \right). \end{align*} $$

Then we have

(8) $$ \begin{align} \exists x \neg {\left( \varphi_2 \right)}_{N} \to {*}. \end{align} $$

By Lemma 4.8.(2), there exists $\psi _2 \in \Pi _k$ such that $\mathrm{FV} \left ({\varphi _2}\right )=\mathrm{FV} \left ({\psi _2}\right )$ and $\neg \varphi _2$ is equivalent to $\psi _2$ . Since $\exists x \psi _2\in \Sigma _{k+1}$ , by ${\Sigma _{k+1}}\text {-}\mathrm{LEM}$ , we have $\exists x \psi _2 \lor \neg \exists x \psi _2$ , and hence,

$$ \begin{align*} \exists x \neg \varphi_2 \lor \forall x \neg \neg \varphi_2. \end{align*} $$

Then, by (5), we obtain

$$ \begin{align*} \exists x \neg {\left( \varphi_2 \right)}_{N} \lor \forall x \neg \neg {\left( \varphi_2 \right)}_{N}. \end{align*} $$

In the former case, we have ${*}$ by (8). In the latter case, we have ${\left ( \forall x {\varphi _2} \right )}_{N}$ . Thus we have shown (6).⊣

Lemma 6.12. Let $\varphi $ be a $\mathsf {HA}^{*} $ -formula.

  1. 1. $\mathsf {HA}^{*} \vdash \varphi \to \neg _{*} \neg _{*} \varphi $ .

  2. 2. $\mathsf {HA}^{*} \vdash \forall x \neg _{*} \varphi \leftrightarrow \neg _{*} \exists x \varphi $ .

  3. 3. $\mathsf {HA}^{*} \vdash \neg _{*} \neg _{*} \neg _{*} \varphi \to \neg _{*} \varphi $ .

  4. 4. $\mathsf {HA}^{*} \vdash \exists x \neg _{*} \neg _{*} \varphi \to \neg _{*} \neg _{*} \exists x \varphi $ .

Proof (1)–(3) are immediate from the definition of $\neg _{*} $ (see Definition 6.6). (4) follows from (1)–(3).⊣

Lemma 6.13. For any $\mathsf {HA}$ -formula $\varphi $ in prenex normal form, $\mathsf {HA}^{*} \vdash \varphi \to \left (\varphi ^N \right )^{*} $ .

Proof Since there exists a closed term t such that $\mathsf {HA} \vdash {\varphi }_{\text {qf}}(x_1, \dots , x_k) \leftrightarrow t(x_1, \dots , x_k) =0 $ for each quantifier-free formula ${\varphi }_{\text {qf}}$ such that $\mathrm{FV} \left ({{\varphi }_{\text {qf}}}\right )=\{x_1, \dots , x_k \}$ (see e.g., [Reference Kohlenbach6, Proposition 3.8]), by Proposition 6.4 and Proposition 6.8, one can assume that formulas in prenex normal form consist of the formulas of form $Q_{1}x_1 \dots Q_{k}x_k \, \varphi _{\text {p}} $ where $Q_i$ s are quantifiers and $\varphi _{\text {p}}$ is prime. We show our assertion by induction on the structure of formulas of this form.

For a prime formula $\varphi _{\text {p}}$ , it is trivial to see that $\mathsf {HA}^{*} $ proves

$$ \begin{align*}\varphi_{\text{p}} \to \varphi_{\text{p}} \lor {*} \to \neg_{*} \neg_{*} \left( \varphi_{\text{p}} \lor {*} \right) \leftrightarrow \left(\left(\varphi_{\text{p}}\right)^N \right)^{*}. \end{align*} $$

Assume the assertion for $\varphi $ . Then, using Lemma 6.12, $\mathsf {HA}^{*}$ proves

$$ \begin{align*}\begin{array}{r} \exists x \varphi \underset{\text{[I.H.]}}{\longrightarrow} \exists x \left(\varphi^N \right)^{*} \equiv \exists x \left(\neg \neg {\varphi}_{N} \right)^{*} \leftrightarrow \exists x \neg_{*} \neg_{*} \left({\varphi}_{N} \right)^{*} \to \neg_{*} \neg_{*} \exists x \left({\varphi}_{N} \right)^{*} \\ \leftrightarrow \left(\left( \exists x \varphi \right)^N \right)^{*} \end{array} \end{align*} $$

and

$$ \begin{align*}\begin{array}{r} \forall x \varphi \underset{\text{[I.H.]}}{\longrightarrow} \forall x \left(\varphi^N \right)^{*} \equiv \forall x \left(\neg \neg {\varphi}_{N} \right)^{*} \leftrightarrow \forall x \neg_{*} \neg_{*} \left({\varphi}_{N} \right)^{*} \to \neg_{*} \neg_{*} \forall x \neg_{*} \neg_{*} \left({\varphi}_{N} \right)^{*} \\ \leftrightarrow \left(\left( \forall x \varphi \right)^N \right)^{*}. \end{array} \end{align*} $$

Theorem 6.14. Let k be a natural number. For any $\varphi \in \Pi _{k+2}$ and any $\psi $ in prenex normal form, if $\mathsf {PA} \vdash \psi \to \varphi $ , then $\mathsf {HA} + {\Sigma _k}\text {-}\mathrm{LEM} \vdash \psi \to \varphi $ .

Proof Let $\varphi :\equiv \forall x \exists y \varphi _1$ where $\varphi _1\in \Pi _k$ . Since one can freely replace the bound variables, assume that the free variables of $ \exists y \varphi _1 $ are not bounded in $\psi $ and x does not occur in $\psi $ without loss of generality.

Suppose $\mathsf {PA} \vdash \psi \to \forall x \exists y \varphi _1$ . By Proposition 6.4, we have that $\mathsf {HA}$ proves $\neg \neg ({\psi }_{N} \to \forall x \neg \neg \exists y\, {\left ( \varphi _1 \right )}_{N})$ , which is intuitionistically equivalent to $\neg \neg {\psi }_{N} \to \forall x \neg \neg \exists y\, {\left ( \varphi _1 \right )}_{N}$ , namely, $\psi ^N \to \forall x \neg \neg \exists y \, {\left ( \varphi _1 \right )}_{N}$ . Then we have

$$ \begin{align*} \mathsf{HA} \vdash \psi^N \to \neg \neg \exists y \, {\left( \varphi_1 \right)}_{N}. \end{align*} $$

By Proposition 6.8, we have

$$ \begin{align*} \mathsf{HA}^{*} \vdash \left(\psi^N\right)^{*} \to \neg_{*} \neg_{*} \exists y \left( {\left( \varphi_1 \right)}_{N}\right)^{*} , \end{align*} $$

and hence,

$$ \begin{align*} \mathsf{HA}^{*} \vdash \psi \to \neg_{*} \neg_{*} \exists y \left( {\left( \varphi_1 \right)}_{N}\right)^{*} \end{align*} $$

by Lemma 6.13. Then, by Lemma 6.11.(2), we have that $\mathsf {HA}^{*} +{\Sigma _k}\text {-}\mathrm{LEM} $ proves

$$ \begin{align*} \psi \to \neg_{*} \neg_{*} \exists y \left( {\left( \varphi_1 \right)}_{N} \lor {*} \right), \end{align*} $$

which is intuitionistically equivalent to

$$ \begin{align*} \psi \to \neg_{*} \neg_{*} \exists y \, {\left( \varphi_1 \right)}_{N}. \end{align*} $$

Since the free variables of $ \exists y \varphi _1 $ are not bounded in $\psi $ , using Lemma 6.10 with Remark 6.2, we have

(9) $$ \begin{align} \mathsf{HA} +{\Sigma_k}\text{-}\text{LEM} \vdash \psi \to \left( (\exists y \, {\left( \varphi_1 \right)}_{N} \to \exists y \varphi_1) \to \exists y \varphi_1\right). \end{align} $$

On the other hand, by Lemma 6.5.(2) and the fact that $ {\Sigma _k}\text {-}\mathrm{LEM}$ derives ${\Sigma _k}\text {-}\mathrm{DNE}$ , we have that $\mathsf {HA} +{\Sigma _k}\text {-}\mathrm{LEM}$ proves

$$ \begin{align*} {\left( \varphi_1 \right)}_{N} \to \neg \neg {\left( \varphi_1 \right)}_{N} \equiv \left(\varphi_1\right)^N \underset{ {\Sigma_{k-1}}\text{-}\mathrm{DNE}}{\longleftrightarrow} \varphi_1. \end{align*} $$

and hence, $\exists y \, {\left ( \varphi _1 \right )}_{N} \to \exists y \varphi _1$ . Then, by (9), we have $\mathsf {HA} +{\Sigma _k}\text {-}\mathrm{LEM} \vdash \psi \to \exists y \varphi _1$ . By our assumption, x does not occur in $\psi $ , and hence, $\mathsf {HA} +{\Sigma _k}\text {-}\mathrm{LEM} \vdash \psi \to \forall x \exists y \varphi _1 $ follows.⊣

Corollary 6.15. $\mathsf {HA} + {\Sigma _k}\text {-}\mathrm{LEM}$ is closed under the $\Sigma _{k+1}$ -generalization of Markov’s rule: If $\mathsf {HA} + {\Sigma _k}\text {-}\mathrm{LEM}$ proves $ \neg \neg \varphi $ where $\varphi \in \Sigma _{k+1}$ , then $\mathsf {HA} + {\Sigma _k}\text {-}\mathrm{LEM}$ proves $ \varphi $ .

Corollary 6.15 is announced in [Reference Hayashi, Nakata, Callaghan, Luo, McKinna and Pollack5, §4.4] without proof and the proof for $k=1$ with using the soundness of the A-translation for $\mathsf {HA} +{\Sigma _1}\text {-}\mathrm{LEM}$ (cf. Remark 6.9) can be found in [Reference Kohlenbach and Safarik7, Proposition 3.2]. In fact, by using the latter, Kohlenbach and Safarik essentially show an instance of Theorem 6.14 for $k=1$ and $\psi \equiv 0=0$ in [Reference Kohlenbach and Safarik7, Proposition 3.3].

In this paper, we have shown Theorem 6.14 in order to prove the optimality of our prenex normal form theorems in §5 (see §7). On the other hand, the conservation result on semi-classical arithmetic itself is interesting. This will be studied comprehensively in [Reference Fujiwara and Kurahashi4].

7 Characterizations

Notation 2. Let ${T} $ be an extension of $\mathsf {HA}$ . Let $\Gamma $ and $\Gamma '$ be classes of $\mathsf {HA}$ -formulas. Then $\mathrm{PNFT}_{T} \left ( {\Gamma }, {\Gamma '} \right ) $ denotes the following statement: for any $\varphi \in \Gamma $ , there exists $\varphi ' \in \Gamma '$ such that $\mathrm{FV} \left ({\varphi }\right )=\mathrm{FV} \left ({\varphi '}\right )$ and ${T} \vdash \varphi \leftrightarrow \varphi ' $ .

Under this notation, Theorem 5.3 asserts (modulo Lemma 2.3) that for a semi-classical theory T containing $\mathsf {HA} + {(\Pi _k\lor \Pi _k)}\text {-}\mathrm{DNE} $ , $\mathrm{PNFT}_{T} \left ( {\mathrm{U}_{k'}}, {\Pi _{k'}} \right ) $ holds for all $k' \leq k$ as well as the analogous assertion for $\mathrm{E}_k$ and $\Sigma _k$ . It is natural to ask whether the verification theories are optimal. In this section, among other things (see Table 1), we show that this is exactly the case:

  1. 1. For a theory ${T}$ in-between $\mathsf {HA} $ and $\mathsf {PA}$ , ${T} \vdash {(\Pi _k\lor \Pi _k)}\text {-}\mathrm{DNE}$ if and only if $\mathrm{PNFT}_{T} \left ( {\mathrm{U}_{k'}}, {\Pi _{k'}} \right ) $ for all $k' \leq k$ . (Theorem 7.3)

  2. 2. For a theory ${T}$ in-between $\mathsf {HA} +{\Pi _{k-1}}\text {-}\mathrm{LEM} $ ( $\mathsf {HA}$ if $k=0$ ) and $\mathsf {PA}$ , ${T} \vdash {\Sigma _k}\text {-}\mathrm{DNE} + {\mathrm{U}_k}\text {-}\mathrm{DNS} $ if and only if $\mathrm{PNFT}_{T} \left ( {\mathrm{E}_{k'}}, {\Sigma _{k'}} \right ) $ for all $k' \leq k$ . (Theorem 7.11)

Table 1 Characterizations of the prenex normal form theorems

Lemma 7.1. Let ${T} $ be a theory in-between $\mathsf {HA} +{\Sigma _{k-2}}\text {-}\mathrm{LEM}\, (\mathsf {HA}$ if $k<2)$ and $\mathsf {PA}$ . If $\mathrm{PNFT}_{T} \left ( {\mathrm{U}_{k}}, {\Pi _{k}} \right ) $ , then ${T} \vdash {(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE}$ .

Proof Fix an instance of ${(\Pi _{k}\lor \Pi _{k})}\text {-}\mathrm{DNE}$

$$ \begin{align*}\varphi :\equiv \forall x \left(\neg\neg \left(\varphi_1(x) \lor \varphi_2(x) \right) \to \varphi_1(x) \lor \varphi_2(x) \right), \end{align*} $$

where $\varphi _1(x), \varphi _2(x) \in \Pi _{k}(x)$ . Since $\neg \neg \left (\varphi _1(x) \lor \varphi _2(x) \right ) $ and $ \varphi _1(x) \lor \varphi _2(x)$ are in $\mathrm{U}_{k}$ , by our assumption, there exist $\rho (x)$ and $\rho '(x)$ in $\Pi _{k}(x)$ such that ${T} $ proves $ \rho (x) \leftrightarrow \varphi _1(x) \lor \varphi _2(x) $ and $\rho '(x) \leftrightarrow \neg \neg \left (\varphi _1(x) \lor \varphi _2(x)\right ) $ . Since $\mathsf {PA} \vdash \varphi $ and $\mathsf {PA}$ is an extension of ${T}$ , we have $\mathsf {PA} \vdash \rho '(x) \to \rho (x) $ . By Theorem 6.14, we have that $\mathsf {HA} +{\Sigma _{k-2}}\text {-}\mathrm{LEM}$ proves $\rho '(x) \to \rho (x)$ , and hence, $\forall x\left ( \rho '(x) \to \rho (x) \right )$ . Since ${T}$ is an extension of $\mathsf {HA} + {\Sigma _{k-2}}\text {-}\mathrm{LEM}$ , we have ${T} \vdash \forall x\left (\rho '(x) \to \rho (x) \right )$ , and hence, ${T} \vdash \varphi $ .⊣

Lemma 7.2. Let ${T} $ be a theory in-between $\mathsf {HA} $ and $\mathsf {PA}$ . If $\mathrm{PNFT}_{T} \left ( {\mathrm{U}_{k'}}, {\Pi _{k'}} \right ) $ for all $k' \leq k$ , then ${T} \vdash {\Sigma _{k-1}}\text {-}\mathrm{LEM}$ .

Proof By induction on k. The base case is trivial. For the induction step, assume $\mathrm{PNFT}_{T} \left ( {\mathrm{U}_{k'}}, {\Pi _{k'}} \right ) $ for all $k' \leq k+1$ . Then, by induction hypothesis, we have ${T} \vdash {\Sigma _{k-1}}\text {-}\mathrm{LEM}$ . Fix an instance of ${\Sigma _{k}}\text {-}\mathrm{LEM}$

$$ \begin{align*}\varphi :\equiv \forall x (\varphi_1(x) \lor \neg \varphi_1(x)), \end{align*} $$

where $\varphi _1(x) \in \Sigma _{k}(x)$ . Since $\varphi \in \mathrm{U}_{k+1}$ , by our assumption, there exists a sentence $\varphi ' \in \Pi _{k+1}$ such that ${T} \vdash \varphi \leftrightarrow \varphi '$ . Since $\mathsf {PA} \vdash \varphi $ , we have $\mathsf {PA} \vdash \varphi '$ . Then, by Theorem 6.14, we have $\mathsf {HA} + {\Sigma _{k-1}}\text {-}\mathrm{LEM} \vdash \varphi '$ , and hence, ${T} \vdash \varphi '$ . Thus we have ${T} \vdash \varphi $ .⊣

Theorem 7.3. Let ${T} $ be a theory in-between $\mathsf {HA} $ and $\mathsf {PA}$ . Then ${T} \vdash {(\Pi _k\lor \Pi _k)}\text {-}\mathrm{DNE}$ if and only if $\mathrm{PNFT}_{T} \left ( {\mathrm{U}_{k'}}, {\Pi _{k'}} \right ) $ for all $k' \leq k$ .

Proof The “only if” direction is immediate from Theorem 5.3.(2). We show the converse direction. Assume $\mathrm{PNFT}_{T} \left ( {\mathrm{U}_{k'}}, {\Pi _{k'}} \right ) $ for all $k' \leq k$ . Let $k>0$ without loss of generality. By Lemma 7.2, ${T} \vdash {\Sigma _{k-1}}\text {-}\mathrm{LEM}$ . Then, by Lemma 7.1, we have ${T} \vdash {(\Pi _k\lor \Pi _k)}\text {-}\mathrm{DNE} $ .⊣

Definition 7.4. Let $\Gamma $ be a class of formulas. Then ${\Gamma }^{\mathrm{dn}}$ denotes the class of $\mathsf {HA}$ -formulas $\neg \neg \varphi $ where $\varphi \in \Gamma $ , and ${\Gamma }^{\mathrm{n}}$ denotes that for $\neg \varphi $ where $\varphi \in \Gamma $ .

Lemma 7.5. Let ${T} $ be a theory in-between $\mathsf {HA} $ and $\mathsf {PA}$ . If $\mathrm{PNFT}_{T} \left ( {{\mathrm{U}_{k'}}^{\mathrm{dn}}}, {{\Pi _{k'}}^{\mathrm{dn}}} \right ) $ for all $k' \leq k$ , then ${T} \vdash \neg \neg {{\Sigma _{k-1}}\text {-}\mathrm{LEM}}$ .

Proof By induction on k. The base case is trivial. For the induction step, assume $\mathrm{PNFT}_{T} \left ( {{\mathrm{U}_{k'}}^{\mathrm{dn}}}, {{\Pi _{k'}}^{\mathrm{dn}}} \right ) $ for all $k' \leq k+1$ . Then, by induction hypothesis, ${T}$ proves $\neg \neg {{\Sigma _{k-1}}\text {-}\mathrm{LEM}}$ . Fix an instance of $\neg \neg {{\Sigma _{k}}\text {-}\mathrm{LEM}}$

$$ \begin{align*} \varphi :\equiv \neg \neg \forall x (\varphi_1(x) \lor \neg \varphi_1(x)), \end{align*} $$

where $\varphi _1(x) \in \Sigma _{k}(x)$ . Since $\varphi \in {\mathrm{U}_{k+1}}^{\mathrm{dn}}$ , by our assumption, there exists a sentence $\varphi ' \in \Pi _{k+1}$ such that ${T} \vdash \varphi \leftrightarrow \neg \neg \varphi '$ . Since $\mathsf {PA} \vdash \forall x (\varphi _1(x) \lor \neg \varphi _1(x))$ , we have $\mathsf {PA} \vdash \varphi '$ . Then, by Theorem 6.14, we have $\mathsf {HA} + {\Sigma _{k-1}}\text {-}\mathrm{LEM} \vdash \varphi '$ , and hence, $\mathsf {HA} + \neg \neg {{\Sigma _{k-1}}\text {-}\mathrm{LEM}} \vdash \neg \neg \varphi '$ by Lemma 4.1. Then ${T} \vdash \varphi $ .⊣

Theorem 7.6. Let ${T} $ be a theory in-between $\mathsf {HA} $ and $\mathsf {PA}$ . The following are pairwise equivalent:

  1. 1. $\mathrm{PNFT}_{T} \left ( {{\mathrm{E}_{k'}}^{\mathrm{n}}}, { {\Pi _{k'}}^{\mathrm{dn}}} \right ) $ for all $k' \leq k$ ;

  2. 2. $\mathrm{PNFT}_{T} \left ( {{\mathrm{U}_{k'}}^{\mathrm{dn}}}, { {\Pi _{k'}}^{\mathrm{dn}}} \right ) $ for all $k' \leq k$ ;

  3. 3. ${T} \vdash {\mathrm{U}_k}\text {-}\mathrm{DNS}$ .

Proof The equivalence of (1) and (2) is trivial (cf. the proof of Lemma 5.1). In addition, (3 $\to $ 2) is immediate from Lemma 5.1 and Proposition 4.6. In what follows, we show (2 $\to $ 3). Assume $\mathrm{PNFT}_{T} \left ( {{\mathrm{U}_{k'}}^{\mathrm{dn}}}, {{\Pi _{k'}}^{\mathrm{dn}}} \right ) $ for all $k' \leq k$ . Since ${\mathrm{U}_{k}}\text {-}\mathrm{DNS}$ is intuitionistically equivalent to $\neg \neg {{\mathrm{U}_{k}}\text {-}\mathrm{DNS}}$ (See Remark 2.8), it suffices to show ${T} \vdash \neg \neg {{\mathrm{U}_{k}}\text {-}\mathrm{DNS}}$ . Let $k>0$ without loss of generality. Fix an instance of $\neg \neg {{\mathrm{U}_k}\text {-}\mathrm{DNS}}$

$$ \begin{align*} \varphi :\equiv \neg \neg \forall x \left( \forall y \neg \neg \varphi_1(x,y) \to \neg \neg \forall y \varphi_1(x,y) \right), \end{align*} $$

where $\varphi _1(x,y)\in \mathrm{U}_k(x,y)$ . Since $i(s) \,{\equiv}\ \kern1pt{-}\,$ for all alternation paths s of $\forall y\neg \neg \varphi _1(x,y)$ and $\forall y \varphi _1(x,y)$ , it is straightforward to show that $\forall y\neg \neg \varphi _1(x,y)$ and $\forall y\varphi _1(x,y)$ are in $\mathrm{U}_{k}(x)$ . Then, by $\mathrm{PNFT}_{T} \left ( {{\mathrm{U}_{k}}^{\mathrm{dn}}}, { {\Pi _{k}}^{\mathrm{dn}}} \right ) $ , there exist $\rho (x), \rho '(x)\in \Pi _{k}(x)$ such that ${T} $ proves $\neg \neg \rho (x) \leftrightarrow \neg \neg \forall y \varphi _1(x,y) $ and $\neg \neg \rho '(x) \leftrightarrow \neg \neg \forall y \neg \neg \varphi _1(x,y) $ . Since $\mathsf {PA}$ is an extension of ${T}$ and $\mathsf {PA} \vdash \varphi $ , we have $\mathsf {PA} \vdash \rho '(x) \to \rho (x)$ . Then, by Theorem 6.14, we have that $\mathsf {HA} + {\Sigma _{k-2}}\text {-}\mathrm{LEM} \, (\mathsf {HA}$ if $k<2)$ proves $\rho '(x) \to \rho (x)$ , and hence, $\forall x \left ( \neg \neg \rho '(x) \to \neg \neg \rho (x) \right )$ . By Lemma 4.1, we have

$$ \begin{align*} \mathsf{HA} + \neg \neg {{\Sigma_{k-2}}\text{-}\text{LEM}} \vdash \neg \neg \forall x \left( \neg \neg \rho'(x) \to \neg \neg \rho(x) \right). \end{align*} $$

On the other hand, by Lemma 7.5 and our assumption, we have ${T} \vdash \neg \neg {{\Sigma _{k-1}}\text {-}\mathrm{LEM}}$ . Then we have

$$ \begin{align*}{T} \vdash \neg \neg \forall x \left( \neg \neg \forall y \neg \neg \varphi_1(x,y) \to \neg \neg \forall y \varphi_1(x,y) \right), \end{align*} $$

and hence, ${T} \vdash \varphi $ .⊣

Remark 7.7. Theorem 7.6 shows that the verification theory for Lemma 5.1 is optimal.

Definition 7.8. Let $\Gamma $ be a class of $\mathsf {HA}$ -formulas. ${\Gamma }^{\mathrm{df}}$ denotes the class of formulas in $\Gamma $ which do not contain $\lor $ .

Lemma 7.9. Let ${T} $ be a theory in-between $\mathsf {HA} +{\Pi _{k-1}}\text {-}\mathrm{LEM}\ (\mathsf {HA}$ if $k=0)$ and $\mathsf {PA}$ . If $\mathrm{PNFT}_{T} \left ( {{\mathrm{E}_{k'}}^{\mathrm{df}}}, {\Sigma _{k'}} \right ) $ for all $k' \leq k$ , then ${T} \vdash {\Sigma _{k}}\text {-}\mathrm{DNE}$ .

Proof By induction on k. The base case is trivial. For the induction step, assume the assertion for k and let ${T}$ be a theory in-between $\mathsf {HA} +{\Pi _{k}}\text {-}\mathrm{LEM}$ and $\mathsf {PA}$ . Assume also that $\mathrm{PNFT}_{T} \left ( {{\mathrm{E}_{k'}}^{\mathrm{df}}}, {\Sigma _{k'}} \right ) $ holds for all $k' \leq k+1$ . Then, by induction hypothesis, ${T}$ proves ${\Sigma _{k}}\text {-}\mathrm{DNE}$ . Since ${T}$ contains $\mathsf {HA} +{\Pi _{k}}\text {-}\mathrm{LEM},$ we have ${T} \vdash {\Sigma _{k}}\text {-}\mathrm{LEM}$ by [Reference Akama, Berardi, Hayashi and Kohlenbach1, Theorem 3.1(ii)]. Fix an instance of ${\Sigma _{k+1}}\text {-}\mathrm{DNE}$

$$ \begin{align*} \varphi :\equiv \forall x ( \neg \neg \varphi_1(x) \to \varphi_1(x)), \end{align*} $$

where $\varphi _1(x) \in \Sigma _{k+1}(x)$ . Without loss of generality, one can assume that $\varphi _1(x)$ does not contain $\lor $ (cf. [Reference Kohlenbach6, Proposition 3.8]). Since $\neg \neg \varphi _1(x)\in {\mathrm{E}_{k+1}}^{\mathrm{df}}$ , By $\mathrm{PNFT}_{T} \left ( {{\mathrm{E}_{k+1}}^{\mathrm{df}}}, {\Sigma _{k+1}} \right ) $ , there exists $\varphi _1'(x) \in \Sigma _{k+1}(x)$ such that ${T} \vdash \neg \neg \varphi _1(x) \leftrightarrow \varphi _1' (x)$ . Since $\mathsf {PA}$ is an extension of ${T}$ and $\mathsf {PA} \vdash \varphi $ , we have $\mathsf {PA} \vdash \varphi _1' (x) \to \varphi _1(x)$ . Then, by Theorem 6.14, we have that $\mathsf {HA} + {\Sigma _{k}}\text {-}\mathrm{LEM} $ proves $ \varphi _1' (x) \to \varphi _1(x)$ , and hence, $\forall x \left ( \varphi _1' (x) \to \varphi _1(x) \right )$ . Since ${T}$ is an extension of $\mathsf {HA} + {\Sigma _{k}}\text {-}\mathrm{LEM}$ , we have ${T} \vdash \varphi $ .⊣

Lemma 7.10. Let ${T} $ be an extension of $\mathsf {HA} + \neg \neg {{\Sigma _{k-1}}\text {-}\mathrm{DNE}}\ (\mathsf {HA}$ if $k=0)$ . If $\mathrm{PNFT}_{T} \left ( {\mathrm{E}_{k'}}, {\Sigma _{k'}} \right ) $ for all $k' \leq k$ , then $\mathrm{PNFT}_{T} \left ( {{\mathrm{E}_{k'}}^{\mathrm{n}}}, {{\Pi _{k'}}^{\mathrm{dn}}} \right ) $ for all $k' \leq k$ .

Proof Assume $\mathrm{PNFT}_{T} \left ( {\mathrm{E}_{k'}}, {\Sigma _{k'}} \right ) $ for all $k' \leq k$ . Fix $k' \leq k$ and $\varphi \in \mathrm{E}_{k'}$ . By $\mathrm{PNFT}_{T} \left ( {\mathrm{E}_{k'}}, {\Sigma _{k'}} \right ) $ , there exists $\varphi '\in \Sigma _{k'}$ such that $\mathrm{FV} \left ({\varphi }\right ) = \mathrm{FV} \left ({\varphi '}\right )$ and ${T} \vdash \varphi \leftrightarrow \varphi '$ . Then

$$ \begin{align*} {T} \vdash \neg \varphi \leftrightarrow \neg \varphi'. \end{align*} $$

On the other hand, by Lemma 4.8.(2), there exists $\varphi '' \in \Pi _{k'}$ such that $\mathrm{FV} \left ({\varphi '}\right ) = \mathrm{FV} \left ({\varphi ''}\right )$ and $\mathsf {HA} + {\Sigma _{k'-1}}\text {-}\mathrm{DNE} \vdash \neg \varphi ' \leftrightarrow \varphi ''$ . Then, by Corollary 4.2, we have

$$ \begin{align*} \mathsf{HA} + \neg \neg {{\Sigma_{k'-1}}\text{-}\mathrm{DNE}} \vdash \neg \varphi' \leftrightarrow \neg \neg \varphi''. \end{align*} $$

Then $\mathrm{FV} \left ({\neg \varphi }\right ) =\mathrm{FV} \left ({\neg \neg \varphi ''}\right )$ and ${T} \vdash \neg \varphi \leftrightarrow \neg \neg \varphi ''$ . Thus we have shown $\mathrm{PNFT}_{T} \left ( {{\mathrm{E}_{k'}}^{\mathrm{n}}}, {{\Sigma _{k'}}^{\mathrm{dn}}} \right ) $ .⊣

Theorem 7.11. Let ${T} $ be a theory in-between $\mathsf {HA} + {\Pi _{k-1}}\text {-}\mathrm{LEM}\ (\mathsf {HA}$ if $k=0)$ and $\mathsf {PA}$ . Then ${T} \vdash {\Sigma _k}\text {-}\mathrm{DNE} + {\mathrm{U}_k}\text {-}\mathrm{DNS}$ if and only if $\mathrm{PNFT}_{T} \left ( {\mathrm{E}_{k'}}, {\Sigma _{k'}} \right ) $ for all $k' \leq k$ .

Proof The “only if” direction is immediate from Theorem 5.3.(1). We show the converse direction. Assume $\mathrm{PNFT}_{T} \left ( {\mathrm{E}_{k'}}, {\Sigma _{k'}} \right ) $ for all $k' \leq k$ . Let $k>0$ without loss of generality. By Lemma 7.9, we have ${T} \vdash {\Sigma _{k}}\text {-}\mathrm{DNE}$ . Then, by Lemma 7.10 and Theorem 7.6, we have ${T} \vdash {\mathrm{U}_k}\text {-}\mathrm{DNS}$ .⊣

Remark 7.12. It is still open whether the assumption that ${T}$ contains ${\Pi _{k-1}}\text {-}\mathrm{LEM}$ can be omitted in Theorem 7.11.

Remark 7.13. Akama et al. [Reference Akama, Berardi, Hayashi and Kohlenbach1] shows that ${\Pi _k}\text {-}\mathrm{LEM}$ does not derive ${\Sigma _k}\text {-}\mathrm{DNE} $ and ${\Sigma _k}\text {-}\mathrm{DNE}$ does not derive ${(\Pi _k \lor \Pi _k)}\text {-}\mathrm{DNE}$ . Theorem 7.11 reveals that the prenex normal form theorem for $\mathrm{E}_k$ and $\Sigma _k$ does not hold in $\mathsf {HA} + {\Pi _k}\text {-}\mathrm{LEM}$ , and Theorem 7.3 reveals that the prenex normal form theorem for $\mathrm{U}_k$ and $\Pi _k$ does not hold in $\mathsf {HA} + {\Sigma _k}\text {-}\mathrm{DNE}$ .

Corollary 7.14. Let ${T} $ be a theory in-between $\mathsf {HA} $ and $\mathsf {PA}$ . Then ${T} \vdash {\Sigma _k}\text {-}\mathrm{DNE} + {\left ( \Pi _k \lor \Pi _k \right )}\text {-}\mathrm{DNE}$ if and only if $\mathrm{PNFT}_{T} \left ( {\mathrm{E}_{k'}}, {\Sigma _{k'}} \right ) $ and $\mathrm{PNFT}_{T} \left ( {\mathrm{U}_{k'}}, {\Pi _{k'}} \right ) $ for all $k' \leq k$ .

Proof Let ${T} $ be a theory in-between $\mathsf {HA} $ and $\mathsf {PA}$ . The “only if” direction follows from Theorem 5.3 and Corollary 5.4.

For the converse direction, assume that $\mathrm{PNFT}_{T} \left ( {\mathrm{E}_{k'}}, {\Sigma _{k'}} \right ) $ and $\mathrm{PNFT}_{T} \left ( {\mathrm{U}_{k'}}, {\Pi _{k'}} \right ) $ hold for all $k' \leq k$ . By Theorems 7.3, we have ${T} \vdash {\left ( \Pi _k \lor \Pi _k \right )}\text {-}\mathrm{DNE}$ . Since ${\Pi _{k-1}}\text {-}\mathrm{LEM}$ is derived from $ {\left ( \Pi _k \lor \Pi _k \right )}\text {-}\mathrm{DNE}$ (cf. [Reference Akama, Berardi, Hayashi and Kohlenbach1, Theorem 3.1(1)]), by Theorem 7.11, we also have ${T} \vdash {\Sigma _k}\text {-}\mathrm{DNE}$ .⊣

In the following, we show the optimality of Theorem 5.7 (see Theorem 7.16).

Lemma 7.15. Let ${T} $ be a theory in-between $\mathsf {HA} +{\Pi _{k-2}}\text {-}\mathrm{LEM}\ (\mathsf {HA}$ if $k<2)$ and $\mathsf {PA}$ . If $\mathrm{PNFT}_{T} \left ( {{\left ({\mathrm{U}_{k'}}^{\mathrm{dn}}\right )}^{\mathrm{df}}}, {\Pi _{k'}} \right ) $ for all $k' \leq k$ , then ${T} \vdash {\Sigma _{k-1}}\text {-}\mathrm{DNE}$ .

Proof By induction on k. The base case is trivial. For the induction step, assume the assertion for k and let ${T}$ be a theory in-between $\mathsf {HA} +{\Pi _{k-1}}\text {-}\mathrm{LEM}$ and $\mathsf {PA}$ . Assume also that $\mathrm{PNFT}_{T} \left ( {{\left ({\mathrm{U}_{k'}}^{\mathrm{dn}}\right )}^{\mathrm{df}}}, {\Pi _{k'}} \right ) $ holds for all $k' \leq k+1$ . Then, by induction hypothesis, ${T}$ proves ${\Sigma _{k-1}}\text {-}\mathrm{DNE}$ . Since ${T}$ contains $\mathsf {HA} +{\Pi _{k-1}}\text {-}\mathrm{LEM},$ we have ${T} \vdash {\Sigma _{k-1}}\text {-}\mathrm{LEM}$ by [Reference Akama, Berardi, Hayashi and Kohlenbach1, Theorem 3.1(ii)]. Fix an instance of ${\Sigma _{k}}\text {-}\mathrm{DNE}$

$$ \begin{align*} \varphi :\equiv \forall x ( \neg \neg \varphi_1(x) \to \varphi_1(x)), \end{align*} $$

where $\varphi _1(x) \in \Sigma _{k} (x)$ . Without loss of generality, one can assume that $\varphi _1(x)$ does not contain $\lor $ (cf. [Reference Kohlenbach6, Proposition 3.8]). From the perspective of Remark 2.5, $\varphi _1(x)$ is in $\Pi _{k+1} (x)$ . Then, by $\mathrm{PNFT}_{T} \left ( {{\left ({\mathrm{U}_{k+1}}^{\mathrm{dn}}\right )}^{\mathrm{df}}}, {\Pi _{k+1}} \right ) $ , there exists $\varphi _1'(x)\in \Pi _{k}(x)$ such that ${T} \vdash \neg \neg \varphi _1(x) \leftrightarrow \varphi _1' (x)$ . Since $\mathsf {PA}$ is an extension of ${T}$ and $\mathsf {PA} \vdash \varphi $ , we have $\mathsf {PA} \vdash \varphi _1' (x) \to \varphi _1(x)$ . By Theorem 6.14, we have that $\mathsf {HA} + {\Sigma _{k-1}}\text {-}\mathrm{LEM}$ proves $\varphi _1' (x) \to \varphi _1(x)$ , and hence, $\forall x \left ( \varphi _1' (x) \to \varphi _1(x) \right )$ . Since ${T}$ is an extension of $\mathsf {HA} + {\Sigma _{k-1}}\text {-}\mathrm{LEM}$ , we have $T \vdash \varphi $ .⊣

Theorem 7.16.

  1. 1. Let ${T} $ be a theory in-between $\mathsf {HA} + {\Pi _{k-1}}\text {-}\mathrm{LEM}\ (\mathsf {HA}$ if $k=0)$ and $\mathsf {PA}$ . Then ${T} \vdash {\Sigma _k}\text {-}\mathrm{DNE}$ if and only if $\mathrm{PNFT}_{T} ( {{\mathrm{E}_{k'}}^{\mathrm{df}}}, {\Sigma _{k'}}) $ for all $k' \leq k$ .

  2. 2. Let ${T} $ be a theory in-between $\mathsf {HA} + {\Pi _{k-2}}\text {-}\mathrm{LEM}\ (\mathsf {HA}$ if $k<2)$ and $\mathsf {PA}$ . Then ${T} \vdash {\Sigma _{k-1}}\text {-}\mathrm{DNE}$ if and only if $\mathrm{PNFT}_{T} ( {{\mathrm{U}_{k'}}^{\mathrm{df}}}, {\Pi _{k'}} ) $ for all $k' \leq k$ .

Proof (1): The “only if” direction is by Theorem 5.7.(1). The converse direction is by Lemma 7.9.

(2): The “only if” direction is by Theorem 5.7.(2). Note that any formula in ${\left ({\mathrm{U}_{k'}}^{\mathrm{dn}}\right )}^{\mathrm{df}}$ is in ${\mathrm{U}_{k'}}^{\mathrm{df}}$ . Then the converse direction follows from Lemma 7.15.⊣

At the end of this section, we characterize some variants of prenex normal form theorems.

Theorem 7.17. Let ${T} $ be a theory in-between $\mathsf {HA} + {\Pi _{k-2}}\text {-}\mathrm{LEM}\ (\mathsf {HA}$ if $k<2)$ and $\mathsf {PA}$ . Then ${T} \vdash {\Sigma _{k-1}}\text {-}\mathrm{DNE} +{\mathrm{U}_k}\text {-}\mathrm{DNS}$ if and only if $\mathrm{PNFT}_{T} \left ( {{\mathrm{U}_{k'}}^{\mathrm{dn}}}, {\Pi _{k'}} \right ) $ for all $k' \leq k$ .

Proof We first show the “only if” direction. Let $k>0$ without loss of generality. Assume ${T} \vdash {\Sigma _{k-1}}\text {-}\mathrm{DNE} +{\mathrm{U}_k}\text {-}\mathrm{DNS}$ and fix $k' \leq k$ . Since ${T} \vdash {\mathrm{U}_{k'}}\text {-}\mathrm{DNS}$ (cf. Proposition 4.6), by Lemma 5.1, for any $\varphi \in \mathrm{U}_{k'}$ , there exists $\varphi '\in \Pi _{k'}$ such that $\mathrm{FV} \left ({\varphi }\right ) = \mathrm{FV} \left ({\varphi '}\right )$ and

$$ \begin{align*}{T} \vdash \neg \neg \varphi \leftrightarrow \neg \neg \varphi'. \end{align*} $$

Since ${T} \vdash {\Sigma _{k-1}}\text {-}\mathrm{DNE}$ , by Lemma 5.2.(4), we have $T \vdash \neg \neg \varphi ' \leftrightarrow \varphi '$ , and hence, $T \vdash \neg \neg \varphi \leftrightarrow \varphi '$ . Thus we have $\mathrm{PNFT}_{T} \left ( {{\mathrm{U}_{k'}}^{\mathrm{dn}}}, {\Pi _{k'}} \right ) $ .

Next, we show the converse direction. Assume that $\mathrm{PNFT}_{T} \left ( {{\mathrm{U}_{k'}}^{\mathrm{dn}}}, {\Pi _{k'}} \right ) $ holds for all $k' \leq k$ . By Lemma 7.15, we have ${T} \vdash {\Sigma _{k-1}}\text {-}\mathrm{DNE}$ . Then, by the assumption, we have $\mathrm{PNFT}_{T} \left ( {{\mathrm{U}_{k'}}^{\mathrm{dn}}}, {{\Pi _{k'}}^{\mathrm{dn}}} \right ) $ for all $k' \leq k$ , and hence, ${T} \vdash {\mathrm{U}_{k}}\text {-}\mathrm{DNS}$ by Theorem 7.6.⊣

Theorem 7.18. Let ${T} $ be a theory in-between $\mathsf {HA} + \neg \neg {\Pi _{k-2}}\text {-}\mathrm{LEM}\ (\mathsf {HA}$ if $k<2)$ and $\mathsf {PA}$ . The following are pairwise equivalent:

  1. 1. $\mathrm{PNFT}_{T} \left ( {{\left ({\mathrm{E}_{k'}}^{\mathrm{n}}\right )}^{\mathrm{df}}}, { {\Pi _{k'}}^{\mathrm{dn}}} \right ) $ for all $k' \leq k$ ;

  2. 2. $\mathrm{PNFT}_{T} \left ( {{\left ({\mathrm{U}_{k'}}^{\mathrm{dn}}\right )}^{\mathrm{df}}}, { {\Pi _{k'}}^{\mathrm{dn}}} \right ) $ for all $k' \leq k$ ; and

  3. 3. ${T} \vdash \neg \neg {\Sigma _{k-1}}\text {-}\mathrm{DNE}$ .

Proof The equivalence of (1) and (2) is trivial (cf. the proof of Lemma 5.1). In addition, (3 $\to $ 1) is immediate from the item 3 in the proof of Theorem 5.7. Then it suffices to show (1 $\to $ 3). We show this by induction on k. The base case is trivial. For the induction step, assume the assertion for k and let ${T}$ be a theory in-between $\mathsf {HA} + \neg \neg {\Pi _{k-1}}\text {-}\mathrm{LEM}$ and $\mathsf {PA}$ . Assume also that $\mathrm{PNFT}_{T} \left ( {{\left ({\mathrm{E}_{k'}}^{\mathrm{n}}\right )}^{\mathrm{df}}}, { {\Pi _{k'}}^{\mathrm{dn}}} \right ) $ holds for all $k' \leq k+1$ . Then, by induction hypothesis, we have ${T} \vdash \neg \neg {\Sigma _{k-1}}\text {-}\mathrm{DNE}$ . Since ${T}$ contains $\mathsf {HA} +\neg \neg {\Pi _{k-1}}\text {-}\mathrm{LEM},$ we have ${T} \vdash \neg \neg {\Sigma _{k-1}}\text {-}\mathrm{LEM}$ by [Reference Akama, Berardi, Hayashi and Kohlenbach1, Theorem 3.1(ii)]. Fix an instance of $\neg \neg {\Sigma _{k}}\text {-}\mathrm{DNE}$

$$ \begin{align*}\varphi :\equiv \neg \neg \forall x ( \neg \neg \varphi_1(x) \to \varphi_1(x)), \end{align*} $$

where $\varphi _1(x) \in \Sigma _{k} (x)$ . Without loss of generality, one can assume that $\varphi _1(x)$ does not contain $\lor $ (cf. [Reference Kohlenbach6, Proposition 3.8]). Note $\forall x ( \neg \neg \varphi _1(x) \to \varphi _1(x)) \in {\mathrm{U}_{k+1}}^{\mathrm{df}}$ , and hence, $\varphi \in {\left ({\mathrm{E}_{k+1}}^{\mathrm{n}}\right )}^{\mathrm{df}}$ . By $\mathrm{PNFT}_{T} \left ( {{\left ({\mathrm{E}_{k+1}}^{\mathrm{n}}\right )}^{\mathrm{df}}}, { {\Pi _{k+1}}^{\mathrm{dn}}} \right ) $ , there exists a sentence $\varphi '\in \Pi _{k+1}$ such that ${T} \vdash \varphi \leftrightarrow \neg \neg \varphi ' $ . Since $\mathsf {PA}$ is an extension of ${T}$ and $\mathsf {PA} \vdash \varphi $ , we have $\mathsf {PA} \vdash \varphi '$ . By Theorem 6.14, we have $ \mathsf {HA} + {\Sigma _{k-1}}\text {-}\mathrm{LEM} \vdash \varphi '$ . Then, by Lemma 4.1, we have

$$ \begin{align*} \mathsf{HA} + \neg \neg {\Sigma_{k-1}}\text{-}\text{LEM} \vdash \neg \neg \varphi'. \end{align*} $$

Since ${T} $ is an extension of $\mathsf {HA} + \neg \neg {\Sigma _{k-1}}\text {-}\mathrm{LEM}$ , we have ${T} \vdash \varphi $ .⊣

All of our characterization results are of the following form: For any theory ${T} $ in-between $\mathsf {HA} + \text {Q}_k$ and $\mathsf {PA}$ , ${T} \vdash \mathrm{P}_k$ if and only if $\mathrm{PNFT}_{T} \left ( {\Gamma _{k'}}, {\Delta _{k'}} \right ) $ holds for all $k' \leq k$ , where $\text { P}_k, \text {Q}_k$ are logical principles and $\Gamma _{k'}, \Delta _{k'}$ are classes of formulas. Based on this representation, our results are summarized in Table 1.

Acknowledgements

The authors thank to Ulrich Kohlenbach for his helpful comments. The first author was supported by JSPS KAKENHI Grant Numbers JP18K13450, JP19J01239 and JP20K14354, and the second author by JP19K14586.

References

REFERENCES

Akama, Y., Berardi, S., Hayashi, S., and Kohlenbach, U., An arithmetical hierarchy of the law of excluded middle and related principles, Proceedings of the 19th Annual IEEE Symposium on Logic in Computer Science (LICS’04), Turku, Finland, IEEE, 2004, pp. 192–201.10.1109/LICS.2004.1319613CrossRefGoogle Scholar
Friedman, H., Classically and intuitionistically provably recursive functions, Higher Set Theory (Müller, G. H. and Scott, D. S., editors), Springer, Berlin/Heidelberg, 1978, pp. 2127.10.1007/BFb0103100CrossRefGoogle Scholar
Fujiwara, M. and Kohlenbach, U., Interrelation between weak fragments of double negation shift and related principles, this Journal, vol. 83 (2018), no. 3, pp. 9911012.Google Scholar
Fujiwara, M. and Kurahashi, T., Conservation theorems on semi-classical arithmetic, preprint, 2020.Google Scholar
Hayashi, S. and Nakata, M., Towards limit computable mathematics, Types for Proofs and Programs (Callaghan, P., Luo, Z., McKinna, J., and Pollack, R., editors), R. Pollack Edn, Springer, Berlin/Heidelberg, 2002, pp. 125144.10.1007/3-540-45842-5_9CrossRefGoogle Scholar
Kohlenbach, U., Applied Proof Theory: Proof Interpretations and their Use in Mathematics, Springer Monographs in Mathematics, Springer-Verlag, Berlin, Germany, 2008.Google Scholar
Kohlenbach, U. and Safarik, P., Fluctuations, effective learnability and metastability in analysis. Annals of Pure and Applied Logic, vol. 165 (2014), no. 1, pp. 266304, The Constructive in Logic and Applications (A. Nerode and M. Fitting, editors).10.1016/j.apal.2013.07.014CrossRefGoogle Scholar
Troelstra, A. S. (editor), Metamathematical Investigation of Intuitionistic Arithmetic and Analysis , Lecture Notes in Mathematics, vol. 344, Springer-Verlag, Berlin, New York, 1973.10.1007/BFb0066739CrossRefGoogle Scholar
Troelstra, A. S. and van Dalen, D., Constructivism in Mathematics, An Introduction, vol. 1, Studies in Logic and the Foundations of Mathematics, vol. 121, North Holland, Amsterdam, 1988.Google Scholar
van Dalen, D., Logic and Structure , fifth Edn., Universitext, Springer-Verlag, London, 2013.10.1007/978-1-4471-4558-5CrossRefGoogle Scholar
Figure 0

Table 1 Characterizations of the prenex normal form theorems