ITP-Course/lectures/04_hol_logic.tex

\part{HOL's Logic}

\frame[plain]{\partpage}


\section{HOL Logic}

\begin{frame}
\frametitle{HOL Logic}
\begin{itemize}
\item the HOL theorem prover uses a version of classical \textbf{h}igher \textbf{o}rder \textbf{l}ogic:\\
classical higher order predicate calculus with \\
terms from the typed lambda calculus (\ie simple type theory)
\item this sounds complicated, but is intuitive for SML programmers
\item (S)ML and HOL logic designed to fit each other
\item if you understand SML, you understand HOL logic

\bigskip
\begin{center}
\emph{HOL = functional programming + logic}
\end{center}
\bigskip

\begin{alertblock}{Ambiguity Warning}
The acronym \textit{HOL} refers to both the \textit{HOL interactive theorem prover} and the \textit{HOL logic} used by it. It's also a common abbreviation for \textit{higher order logic} in general.
\end{alertblock}
\end{itemize}
\end{frame}

\begin{frame}
\frametitle{Types}
\begin{itemize}
\item SML datatype for types
\begin{itemize}
\item \emph{Type Variables} ($\texttt{'a},\ \alpha,\ \texttt{'b},\ \beta,\ \ldots$)\\
Type variables are implicitly universally quantified. Theorems containing type variables
hold for all instantiations of these. Proofs using type variables can be seen as proof schemata.
\item \emph{Atomic Types} ($\texttt{c}$)\\
Atomic types denote fixed types. Examples: \texttt{num}, \texttt{bool}, \texttt{unit}
\item \emph{Compound Types} ($(\sigma_1, \ldots, \sigma_n) \textit{op}$)\\
\textit{op} is a \alert{type operator} of arity \textit{n} and $\sigma_1, \ldots, \sigma_n$ \alert{argument types}. Type operators denote operations for constructing types.\\
Examples: \texttt{num list} or \texttt{'a \# 'b}.
\item \emph{Function Types} ($\sigma_1 \to \sigma_2$)\\
$\sigma_1 \to \sigma_2$ is the type of \alert{total} functions from $\sigma_1$ to $\sigma_2$. 
\end{itemize}
\item types are never empty in HOL, \ie\\
for each type at least one value exists
\item all HOL functions are total
\end{itemize}
\end{frame}


\begin{frame}
\frametitle{Terms}
\begin{itemize}
\item SML datatype for terms
\begin{itemize}
\item \emph{Variables} ($\texttt{x}, \texttt{y}, \ldots$)
\item \emph{Constants} ($\texttt{c}, \ldots$)
\item \emph{Function Application} ($\texttt{f a}$)
\item \emph{Lambda Abstraction} ($\texttt{\textbackslash x.\ f x}$\ \ or\ \ $\lambda x.\ f x$)\\
Lambda abstraction represents anonymous function definition.\\The corresponding SML syntax is \texttt{fn x => f x}.
\end{itemize}
\item terms have to be well-typed
\item same typing rules and same type-inference as in SML take place
\item terms very similar to SML expressions
\item notice: predicates are functions with return type \texttt{bool}, \ie
no distinction between functions and predicates, terms and formulae
\end{itemize}
\end{frame}

\begin{frame}
\frametitle{Terms II}
\begin{tabular}{lll}
\textbf{HOL term} & \textbf{SML expression} & \textbf{type HOL / SML} \\
\hol{0} & \ml{0} & \hol{num} / \ml{int} \\
\hol{x:'a} & \ml{x:'a} & variable of type \hol{'a} \\
\hol{x:bool} & \ml{x:bool} & variable of type \hol{bool}  \\
\hol{x + 5} & \ml{x + 5} & applying function \hol{+} to \hol{x} and \hol{5}  \\
\hol{\textbackslash x.\ x + 5} & \ml{fn x => x + 5} & anonymous (\aka inline) function \\
& & of type \hol{num -> num} \\
\hol{(5, T)} & \ml{(5, true)} & \hol{num \# bool} / \ml{int * bool}\\
\hol{[5;3;2]++[6]} & \ml{[5,3,2]@[6]} & \hol{num list} / \ml{int list}
\end{tabular}
\end{frame}

\begin{frame}
\frametitle{Free and Bound Variables / Alpha Equivalence}
\begin{itemize}
\item in SML, the names of function arguments does not matter (much)
\item similarly in HOL, the names of variables used by lambda-abstractions does not matter (much)
\item the lambda-expression $\lambda x.\ t$ is said to \emph{bind} the variables $x$ in term $t$
\item variables that are guarded by a lambda expression are called \emph{bound}
\item all other variables are \emph{free} 
\item Example: $x$ is free and $y$ is bound in \hol{$(x = 5) \wedge (\lambda y.\ (y < x))\ 3$} 
\item the names of bound variables are unimportant semantically
\item two terms are called \emph{alpha-equivalent} iff they differ only in the names of bound variables
\item Example: \hol{$\lambda{}x.\ x$} and \hol{$\lambda{}y.\ y$} are alpha-equivalent
\item Example: \hol{$x$} and \hol{$y$} are not alpha-equivalent 
\end{itemize}
\end{frame}


\begin{frame}
\frametitle{Theorems}
\begin{itemize}
\item theorems are of the form $\Gamma \entails p$ where
\begin{itemize}
\item $\Gamma$ is a set of hypothesis
\item $p$ is the conclusion of the theorem
\item all elements of $\Gamma$ and $p$ are formulae, \ie terms of type \texttt{bool}
\end{itemize}
\item $\Gamma \entails p$ records that using $\Gamma$ the statement $p$ \alert{has been} proved\\
\item notice difference to logic: there it means \alert{can be} proved
\item the proof itself is not recorded
\item theorems can only be created through a small interface in the \emph{kernel}
\end{itemize}
\end{frame}


\section{Kernel}
\begin{frame}
\frametitle{HOL Light Kernel}

\begin{itemize}
\item the HOL kernel is hard to explain
\begin{itemize}
\item for historic reasons some concepts are represented rather complicated
\item for speed reasons some derivable concepts have been added
\end{itemize}
\item instead consider the HOL Light kernel, which is a cleaned-up version
\item there are two predefined constants
\begin{itemize}
\item \texttt{= :\ 'a -> 'a -> bool}
\item \texttt{@ :\ ('a -> bool) -> 'a}
\end{itemize}
\item there are two predefined types
\begin{itemize}
\item \texttt{bool}
\item \texttt{ind}
\end{itemize}
\item the meaning of these types and constants is given by inference rules and axioms
\end{itemize}
\end{frame}

\begin{frame}
\frametitle{HOL Light Inferences I}
\begin{columns}
\begin{column}{.45\textwidth}
\begin{center}
$\inferrule* [right=REFL] {\ }{\entails t = t}$\\[1em]
$\inferrule*[right=TRANS] {\Gamma \entails s = t\\\Delta \entails t = u}{\Gamma \cup \Delta \entails s = u}$\\[1em]
$\inferrule*[right=COMB]{\Gamma \entails s = t\\\Delta \entails u = v \\\\ \textit{types fit}}{\Gamma \cup \Delta \entails s(u) = t(v)}$\\[1em]
\end{center}
\end{column}
\begin{column}{.45\textwidth}
\begin{center}
$\inferrule*[right=ABS]{\Gamma \entails s = t\\x\ \textit{not free in}\ \Gamma}{\Gamma \entails \lambda{}x.\ s = \lambda{}x.\ t}$\\[1em]
$\inferrule*[right=BETA]{\ }{\entails (\lambda{}x.\ t)\, x = t}$\\[1em]
$\inferrule*[right=ASSUME]{\ }{\{p\}\entails p}$
\end{center}
\end{column}
\end{columns}

\end{frame}


\begin{frame}
\frametitle{HOL Light Inferences II}
\begin{center}
$\inferrule*[right=EQ\_MP]{\Gamma \entails p \Leftrightarrow q\\\Delta \entails p}{\Gamma \cup \Delta \entails q}$\\[1em]
$\inferrule*[right=DEDUCT\_ANTISYM\_RULE]{\Gamma \entails p\\\Delta \entails q}
{(\Gamma-\{q\}) \cup (\Delta - \{p\}) \entails p \Leftrightarrow q}$\\[1em]
$\inferrule*[right=INST]{\Gamma[x_1, \ldots, x_n] \entails p[x_1, \ldots, x_n]}
{\Gamma[t_1, \ldots, t_n] \entails p[t_1, \ldots, t_n]}$\\[1em]
$\inferrule*[right=INST\_TYPE]{\Gamma[\alpha_1, \ldots, \alpha_n] \entails p[\alpha_1, \ldots, \alpha_n]}
{\Gamma[\gamma_1, \ldots, \gamma_n] \entails p[\gamma_1, \ldots, \gamma_n]}$\\[1em]
\end{center}
\end{frame}

\newcommand{\tabitem}{~~\llap{\textbullet}~~}

\begin{frame}
\frametitle{HOL Light Axioms and Definition Principles}
\begin{itemize}
\item 3 axioms needed\medskip\\\qquad
\begin{tabular}{ll}
ETA\_AX & $|- (\lambda{}x.\ t\ x) = t$ \\
SELECT\_AX & $|- P\ x \Longrightarrow P ((@) P))$ \\
INFINITY\_AX & predefined type \texttt{ind} is infinite 
\end{tabular}
\item definition principle for constants
\begin{itemize}
\item constants can be introduced as abbreviations
\item constraint: no free vars and no new type vars
\end{itemize}
\item definition principle for types
\begin{itemize}
\item new types can be defined as non-empty subtypes of existing types
\end{itemize}
\item both principles 
\begin{itemize}
\item lead to conservative extensions
\item preserve consistency 
\end{itemize}
\end{itemize}
\end{frame}


\begin{frame}
\frametitle{HOL Light derived concepts}
Everything else is derived from this small kernel. 
\[
\begin{array}{ccl}
T & =_{\textit{def}} & (\lambda{}p.\ p) = (\lambda{}p.\ p)\\
\wedge & =_{\textit{def}} & \lambda{}p\,q.\ (\lambda f.\ f\ p\ q) = (\lambda{}f.\ f\ T\ T) \\
\Longrightarrow & =_{\textit{def}} & \lambda{}p\,q.\ (p \wedge q \Leftrightarrow p) \\
\forall & =_{\textit{def}} & \lambda{}P.\ (P = \lambda{}x.\ T) \\
\exists & =_{\textit{def}} & \lambda{}P.\ (\forall{}q.\ (\forall{}x.\ P(x) \Longrightarrow q) \Longrightarrow q) \\
\ldots \\
\end{array}
\]
\end{frame}


\begin{frame}
\frametitle{Multiple Kernels}
\begin{itemize}
\item Kernel defines abstract datatypes for types, terms and theorems
\item one does not need to look at the internal implementation
\item therefore, easy to exchange
\item there are at least 3 different kernels for HOL
\begin{itemize}
\item standard kernel (de Bruijn indices)
\item experimental kernel (name / type pairs)
\item OpenTheory kernel (for proof recording) 
\end{itemize}
\end{itemize}
\end{frame}


\section{HOL Logic Summary}

\begin{frame}
\frametitle{HOL Logic Summary}
\begin{itemize}
\item HOL theorem prover uses classical higher order logic
\item HOL logic is very similar to SML
\begin{itemize}
\item syntax 
\item type system 
\item type inference
\end{itemize}
\item HOL theorem prover very trustworthy because of LCF approach
\begin{itemize}
\item there is a small kernel
\item proofs are not stored explicitly
\end{itemize}
\item you don't need to know the details of the kernel
\item usually one works at a much higher level of abstraction
\end{itemize}
\end{frame}

%%% Local Variables:
%%% mode: latex
%%% TeX-master: "current"
%%% End: