axiomatic derivations

PHIL 452 · Modal Logic

Gabriel Uzquiano

uniform substitution

Given a formula \(\varphi\) with propositional variables \(p_1, \dots, p_n\) and formulas \(\psi_1, \dots, \psi_n\), we define

\[ \varphi[\psi_1/p_1, \dots, \psi_n/p_n] \]

by the recursion:

\[ \begin{array}{lll} p_i[\psi_1/p_1, \dots, \psi_n/p_n] & := & \psi_i \\ (\neg \chi) [\psi_1/p_1, \dots, \psi_n/p_n] & := & \neg \chi[\psi_1/p_1, \dots, \psi_n/p_n] \\ (\chi \to \rho)[\psi_1/p_1, \dots, \psi_n/p_n] & := & (\chi[\psi_1/p_1, \dots, \psi_n/p_n] \to \rho[\psi_1/p_1, \dots, \psi_n/p_n]) \end{array} \]

how to use uniform substitution

\[ \begin{array}{lll} p[\neg q] & := & \neg q \\ \neg (p \to \neg q)[\neg p/p, (p \to q)/q] & := & \neg (\neg p \to \neg (p \to q)) \\ (p \to (q \to p))[(q \to p)/p, (p \to q)/q] & := & ((q \to p) \to ((p \to q) \to (q \to p))) \\ \end{array} \]

Unpack (p → (q → ¬p))[¬q/p, ¬p/q]:

axioms

Axioms are substitution instances of one of the three formulas below:

\[\begin{array}{lll} \textsf{A1} & & p \to (q \to p)\\ \textsf{A2} & & (p \to (q \to r)) \to ((p \to q) \to (p \to r))\\ \textsf{A3} & & (\neg p \to \neg q) \to ((\neg p \to q) \to p) \end{array}\]

Which formulas below are axioms?

rule of inference

There is one rule of inference:

\[\begin{array}{lll} \textsf{Modus Ponens} & & \varphi, \ (\varphi \to \psi)/\psi\\ \end{array}\]

  • \(p \to (q \to p)\)
  • \(p\)
  • \((q \to p)\)
  • \((p \to q) \to (q \to p)\)
  • \((p \to q)\)
  • \((q \to p)\)

derivability

A formula \(\varphi\) is derivable from a set of well-formed formulas \(\Gamma\), written \(\Gamma \vdash \varphi\), iff, there is a finite sequence of well-formed formulas \((\chi_1, \dots , \chi_n)\) such that

  • \(\chi_n = \varphi\),

and for each \(m \leq n\), either

  • \(\chi_m\) is an axiom, or

  • \(\chi_m \in \Gamma\), or

  • \(\chi_k = (\chi_l \to \chi_m)\), for some \(k, l <m\).

Such a finite sequence \(( \chi_1, ..., \chi_n )\) is called a derivation (or a proof) of \(\varphi\) from \(\Gamma\).

theorem

A formula \(\varphi\) is a theorem, written \(\vdash \varphi\), iff there is a derivation of \(\varphi\) from the empty set.

\(\vdash p \to p\)

\[\begin{array}{lll} 1 & p \to ((p \to p) \to p) & \textsf{A1}[p/p, (p \to p)/q]\\ 2 & (p \to ((p \to p) \to p))\to & \\ & ((p \to (p \to p)) \to (p \to p)) & \textsf{A2}[p/p, (p \to p)/q, p/r]\\ 3 & (p \to (p \to p)) \to (p \to p) & \textsf{MP} \ 1, 2\\ 4 & p \to (p \to p) & \textsf{A1}[p/p, p/q]\\ 5 & p \to p & \textsf{MP} \ 3, 4\\ \end{array}\]

The argument generalizes: \(\vdash \varphi \to \varphi\)

We now develop a library of subroutines we can use in the course of more complex proofs.

the deduction theorem

For all sets of formulas \(\Gamma\) and formulas \(\varphi\), and \(\psi\), \[ \begin{array}{lll} \Gamma, \varphi \vdash \psi & \Leftrightarrow & \Gamma \vdash \varphi \to \psi\\ \end{array} \]

Notation: \[ \begin{array}{lll} \Gamma, \varphi & := & \Gamma \cup \{\varphi\} \\ \varphi \vdash \psi & := & \{\varphi\} \vdash \psi \\ \end{array} \]

the simple direction

\[\begin{array}{lll} \Gamma \vdash \varphi \to \psi & \Rightarrow & \Gamma, \varphi \vdash \psi\\ \end{array}\]
  • Assume \(\Gamma \vdash \varphi \to \psi\). We argue that there is a derivation of \(\psi\) from \(\Gamma, \varphi\).
  • There is, by assumption, a derivation of \(\psi\) from \(\Gamma\) of some finite length \(n\), e.g. \[\begin{array}{lll} 1 & \cdots & \cdots \\ \vdots & \vdots & \vdots \\ n & \varphi \to \psi & \cdots \\ \end{array}\]

the key step

  • We expand a derivation of \(\psi\) from \(\Gamma\) with two lines: \[\begin{array}{lll} 1 & \cdots & \cdots \\ \vdots & \vdots & \vdots \\ n & \varphi \to \psi & \cdots \\ n + 1 & \varphi & \textsf{premise} \\ n + 2 & \psi & \textsf{MP} \ n, n+1 \\ \end{array}\]

the crucial direction

\[\begin{array}{lll} \Gamma, \varphi \vdash \psi & \Rightarrow & \Gamma \vdash \varphi \to \psi\\ \end{array}\]
  • Use a complete induction on the length of derivations:

Given a condition \(\Phi(n)\) on positive integers, if

if for all \(m < n\), \(m\) satisfies the condition, then \(n\) does,

then all positive integers satisfy the condition.

Set the condition \(\Phi(n)\) on length as:

For all sets of formulas \(\Gamma\), and formulas \(\chi_1\), … \(\chi_n\), \(\varphi\), and \(\psi\),

if \(\langle \chi_1, \cdots \chi_n\rangle\) is a derivation of \(\psi\) from \(\Gamma, \varphi\) of length \(n\), then \(\Gamma \vdash \varphi \to \psi\).

the induction

We argue that for each \(n\), if for all \(m < n\), \(m\) satisfies the condition, then \(n\) does.

  • Assume that for all \(m < n\), \(m\) satisfies the condition. That is, for all \(m < n\):

For all sets of formulas \(\Gamma\) and formulas \(\chi_1\), … \(\chi_m\), \(\varphi\), and \(\psi\),

if \(\langle \chi_1, \cdots \chi_m\rangle\) is a derivation of \(\psi\) from \(\Gamma, \varphi\) of length \(m\), then \(\Gamma \vdash \varphi \to \psi\).

  • We now prove:

For all sets of formulas \(\Gamma\) and formulas \(\chi_1\), … \(\chi_n\), \(\varphi\), and \(\psi\),

if \(\langle \chi_1, \cdots \chi_n\rangle\) is a derivation of \(\psi\) from \(\Gamma, \varphi\) of length \(n\), then \(\Gamma \vdash \varphi \to \psi\).

distinguish four cases

Consider a derivation of \(\psi\) from \(\Gamma, \varphi\) of the form: \[( \chi_1, ..., \chi_{n}).\] Thus \(\chi_{n}= \psi\), and for each \(1 \leq i \leq n\), \(\chi_{i}\) is either:

  1. an axiom, or
  2. a member of \(\Gamma\), or
  3. \(\varphi\), or
  4. the outcome of an application \(\textsf{MP}\) on two earlier lines.

two cases

  • Case 1. \(\psi\) is an axiom. Then \(\Gamma \vdash \varphi \to \psi\):

\[\begin{array}{lll} 1 & \psi \to (\varphi \to \psi) & \textsf{A1}[\psi/p, \varphi/q]\\ 2 & \psi & \textsf{axiom}\\ 3 & \varphi \to \psi & \textsf{MP} \ 1, 2\\ \end{array}\]


  • Case 2. \(\psi\) is a member of \(\Gamma\). Then \(\Gamma \vdash \varphi \to \psi\):

\[\begin{array}{lll} 1 & \psi \to (\varphi \to \psi) & \textsf{A1}[\psi/p, \varphi/q]\\ 2 & \psi & \textsf{premise}\\ 3 & \varphi \to \psi & \textsf{MP} \ 1, 2\\ \end{array}\]

two more cases

  • Case 3. \(\psi\) is \(\varphi\). Then \(\vdash \varphi \to \psi\).
    This is because \(\varphi \to \psi\) is just \(\varphi \to \varphi\) and \(\vdash \varphi \to \varphi\). So, \(\Gamma \vdash \varphi \to \varphi\)

  • Case 4. \(\psi\) is the output of an application of \(\textsf{MP}\) to two earlier lines \(\chi_l\) and \(\chi_k\). That means that \(\chi_k\) is \((\chi_l \to \psi)\) for some \(k, l < n\). Because both \(k, l < n\), there are derivations of \(\chi_l\) and \(\chi_l \to \psi\), respectively, of length less than \(n\).

    By inductive hypothesis:
    \(\Gamma \vdash \varphi \to \chi_l\)
    \(\Gamma \vdash \varphi \to (\chi_l \to \psi)\).

a proof of \(\varphi \to \psi\) from \(\Gamma\)

Combine the derivations of \(\varphi \to \chi_l\) and \(\varphi \to (\chi_l \to \psi)\) from \(\Gamma\) into a derivation of \(\varphi \to \psi\) from \(\Gamma\):

\[ \begin{array}{lll} 1 & \cdots & \cdots\\ \vdots & \vdots & \vdots\\ k & \varphi \to (\chi_l \to \psi) & \Gamma \vdash \varphi \to (\chi_l \to \psi) \\ \vdots & \vdots & \vdots \\ m & \varphi \to \chi_l & \Gamma \vdash \varphi \to \chi_l\\ %\vdots & \vdots & \vdots \\ m+1 & (\varphi \to (\chi_1 \to \psi)) \to & \textsf{A2}[\varphi/p,\chi_l/q,\psi/r]\\ & ((\varphi \to \chi_l)\to (\varphi \to \psi)) & \\ n+2 & (\varphi \to \chi_l)\to (\varphi \to \psi) & \textsf{MP} \ k, n\\ n+3 & \varphi \to \psi & \textsf{MP} \ m, n+1\\ \end{array} \]

applications of the deduction theorem

To prove a conditional, we adjoin the antecedent to our assumptions and set out to derive the consequent from them. If we can, we are entitled to the conditional on the basis of the original assumptions.

transitivity

\(\vdash (\varphi \to \psi) \to ((\psi \to \chi)\to (\varphi \to \chi))\)

Proof. Three applications of the Deduction Theorem deliver: \[\{\varphi \to \psi \} \vdash (\psi \to \chi) \to (\varphi \to \chi).\] \[\{ \varphi \to \psi, \psi \to \chi\} \vdash \varphi \to \chi,\] \[\{\varphi \to \psi, \psi \to \chi, \varphi\} \vdash \chi.\] Here is a derivation of \(\chi\) from \(\{\varphi \to \psi, \psi \to \chi, \varphi\}\): \[ \begin{array}{lll} 1 & \varphi \to \psi & \textsf{premise} \\ 2 & \psi \to \chi & \textsf{premise}\\ 3 & \varphi & \textsf{premise} \\ 4 & \psi & \textsf{MP} \ 1, 3\\ 5 & \chi & \textsf{MP} \ 2, 4\\ \end{array} \]

reverse contraposition

\(\vdash (\neg \psi \to \neg \varphi) \to (\varphi \to \psi)\)

Proof. Two applications of the Deduction Theorem deliver: \[\neg \psi \to \neg \varphi \vdash \varphi \to \psi \] \[\{\neg \psi \to \neg \varphi, \varphi\} \vdash \psi\] Here is a derivation of \(\psi\) from \(\{\neg \psi \to \neg \varphi, \varphi\}\):

\[ \begin{array}{lll} 1 & \neg \psi \to \neg \varphi & \\ 2 & \varphi \to (\neg \psi \to \varphi) & \textsf{A1}[\varphi/p,\neg \psi/q]\\ 3 & \varphi & \textsf{premise} \\ 4 & \neg \psi \to \varphi & \textsf{MP} \ 2, 3\\ 5 & (\neg \psi \to \neg \varphi) \to ((\neg \psi \to \varphi) \to \psi) & \textsf{A3}[\psi/p, \varphi/q]\\ 6 & (\neg \psi \to \varphi) \to \psi & \textsf{MP} \ 1, 5\\ 7 & \psi & \textsf{MP} \ 4, 6 \end{array} \]

double negation elimination

\(\vdash \neg \neg \varphi \to \varphi\)

Proof. Given the Deduction Theorem, it suffices to prove: \[\neg \neg \varphi \vdash \varphi\] Here is a derivation of \(\varphi\) from \(\neg \neg \varphi\):

\[\begin{array}{lll} 1 & \neg \neg \varphi \to (\neg \varphi \to \neg \neg \varphi) & \textsf{A1}[\neg\neg\varphi/p, \neg \varphi/q]\\ 2 & \neg \neg \varphi & \textsf{premise} \\ 3 & \neg \varphi \to \neg \neg \varphi & \textsf{MP} \ 1, 2 \\ 4 & (\neg \varphi \to \neg \neg \varphi) \to ((\neg \varphi \to \neg \varphi) \to \varphi) & \textsf{A3}[\varphi/p, \neg \varphi/q]\\ 5 & (\neg \varphi \to \neg \varphi) \to \varphi & \textsf{MP} \ 3, 4\\ 6 & \neg \varphi \to \neg \varphi & \textsf{theorem} \ (4.1)\\ 7 & \varphi & \textsf{MP} \ 5, 6 \end{array}\]

double negation introduction

\(\vdash \varphi \to \neg \neg \varphi\)

Proof. Combine two observations above:

\[\begin{array}{lll} 1 & \neg \neg \neg \varphi \to \neg \varphi & \textsf{theorem} \ (4.4)\\ 2 & (\neg \neg \neg \varphi \to \neg \varphi) \to (\varphi \to \neg \neg \varphi) & \textsf{theorem} \ (4.3) \\ 3 & \varphi \to \neg \neg \varphi & \textsf{MP} \ 1, 2 \end{array}\]

contraposition

\(\vdash (\varphi \to \psi) \to (\neg \psi \to \neg \varphi)\)

Proof. Given the Deduction Theorem, it suffices to prove: \[\varphi \to \psi \vdash \neg \psi \to \neg \varphi\]

\[ \begin{array}{llll} 1 & & \varphi \to \psi & \textsf{premise} \\ 2 & & \psi \to \neg \neg \psi & \textsf{theorem} \ (4.5)\\ 3 & & \neg \neg \varphi \to \varphi & \textsf{theorem} \ (4.4)\\ 4 & & (\neg \neg \varphi \to \varphi) \to ((\varphi \to \psi) \to (\neg \neg \varphi \to \psi)) & \textsf{theorem} \ (4.2) \\ 5 & & (\varphi \to \psi) \to (\neg \neg \varphi \to \psi) & \textsf{MP} \ 3, 4 \\ 6 & & \neg \neg \varphi \to \psi & \textsf{MP} \ 1, 4 \\ 7 & & (\neg \neg \varphi \to \psi) \to ((\psi \to \neg \neg \psi) \to (\neg \neg \varphi \to \neg \neg \psi)) & \textsf{theorem} \ (4.2) \\ 8 & & (\psi \to \neg \neg \psi) \to (\neg \neg \varphi \to \neg \neg \psi) & \textsf{MP} \ 6, 7\\ 9 & & \neg \neg \varphi \to \neg \neg \psi & \textsf{MP} \ 2, 8\\ 10 & & (\neg \neg \varphi \to \neg \neg \psi) \to (\neg \psi \to \neg \varphi) & \textsf{theorem} \ (4.3)\\ 11 & & \neg \psi \to \neg \varphi & \textsf{MP} \ 9, 10 \end{array} \]

exercise

Use the Deduction Theorem to justify:

\[ (\varphi \to \neg \varphi) \to \neg \varphi \]

Proof. Given the Deduction Theorem, we target \(\{\varphi \to \neg \varphi\} \vdash \neg \varphi\).

\[\begin{array}{lll} 1 & \varphi \to \neg \varphi & \textsf{premise} \\ 2 & (\neg \neg \varphi \to \neg \varphi) \to ((\neg \neg \varphi \to \varphi) \to \neg \neg \neg \varphi) & \textsf{A3} \\ 3 & (\varphi \to \neg \varphi) \to (\neg \neg \varphi \to \neg \varphi) & \textsf{theorem} \ (4.6)\\ 4 & \neg \neg \varphi \to \neg \varphi & \textsf{MP} \ 1, 3\\ 5 & (\neg \neg \varphi \to \varphi) \to \neg \neg \neg \varphi & \textsf{MP} \ 2, 4\\ 6 & \neg \neg \varphi \to \varphi & \textsf{theorem} \ (4.4)\\ 7 & \neg \neg \neg \varphi & \textsf{MP} \ 5, 6\\ 8 & \neg \neg \neg \varphi \to \neg \varphi & \textsf{theorem} \ (4.4)\\ 9 & \neg \varphi & \textsf{MP} \ 7, 8\\ \end{array}\]

ex falso quodlibet

\(\vdash \neg \varphi \to (\varphi \to \psi)\)

Proof. Two applications of the Deduction Theorem deliver: \[\neg \varphi \vdash \varphi \to \psi\] \[\{\neg \varphi, \varphi\} \vdash \psi\] Here is a derivation of \(\psi\) from \(\{\neg \varphi, \varphi\}\): \[\begin{array}{lll} 1 & \neg \varphi \to (\neg \psi \to \neg \varphi) & \textsf{A1}[\neg \varphi/p, \neg \psi/q] \\ 2 & \neg \varphi & \textsf{premise} \\ 3 & \neg \psi \to \neg \varphi & \textsf{MP} \ 1, 2\\ 4 & (\neg \psi \to \neg \varphi) \to (\varphi \to \psi) & \textsf{theorem} \ (4.3)\\ 5 & \varphi \to \psi & \textsf{MP} \ 3, 4\\ 6 & \varphi & \textsf{premise} \\ 7 & \psi & \textsf{MP} \ 5, 6\\ \end{array}\]