1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/doc-src/ZF/ZF.tex Wed Jan 13 16:36:36 1999 +0100
1.3 @@ -0,0 +1,2616 @@
1.4 +%% $Id$
1.5 +\chapter{Zermelo-Fraenkel Set Theory}
1.6 +\index{set theory|(}
1.7 +
1.8 +The theory~\thydx{ZF} implements Zermelo-Fraenkel set
1.9 +theory~\cite{halmos60,suppes72} as an extension of~\texttt{FOL}, classical
1.10 +first-order logic. The theory includes a collection of derived natural
1.11 +deduction rules, for use with Isabelle's classical reasoner. Much
1.12 +of it is based on the work of No\"el~\cite{noel}.
1.13 +
1.14 +A tremendous amount of set theory has been formally developed, including the
1.15 +basic properties of relations, functions, ordinals and cardinals. Significant
1.16 +results have been proved, such as the Schr\"oder-Bernstein Theorem, the
1.17 +Wellordering Theorem and a version of Ramsey's Theorem. \texttt{ZF} provides
1.18 +both the integers and the natural numbers. General methods have been
1.19 +developed for solving recursion equations over monotonic functors; these have
1.20 +been applied to yield constructions of lists, trees, infinite lists, etc.
1.21 +
1.22 +\texttt{ZF} has a flexible package for handling inductive definitions,
1.23 +such as inference systems, and datatype definitions, such as lists and
1.24 +trees. Moreover it handles coinductive definitions, such as
1.25 +bisimulation relations, and codatatype definitions, such as streams. It
1.26 +provides a streamlined syntax for defining primitive recursive functions over
1.27 +datatypes.
1.28 +
1.29 +Because {\ZF} is an extension of {\FOL}, it provides the same
1.30 +packages, namely \texttt{hyp_subst_tac}, the simplifier, and the
1.31 +classical reasoner. The default simpset and claset are usually
1.32 +satisfactory.
1.33 +
1.34 +Published articles~\cite{paulson-set-I,paulson-set-II} describe \texttt{ZF}
1.35 +less formally than this chapter. Isabelle employs a novel treatment of
1.36 +non-well-founded data structures within the standard {\sc zf} axioms including
1.37 +the Axiom of Foundation~\cite{paulson-final}.
1.38 +
1.39 +
1.40 +\section{Which version of axiomatic set theory?}
1.41 +The two main axiom systems for set theory are Bernays-G\"odel~({\sc bg})
1.42 +and Zermelo-Fraenkel~({\sc zf}). Resolution theorem provers can use {\sc
1.43 + bg} because it is finite~\cite{boyer86,quaife92}. {\sc zf} does not
1.44 +have a finite axiom system because of its Axiom Scheme of Replacement.
1.45 +This makes it awkward to use with many theorem provers, since instances
1.46 +of the axiom scheme have to be invoked explicitly. Since Isabelle has no
1.47 +difficulty with axiom schemes, we may adopt either axiom system.
1.48 +
1.49 +These two theories differ in their treatment of {\bf classes}, which are
1.50 +collections that are `too big' to be sets. The class of all sets,~$V$,
1.51 +cannot be a set without admitting Russell's Paradox. In {\sc bg}, both
1.52 +classes and sets are individuals; $x\in V$ expresses that $x$ is a set. In
1.53 +{\sc zf}, all variables denote sets; classes are identified with unary
1.54 +predicates. The two systems define essentially the same sets and classes,
1.55 +with similar properties. In particular, a class cannot belong to another
1.56 +class (let alone a set).
1.57 +
1.58 +Modern set theorists tend to prefer {\sc zf} because they are mainly concerned
1.59 +with sets, rather than classes. {\sc bg} requires tiresome proofs that various
1.60 +collections are sets; for instance, showing $x\in\{x\}$ requires showing that
1.61 +$x$ is a set.
1.62 +
1.63 +
1.64 +\begin{figure} \small
1.65 +\begin{center}
1.66 +\begin{tabular}{rrr}
1.67 + \it name &\it meta-type & \it description \\
1.68 + \cdx{Let} & $[\alpha,\alpha\To\beta]\To\beta$ & let binder\\
1.69 + \cdx{0} & $i$ & empty set\\
1.70 + \cdx{cons} & $[i,i]\To i$ & finite set constructor\\
1.71 + \cdx{Upair} & $[i,i]\To i$ & unordered pairing\\
1.72 + \cdx{Pair} & $[i,i]\To i$ & ordered pairing\\
1.73 + \cdx{Inf} & $i$ & infinite set\\
1.74 + \cdx{Pow} & $i\To i$ & powerset\\
1.75 + \cdx{Union} \cdx{Inter} & $i\To i$ & set union/intersection \\
1.76 + \cdx{split} & $[[i,i]\To i, i] \To i$ & generalized projection\\
1.77 + \cdx{fst} \cdx{snd} & $i\To i$ & projections\\
1.78 + \cdx{converse}& $i\To i$ & converse of a relation\\
1.79 + \cdx{succ} & $i\To i$ & successor\\
1.80 + \cdx{Collect} & $[i,i\To o]\To i$ & separation\\
1.81 + \cdx{Replace} & $[i, [i,i]\To o] \To i$ & replacement\\
1.82 + \cdx{PrimReplace} & $[i, [i,i]\To o] \To i$ & primitive replacement\\
1.83 + \cdx{RepFun} & $[i, i\To i] \To i$ & functional replacement\\
1.84 + \cdx{Pi} \cdx{Sigma} & $[i,i\To i]\To i$ & general product/sum\\
1.85 + \cdx{domain} & $i\To i$ & domain of a relation\\
1.86 + \cdx{range} & $i\To i$ & range of a relation\\
1.87 + \cdx{field} & $i\To i$ & field of a relation\\
1.88 + \cdx{Lambda} & $[i, i\To i]\To i$ & $\lambda$-abstraction\\
1.89 + \cdx{restrict}& $[i, i] \To i$ & restriction of a function\\
1.90 + \cdx{The} & $[i\To o]\To i$ & definite description\\
1.91 + \cdx{if} & $[o,i,i]\To i$ & conditional\\
1.92 + \cdx{Ball} \cdx{Bex} & $[i, i\To o]\To o$ & bounded quantifiers
1.93 +\end{tabular}
1.94 +\end{center}
1.95 +\subcaption{Constants}
1.96 +
1.97 +\begin{center}
1.98 +\index{*"`"` symbol}
1.99 +\index{*"-"`"` symbol}
1.100 +\index{*"` symbol}\index{function applications!in \ZF}
1.101 +\index{*"- symbol}
1.102 +\index{*": symbol}
1.103 +\index{*"<"= symbol}
1.104 +\begin{tabular}{rrrr}
1.105 + \it symbol & \it meta-type & \it priority & \it description \\
1.106 + \tt `` & $[i,i]\To i$ & Left 90 & image \\
1.107 + \tt -`` & $[i,i]\To i$ & Left 90 & inverse image \\
1.108 + \tt ` & $[i,i]\To i$ & Left 90 & application \\
1.109 + \sdx{Int} & $[i,i]\To i$ & Left 70 & intersection ($\int$) \\
1.110 + \sdx{Un} & $[i,i]\To i$ & Left 65 & union ($\un$) \\
1.111 + \tt - & $[i,i]\To i$ & Left 65 & set difference ($-$) \\[1ex]
1.112 + \tt: & $[i,i]\To o$ & Left 50 & membership ($\in$) \\
1.113 + \tt <= & $[i,i]\To o$ & Left 50 & subset ($\subseteq$)
1.114 +\end{tabular}
1.115 +\end{center}
1.116 +\subcaption{Infixes}
1.117 +\caption{Constants of {\ZF}} \label{zf-constants}
1.118 +\end{figure}
1.119 +
1.120 +
1.121 +\section{The syntax of set theory}
1.122 +The language of set theory, as studied by logicians, has no constants. The
1.123 +traditional axioms merely assert the existence of empty sets, unions,
1.124 +powersets, etc.; this would be intolerable for practical reasoning. The
1.125 +Isabelle theory declares constants for primitive sets. It also extends
1.126 +\texttt{FOL} with additional syntax for finite sets, ordered pairs,
1.127 +comprehension, general union/intersection, general sums/products, and
1.128 +bounded quantifiers. In most other respects, Isabelle implements precisely
1.129 +Zermelo-Fraenkel set theory.
1.130 +
1.131 +Figure~\ref{zf-constants} lists the constants and infixes of~\ZF, while
1.132 +Figure~\ref{zf-trans} presents the syntax translations. Finally,
1.133 +Figure~\ref{zf-syntax} presents the full grammar for set theory, including
1.134 +the constructs of \FOL.
1.135 +
1.136 +Local abbreviations can be introduced by a \texttt{let} construct whose
1.137 +syntax appears in Fig.\ts\ref{zf-syntax}. Internally it is translated into
1.138 +the constant~\cdx{Let}. It can be expanded by rewriting with its
1.139 +definition, \tdx{Let_def}.
1.140 +
1.141 +Apart from \texttt{let}, set theory does not use polymorphism. All terms in
1.142 +{\ZF} have type~\tydx{i}, which is the type of individuals and has class~{\tt
1.143 + term}. The type of first-order formulae, remember, is~\textit{o}.
1.144 +
1.145 +Infix operators include binary union and intersection ($A\un B$ and
1.146 +$A\int B$), set difference ($A-B$), and the subset and membership
1.147 +relations. Note that $a$\verb|~:|$b$ is translated to $\neg(a\in b)$. The
1.148 +union and intersection operators ($\bigcup A$ and $\bigcap A$) form the
1.149 +union or intersection of a set of sets; $\bigcup A$ means the same as
1.150 +$\bigcup@{x\in A}x$. Of these operators, only $\bigcup A$ is primitive.
1.151 +
1.152 +The constant \cdx{Upair} constructs unordered pairs; thus {\tt
1.153 + Upair($A$,$B$)} denotes the set~$\{A,B\}$ and \texttt{Upair($A$,$A$)}
1.154 +denotes the singleton~$\{A\}$. General union is used to define binary
1.155 +union. The Isabelle version goes on to define the constant
1.156 +\cdx{cons}:
1.157 +\begin{eqnarray*}
1.158 + A\cup B & \equiv & \bigcup(\texttt{Upair}(A,B)) \\
1.159 + \texttt{cons}(a,B) & \equiv & \texttt{Upair}(a,a) \un B
1.160 +\end{eqnarray*}
1.161 +The $\{a@1, \ldots\}$ notation abbreviates finite sets constructed in the
1.162 +obvious manner using~\texttt{cons} and~$\emptyset$ (the empty set):
1.163 +\begin{eqnarray*}
1.164 + \{a,b,c\} & \equiv & \texttt{cons}(a,\texttt{cons}(b,\texttt{cons}(c,\emptyset)))
1.165 +\end{eqnarray*}
1.166 +
1.167 +The constant \cdx{Pair} constructs ordered pairs, as in {\tt
1.168 +Pair($a$,$b$)}. Ordered pairs may also be written within angle brackets,
1.169 +as {\tt<$a$,$b$>}. The $n$-tuple {\tt<$a@1$,\ldots,$a@{n-1}$,$a@n$>}
1.170 +abbreviates the nest of pairs\par\nobreak
1.171 +\centerline{\texttt{Pair($a@1$,\ldots,Pair($a@{n-1}$,$a@n$)\ldots).}}
1.172 +
1.173 +In {\ZF}, a function is a set of pairs. A {\ZF} function~$f$ is simply an
1.174 +individual as far as Isabelle is concerned: its Isabelle type is~$i$, not
1.175 +say $i\To i$. The infix operator~{\tt`} denotes the application of a
1.176 +function set to its argument; we must write~$f{\tt`}x$, not~$f(x)$. The
1.177 +syntax for image is~$f{\tt``}A$ and that for inverse image is~$f{\tt-``}A$.
1.178 +
1.179 +
1.180 +\begin{figure}
1.181 +\index{lambda abs@$\lambda$-abstractions!in \ZF}
1.182 +\index{*"-"> symbol}
1.183 +\index{*"* symbol}
1.184 +\begin{center} \footnotesize\tt\frenchspacing
1.185 +\begin{tabular}{rrr}
1.186 + \it external & \it internal & \it description \\
1.187 + $a$ \ttilde: $b$ & \ttilde($a$ : $b$) & \rm negated membership\\
1.188 + \ttlbrace$a@1$, $\ldots$, $a@n$\ttrbrace & cons($a@1$,$\ldots$,cons($a@n$,0)) &
1.189 + \rm finite set \\
1.190 + <$a@1$, $\ldots$, $a@{n-1}$, $a@n$> &
1.191 + Pair($a@1$,\ldots,Pair($a@{n-1}$,$a@n$)\ldots) &
1.192 + \rm ordered $n$-tuple \\
1.193 + \ttlbrace$x$:$A . P[x]$\ttrbrace & Collect($A$,$\lambda x. P[x]$) &
1.194 + \rm separation \\
1.195 + \ttlbrace$y . x$:$A$, $Q[x,y]$\ttrbrace & Replace($A$,$\lambda x\,y. Q[x,y]$) &
1.196 + \rm replacement \\
1.197 + \ttlbrace$b[x] . x$:$A$\ttrbrace & RepFun($A$,$\lambda x. b[x]$) &
1.198 + \rm functional replacement \\
1.199 + \sdx{INT} $x$:$A . B[x]$ & Inter(\ttlbrace$B[x] . x$:$A$\ttrbrace) &
1.200 + \rm general intersection \\
1.201 + \sdx{UN} $x$:$A . B[x]$ & Union(\ttlbrace$B[x] . x$:$A$\ttrbrace) &
1.202 + \rm general union \\
1.203 + \sdx{PROD} $x$:$A . B[x]$ & Pi($A$,$\lambda x. B[x]$) &
1.204 + \rm general product \\
1.205 + \sdx{SUM} $x$:$A . B[x]$ & Sigma($A$,$\lambda x. B[x]$) &
1.206 + \rm general sum \\
1.207 + $A$ -> $B$ & Pi($A$,$\lambda x. B$) &
1.208 + \rm function space \\
1.209 + $A$ * $B$ & Sigma($A$,$\lambda x. B$) &
1.210 + \rm binary product \\
1.211 + \sdx{THE} $x . P[x]$ & The($\lambda x. P[x]$) &
1.212 + \rm definite description \\
1.213 + \sdx{lam} $x$:$A . b[x]$ & Lambda($A$,$\lambda x. b[x]$) &
1.214 + \rm $\lambda$-abstraction\\[1ex]
1.215 + \sdx{ALL} $x$:$A . P[x]$ & Ball($A$,$\lambda x. P[x]$) &
1.216 + \rm bounded $\forall$ \\
1.217 + \sdx{EX} $x$:$A . P[x]$ & Bex($A$,$\lambda x. P[x]$) &
1.218 + \rm bounded $\exists$
1.219 +\end{tabular}
1.220 +\end{center}
1.221 +\caption{Translations for {\ZF}} \label{zf-trans}
1.222 +\end{figure}
1.223 +
1.224 +
1.225 +\begin{figure}
1.226 +\index{*let symbol}
1.227 +\index{*in symbol}
1.228 +\dquotes
1.229 +\[\begin{array}{rcl}
1.230 + term & = & \hbox{expression of type~$i$} \\
1.231 + & | & "let"~id~"="~term";"\dots";"~id~"="~term~"in"~term \\
1.232 + & | & "if"~term~"then"~term~"else"~term \\
1.233 + & | & "{\ttlbrace} " term\; ("," term)^* " {\ttrbrace}" \\
1.234 + & | & "< " term\; ("," term)^* " >" \\
1.235 + & | & "{\ttlbrace} " id ":" term " . " formula " {\ttrbrace}" \\
1.236 + & | & "{\ttlbrace} " id " . " id ":" term ", " formula " {\ttrbrace}" \\
1.237 + & | & "{\ttlbrace} " term " . " id ":" term " {\ttrbrace}" \\
1.238 + & | & term " `` " term \\
1.239 + & | & term " -`` " term \\
1.240 + & | & term " ` " term \\
1.241 + & | & term " * " term \\
1.242 + & | & term " Int " term \\
1.243 + & | & term " Un " term \\
1.244 + & | & term " - " term \\
1.245 + & | & term " -> " term \\
1.246 + & | & "THE~~" id " . " formula\\
1.247 + & | & "lam~~" id ":" term " . " term \\
1.248 + & | & "INT~~" id ":" term " . " term \\
1.249 + & | & "UN~~~" id ":" term " . " term \\
1.250 + & | & "PROD~" id ":" term " . " term \\
1.251 + & | & "SUM~~" id ":" term " . " term \\[2ex]
1.252 + formula & = & \hbox{expression of type~$o$} \\
1.253 + & | & term " : " term \\
1.254 + & | & term " \ttilde: " term \\
1.255 + & | & term " <= " term \\
1.256 + & | & term " = " term \\
1.257 + & | & term " \ttilde= " term \\
1.258 + & | & "\ttilde\ " formula \\
1.259 + & | & formula " \& " formula \\
1.260 + & | & formula " | " formula \\
1.261 + & | & formula " --> " formula \\
1.262 + & | & formula " <-> " formula \\
1.263 + & | & "ALL " id ":" term " . " formula \\
1.264 + & | & "EX~~" id ":" term " . " formula \\
1.265 + & | & "ALL~" id~id^* " . " formula \\
1.266 + & | & "EX~~" id~id^* " . " formula \\
1.267 + & | & "EX!~" id~id^* " . " formula
1.268 + \end{array}
1.269 +\]
1.270 +\caption{Full grammar for {\ZF}} \label{zf-syntax}
1.271 +\end{figure}
1.272 +
1.273 +
1.274 +\section{Binding operators}
1.275 +The constant \cdx{Collect} constructs sets by the principle of {\bf
1.276 + separation}. The syntax for separation is
1.277 +\hbox{\tt\ttlbrace$x$:$A$.\ $P[x]$\ttrbrace}, where $P[x]$ is a formula
1.278 +that may contain free occurrences of~$x$. It abbreviates the set {\tt
1.279 + Collect($A$,$\lambda x. P[x]$)}, which consists of all $x\in A$ that
1.280 +satisfy~$P[x]$. Note that \texttt{Collect} is an unfortunate choice of
1.281 +name: some set theories adopt a set-formation principle, related to
1.282 +replacement, called collection.
1.283 +
1.284 +The constant \cdx{Replace} constructs sets by the principle of {\bf
1.285 + replacement}. The syntax
1.286 +\hbox{\tt\ttlbrace$y$.\ $x$:$A$,$Q[x,y]$\ttrbrace} denotes the set {\tt
1.287 + Replace($A$,$\lambda x\,y. Q[x,y]$)}, which consists of all~$y$ such
1.288 +that there exists $x\in A$ satisfying~$Q[x,y]$. The Replacement Axiom
1.289 +has the condition that $Q$ must be single-valued over~$A$: for
1.290 +all~$x\in A$ there exists at most one $y$ satisfying~$Q[x,y]$. A
1.291 +single-valued binary predicate is also called a {\bf class function}.
1.292 +
1.293 +The constant \cdx{RepFun} expresses a special case of replacement,
1.294 +where $Q[x,y]$ has the form $y=b[x]$. Such a $Q$ is trivially
1.295 +single-valued, since it is just the graph of the meta-level
1.296 +function~$\lambda x. b[x]$. The resulting set consists of all $b[x]$
1.297 +for~$x\in A$. This is analogous to the \ML{} functional \texttt{map},
1.298 +since it applies a function to every element of a set. The syntax is
1.299 +\hbox{\tt\ttlbrace$b[x]$.\ $x$:$A$\ttrbrace}, which expands to {\tt
1.300 + RepFun($A$,$\lambda x. b[x]$)}.
1.301 +
1.302 +\index{*INT symbol}\index{*UN symbol}
1.303 +General unions and intersections of indexed
1.304 +families of sets, namely $\bigcup@{x\in A}B[x]$ and $\bigcap@{x\in A}B[x]$,
1.305 +are written \hbox{\tt UN $x$:$A$.\ $B[x]$} and \hbox{\tt INT $x$:$A$.\ $B[x]$}.
1.306 +Their meaning is expressed using \texttt{RepFun} as
1.307 +\[
1.308 +\bigcup(\{B[x]. x\in A\}) \qquad\hbox{and}\qquad
1.309 +\bigcap(\{B[x]. x\in A\}).
1.310 +\]
1.311 +General sums $\sum@{x\in A}B[x]$ and products $\prod@{x\in A}B[x]$ can be
1.312 +constructed in set theory, where $B[x]$ is a family of sets over~$A$. They
1.313 +have as special cases $A\times B$ and $A\to B$, where $B$ is simply a set.
1.314 +This is similar to the situation in Constructive Type Theory (set theory
1.315 +has `dependent sets') and calls for similar syntactic conventions. The
1.316 +constants~\cdx{Sigma} and~\cdx{Pi} construct general sums and
1.317 +products. Instead of \texttt{Sigma($A$,$B$)} and \texttt{Pi($A$,$B$)} we may
1.318 +write
1.319 +\hbox{\tt SUM $x$:$A$.\ $B[x]$} and \hbox{\tt PROD $x$:$A$.\ $B[x]$}.
1.320 +\index{*SUM symbol}\index{*PROD symbol}%
1.321 +The special cases as \hbox{\tt$A$*$B$} and \hbox{\tt$A$->$B$} abbreviate
1.322 +general sums and products over a constant family.\footnote{Unlike normal
1.323 +infix operators, {\tt*} and {\tt->} merely define abbreviations; there are
1.324 +no constants~\texttt{op~*} and~\hbox{\tt op~->}.} Isabelle accepts these
1.325 +abbreviations in parsing and uses them whenever possible for printing.
1.326 +
1.327 +\index{*THE symbol}
1.328 +As mentioned above, whenever the axioms assert the existence and uniqueness
1.329 +of a set, Isabelle's set theory declares a constant for that set. These
1.330 +constants can express the {\bf definite description} operator~$\iota
1.331 +x. P[x]$, which stands for the unique~$a$ satisfying~$P[a]$, if such exists.
1.332 +Since all terms in {\ZF} denote something, a description is always
1.333 +meaningful, but we do not know its value unless $P[x]$ defines it uniquely.
1.334 +Using the constant~\cdx{The}, we may write descriptions as {\tt
1.335 + The($\lambda x. P[x]$)} or use the syntax \hbox{\tt THE $x$.\ $P[x]$}.
1.336 +
1.337 +\index{*lam symbol}
1.338 +Function sets may be written in $\lambda$-notation; $\lambda x\in A. b[x]$
1.339 +stands for the set of all pairs $\pair{x,b[x]}$ for $x\in A$. In order for
1.340 +this to be a set, the function's domain~$A$ must be given. Using the
1.341 +constant~\cdx{Lambda}, we may express function sets as {\tt
1.342 +Lambda($A$,$\lambda x. b[x]$)} or use the syntax \hbox{\tt lam $x$:$A$.\ $b[x]$}.
1.343 +
1.344 +Isabelle's set theory defines two {\bf bounded quantifiers}:
1.345 +\begin{eqnarray*}
1.346 + \forall x\in A. P[x] &\hbox{abbreviates}& \forall x. x\in A\imp P[x] \\
1.347 + \exists x\in A. P[x] &\hbox{abbreviates}& \exists x. x\in A\conj P[x]
1.348 +\end{eqnarray*}
1.349 +The constants~\cdx{Ball} and~\cdx{Bex} are defined
1.350 +accordingly. Instead of \texttt{Ball($A$,$P$)} and \texttt{Bex($A$,$P$)} we may
1.351 +write
1.352 +\hbox{\tt ALL $x$:$A$.\ $P[x]$} and \hbox{\tt EX $x$:$A$.\ $P[x]$}.
1.353 +
1.354 +
1.355 +%%%% ZF.thy
1.356 +
1.357 +\begin{figure}
1.358 +\begin{ttbox}
1.359 +\tdx{Let_def} Let(s, f) == f(s)
1.360 +
1.361 +\tdx{Ball_def} Ball(A,P) == ALL x. x:A --> P(x)
1.362 +\tdx{Bex_def} Bex(A,P) == EX x. x:A & P(x)
1.363 +
1.364 +\tdx{subset_def} A <= B == ALL x:A. x:B
1.365 +\tdx{extension} A = B <-> A <= B & B <= A
1.366 +
1.367 +\tdx{Union_iff} A : Union(C) <-> (EX B:C. A:B)
1.368 +\tdx{Pow_iff} A : Pow(B) <-> A <= B
1.369 +\tdx{foundation} A=0 | (EX x:A. ALL y:x. ~ y:A)
1.370 +
1.371 +\tdx{replacement} (ALL x:A. ALL y z. P(x,y) & P(x,z) --> y=z) ==>
1.372 + b : PrimReplace(A,P) <-> (EX x:A. P(x,b))
1.373 +\subcaption{The Zermelo-Fraenkel Axioms}
1.374 +
1.375 +\tdx{Replace_def} Replace(A,P) ==
1.376 + PrimReplace(A, \%x y. (EX!z. P(x,z)) & P(x,y))
1.377 +\tdx{RepFun_def} RepFun(A,f) == {\ttlbrace}y . x:A, y=f(x)\ttrbrace
1.378 +\tdx{the_def} The(P) == Union({\ttlbrace}y . x:{\ttlbrace}0{\ttrbrace}, P(y){\ttrbrace})
1.379 +\tdx{if_def} if(P,a,b) == THE z. P & z=a | ~P & z=b
1.380 +\tdx{Collect_def} Collect(A,P) == {\ttlbrace}y . x:A, x=y & P(x){\ttrbrace}
1.381 +\tdx{Upair_def} Upair(a,b) ==
1.382 + {\ttlbrace}y. x:Pow(Pow(0)), (x=0 & y=a) | (x=Pow(0) & y=b){\ttrbrace}
1.383 +\subcaption{Consequences of replacement}
1.384 +
1.385 +\tdx{Inter_def} Inter(A) == {\ttlbrace}x:Union(A) . ALL y:A. x:y{\ttrbrace}
1.386 +\tdx{Un_def} A Un B == Union(Upair(A,B))
1.387 +\tdx{Int_def} A Int B == Inter(Upair(A,B))
1.388 +\tdx{Diff_def} A - B == {\ttlbrace}x:A . x~:B{\ttrbrace}
1.389 +\subcaption{Union, intersection, difference}
1.390 +\end{ttbox}
1.391 +\caption{Rules and axioms of {\ZF}} \label{zf-rules}
1.392 +\end{figure}
1.393 +
1.394 +
1.395 +\begin{figure}
1.396 +\begin{ttbox}
1.397 +\tdx{cons_def} cons(a,A) == Upair(a,a) Un A
1.398 +\tdx{succ_def} succ(i) == cons(i,i)
1.399 +\tdx{infinity} 0:Inf & (ALL y:Inf. succ(y): Inf)
1.400 +\subcaption{Finite and infinite sets}
1.401 +
1.402 +\tdx{Pair_def} <a,b> == {\ttlbrace}{\ttlbrace}a,a{\ttrbrace}, {\ttlbrace}a,b{\ttrbrace}{\ttrbrace}
1.403 +\tdx{split_def} split(c,p) == THE y. EX a b. p=<a,b> & y=c(a,b)
1.404 +\tdx{fst_def} fst(A) == split(\%x y. x, p)
1.405 +\tdx{snd_def} snd(A) == split(\%x y. y, p)
1.406 +\tdx{Sigma_def} Sigma(A,B) == UN x:A. UN y:B(x). {\ttlbrace}<x,y>{\ttrbrace}
1.407 +\subcaption{Ordered pairs and Cartesian products}
1.408 +
1.409 +\tdx{converse_def} converse(r) == {\ttlbrace}z. w:r, EX x y. w=<x,y> & z=<y,x>{\ttrbrace}
1.410 +\tdx{domain_def} domain(r) == {\ttlbrace}x. w:r, EX y. w=<x,y>{\ttrbrace}
1.411 +\tdx{range_def} range(r) == domain(converse(r))
1.412 +\tdx{field_def} field(r) == domain(r) Un range(r)
1.413 +\tdx{image_def} r `` A == {\ttlbrace}y : range(r) . EX x:A. <x,y> : r{\ttrbrace}
1.414 +\tdx{vimage_def} r -`` A == converse(r)``A
1.415 +\subcaption{Operations on relations}
1.416 +
1.417 +\tdx{lam_def} Lambda(A,b) == {\ttlbrace}<x,b(x)> . x:A{\ttrbrace}
1.418 +\tdx{apply_def} f`a == THE y. <a,y> : f
1.419 +\tdx{Pi_def} Pi(A,B) == {\ttlbrace}f: Pow(Sigma(A,B)). ALL x:A. EX! y. <x,y>: f{\ttrbrace}
1.420 +\tdx{restrict_def} restrict(f,A) == lam x:A. f`x
1.421 +\subcaption{Functions and general product}
1.422 +\end{ttbox}
1.423 +\caption{Further definitions of {\ZF}} \label{zf-defs}
1.424 +\end{figure}
1.425 +
1.426 +
1.427 +
1.428 +\section{The Zermelo-Fraenkel axioms}
1.429 +The axioms appear in Fig.\ts \ref{zf-rules}. They resemble those
1.430 +presented by Suppes~\cite{suppes72}. Most of the theory consists of
1.431 +definitions. In particular, bounded quantifiers and the subset relation
1.432 +appear in other axioms. Object-level quantifiers and implications have
1.433 +been replaced by meta-level ones wherever possible, to simplify use of the
1.434 +axioms. See the file \texttt{ZF/ZF.thy} for details.
1.435 +
1.436 +The traditional replacement axiom asserts
1.437 +\[ y \in \texttt{PrimReplace}(A,P) \bimp (\exists x\in A. P(x,y)) \]
1.438 +subject to the condition that $P(x,y)$ is single-valued for all~$x\in A$.
1.439 +The Isabelle theory defines \cdx{Replace} to apply
1.440 +\cdx{PrimReplace} to the single-valued part of~$P$, namely
1.441 +\[ (\exists!z. P(x,z)) \conj P(x,y). \]
1.442 +Thus $y\in \texttt{Replace}(A,P)$ if and only if there is some~$x$ such that
1.443 +$P(x,-)$ holds uniquely for~$y$. Because the equivalence is unconditional,
1.444 +\texttt{Replace} is much easier to use than \texttt{PrimReplace}; it defines the
1.445 +same set, if $P(x,y)$ is single-valued. The nice syntax for replacement
1.446 +expands to \texttt{Replace}.
1.447 +
1.448 +Other consequences of replacement include functional replacement
1.449 +(\cdx{RepFun}) and definite descriptions (\cdx{The}).
1.450 +Axioms for separation (\cdx{Collect}) and unordered pairs
1.451 +(\cdx{Upair}) are traditionally assumed, but they actually follow
1.452 +from replacement~\cite[pages 237--8]{suppes72}.
1.453 +
1.454 +The definitions of general intersection, etc., are straightforward. Note
1.455 +the definition of \texttt{cons}, which underlies the finite set notation.
1.456 +The axiom of infinity gives us a set that contains~0 and is closed under
1.457 +successor (\cdx{succ}). Although this set is not uniquely defined,
1.458 +the theory names it (\cdx{Inf}) in order to simplify the
1.459 +construction of the natural numbers.
1.460 +
1.461 +Further definitions appear in Fig.\ts\ref{zf-defs}. Ordered pairs are
1.462 +defined in the standard way, $\pair{a,b}\equiv\{\{a\},\{a,b\}\}$. Recall
1.463 +that \cdx{Sigma}$(A,B)$ generalizes the Cartesian product of two
1.464 +sets. It is defined to be the union of all singleton sets
1.465 +$\{\pair{x,y}\}$, for $x\in A$ and $y\in B(x)$. This is a typical usage of
1.466 +general union.
1.467 +
1.468 +The projections \cdx{fst} and~\cdx{snd} are defined in terms of the
1.469 +generalized projection \cdx{split}. The latter has been borrowed from
1.470 +Martin-L\"of's Type Theory, and is often easier to use than \cdx{fst}
1.471 +and~\cdx{snd}.
1.472 +
1.473 +Operations on relations include converse, domain, range, and image. The
1.474 +set ${\tt Pi}(A,B)$ generalizes the space of functions between two sets.
1.475 +Note the simple definitions of $\lambda$-abstraction (using
1.476 +\cdx{RepFun}) and application (using a definite description). The
1.477 +function \cdx{restrict}$(f,A)$ has the same values as~$f$, but only
1.478 +over the domain~$A$.
1.479 +
1.480 +
1.481 +%%%% zf.ML
1.482 +
1.483 +\begin{figure}
1.484 +\begin{ttbox}
1.485 +\tdx{ballI} [| !!x. x:A ==> P(x) |] ==> ALL x:A. P(x)
1.486 +\tdx{bspec} [| ALL x:A. P(x); x: A |] ==> P(x)
1.487 +\tdx{ballE} [| ALL x:A. P(x); P(x) ==> Q; ~ x:A ==> Q |] ==> Q
1.488 +
1.489 +\tdx{ball_cong} [| A=A'; !!x. x:A' ==> P(x) <-> P'(x) |] ==>
1.490 + (ALL x:A. P(x)) <-> (ALL x:A'. P'(x))
1.491 +
1.492 +\tdx{bexI} [| P(x); x: A |] ==> EX x:A. P(x)
1.493 +\tdx{bexCI} [| ALL x:A. ~P(x) ==> P(a); a: A |] ==> EX x:A. P(x)
1.494 +\tdx{bexE} [| EX x:A. P(x); !!x. [| x:A; P(x) |] ==> Q |] ==> Q
1.495 +
1.496 +\tdx{bex_cong} [| A=A'; !!x. x:A' ==> P(x) <-> P'(x) |] ==>
1.497 + (EX x:A. P(x)) <-> (EX x:A'. P'(x))
1.498 +\subcaption{Bounded quantifiers}
1.499 +
1.500 +\tdx{subsetI} (!!x. x:A ==> x:B) ==> A <= B
1.501 +\tdx{subsetD} [| A <= B; c:A |] ==> c:B
1.502 +\tdx{subsetCE} [| A <= B; ~(c:A) ==> P; c:B ==> P |] ==> P
1.503 +\tdx{subset_refl} A <= A
1.504 +\tdx{subset_trans} [| A<=B; B<=C |] ==> A<=C
1.505 +
1.506 +\tdx{equalityI} [| A <= B; B <= A |] ==> A = B
1.507 +\tdx{equalityD1} A = B ==> A<=B
1.508 +\tdx{equalityD2} A = B ==> B<=A
1.509 +\tdx{equalityE} [| A = B; [| A<=B; B<=A |] ==> P |] ==> P
1.510 +\subcaption{Subsets and extensionality}
1.511 +
1.512 +\tdx{emptyE} a:0 ==> P
1.513 +\tdx{empty_subsetI} 0 <= A
1.514 +\tdx{equals0I} [| !!y. y:A ==> False |] ==> A=0
1.515 +\tdx{equals0D} [| A=0; a:A |] ==> P
1.516 +
1.517 +\tdx{PowI} A <= B ==> A : Pow(B)
1.518 +\tdx{PowD} A : Pow(B) ==> A<=B
1.519 +\subcaption{The empty set; power sets}
1.520 +\end{ttbox}
1.521 +\caption{Basic derived rules for {\ZF}} \label{zf-lemmas1}
1.522 +\end{figure}
1.523 +
1.524 +
1.525 +\section{From basic lemmas to function spaces}
1.526 +Faced with so many definitions, it is essential to prove lemmas. Even
1.527 +trivial theorems like $A \int B = B \int A$ would be difficult to
1.528 +prove from the definitions alone. Isabelle's set theory derives many
1.529 +rules using a natural deduction style. Ideally, a natural deduction
1.530 +rule should introduce or eliminate just one operator, but this is not
1.531 +always practical. For most operators, we may forget its definition
1.532 +and use its derived rules instead.
1.533 +
1.534 +\subsection{Fundamental lemmas}
1.535 +Figure~\ref{zf-lemmas1} presents the derived rules for the most basic
1.536 +operators. The rules for the bounded quantifiers resemble those for the
1.537 +ordinary quantifiers, but note that \tdx{ballE} uses a negated assumption
1.538 +in the style of Isabelle's classical reasoner. The \rmindex{congruence
1.539 + rules} \tdx{ball_cong} and \tdx{bex_cong} are required by Isabelle's
1.540 +simplifier, but have few other uses. Congruence rules must be specially
1.541 +derived for all binding operators, and henceforth will not be shown.
1.542 +
1.543 +Figure~\ref{zf-lemmas1} also shows rules for the subset and equality
1.544 +relations (proof by extensionality), and rules about the empty set and the
1.545 +power set operator.
1.546 +
1.547 +Figure~\ref{zf-lemmas2} presents rules for replacement and separation.
1.548 +The rules for \cdx{Replace} and \cdx{RepFun} are much simpler than
1.549 +comparable rules for \texttt{PrimReplace} would be. The principle of
1.550 +separation is proved explicitly, although most proofs should use the
1.551 +natural deduction rules for \texttt{Collect}. The elimination rule
1.552 +\tdx{CollectE} is equivalent to the two destruction rules
1.553 +\tdx{CollectD1} and \tdx{CollectD2}, but each rule is suited to
1.554 +particular circumstances. Although too many rules can be confusing, there
1.555 +is no reason to aim for a minimal set of rules. See the file
1.556 +\texttt{ZF/ZF.ML} for a complete listing.
1.557 +
1.558 +Figure~\ref{zf-lemmas3} presents rules for general union and intersection.
1.559 +The empty intersection should be undefined. We cannot have
1.560 +$\bigcap(\emptyset)=V$ because $V$, the universal class, is not a set. All
1.561 +expressions denote something in {\ZF} set theory; the definition of
1.562 +intersection implies $\bigcap(\emptyset)=\emptyset$, but this value is
1.563 +arbitrary. The rule \tdx{InterI} must have a premise to exclude
1.564 +the empty intersection. Some of the laws governing intersections require
1.565 +similar premises.
1.566 +
1.567 +
1.568 +%the [p] gives better page breaking for the book
1.569 +\begin{figure}[p]
1.570 +\begin{ttbox}
1.571 +\tdx{ReplaceI} [| x: A; P(x,b); !!y. P(x,y) ==> y=b |] ==>
1.572 + b : {\ttlbrace}y. x:A, P(x,y){\ttrbrace}
1.573 +
1.574 +\tdx{ReplaceE} [| b : {\ttlbrace}y. x:A, P(x,y){\ttrbrace};
1.575 + !!x. [| x: A; P(x,b); ALL y. P(x,y)-->y=b |] ==> R
1.576 + |] ==> R
1.577 +
1.578 +\tdx{RepFunI} [| a : A |] ==> f(a) : {\ttlbrace}f(x). x:A{\ttrbrace}
1.579 +\tdx{RepFunE} [| b : {\ttlbrace}f(x). x:A{\ttrbrace};
1.580 + !!x.[| x:A; b=f(x) |] ==> P |] ==> P
1.581 +
1.582 +\tdx{separation} a : {\ttlbrace}x:A. P(x){\ttrbrace} <-> a:A & P(a)
1.583 +\tdx{CollectI} [| a:A; P(a) |] ==> a : {\ttlbrace}x:A. P(x){\ttrbrace}
1.584 +\tdx{CollectE} [| a : {\ttlbrace}x:A. P(x){\ttrbrace}; [| a:A; P(a) |] ==> R |] ==> R
1.585 +\tdx{CollectD1} a : {\ttlbrace}x:A. P(x){\ttrbrace} ==> a:A
1.586 +\tdx{CollectD2} a : {\ttlbrace}x:A. P(x){\ttrbrace} ==> P(a)
1.587 +\end{ttbox}
1.588 +\caption{Replacement and separation} \label{zf-lemmas2}
1.589 +\end{figure}
1.590 +
1.591 +
1.592 +\begin{figure}
1.593 +\begin{ttbox}
1.594 +\tdx{UnionI} [| B: C; A: B |] ==> A: Union(C)
1.595 +\tdx{UnionE} [| A : Union(C); !!B.[| A: B; B: C |] ==> R |] ==> R
1.596 +
1.597 +\tdx{InterI} [| !!x. x: C ==> A: x; c:C |] ==> A : Inter(C)
1.598 +\tdx{InterD} [| A : Inter(C); B : C |] ==> A : B
1.599 +\tdx{InterE} [| A : Inter(C); A:B ==> R; ~ B:C ==> R |] ==> R
1.600 +
1.601 +\tdx{UN_I} [| a: A; b: B(a) |] ==> b: (UN x:A. B(x))
1.602 +\tdx{UN_E} [| b : (UN x:A. B(x)); !!x.[| x: A; b: B(x) |] ==> R
1.603 + |] ==> R
1.604 +
1.605 +\tdx{INT_I} [| !!x. x: A ==> b: B(x); a: A |] ==> b: (INT x:A. B(x))
1.606 +\tdx{INT_E} [| b : (INT x:A. B(x)); a: A |] ==> b : B(a)
1.607 +\end{ttbox}
1.608 +\caption{General union and intersection} \label{zf-lemmas3}
1.609 +\end{figure}
1.610 +
1.611 +
1.612 +%%% upair.ML
1.613 +
1.614 +\begin{figure}
1.615 +\begin{ttbox}
1.616 +\tdx{pairing} a:Upair(b,c) <-> (a=b | a=c)
1.617 +\tdx{UpairI1} a : Upair(a,b)
1.618 +\tdx{UpairI2} b : Upair(a,b)
1.619 +\tdx{UpairE} [| a : Upair(b,c); a = b ==> P; a = c ==> P |] ==> P
1.620 +\end{ttbox}
1.621 +\caption{Unordered pairs} \label{zf-upair1}
1.622 +\end{figure}
1.623 +
1.624 +
1.625 +\begin{figure}
1.626 +\begin{ttbox}
1.627 +\tdx{UnI1} c : A ==> c : A Un B
1.628 +\tdx{UnI2} c : B ==> c : A Un B
1.629 +\tdx{UnCI} (~c : B ==> c : A) ==> c : A Un B
1.630 +\tdx{UnE} [| c : A Un B; c:A ==> P; c:B ==> P |] ==> P
1.631 +
1.632 +\tdx{IntI} [| c : A; c : B |] ==> c : A Int B
1.633 +\tdx{IntD1} c : A Int B ==> c : A
1.634 +\tdx{IntD2} c : A Int B ==> c : B
1.635 +\tdx{IntE} [| c : A Int B; [| c:A; c:B |] ==> P |] ==> P
1.636 +
1.637 +\tdx{DiffI} [| c : A; ~ c : B |] ==> c : A - B
1.638 +\tdx{DiffD1} c : A - B ==> c : A
1.639 +\tdx{DiffD2} c : A - B ==> c ~: B
1.640 +\tdx{DiffE} [| c : A - B; [| c:A; ~ c:B |] ==> P |] ==> P
1.641 +\end{ttbox}
1.642 +\caption{Union, intersection, difference} \label{zf-Un}
1.643 +\end{figure}
1.644 +
1.645 +
1.646 +\begin{figure}
1.647 +\begin{ttbox}
1.648 +\tdx{consI1} a : cons(a,B)
1.649 +\tdx{consI2} a : B ==> a : cons(b,B)
1.650 +\tdx{consCI} (~ a:B ==> a=b) ==> a: cons(b,B)
1.651 +\tdx{consE} [| a : cons(b,A); a=b ==> P; a:A ==> P |] ==> P
1.652 +
1.653 +\tdx{singletonI} a : {\ttlbrace}a{\ttrbrace}
1.654 +\tdx{singletonE} [| a : {\ttlbrace}b{\ttrbrace}; a=b ==> P |] ==> P
1.655 +\end{ttbox}
1.656 +\caption{Finite and singleton sets} \label{zf-upair2}
1.657 +\end{figure}
1.658 +
1.659 +
1.660 +\begin{figure}
1.661 +\begin{ttbox}
1.662 +\tdx{succI1} i : succ(i)
1.663 +\tdx{succI2} i : j ==> i : succ(j)
1.664 +\tdx{succCI} (~ i:j ==> i=j) ==> i: succ(j)
1.665 +\tdx{succE} [| i : succ(j); i=j ==> P; i:j ==> P |] ==> P
1.666 +\tdx{succ_neq_0} [| succ(n)=0 |] ==> P
1.667 +\tdx{succ_inject} succ(m) = succ(n) ==> m=n
1.668 +\end{ttbox}
1.669 +\caption{The successor function} \label{zf-succ}
1.670 +\end{figure}
1.671 +
1.672 +
1.673 +\begin{figure}
1.674 +\begin{ttbox}
1.675 +\tdx{the_equality} [| P(a); !!x. P(x) ==> x=a |] ==> (THE x. P(x)) = a
1.676 +\tdx{theI} EX! x. P(x) ==> P(THE x. P(x))
1.677 +
1.678 +\tdx{if_P} P ==> (if P then a else b) = a
1.679 +\tdx{if_not_P} ~P ==> (if P then a else b) = b
1.680 +
1.681 +\tdx{mem_asym} [| a:b; b:a |] ==> P
1.682 +\tdx{mem_irrefl} a:a ==> P
1.683 +\end{ttbox}
1.684 +\caption{Descriptions; non-circularity} \label{zf-the}
1.685 +\end{figure}
1.686 +
1.687 +
1.688 +\subsection{Unordered pairs and finite sets}
1.689 +Figure~\ref{zf-upair1} presents the principle of unordered pairing, along
1.690 +with its derived rules. Binary union and intersection are defined in terms
1.691 +of ordered pairs (Fig.\ts\ref{zf-Un}). Set difference is also included. The
1.692 +rule \tdx{UnCI} is useful for classical reasoning about unions,
1.693 +like \texttt{disjCI}\@; it supersedes \tdx{UnI1} and
1.694 +\tdx{UnI2}, but these rules are often easier to work with. For
1.695 +intersection and difference we have both elimination and destruction rules.
1.696 +Again, there is no reason to provide a minimal rule set.
1.697 +
1.698 +Figure~\ref{zf-upair2} is concerned with finite sets: it presents rules
1.699 +for~\texttt{cons}, the finite set constructor, and rules for singleton
1.700 +sets. Figure~\ref{zf-succ} presents derived rules for the successor
1.701 +function, which is defined in terms of~\texttt{cons}. The proof that {\tt
1.702 + succ} is injective appears to require the Axiom of Foundation.
1.703 +
1.704 +Definite descriptions (\sdx{THE}) are defined in terms of the singleton
1.705 +set~$\{0\}$, but their derived rules fortunately hide this
1.706 +(Fig.\ts\ref{zf-the}). The rule~\tdx{theI} is difficult to apply
1.707 +because of the two occurrences of~$\Var{P}$. However,
1.708 +\tdx{the_equality} does not have this problem and the files contain
1.709 +many examples of its use.
1.710 +
1.711 +Finally, the impossibility of having both $a\in b$ and $b\in a$
1.712 +(\tdx{mem_asym}) is proved by applying the Axiom of Foundation to
1.713 +the set $\{a,b\}$. The impossibility of $a\in a$ is a trivial consequence.
1.714 +
1.715 +See the file \texttt{ZF/upair.ML} for full proofs of the rules discussed in
1.716 +this section.
1.717 +
1.718 +
1.719 +%%% subset.ML
1.720 +
1.721 +\begin{figure}
1.722 +\begin{ttbox}
1.723 +\tdx{Union_upper} B:A ==> B <= Union(A)
1.724 +\tdx{Union_least} [| !!x. x:A ==> x<=C |] ==> Union(A) <= C
1.725 +
1.726 +\tdx{Inter_lower} B:A ==> Inter(A) <= B
1.727 +\tdx{Inter_greatest} [| a:A; !!x. x:A ==> C<=x |] ==> C <= Inter(A)
1.728 +
1.729 +\tdx{Un_upper1} A <= A Un B
1.730 +\tdx{Un_upper2} B <= A Un B
1.731 +\tdx{Un_least} [| A<=C; B<=C |] ==> A Un B <= C
1.732 +
1.733 +\tdx{Int_lower1} A Int B <= A
1.734 +\tdx{Int_lower2} A Int B <= B
1.735 +\tdx{Int_greatest} [| C<=A; C<=B |] ==> C <= A Int B
1.736 +
1.737 +\tdx{Diff_subset} A-B <= A
1.738 +\tdx{Diff_contains} [| C<=A; C Int B = 0 |] ==> C <= A-B
1.739 +
1.740 +\tdx{Collect_subset} Collect(A,P) <= A
1.741 +\end{ttbox}
1.742 +\caption{Subset and lattice properties} \label{zf-subset}
1.743 +\end{figure}
1.744 +
1.745 +
1.746 +\subsection{Subset and lattice properties}
1.747 +The subset relation is a complete lattice. Unions form least upper bounds;
1.748 +non-empty intersections form greatest lower bounds. Figure~\ref{zf-subset}
1.749 +shows the corresponding rules. A few other laws involving subsets are
1.750 +included. Proofs are in the file \texttt{ZF/subset.ML}.
1.751 +
1.752 +Reasoning directly about subsets often yields clearer proofs than
1.753 +reasoning about the membership relation. Section~\ref{sec:ZF-pow-example}
1.754 +below presents an example of this, proving the equation ${{\tt Pow}(A)\cap
1.755 + {\tt Pow}(B)}= {\tt Pow}(A\cap B)$.
1.756 +
1.757 +%%% pair.ML
1.758 +
1.759 +\begin{figure}
1.760 +\begin{ttbox}
1.761 +\tdx{Pair_inject1} <a,b> = <c,d> ==> a=c
1.762 +\tdx{Pair_inject2} <a,b> = <c,d> ==> b=d
1.763 +\tdx{Pair_inject} [| <a,b> = <c,d>; [| a=c; b=d |] ==> P |] ==> P
1.764 +\tdx{Pair_neq_0} <a,b>=0 ==> P
1.765 +
1.766 +\tdx{fst_conv} fst(<a,b>) = a
1.767 +\tdx{snd_conv} snd(<a,b>) = b
1.768 +\tdx{split} split(\%x y. c(x,y), <a,b>) = c(a,b)
1.769 +
1.770 +\tdx{SigmaI} [| a:A; b:B(a) |] ==> <a,b> : Sigma(A,B)
1.771 +
1.772 +\tdx{SigmaE} [| c: Sigma(A,B);
1.773 + !!x y.[| x:A; y:B(x); c=<x,y> |] ==> P |] ==> P
1.774 +
1.775 +\tdx{SigmaE2} [| <a,b> : Sigma(A,B);
1.776 + [| a:A; b:B(a) |] ==> P |] ==> P
1.777 +\end{ttbox}
1.778 +\caption{Ordered pairs; projections; general sums} \label{zf-pair}
1.779 +\end{figure}
1.780 +
1.781 +
1.782 +\subsection{Ordered pairs} \label{sec:pairs}
1.783 +
1.784 +Figure~\ref{zf-pair} presents the rules governing ordered pairs,
1.785 +projections and general sums. File \texttt{ZF/pair.ML} contains the
1.786 +full (and tedious) proof that $\{\{a\},\{a,b\}\}$ functions as an ordered
1.787 +pair. This property is expressed as two destruction rules,
1.788 +\tdx{Pair_inject1} and \tdx{Pair_inject2}, and equivalently
1.789 +as the elimination rule \tdx{Pair_inject}.
1.790 +
1.791 +The rule \tdx{Pair_neq_0} asserts $\pair{a,b}\neq\emptyset$. This
1.792 +is a property of $\{\{a\},\{a,b\}\}$, and need not hold for other
1.793 +encodings of ordered pairs. The non-standard ordered pairs mentioned below
1.794 +satisfy $\pair{\emptyset;\emptyset}=\emptyset$.
1.795 +
1.796 +The natural deduction rules \tdx{SigmaI} and \tdx{SigmaE}
1.797 +assert that \cdx{Sigma}$(A,B)$ consists of all pairs of the form
1.798 +$\pair{x,y}$, for $x\in A$ and $y\in B(x)$. The rule \tdx{SigmaE2}
1.799 +merely states that $\pair{a,b}\in \texttt{Sigma}(A,B)$ implies $a\in A$ and
1.800 +$b\in B(a)$.
1.801 +
1.802 +In addition, it is possible to use tuples as patterns in abstractions:
1.803 +\begin{center}
1.804 +{\tt\%<$x$,$y$>. $t$} \quad stands for\quad \texttt{split(\%$x$ $y$.\ $t$)}
1.805 +\end{center}
1.806 +Nested patterns are translated recursively:
1.807 +{\tt\%<$x$,$y$,$z$>. $t$} $\leadsto$ {\tt\%<$x$,<$y$,$z$>>. $t$} $\leadsto$
1.808 +\texttt{split(\%$x$.\%<$y$,$z$>. $t$)} $\leadsto$ \texttt{split(\%$x$. split(\%$y$
1.809 + $z$.\ $t$))}. The reverse translation is performed upon printing.
1.810 +\begin{warn}
1.811 + The translation between patterns and \texttt{split} is performed automatically
1.812 + by the parser and printer. Thus the internal and external form of a term
1.813 + may differ, which affects proofs. For example the term {\tt
1.814 + (\%<x,y>.<y,x>)<a,b>} requires the theorem \texttt{split} to rewrite to
1.815 + {\tt<b,a>}.
1.816 +\end{warn}
1.817 +In addition to explicit $\lambda$-abstractions, patterns can be used in any
1.818 +variable binding construct which is internally described by a
1.819 +$\lambda$-abstraction. Here are some important examples:
1.820 +\begin{description}
1.821 +\item[Let:] \texttt{let {\it pattern} = $t$ in $u$}
1.822 +\item[Choice:] \texttt{THE~{\it pattern}~.~$P$}
1.823 +\item[Set operations:] \texttt{UN~{\it pattern}:$A$.~$B$}
1.824 +\item[Comprehension:] \texttt{{\ttlbrace}~{\it pattern}:$A$~.~$P$~{\ttrbrace}}
1.825 +\end{description}
1.826 +
1.827 +
1.828 +%%% domrange.ML
1.829 +
1.830 +\begin{figure}
1.831 +\begin{ttbox}
1.832 +\tdx{domainI} <a,b>: r ==> a : domain(r)
1.833 +\tdx{domainE} [| a : domain(r); !!y. <a,y>: r ==> P |] ==> P
1.834 +\tdx{domain_subset} domain(Sigma(A,B)) <= A
1.835 +
1.836 +\tdx{rangeI} <a,b>: r ==> b : range(r)
1.837 +\tdx{rangeE} [| b : range(r); !!x. <x,b>: r ==> P |] ==> P
1.838 +\tdx{range_subset} range(A*B) <= B
1.839 +
1.840 +\tdx{fieldI1} <a,b>: r ==> a : field(r)
1.841 +\tdx{fieldI2} <a,b>: r ==> b : field(r)
1.842 +\tdx{fieldCI} (~ <c,a>:r ==> <a,b>: r) ==> a : field(r)
1.843 +
1.844 +\tdx{fieldE} [| a : field(r);
1.845 + !!x. <a,x>: r ==> P;
1.846 + !!x. <x,a>: r ==> P
1.847 + |] ==> P
1.848 +
1.849 +\tdx{field_subset} field(A*A) <= A
1.850 +\end{ttbox}
1.851 +\caption{Domain, range and field of a relation} \label{zf-domrange}
1.852 +\end{figure}
1.853 +
1.854 +\begin{figure}
1.855 +\begin{ttbox}
1.856 +\tdx{imageI} [| <a,b>: r; a:A |] ==> b : r``A
1.857 +\tdx{imageE} [| b: r``A; !!x.[| <x,b>: r; x:A |] ==> P |] ==> P
1.858 +
1.859 +\tdx{vimageI} [| <a,b>: r; b:B |] ==> a : r-``B
1.860 +\tdx{vimageE} [| a: r-``B; !!x.[| <a,x>: r; x:B |] ==> P |] ==> P
1.861 +\end{ttbox}
1.862 +\caption{Image and inverse image} \label{zf-domrange2}
1.863 +\end{figure}
1.864 +
1.865 +
1.866 +\subsection{Relations}
1.867 +Figure~\ref{zf-domrange} presents rules involving relations, which are sets
1.868 +of ordered pairs. The converse of a relation~$r$ is the set of all pairs
1.869 +$\pair{y,x}$ such that $\pair{x,y}\in r$; if $r$ is a function, then
1.870 +{\cdx{converse}$(r)$} is its inverse. The rules for the domain
1.871 +operation, namely \tdx{domainI} and~\tdx{domainE}, assert that
1.872 +\cdx{domain}$(r)$ consists of all~$x$ such that $r$ contains
1.873 +some pair of the form~$\pair{x,y}$. The range operation is similar, and
1.874 +the field of a relation is merely the union of its domain and range.
1.875 +
1.876 +Figure~\ref{zf-domrange2} presents rules for images and inverse images.
1.877 +Note that these operations are generalisations of range and domain,
1.878 +respectively. See the file \texttt{ZF/domrange.ML} for derivations of the
1.879 +rules.
1.880 +
1.881 +
1.882 +%%% func.ML
1.883 +
1.884 +\begin{figure}
1.885 +\begin{ttbox}
1.886 +\tdx{fun_is_rel} f: Pi(A,B) ==> f <= Sigma(A,B)
1.887 +
1.888 +\tdx{apply_equality} [| <a,b>: f; f: Pi(A,B) |] ==> f`a = b
1.889 +\tdx{apply_equality2} [| <a,b>: f; <a,c>: f; f: Pi(A,B) |] ==> b=c
1.890 +
1.891 +\tdx{apply_type} [| f: Pi(A,B); a:A |] ==> f`a : B(a)
1.892 +\tdx{apply_Pair} [| f: Pi(A,B); a:A |] ==> <a,f`a>: f
1.893 +\tdx{apply_iff} f: Pi(A,B) ==> <a,b>: f <-> a:A & f`a = b
1.894 +
1.895 +\tdx{fun_extension} [| f : Pi(A,B); g: Pi(A,D);
1.896 + !!x. x:A ==> f`x = g`x |] ==> f=g
1.897 +
1.898 +\tdx{domain_type} [| <a,b> : f; f: Pi(A,B) |] ==> a : A
1.899 +\tdx{range_type} [| <a,b> : f; f: Pi(A,B) |] ==> b : B(a)
1.900 +
1.901 +\tdx{Pi_type} [| f: A->C; !!x. x:A ==> f`x: B(x) |] ==> f: Pi(A,B)
1.902 +\tdx{domain_of_fun} f: Pi(A,B) ==> domain(f)=A
1.903 +\tdx{range_of_fun} f: Pi(A,B) ==> f: A->range(f)
1.904 +
1.905 +\tdx{restrict} a : A ==> restrict(f,A) ` a = f`a
1.906 +\tdx{restrict_type} [| !!x. x:A ==> f`x: B(x) |] ==>
1.907 + restrict(f,A) : Pi(A,B)
1.908 +\end{ttbox}
1.909 +\caption{Functions} \label{zf-func1}
1.910 +\end{figure}
1.911 +
1.912 +
1.913 +\begin{figure}
1.914 +\begin{ttbox}
1.915 +\tdx{lamI} a:A ==> <a,b(a)> : (lam x:A. b(x))
1.916 +\tdx{lamE} [| p: (lam x:A. b(x)); !!x.[| x:A; p=<x,b(x)> |] ==> P
1.917 + |] ==> P
1.918 +
1.919 +\tdx{lam_type} [| !!x. x:A ==> b(x): B(x) |] ==> (lam x:A. b(x)) : Pi(A,B)
1.920 +
1.921 +\tdx{beta} a : A ==> (lam x:A. b(x)) ` a = b(a)
1.922 +\tdx{eta} f : Pi(A,B) ==> (lam x:A. f`x) = f
1.923 +\end{ttbox}
1.924 +\caption{$\lambda$-abstraction} \label{zf-lam}
1.925 +\end{figure}
1.926 +
1.927 +
1.928 +\begin{figure}
1.929 +\begin{ttbox}
1.930 +\tdx{fun_empty} 0: 0->0
1.931 +\tdx{fun_single} {\ttlbrace}<a,b>{\ttrbrace} : {\ttlbrace}a{\ttrbrace} -> {\ttlbrace}b{\ttrbrace}
1.932 +
1.933 +\tdx{fun_disjoint_Un} [| f: A->B; g: C->D; A Int C = 0 |] ==>
1.934 + (f Un g) : (A Un C) -> (B Un D)
1.935 +
1.936 +\tdx{fun_disjoint_apply1} [| a:A; f: A->B; g: C->D; A Int C = 0 |] ==>
1.937 + (f Un g)`a = f`a
1.938 +
1.939 +\tdx{fun_disjoint_apply2} [| c:C; f: A->B; g: C->D; A Int C = 0 |] ==>
1.940 + (f Un g)`c = g`c
1.941 +\end{ttbox}
1.942 +\caption{Constructing functions from smaller sets} \label{zf-func2}
1.943 +\end{figure}
1.944 +
1.945 +
1.946 +\subsection{Functions}
1.947 +Functions, represented by graphs, are notoriously difficult to reason
1.948 +about. The file \texttt{ZF/func.ML} derives many rules, which overlap more
1.949 +than they ought. This section presents the more important rules.
1.950 +
1.951 +Figure~\ref{zf-func1} presents the basic properties of \cdx{Pi}$(A,B)$,
1.952 +the generalized function space. For example, if $f$ is a function and
1.953 +$\pair{a,b}\in f$, then $f`a=b$ (\tdx{apply_equality}). Two functions
1.954 +are equal provided they have equal domains and deliver equals results
1.955 +(\tdx{fun_extension}).
1.956 +
1.957 +By \tdx{Pi_type}, a function typing of the form $f\in A\to C$ can be
1.958 +refined to the dependent typing $f\in\prod@{x\in A}B(x)$, given a suitable
1.959 +family of sets $\{B(x)\}@{x\in A}$. Conversely, by \tdx{range_of_fun},
1.960 +any dependent typing can be flattened to yield a function type of the form
1.961 +$A\to C$; here, $C={\tt range}(f)$.
1.962 +
1.963 +Among the laws for $\lambda$-abstraction, \tdx{lamI} and \tdx{lamE}
1.964 +describe the graph of the generated function, while \tdx{beta} and
1.965 +\tdx{eta} are the standard conversions. We essentially have a
1.966 +dependently-typed $\lambda$-calculus (Fig.\ts\ref{zf-lam}).
1.967 +
1.968 +Figure~\ref{zf-func2} presents some rules that can be used to construct
1.969 +functions explicitly. We start with functions consisting of at most one
1.970 +pair, and may form the union of two functions provided their domains are
1.971 +disjoint.
1.972 +
1.973 +
1.974 +\begin{figure}
1.975 +\begin{ttbox}
1.976 +\tdx{Int_absorb} A Int A = A
1.977 +\tdx{Int_commute} A Int B = B Int A
1.978 +\tdx{Int_assoc} (A Int B) Int C = A Int (B Int C)
1.979 +\tdx{Int_Un_distrib} (A Un B) Int C = (A Int C) Un (B Int C)
1.980 +
1.981 +\tdx{Un_absorb} A Un A = A
1.982 +\tdx{Un_commute} A Un B = B Un A
1.983 +\tdx{Un_assoc} (A Un B) Un C = A Un (B Un C)
1.984 +\tdx{Un_Int_distrib} (A Int B) Un C = (A Un C) Int (B Un C)
1.985 +
1.986 +\tdx{Diff_cancel} A-A = 0
1.987 +\tdx{Diff_disjoint} A Int (B-A) = 0
1.988 +\tdx{Diff_partition} A<=B ==> A Un (B-A) = B
1.989 +\tdx{double_complement} [| A<=B; B<= C |] ==> (B - (C-A)) = A
1.990 +\tdx{Diff_Un} A - (B Un C) = (A-B) Int (A-C)
1.991 +\tdx{Diff_Int} A - (B Int C) = (A-B) Un (A-C)
1.992 +
1.993 +\tdx{Union_Un_distrib} Union(A Un B) = Union(A) Un Union(B)
1.994 +\tdx{Inter_Un_distrib} [| a:A; b:B |] ==>
1.995 + Inter(A Un B) = Inter(A) Int Inter(B)
1.996 +
1.997 +\tdx{Int_Union_RepFun} A Int Union(B) = (UN C:B. A Int C)
1.998 +
1.999 +\tdx{Un_Inter_RepFun} b:B ==>
1.1000 + A Un Inter(B) = (INT C:B. A Un C)
1.1001 +
1.1002 +\tdx{SUM_Un_distrib1} (SUM x:A Un B. C(x)) =
1.1003 + (SUM x:A. C(x)) Un (SUM x:B. C(x))
1.1004 +
1.1005 +\tdx{SUM_Un_distrib2} (SUM x:C. A(x) Un B(x)) =
1.1006 + (SUM x:C. A(x)) Un (SUM x:C. B(x))
1.1007 +
1.1008 +\tdx{SUM_Int_distrib1} (SUM x:A Int B. C(x)) =
1.1009 + (SUM x:A. C(x)) Int (SUM x:B. C(x))
1.1010 +
1.1011 +\tdx{SUM_Int_distrib2} (SUM x:C. A(x) Int B(x)) =
1.1012 + (SUM x:C. A(x)) Int (SUM x:C. B(x))
1.1013 +\end{ttbox}
1.1014 +\caption{Equalities} \label{zf-equalities}
1.1015 +\end{figure}
1.1016 +
1.1017 +
1.1018 +\begin{figure}
1.1019 +%\begin{constants}
1.1020 +% \cdx{1} & $i$ & & $\{\emptyset\}$ \\
1.1021 +% \cdx{bool} & $i$ & & the set $\{\emptyset,1\}$ \\
1.1022 +% \cdx{cond} & $[i,i,i]\To i$ & & conditional for \texttt{bool} \\
1.1023 +% \cdx{not} & $i\To i$ & & negation for \texttt{bool} \\
1.1024 +% \sdx{and} & $[i,i]\To i$ & Left 70 & conjunction for \texttt{bool} \\
1.1025 +% \sdx{or} & $[i,i]\To i$ & Left 65 & disjunction for \texttt{bool} \\
1.1026 +% \sdx{xor} & $[i,i]\To i$ & Left 65 & exclusive-or for \texttt{bool}
1.1027 +%\end{constants}
1.1028 +%
1.1029 +\begin{ttbox}
1.1030 +\tdx{bool_def} bool == {\ttlbrace}0,1{\ttrbrace}
1.1031 +\tdx{cond_def} cond(b,c,d) == if b=1 then c else d
1.1032 +\tdx{not_def} not(b) == cond(b,0,1)
1.1033 +\tdx{and_def} a and b == cond(a,b,0)
1.1034 +\tdx{or_def} a or b == cond(a,1,b)
1.1035 +\tdx{xor_def} a xor b == cond(a,not(b),b)
1.1036 +
1.1037 +\tdx{bool_1I} 1 : bool
1.1038 +\tdx{bool_0I} 0 : bool
1.1039 +\tdx{boolE} [| c: bool; c=1 ==> P; c=0 ==> P |] ==> P
1.1040 +\tdx{cond_1} cond(1,c,d) = c
1.1041 +\tdx{cond_0} cond(0,c,d) = d
1.1042 +\end{ttbox}
1.1043 +\caption{The booleans} \label{zf-bool}
1.1044 +\end{figure}
1.1045 +
1.1046 +
1.1047 +\section{Further developments}
1.1048 +The next group of developments is complex and extensive, and only
1.1049 +highlights can be covered here. It involves many theories and ML files of
1.1050 +proofs.
1.1051 +
1.1052 +Figure~\ref{zf-equalities} presents commutative, associative, distributive,
1.1053 +and idempotency laws of union and intersection, along with other equations.
1.1054 +See file \texttt{ZF/equalities.ML}.
1.1055 +
1.1056 +Theory \thydx{Bool} defines $\{0,1\}$ as a set of booleans, with the usual
1.1057 +operators including a conditional (Fig.\ts\ref{zf-bool}). Although {\ZF} is a
1.1058 +first-order theory, you can obtain the effect of higher-order logic using
1.1059 +\texttt{bool}-valued functions, for example. The constant~\texttt{1} is
1.1060 +translated to \texttt{succ(0)}.
1.1061 +
1.1062 +\begin{figure}
1.1063 +\index{*"+ symbol}
1.1064 +\begin{constants}
1.1065 + \it symbol & \it meta-type & \it priority & \it description \\
1.1066 + \tt + & $[i,i]\To i$ & Right 65 & disjoint union operator\\
1.1067 + \cdx{Inl}~~\cdx{Inr} & $i\To i$ & & injections\\
1.1068 + \cdx{case} & $[i\To i,i\To i, i]\To i$ & & conditional for $A+B$
1.1069 +\end{constants}
1.1070 +\begin{ttbox}
1.1071 +\tdx{sum_def} A+B == {\ttlbrace}0{\ttrbrace}*A Un {\ttlbrace}1{\ttrbrace}*B
1.1072 +\tdx{Inl_def} Inl(a) == <0,a>
1.1073 +\tdx{Inr_def} Inr(b) == <1,b>
1.1074 +\tdx{case_def} case(c,d,u) == split(\%y z. cond(y, d(z), c(z)), u)
1.1075 +
1.1076 +\tdx{sum_InlI} a : A ==> Inl(a) : A+B
1.1077 +\tdx{sum_InrI} b : B ==> Inr(b) : A+B
1.1078 +
1.1079 +\tdx{Inl_inject} Inl(a)=Inl(b) ==> a=b
1.1080 +\tdx{Inr_inject} Inr(a)=Inr(b) ==> a=b
1.1081 +\tdx{Inl_neq_Inr} Inl(a)=Inr(b) ==> P
1.1082 +
1.1083 +\tdx{sumE2} u: A+B ==> (EX x. x:A & u=Inl(x)) | (EX y. y:B & u=Inr(y))
1.1084 +
1.1085 +\tdx{case_Inl} case(c,d,Inl(a)) = c(a)
1.1086 +\tdx{case_Inr} case(c,d,Inr(b)) = d(b)
1.1087 +\end{ttbox}
1.1088 +\caption{Disjoint unions} \label{zf-sum}
1.1089 +\end{figure}
1.1090 +
1.1091 +
1.1092 +Theory \thydx{Sum} defines the disjoint union of two sets, with
1.1093 +injections and a case analysis operator (Fig.\ts\ref{zf-sum}). Disjoint
1.1094 +unions play a role in datatype definitions, particularly when there is
1.1095 +mutual recursion~\cite{paulson-set-II}.
1.1096 +
1.1097 +\begin{figure}
1.1098 +\begin{ttbox}
1.1099 +\tdx{QPair_def} <a;b> == a+b
1.1100 +\tdx{qsplit_def} qsplit(c,p) == THE y. EX a b. p=<a;b> & y=c(a,b)
1.1101 +\tdx{qfsplit_def} qfsplit(R,z) == EX x y. z=<x;y> & R(x,y)
1.1102 +\tdx{qconverse_def} qconverse(r) == {\ttlbrace}z. w:r, EX x y. w=<x;y> & z=<y;x>{\ttrbrace}
1.1103 +\tdx{QSigma_def} QSigma(A,B) == UN x:A. UN y:B(x). {\ttlbrace}<x;y>{\ttrbrace}
1.1104 +
1.1105 +\tdx{qsum_def} A <+> B == ({\ttlbrace}0{\ttrbrace} <*> A) Un ({\ttlbrace}1{\ttrbrace} <*> B)
1.1106 +\tdx{QInl_def} QInl(a) == <0;a>
1.1107 +\tdx{QInr_def} QInr(b) == <1;b>
1.1108 +\tdx{qcase_def} qcase(c,d) == qsplit(\%y z. cond(y, d(z), c(z)))
1.1109 +\end{ttbox}
1.1110 +\caption{Non-standard pairs, products and sums} \label{zf-qpair}
1.1111 +\end{figure}
1.1112 +
1.1113 +Theory \thydx{QPair} defines a notion of ordered pair that admits
1.1114 +non-well-founded tupling (Fig.\ts\ref{zf-qpair}). Such pairs are written
1.1115 +{\tt<$a$;$b$>}. It also defines the eliminator \cdx{qsplit}, the
1.1116 +converse operator \cdx{qconverse}, and the summation operator
1.1117 +\cdx{QSigma}. These are completely analogous to the corresponding
1.1118 +versions for standard ordered pairs. The theory goes on to define a
1.1119 +non-standard notion of disjoint sum using non-standard pairs. All of these
1.1120 +concepts satisfy the same properties as their standard counterparts; in
1.1121 +addition, {\tt<$a$;$b$>} is continuous. The theory supports coinductive
1.1122 +definitions, for example of infinite lists~\cite{paulson-final}.
1.1123 +
1.1124 +\begin{figure}
1.1125 +\begin{ttbox}
1.1126 +\tdx{bnd_mono_def} bnd_mono(D,h) ==
1.1127 + h(D)<=D & (ALL W X. W<=X --> X<=D --> h(W) <= h(X))
1.1128 +
1.1129 +\tdx{lfp_def} lfp(D,h) == Inter({\ttlbrace}X: Pow(D). h(X) <= X{\ttrbrace})
1.1130 +\tdx{gfp_def} gfp(D,h) == Union({\ttlbrace}X: Pow(D). X <= h(X){\ttrbrace})
1.1131 +
1.1132 +
1.1133 +\tdx{lfp_lowerbound} [| h(A) <= A; A<=D |] ==> lfp(D,h) <= A
1.1134 +
1.1135 +\tdx{lfp_subset} lfp(D,h) <= D
1.1136 +
1.1137 +\tdx{lfp_greatest} [| bnd_mono(D,h);
1.1138 + !!X. [| h(X) <= X; X<=D |] ==> A<=X
1.1139 + |] ==> A <= lfp(D,h)
1.1140 +
1.1141 +\tdx{lfp_Tarski} bnd_mono(D,h) ==> lfp(D,h) = h(lfp(D,h))
1.1142 +
1.1143 +\tdx{induct} [| a : lfp(D,h); bnd_mono(D,h);
1.1144 + !!x. x : h(Collect(lfp(D,h),P)) ==> P(x)
1.1145 + |] ==> P(a)
1.1146 +
1.1147 +\tdx{lfp_mono} [| bnd_mono(D,h); bnd_mono(E,i);
1.1148 + !!X. X<=D ==> h(X) <= i(X)
1.1149 + |] ==> lfp(D,h) <= lfp(E,i)
1.1150 +
1.1151 +\tdx{gfp_upperbound} [| A <= h(A); A<=D |] ==> A <= gfp(D,h)
1.1152 +
1.1153 +\tdx{gfp_subset} gfp(D,h) <= D
1.1154 +
1.1155 +\tdx{gfp_least} [| bnd_mono(D,h);
1.1156 + !!X. [| X <= h(X); X<=D |] ==> X<=A
1.1157 + |] ==> gfp(D,h) <= A
1.1158 +
1.1159 +\tdx{gfp_Tarski} bnd_mono(D,h) ==> gfp(D,h) = h(gfp(D,h))
1.1160 +
1.1161 +\tdx{coinduct} [| bnd_mono(D,h); a: X; X <= h(X Un gfp(D,h)); X <= D
1.1162 + |] ==> a : gfp(D,h)
1.1163 +
1.1164 +\tdx{gfp_mono} [| bnd_mono(D,h); D <= E;
1.1165 + !!X. X<=D ==> h(X) <= i(X)
1.1166 + |] ==> gfp(D,h) <= gfp(E,i)
1.1167 +\end{ttbox}
1.1168 +\caption{Least and greatest fixedpoints} \label{zf-fixedpt}
1.1169 +\end{figure}
1.1170 +
1.1171 +The Knaster-Tarski Theorem states that every monotone function over a
1.1172 +complete lattice has a fixedpoint. Theory \thydx{Fixedpt} proves the
1.1173 +Theorem only for a particular lattice, namely the lattice of subsets of a
1.1174 +set (Fig.\ts\ref{zf-fixedpt}). The theory defines least and greatest
1.1175 +fixedpoint operators with corresponding induction and coinduction rules.
1.1176 +These are essential to many definitions that follow, including the natural
1.1177 +numbers and the transitive closure operator. The (co)inductive definition
1.1178 +package also uses the fixedpoint operators~\cite{paulson-CADE}. See
1.1179 +Davey and Priestley~\cite{davey&priestley} for more on the Knaster-Tarski
1.1180 +Theorem and my paper~\cite{paulson-set-II} for discussion of the Isabelle
1.1181 +proofs.
1.1182 +
1.1183 +Monotonicity properties are proved for most of the set-forming operations:
1.1184 +union, intersection, Cartesian product, image, domain, range, etc. These
1.1185 +are useful for applying the Knaster-Tarski Fixedpoint Theorem. The proofs
1.1186 +themselves are trivial applications of Isabelle's classical reasoner. See
1.1187 +file \texttt{ZF/mono.ML}.
1.1188 +
1.1189 +
1.1190 +\begin{figure}
1.1191 +\begin{constants}
1.1192 + \it symbol & \it meta-type & \it priority & \it description \\
1.1193 + \sdx{O} & $[i,i]\To i$ & Right 60 & composition ($\circ$) \\
1.1194 + \cdx{id} & $i\To i$ & & identity function \\
1.1195 + \cdx{inj} & $[i,i]\To i$ & & injective function space\\
1.1196 + \cdx{surj} & $[i,i]\To i$ & & surjective function space\\
1.1197 + \cdx{bij} & $[i,i]\To i$ & & bijective function space
1.1198 +\end{constants}
1.1199 +
1.1200 +\begin{ttbox}
1.1201 +\tdx{comp_def} r O s == {\ttlbrace}xz : domain(s)*range(r) .
1.1202 + EX x y z. xz=<x,z> & <x,y>:s & <y,z>:r{\ttrbrace}
1.1203 +\tdx{id_def} id(A) == (lam x:A. x)
1.1204 +\tdx{inj_def} inj(A,B) == {\ttlbrace} f: A->B. ALL w:A. ALL x:A. f`w=f`x --> w=x {\ttrbrace}
1.1205 +\tdx{surj_def} surj(A,B) == {\ttlbrace} f: A->B . ALL y:B. EX x:A. f`x=y {\ttrbrace}
1.1206 +\tdx{bij_def} bij(A,B) == inj(A,B) Int surj(A,B)
1.1207 +
1.1208 +
1.1209 +\tdx{left_inverse} [| f: inj(A,B); a: A |] ==> converse(f)`(f`a) = a
1.1210 +\tdx{right_inverse} [| f: inj(A,B); b: range(f) |] ==>
1.1211 + f`(converse(f)`b) = b
1.1212 +
1.1213 +\tdx{inj_converse_inj} f: inj(A,B) ==> converse(f): inj(range(f), A)
1.1214 +\tdx{bij_converse_bij} f: bij(A,B) ==> converse(f): bij(B,A)
1.1215 +
1.1216 +\tdx{comp_type} [| s<=A*B; r<=B*C |] ==> (r O s) <= A*C
1.1217 +\tdx{comp_assoc} (r O s) O t = r O (s O t)
1.1218 +
1.1219 +\tdx{left_comp_id} r<=A*B ==> id(B) O r = r
1.1220 +\tdx{right_comp_id} r<=A*B ==> r O id(A) = r
1.1221 +
1.1222 +\tdx{comp_func} [| g:A->B; f:B->C |] ==> (f O g):A->C
1.1223 +\tdx{comp_func_apply} [| g:A->B; f:B->C; a:A |] ==> (f O g)`a = f`(g`a)
1.1224 +
1.1225 +\tdx{comp_inj} [| g:inj(A,B); f:inj(B,C) |] ==> (f O g):inj(A,C)
1.1226 +\tdx{comp_surj} [| g:surj(A,B); f:surj(B,C) |] ==> (f O g):surj(A,C)
1.1227 +\tdx{comp_bij} [| g:bij(A,B); f:bij(B,C) |] ==> (f O g):bij(A,C)
1.1228 +
1.1229 +\tdx{left_comp_inverse} f: inj(A,B) ==> converse(f) O f = id(A)
1.1230 +\tdx{right_comp_inverse} f: surj(A,B) ==> f O converse(f) = id(B)
1.1231 +
1.1232 +\tdx{bij_disjoint_Un}
1.1233 + [| f: bij(A,B); g: bij(C,D); A Int C = 0; B Int D = 0 |] ==>
1.1234 + (f Un g) : bij(A Un C, B Un D)
1.1235 +
1.1236 +\tdx{restrict_bij} [| f:inj(A,B); C<=A |] ==> restrict(f,C): bij(C, f``C)
1.1237 +\end{ttbox}
1.1238 +\caption{Permutations} \label{zf-perm}
1.1239 +\end{figure}
1.1240 +
1.1241 +The theory \thydx{Perm} is concerned with permutations (bijections) and
1.1242 +related concepts. These include composition of relations, the identity
1.1243 +relation, and three specialized function spaces: injective, surjective and
1.1244 +bijective. Figure~\ref{zf-perm} displays many of their properties that
1.1245 +have been proved. These results are fundamental to a treatment of
1.1246 +equipollence and cardinality.
1.1247 +
1.1248 +\begin{figure}\small
1.1249 +\index{#*@{\tt\#*} symbol}
1.1250 +\index{*div symbol}
1.1251 +\index{*mod symbol}
1.1252 +\index{#+@{\tt\#+} symbol}
1.1253 +\index{#-@{\tt\#-} symbol}
1.1254 +\begin{constants}
1.1255 + \it symbol & \it meta-type & \it priority & \it description \\
1.1256 + \cdx{nat} & $i$ & & set of natural numbers \\
1.1257 + \cdx{nat_case}& $[i,i\To i,i]\To i$ & & conditional for $nat$\\
1.1258 + \tt \#* & $[i,i]\To i$ & Left 70 & multiplication \\
1.1259 + \tt div & $[i,i]\To i$ & Left 70 & division\\
1.1260 + \tt mod & $[i,i]\To i$ & Left 70 & modulus\\
1.1261 + \tt \#+ & $[i,i]\To i$ & Left 65 & addition\\
1.1262 + \tt \#- & $[i,i]\To i$ & Left 65 & subtraction
1.1263 +\end{constants}
1.1264 +
1.1265 +\begin{ttbox}
1.1266 +\tdx{nat_def} nat == lfp(lam r: Pow(Inf). {\ttlbrace}0{\ttrbrace} Un {\ttlbrace}succ(x). x:r{\ttrbrace}
1.1267 +
1.1268 +\tdx{mod_def} m mod n == transrec(m, \%j f. if j:n then j else f`(j#-n))
1.1269 +\tdx{div_def} m div n == transrec(m, \%j f. if j:n then 0 else succ(f`(j#-n)))
1.1270 +
1.1271 +\tdx{nat_case_def} nat_case(a,b,k) ==
1.1272 + THE y. k=0 & y=a | (EX x. k=succ(x) & y=b(x))
1.1273 +
1.1274 +\tdx{nat_0I} 0 : nat
1.1275 +\tdx{nat_succI} n : nat ==> succ(n) : nat
1.1276 +
1.1277 +\tdx{nat_induct}
1.1278 + [| n: nat; P(0); !!x. [| x: nat; P(x) |] ==> P(succ(x))
1.1279 + |] ==> P(n)
1.1280 +
1.1281 +\tdx{nat_case_0} nat_case(a,b,0) = a
1.1282 +\tdx{nat_case_succ} nat_case(a,b,succ(m)) = b(m)
1.1283 +
1.1284 +\tdx{add_0} 0 #+ n = n
1.1285 +\tdx{add_succ} succ(m) #+ n = succ(m #+ n)
1.1286 +
1.1287 +\tdx{mult_type} [| m:nat; n:nat |] ==> m #* n : nat
1.1288 +\tdx{mult_0} 0 #* n = 0
1.1289 +\tdx{mult_succ} succ(m) #* n = n #+ (m #* n)
1.1290 +\tdx{mult_commute} [| m:nat; n:nat |] ==> m #* n = n #* m
1.1291 +\tdx{add_mult_dist} [| m:nat; k:nat |] ==> (m #+ n) #* k = (m #* k){\thinspace}#+{\thinspace}(n #* k)
1.1292 +\tdx{mult_assoc}
1.1293 + [| m:nat; n:nat; k:nat |] ==> (m #* n) #* k = m #* (n #* k)
1.1294 +\tdx{mod_quo_equality}
1.1295 + [| 0:n; m:nat; n:nat |] ==> (m div n)#*n #+ m mod n = m
1.1296 +\end{ttbox}
1.1297 +\caption{The natural numbers} \label{zf-nat}
1.1298 +\end{figure}
1.1299 +
1.1300 +Theory \thydx{Nat} defines the natural numbers and mathematical
1.1301 +induction, along with a case analysis operator. The set of natural
1.1302 +numbers, here called \texttt{nat}, is known in set theory as the ordinal~$\omega$.
1.1303 +
1.1304 +Theory \thydx{Arith} develops arithmetic on the natural numbers
1.1305 +(Fig.\ts\ref{zf-nat}). Addition, multiplication and subtraction are defined
1.1306 +by primitive recursion. Division and remainder are defined by repeated
1.1307 +subtraction, which requires well-founded recursion; the termination argument
1.1308 +relies on the divisor's being non-zero. Many properties are proved:
1.1309 +commutative, associative and distributive laws, identity and cancellation
1.1310 +laws, etc. The most interesting result is perhaps the theorem $a \bmod b +
1.1311 +(a/b)\times b = a$.
1.1312 +
1.1313 +Theory \thydx{Univ} defines a `universe' $\texttt{univ}(A)$, which is used by
1.1314 +the datatype package. This set contains $A$ and the
1.1315 +natural numbers. Vitally, it is closed under finite products: ${\tt
1.1316 + univ}(A)\times{\tt univ}(A)\subseteq{\tt univ}(A)$. This theory also
1.1317 +defines the cumulative hierarchy of axiomatic set theory, which
1.1318 +traditionally is written $V@\alpha$ for an ordinal~$\alpha$. The
1.1319 +`universe' is a simple generalization of~$V@\omega$.
1.1320 +
1.1321 +Theory \thydx{QUniv} defines a `universe' ${\tt quniv}(A)$, which is used by
1.1322 +the datatype package to construct codatatypes such as streams. It is
1.1323 +analogous to ${\tt univ}(A)$ (and is defined in terms of it) but is closed
1.1324 +under the non-standard product and sum.
1.1325 +
1.1326 +Theory \texttt{Finite} (Figure~\ref{zf-fin}) defines the finite set operator;
1.1327 +${\tt Fin}(A)$ is the set of all finite sets over~$A$. The theory employs
1.1328 +Isabelle's inductive definition package, which proves various rules
1.1329 +automatically. The induction rule shown is stronger than the one proved by
1.1330 +the package. The theory also defines the set of all finite functions
1.1331 +between two given sets.
1.1332 +
1.1333 +\begin{figure}
1.1334 +\begin{ttbox}
1.1335 +\tdx{Fin.emptyI} 0 : Fin(A)
1.1336 +\tdx{Fin.consI} [| a: A; b: Fin(A) |] ==> cons(a,b) : Fin(A)
1.1337 +
1.1338 +\tdx{Fin_induct}
1.1339 + [| b: Fin(A);
1.1340 + P(0);
1.1341 + !!x y. [| x: A; y: Fin(A); x~:y; P(y) |] ==> P(cons(x,y))
1.1342 + |] ==> P(b)
1.1343 +
1.1344 +\tdx{Fin_mono} A<=B ==> Fin(A) <= Fin(B)
1.1345 +\tdx{Fin_UnI} [| b: Fin(A); c: Fin(A) |] ==> b Un c : Fin(A)
1.1346 +\tdx{Fin_UnionI} C : Fin(Fin(A)) ==> Union(C) : Fin(A)
1.1347 +\tdx{Fin_subset} [| c<=b; b: Fin(A) |] ==> c: Fin(A)
1.1348 +\end{ttbox}
1.1349 +\caption{The finite set operator} \label{zf-fin}
1.1350 +\end{figure}
1.1351 +
1.1352 +\begin{figure}
1.1353 +\begin{constants}
1.1354 + \it symbol & \it meta-type & \it priority & \it description \\
1.1355 + \cdx{list} & $i\To i$ && lists over some set\\
1.1356 + \cdx{list_case} & $[i, [i,i]\To i, i] \To i$ && conditional for $list(A)$ \\
1.1357 + \cdx{map} & $[i\To i, i] \To i$ & & mapping functional\\
1.1358 + \cdx{length} & $i\To i$ & & length of a list\\
1.1359 + \cdx{rev} & $i\To i$ & & reverse of a list\\
1.1360 + \tt \at & $[i,i]\To i$ & Right 60 & append for lists\\
1.1361 + \cdx{flat} & $i\To i$ & & append of list of lists
1.1362 +\end{constants}
1.1363 +
1.1364 +\underscoreon %%because @ is used here
1.1365 +\begin{ttbox}
1.1366 +\tdx{NilI} Nil : list(A)
1.1367 +\tdx{ConsI} [| a: A; l: list(A) |] ==> Cons(a,l) : list(A)
1.1368 +
1.1369 +\tdx{List.induct}
1.1370 + [| l: list(A);
1.1371 + P(Nil);
1.1372 + !!x y. [| x: A; y: list(A); P(y) |] ==> P(Cons(x,y))
1.1373 + |] ==> P(l)
1.1374 +
1.1375 +\tdx{Cons_iff} Cons(a,l)=Cons(a',l') <-> a=a' & l=l'
1.1376 +\tdx{Nil_Cons_iff} ~ Nil=Cons(a,l)
1.1377 +
1.1378 +\tdx{list_mono} A<=B ==> list(A) <= list(B)
1.1379 +
1.1380 +\tdx{map_ident} l: list(A) ==> map(\%u. u, l) = l
1.1381 +\tdx{map_compose} l: list(A) ==> map(h, map(j,l)) = map(\%u. h(j(u)), l)
1.1382 +\tdx{map_app_distrib} xs: list(A) ==> map(h, xs@ys) = map(h,xs) @ map(h,ys)
1.1383 +\tdx{map_type}
1.1384 + [| l: list(A); !!x. x: A ==> h(x): B |] ==> map(h,l) : list(B)
1.1385 +\tdx{map_flat}
1.1386 + ls: list(list(A)) ==> map(h, flat(ls)) = flat(map(map(h),ls))
1.1387 +\end{ttbox}
1.1388 +\caption{Lists} \label{zf-list}
1.1389 +\end{figure}
1.1390 +
1.1391 +
1.1392 +Figure~\ref{zf-list} presents the set of lists over~$A$, ${\tt list}(A)$. The
1.1393 +definition employs Isabelle's datatype package, which defines the introduction
1.1394 +and induction rules automatically, as well as the constructors, case operator
1.1395 +(\verb|list_case|) and recursion operator. The theory then defines the usual
1.1396 +list functions by primitive recursion. See theory \texttt{List}.
1.1397 +
1.1398 +
1.1399 +\section{Simplification and classical reasoning}
1.1400 +
1.1401 +{\ZF} inherits simplification from {\FOL} but adopts it for set theory. The
1.1402 +extraction of rewrite rules takes the {\ZF} primitives into account. It can
1.1403 +strip bounded universal quantifiers from a formula; for example, ${\forall
1.1404 + x\in A. f(x)=g(x)}$ yields the conditional rewrite rule $x\in A \Imp
1.1405 +f(x)=g(x)$. Given $a\in\{x\in A. P(x)\}$ it extracts rewrite rules from $a\in
1.1406 +A$ and~$P(a)$. It can also break down $a\in A\int B$ and $a\in A-B$.
1.1407 +
1.1408 +Simplification tactics tactics such as \texttt{Asm_simp_tac} and
1.1409 +\texttt{Full_simp_tac} use the default simpset (\texttt{simpset()}), which
1.1410 +works for most purposes. A small simplification set for set theory is
1.1411 +called~\ttindexbold{ZF_ss}, and you can even use \ttindex{FOL_ss} as a minimal
1.1412 +starting point. \texttt{ZF_ss} contains congruence rules for all the binding
1.1413 +operators of {\ZF}\@. It contains all the conversion rules, such as
1.1414 +\texttt{fst} and \texttt{snd}, as well as the rewrites shown in
1.1415 +Fig.\ts\ref{zf-simpdata}. See the file \texttt{ZF/simpdata.ML} for a fuller
1.1416 +list.
1.1417 +
1.1418 +As for the classical reasoner, tactics such as \texttt{Blast_tac} and {\tt
1.1419 + Best_tac} refer to the default claset (\texttt{claset()}). This works for
1.1420 +most purposes. Named clasets include \ttindexbold{ZF_cs} (basic set theory)
1.1421 +and \ttindexbold{le_cs} (useful for reasoning about the relations $<$ and
1.1422 +$\le$). You can use \ttindex{FOL_cs} as a minimal basis for building your own
1.1423 +clasets. See \iflabelundefined{chap:classical}{the {\em Reference Manual\/}}%
1.1424 +{Chap.\ts\ref{chap:classical}} for more discussion of classical proof methods.
1.1425 +
1.1426 +
1.1427 +\begin{figure}
1.1428 +\begin{eqnarray*}
1.1429 + a\in \emptyset & \bimp & \bot\\
1.1430 + a \in A \un B & \bimp & a\in A \disj a\in B\\
1.1431 + a \in A \int B & \bimp & a\in A \conj a\in B\\
1.1432 + a \in A-B & \bimp & a\in A \conj \neg (a\in B)\\
1.1433 + \pair{a,b}\in {\tt Sigma}(A,B)
1.1434 + & \bimp & a\in A \conj b\in B(a)\\
1.1435 + a \in {\tt Collect}(A,P) & \bimp & a\in A \conj P(a)\\
1.1436 + (\forall x \in \emptyset. P(x)) & \bimp & \top\\
1.1437 + (\forall x \in A. \top) & \bimp & \top
1.1438 +\end{eqnarray*}
1.1439 +\caption{Some rewrite rules for set theory} \label{zf-simpdata}
1.1440 +\end{figure}
1.1441 +
1.1442 +
1.1443 +\section{Datatype definitions}
1.1444 +\label{sec:ZF:datatype}
1.1445 +\index{*datatype|(}
1.1446 +
1.1447 +The \ttindex{datatype} definition package of \ZF\ constructs inductive
1.1448 +datatypes similar to those of \ML. It can also construct coinductive
1.1449 +datatypes (codatatypes), which are non-well-founded structures such as
1.1450 +streams. It defines the set using a fixed-point construction and proves
1.1451 +induction rules, as well as theorems for recursion and case combinators. It
1.1452 +supplies mechanisms for reasoning about freeness. The datatype package can
1.1453 +handle both mutual and indirect recursion.
1.1454 +
1.1455 +
1.1456 +\subsection{Basics}
1.1457 +\label{subsec:datatype:basics}
1.1458 +
1.1459 +A \texttt{datatype} definition has the following form:
1.1460 +\[
1.1461 +\begin{array}{llcl}
1.1462 +\mathtt{datatype} & t@1(A@1,\ldots,A@h) & = &
1.1463 + constructor^1@1 ~\mid~ \ldots ~\mid~ constructor^1@{k@1} \\
1.1464 + & & \vdots \\
1.1465 +\mathtt{and} & t@n(A@1,\ldots,A@h) & = &
1.1466 + constructor^n@1~ ~\mid~ \ldots ~\mid~ constructor^n@{k@n}
1.1467 +\end{array}
1.1468 +\]
1.1469 +Here $t@1$, \ldots,~$t@n$ are identifiers and $A@1$, \ldots,~$A@h$ are
1.1470 +variables: the datatype's parameters. Each constructor specification has the
1.1471 +form \dquotesoff
1.1472 +\[ C \hbox{\tt~( } \hbox{\tt"} x@1 \hbox{\tt:} T@1 \hbox{\tt"},\;
1.1473 + \ldots,\;
1.1474 + \hbox{\tt"} x@m \hbox{\tt:} T@m \hbox{\tt"}
1.1475 + \hbox{\tt~)}
1.1476 +\]
1.1477 +Here $C$ is the constructor name, and variables $x@1$, \ldots,~$x@m$ are the
1.1478 +constructor arguments, belonging to the sets $T@1$, \ldots, $T@m$,
1.1479 +respectively. Typically each $T@j$ is either a constant set, a datatype
1.1480 +parameter (one of $A@1$, \ldots, $A@h$) or a recursive occurrence of one of
1.1481 +the datatypes, say $t@i(A@1,\ldots,A@h)$. More complex possibilities exist,
1.1482 +but they are much harder to realize. Often, additional information must be
1.1483 +supplied in the form of theorems.
1.1484 +
1.1485 +A datatype can occur recursively as the argument of some function~$F$. This
1.1486 +is called a {\em nested} (or \emph{indirect}) occurrence. It is only allowed
1.1487 +if the datatype package is given a theorem asserting that $F$ is monotonic.
1.1488 +If the datatype has indirect occurrences, then Isabelle/ZF does not support
1.1489 +recursive function definitions.
1.1490 +
1.1491 +A simple example of a datatype is \texttt{list}, which is built-in, and is
1.1492 +defined by
1.1493 +\begin{ttbox}
1.1494 +consts list :: i=>i
1.1495 +datatype "list(A)" = Nil | Cons ("a:A", "l: list(A)")
1.1496 +\end{ttbox}
1.1497 +Note that the datatype operator must be declared as a constant first.
1.1498 +However, the package declares the constructors. Here, \texttt{Nil} gets type
1.1499 +$i$ and \texttt{Cons} gets type $[i,i]\To i$.
1.1500 +
1.1501 +Trees and forests can be modelled by the mutually recursive datatype
1.1502 +definition
1.1503 +\begin{ttbox}
1.1504 +consts tree, forest, tree_forest :: i=>i
1.1505 +datatype "tree(A)" = Tcons ("a: A", "f: forest(A)")
1.1506 +and "forest(A)" = Fnil | Fcons ("t: tree(A)", "f: forest(A)")
1.1507 +\end{ttbox}
1.1508 +Here $\texttt{tree}(A)$ is the set of trees over $A$, $\texttt{forest}(A)$ is
1.1509 +the set of forests over $A$, and $\texttt{tree_forest}(A)$ is the union of
1.1510 +the previous two sets. All three operators must be declared first.
1.1511 +
1.1512 +The datatype \texttt{term}, which is defined by
1.1513 +\begin{ttbox}
1.1514 +consts term :: i=>i
1.1515 +datatype "term(A)" = Apply ("a: A", "l: list(term(A))")
1.1516 + monos "[list_mono]"
1.1517 +\end{ttbox}
1.1518 +is an example of nested recursion. (The theorem \texttt{list_mono} is proved
1.1519 +in file \texttt{List.ML}, and the \texttt{term} example is devaloped in theory
1.1520 +\thydx{ex/Term}.)
1.1521 +
1.1522 +\subsubsection{Freeness of the constructors}
1.1523 +
1.1524 +Constructors satisfy {\em freeness} properties. Constructions are distinct,
1.1525 +for example $\texttt{Nil}\not=\texttt{Cons}(a,l)$, and they are injective, for
1.1526 +example $\texttt{Cons}(a,l)=\texttt{Cons}(a',l') \bimp a=a' \conj l=l'$.
1.1527 +Because the number of freeness is quadratic in the number of constructors, the
1.1528 +datatype package does not prove them, but instead provides several means of
1.1529 +proving them dynamically. For the \texttt{list} datatype, freeness reasoning
1.1530 +can be done in two ways: by simplifying with the theorems
1.1531 +\texttt{list.free_iffs} or by invoking the classical reasoner with
1.1532 +\texttt{list.free_SEs} as safe elimination rules. Occasionally this exposes
1.1533 +the underlying representation of some constructor, which can be rectified
1.1534 +using the command \hbox{\tt fold_tac list.con_defs}.
1.1535 +
1.1536 +\subsubsection{Structural induction}
1.1537 +
1.1538 +The datatype package also provides structural induction rules. For datatypes
1.1539 +without mutual or nested recursion, the rule has the form exemplified by
1.1540 +\texttt{list.induct} in Fig.\ts\ref{zf-list}. For mutually recursive
1.1541 +datatypes, the induction rule is supplied in two forms. Consider datatype
1.1542 +\texttt{TF}. The rule \texttt{tree_forest.induct} performs induction over a
1.1543 +single predicate~\texttt{P}, which is presumed to be defined for both trees
1.1544 +and forests:
1.1545 +\begin{ttbox}
1.1546 +[| x : tree_forest(A);
1.1547 + !!a f. [| a : A; f : forest(A); P(f) |] ==> P(Tcons(a, f)); P(Fnil);
1.1548 + !!f t. [| t : tree(A); P(t); f : forest(A); P(f) |]
1.1549 + ==> P(Fcons(t, f))
1.1550 +|] ==> P(x)
1.1551 +\end{ttbox}
1.1552 +The rule \texttt{tree_forest.mutual_induct} performs induction over two
1.1553 +distinct predicates, \texttt{P_tree} and \texttt{P_forest}.
1.1554 +\begin{ttbox}
1.1555 +[| !!a f.
1.1556 + [| a : A; f : forest(A); P_forest(f) |] ==> P_tree(Tcons(a, f));
1.1557 + P_forest(Fnil);
1.1558 + !!f t. [| t : tree(A); P_tree(t); f : forest(A); P_forest(f) |]
1.1559 + ==> P_forest(Fcons(t, f))
1.1560 +|] ==> (ALL za. za : tree(A) --> P_tree(za)) &
1.1561 + (ALL za. za : forest(A) --> P_forest(za))
1.1562 +\end{ttbox}
1.1563 +
1.1564 +For datatypes with nested recursion, such as the \texttt{term} example from
1.1565 +above, things are a bit more complicated. The rule \texttt{term.induct}
1.1566 +refers to the monotonic operator, \texttt{list}:
1.1567 +\begin{ttbox}
1.1568 +[| x : term(A);
1.1569 + !!a l. [| a : A; l : list(Collect(term(A), P)) |] ==> P(Apply(a, l))
1.1570 +|] ==> P(x)
1.1571 +\end{ttbox}
1.1572 +The file \texttt{ex/Term.ML} derives two higher-level induction rules, one of
1.1573 +which is particularly useful for proving equations:
1.1574 +\begin{ttbox}
1.1575 +[| t : term(A);
1.1576 + !!x zs. [| x : A; zs : list(term(A)); map(f, zs) = map(g, zs) |]
1.1577 + ==> f(Apply(x, zs)) = g(Apply(x, zs))
1.1578 +|] ==> f(t) = g(t)
1.1579 +\end{ttbox}
1.1580 +How this can be generalized to other nested datatypes is a matter for future
1.1581 +research.
1.1582 +
1.1583 +
1.1584 +\subsubsection{The \texttt{case} operator}
1.1585 +
1.1586 +The package defines an operator for performing case analysis over the
1.1587 +datatype. For \texttt{list}, it is called \texttt{list_case} and satisfies
1.1588 +the equations
1.1589 +\begin{ttbox}
1.1590 +list_case(f_Nil, f_Cons, []) = f_Nil
1.1591 +list_case(f_Nil, f_Cons, Cons(a, l)) = f_Cons(a, l)
1.1592 +\end{ttbox}
1.1593 +Here \texttt{f_Nil} is the value to return if the argument is \texttt{Nil} and
1.1594 +\texttt{f_Cons} is a function that computes the value to return if the
1.1595 +argument has the form $\texttt{Cons}(a,l)$. The function can be expressed as
1.1596 +an abstraction, over patterns if desired (\S\ref{sec:pairs}).
1.1597 +
1.1598 +For mutually recursive datatypes, there is a single \texttt{case} operator.
1.1599 +In the tree/forest example, the constant \texttt{tree_forest_case} handles all
1.1600 +of the constructors of the two datatypes.
1.1601 +
1.1602 +
1.1603 +
1.1604 +
1.1605 +\subsection{Defining datatypes}
1.1606 +
1.1607 +The theory syntax for datatype definitions is shown in
1.1608 +Fig.~\ref{datatype-grammar}. In order to be well-formed, a datatype
1.1609 +definition has to obey the rules stated in the previous section. As a result
1.1610 +the theory is extended with the new types, the constructors, and the theorems
1.1611 +listed in the previous section. The quotation marks are necessary because
1.1612 +they enclose general Isabelle formul\ae.
1.1613 +
1.1614 +\begin{figure}
1.1615 +\begin{rail}
1.1616 +datatype : ( 'datatype' | 'codatatype' ) datadecls;
1.1617 +
1.1618 +datadecls: ( '"' id arglist '"' '=' (constructor + '|') ) + 'and'
1.1619 + ;
1.1620 +constructor : name ( () | consargs ) ( () | ( '(' mixfix ')' ) )
1.1621 + ;
1.1622 +consargs : '(' ('"' var ':' term '"' + ',') ')'
1.1623 + ;
1.1624 +\end{rail}
1.1625 +\caption{Syntax of datatype declarations}
1.1626 +\label{datatype-grammar}
1.1627 +\end{figure}
1.1628 +
1.1629 +Codatatypes are declared like datatypes and are identical to them in every
1.1630 +respect except that they have a coinduction rule instead of an induction rule.
1.1631 +Note that while an induction rule has the effect of limiting the values
1.1632 +contained in the set, a coinduction rule gives a way of constructing new
1.1633 +values of the set.
1.1634 +
1.1635 +Most of the theorems about datatypes become part of the default simpset. You
1.1636 +never need to see them again because the simplifier applies them
1.1637 +automatically. Add freeness properties (\texttt{free_iffs}) to the simpset
1.1638 +when you want them. Induction or exhaustion are usually invoked by hand,
1.1639 +usually via these special-purpose tactics:
1.1640 +\begin{ttdescription}
1.1641 +\item[\ttindexbold{induct_tac} {\tt"}$x${\tt"} $i$] applies structural
1.1642 + induction on variable $x$ to subgoal $i$, provided the type of $x$ is a
1.1643 + datatype. The induction variable should not occur among other assumptions
1.1644 + of the subgoal.
1.1645 +\end{ttdescription}
1.1646 +In some cases, induction is overkill and a case distinction over all
1.1647 +constructors of the datatype suffices.
1.1648 +\begin{ttdescription}
1.1649 +\item[\ttindexbold{exhaust_tac} {\tt"}$x${\tt"} $i$]
1.1650 + performs an exhaustive case analysis for the variable~$x$.
1.1651 +\end{ttdescription}
1.1652 +
1.1653 +Both tactics can only be applied to a variable, whose typing must be given in
1.1654 +some assumption, for example the assumption \texttt{x:\ list(A)}. The tactics
1.1655 +also work for the natural numbers (\texttt{nat}) and disjoint sums, although
1.1656 +these sets were not defined using the datatype package. (Disjoint sums are
1.1657 +not recursive, so only \texttt{exhaust_tac} is available.)
1.1658 +
1.1659 +\bigskip
1.1660 +Here are some more details for the technically minded. Processing the
1.1661 +theory file produces an \ML\ structure which, in addition to the usual
1.1662 +components, contains a structure named $t$ for each datatype $t$ defined in
1.1663 +the file. Each structure $t$ contains the following elements:
1.1664 +\begin{ttbox}
1.1665 +val intrs : thm list \textrm{the introduction rules}
1.1666 +val elim : thm \textrm{the elimination (case analysis) rule}
1.1667 +val induct : thm \textrm{the standard induction rule}
1.1668 +val mutual_induct : thm \textrm{the mutual induction rule, or \texttt{True}}
1.1669 +val case_eqns : thm list \textrm{equations for the case operator}
1.1670 +val recursor_eqns : thm list \textrm{equations for the recursor}
1.1671 +val con_defs : thm list \textrm{definitions of the case operator and constructors}
1.1672 +val free_iffs : thm list \textrm{logical equivalences for proving freeness}
1.1673 +val free_SEs : thm list \textrm{elimination rules for proving freeness}
1.1674 +val mk_free : string -> thm \textrm{A function for proving freeness theorems}
1.1675 +val mk_cases : thm list -> string -> thm \textrm{case analysis, see below}
1.1676 +val defs : thm list \textrm{definitions of operators}
1.1677 +val bnd_mono : thm list \textrm{monotonicity property}
1.1678 +val dom_subset : thm list \textrm{inclusion in `bounding set'}
1.1679 +\end{ttbox}
1.1680 +Furthermore there is the theorem $C$\texttt{_I} for every constructor~$C$; for
1.1681 +example, the \texttt{list} datatype's introduction rules are bound to the
1.1682 +identifiers \texttt{Nil_I} and \texttt{Cons_I}.
1.1683 +
1.1684 +For a codatatype, the component \texttt{coinduct} is the coinduction rule,
1.1685 +replacing the \texttt{induct} component.
1.1686 +
1.1687 +See the theories \texttt{ex/Ntree} and \texttt{ex/Brouwer} for examples of
1.1688 +infinitely branching datatypes. See theory \texttt{ex/LList} for an example
1.1689 +of a codatatype. Some of these theories illustrate the use of additional,
1.1690 +undocumented features of the datatype package. Datatype definitions are
1.1691 +reduced to inductive definitions, and the advanced features should be
1.1692 +understood in that light.
1.1693 +
1.1694 +
1.1695 +\subsection{Examples}
1.1696 +
1.1697 +\subsubsection{The datatype of binary trees}
1.1698 +
1.1699 +Let us define the set $\texttt{bt}(A)$ of binary trees over~$A$. The theory
1.1700 +must contain these lines:
1.1701 +\begin{ttbox}
1.1702 +consts bt :: i=>i
1.1703 +datatype "bt(A)" = Lf | Br ("a: A", "t1: bt(A)", "t2: bt(A)")
1.1704 +\end{ttbox}
1.1705 +After loading the theory, we can prove, for example, that no tree equals its
1.1706 +left branch. To ease the induction, we state the goal using quantifiers.
1.1707 +\begin{ttbox}
1.1708 +Goal "l : bt(A) ==> ALL x r. Br(x,l,r) ~= l";
1.1709 +{\out Level 0}
1.1710 +{\out l : bt(A) ==> ALL x r. Br(x, l, r) ~= l}
1.1711 +{\out 1. l : bt(A) ==> ALL x r. Br(x, l, r) ~= l}
1.1712 +\end{ttbox}
1.1713 +This can be proved by the structural induction tactic:
1.1714 +\begin{ttbox}
1.1715 +by (induct_tac "l" 1);
1.1716 +{\out Level 1}
1.1717 +{\out l : bt(A) ==> ALL x r. Br(x, l, r) ~= l}
1.1718 +{\out 1. ALL x r. Br(x, Lf, r) ~= Lf}
1.1719 +{\out 2. !!a t1 t2.}
1.1720 +{\out [| a : A; t1 : bt(A); ALL x r. Br(x, t1, r) ~= t1; t2 : bt(A);}
1.1721 +{\out ALL x r. Br(x, t2, r) ~= t2 |]}
1.1722 +{\out ==> ALL x r. Br(x, Br(a, t1, t2), r) ~= Br(a, t1, t2)}
1.1723 +\end{ttbox}
1.1724 +Both subgoals are proved using the simplifier. Tactic
1.1725 +\texttt{asm_full_simp_tac} is used, rewriting the assumptions.
1.1726 +This is because simplification using the freeness properties can unfold the
1.1727 +definition of constructor~\texttt{Br}, so we arrange that all occurrences are
1.1728 +unfolded.
1.1729 +\begin{ttbox}
1.1730 +by (ALLGOALS (asm_full_simp_tac (simpset() addsimps bt.free_iffs)));
1.1731 +{\out Level 2}
1.1732 +{\out l : bt(A) ==> ALL x r. Br(x, l, r) ~= l}
1.1733 +{\out No subgoals!}
1.1734 +\end{ttbox}
1.1735 +To remove the quantifiers from the induction formula, we save the theorem using
1.1736 +\ttindex{qed_spec_mp}.
1.1737 +\begin{ttbox}
1.1738 +qed_spec_mp "Br_neq_left";
1.1739 +{\out val Br_neq_left = "?l : bt(?A) ==> Br(?x, ?l, ?r) ~= ?l" : thm}
1.1740 +\end{ttbox}
1.1741 +
1.1742 +When there are only a few constructors, we might prefer to prove the freenness
1.1743 +theorems for each constructor. This is trivial, using the function given us
1.1744 +for that purpose:
1.1745 +\begin{ttbox}
1.1746 +val Br_iff = bt.mk_free "Br(a,l,r)=Br(a',l',r') <-> a=a' & l=l' & r=r'";
1.1747 +{\out val Br_iff =}
1.1748 +{\out "Br(?a, ?l, ?r) = Br(?a', ?l', ?r') <->}
1.1749 +{\out ?a = ?a' & ?l = ?l' & ?r = ?r'" : thm}
1.1750 +\end{ttbox}
1.1751 +
1.1752 +The purpose of \ttindex{mk_cases} is to generate simplified instances of the
1.1753 +elimination (case analysis) rule. Its theorem list argument is a list of
1.1754 +constructor definitions, which it uses for freeness reasoning. For example,
1.1755 +this instance of the elimination rule propagates type-checking information
1.1756 +from the premise $\texttt{Br}(a,l,r)\in\texttt{bt}(A)$:
1.1757 +\begin{ttbox}
1.1758 +val BrE = bt.mk_cases bt.con_defs "Br(a,l,r) : bt(A)";
1.1759 +{\out val BrE =}
1.1760 +{\out "[| Br(?a, ?l, ?r) : bt(?A);}
1.1761 +{\out [| ?a : ?A; ?l : bt(?A); ?r : bt(?A) |] ==> ?Q |] ==> ?Q" : thm}
1.1762 +\end{ttbox}
1.1763 +
1.1764 +
1.1765 +\subsubsection{Mixfix syntax in datatypes}
1.1766 +
1.1767 +Mixfix syntax is sometimes convenient. The theory \texttt{ex/PropLog} makes a
1.1768 +deep embedding of propositional logic:
1.1769 +\begin{ttbox}
1.1770 +consts prop :: i
1.1771 +datatype "prop" = Fls
1.1772 + | Var ("n: nat") ("#_" [100] 100)
1.1773 + | "=>" ("p: prop", "q: prop") (infixr 90)
1.1774 +\end{ttbox}
1.1775 +The second constructor has a special $\#n$ syntax, while the third constructor
1.1776 +is an infixed arrow.
1.1777 +
1.1778 +
1.1779 +\subsubsection{A giant enumeration type}
1.1780 +
1.1781 +This example shows a datatype that consists of 60 constructors:
1.1782 +\begin{ttbox}
1.1783 +consts enum :: i
1.1784 +datatype
1.1785 + "enum" = C00 | C01 | C02 | C03 | C04 | C05 | C06 | C07 | C08 | C09
1.1786 + | C10 | C11 | C12 | C13 | C14 | C15 | C16 | C17 | C18 | C19
1.1787 + | C20 | C21 | C22 | C23 | C24 | C25 | C26 | C27 | C28 | C29
1.1788 + | C30 | C31 | C32 | C33 | C34 | C35 | C36 | C37 | C38 | C39
1.1789 + | C40 | C41 | C42 | C43 | C44 | C45 | C46 | C47 | C48 | C49
1.1790 + | C50 | C51 | C52 | C53 | C54 | C55 | C56 | C57 | C58 | C59
1.1791 +end
1.1792 +\end{ttbox}
1.1793 +The datatype package scales well. Even though all properties are proved
1.1794 +rather than assumed, full processing of this definition takes under 15 seconds
1.1795 +(on a 300 MHz Pentium). The constructors have a balanced representation,
1.1796 +essentially binary notation, so freeness properties can be proved fast.
1.1797 +\begin{ttbox}
1.1798 +Goal "C00 ~= C01";
1.1799 +by (simp_tac (simpset() addsimps enum.free_iffs) 1);
1.1800 +\end{ttbox}
1.1801 +You need not derive such inequalities explicitly. The simplifier will dispose
1.1802 +of them automatically, given the theorem list \texttt{free_iffs}.
1.1803 +
1.1804 +\index{*datatype|)}
1.1805 +
1.1806 +
1.1807 +\subsection{Recursive function definitions}\label{sec:ZF:recursive}
1.1808 +\index{recursive functions|see{recursion}}
1.1809 +\index{*primrec|(}
1.1810 +
1.1811 +Datatypes come with a uniform way of defining functions, {\bf primitive
1.1812 + recursion}. Such definitions rely on the recursion operator defined by the
1.1813 +datatype package. Isabelle proves the desired recursion equations as
1.1814 +theorems.
1.1815 +
1.1816 +In principle, one could introduce primitive recursive functions by asserting
1.1817 +their reduction rules as new axioms. Here is a dangerous way of defining the
1.1818 +append function for lists:
1.1819 +\begin{ttbox}\slshape
1.1820 +consts "\at" :: [i,i]=>i (infixr 60)
1.1821 +rules
1.1822 + app_Nil "[] \at ys = ys"
1.1823 + app_Cons "(Cons(a,l)) \at ys = Cons(a, l \at ys)"
1.1824 +\end{ttbox}
1.1825 +Asserting axioms brings the danger of accidentally asserting nonsense. It
1.1826 +should be avoided at all costs!
1.1827 +
1.1828 +The \ttindex{primrec} declaration is a safe means of defining primitive
1.1829 +recursive functions on datatypes:
1.1830 +\begin{ttbox}
1.1831 +consts "\at" :: [i,i]=>i (infixr 60)
1.1832 +primrec
1.1833 + "[] \at ys = ys"
1.1834 + "(Cons(a,l)) \at ys = Cons(a, l \at ys)"
1.1835 +\end{ttbox}
1.1836 +Isabelle will now check that the two rules do indeed form a primitive
1.1837 +recursive definition. For example, the declaration
1.1838 +\begin{ttbox}
1.1839 +primrec
1.1840 + "[] \at ys = us"
1.1841 +\end{ttbox}
1.1842 +is rejected with an error message ``\texttt{Extra variables on rhs}''.
1.1843 +
1.1844 +
1.1845 +\subsubsection{Syntax of recursive definitions}
1.1846 +
1.1847 +The general form of a primitive recursive definition is
1.1848 +\begin{ttbox}
1.1849 +primrec
1.1850 + {\it reduction rules}
1.1851 +\end{ttbox}
1.1852 +where \textit{reduction rules} specify one or more equations of the form
1.1853 +\[ f \, x@1 \, \dots \, x@m \, (C \, y@1 \, \dots \, y@k) \, z@1 \,
1.1854 +\dots \, z@n = r \] such that $C$ is a constructor of the datatype, $r$
1.1855 +contains only the free variables on the left-hand side, and all recursive
1.1856 +calls in $r$ are of the form $f \, \dots \, y@i \, \dots$ for some $i$.
1.1857 +There must be at most one reduction rule for each constructor. The order is
1.1858 +immaterial. For missing constructors, the function is defined to return zero.
1.1859 +
1.1860 +All reduction rules are added to the default simpset.
1.1861 +If you would like to refer to some rule by name, then you must prefix
1.1862 +the rule with an identifier. These identifiers, like those in the
1.1863 +\texttt{rules} section of a theory, will be visible at the \ML\ level.
1.1864 +
1.1865 +The reduction rules for {\tt\at} become part of the default simpset, which
1.1866 +leads to short proof scripts:
1.1867 +\begin{ttbox}\underscoreon
1.1868 +Goal "xs: list(A) ==> (xs @ ys) @ zs = xs @ (ys @ zs)";
1.1869 +by (induct\_tac "xs" 1);
1.1870 +by (ALLGOALS Asm\_simp\_tac);
1.1871 +\end{ttbox}
1.1872 +
1.1873 +You can even use the \texttt{primrec} form with non-recursive datatypes and
1.1874 +with codatatypes. Recursion is not allowed, but it provides a convenient
1.1875 +syntax for defining functions by cases.
1.1876 +
1.1877 +
1.1878 +\subsubsection{Example: varying arguments}
1.1879 +
1.1880 +All arguments, other than the recursive one, must be the same in each equation
1.1881 +and in each recursive call. To get around this restriction, use explict
1.1882 +$\lambda$-abstraction and function application. Here is an example, drawn
1.1883 +from the theory \texttt{Resid/Substitution}. The type of redexes is declared
1.1884 +as follows:
1.1885 +\begin{ttbox}
1.1886 +consts redexes :: i
1.1887 +datatype
1.1888 + "redexes" = Var ("n: nat")
1.1889 + | Fun ("t: redexes")
1.1890 + | App ("b:bool" ,"f:redexes" , "a:redexes")
1.1891 +\end{ttbox}
1.1892 +
1.1893 +The function \texttt{lift} takes a second argument, $k$, which varies in
1.1894 +recursive calls.
1.1895 +\begin{ttbox}
1.1896 +primrec
1.1897 + "lift(Var(i)) = (lam k:nat. if i<k then Var(i) else Var(succ(i)))"
1.1898 + "lift(Fun(t)) = (lam k:nat. Fun(lift(t) ` succ(k)))"
1.1899 + "lift(App(b,f,a)) = (lam k:nat. App(b, lift(f)`k, lift(a)`k))"
1.1900 +\end{ttbox}
1.1901 +Now \texttt{lift(r)`k} satisfies the required recursion equations.
1.1902 +
1.1903 +\index{recursion!primitive|)}
1.1904 +\index{*primrec|)}
1.1905 +
1.1906 +
1.1907 +\section{Inductive and coinductive definitions}
1.1908 +\index{*inductive|(}
1.1909 +\index{*coinductive|(}
1.1910 +
1.1911 +An {\bf inductive definition} specifies the least set~$R$ closed under given
1.1912 +rules. (Applying a rule to elements of~$R$ yields a result within~$R$.) For
1.1913 +example, a structural operational semantics is an inductive definition of an
1.1914 +evaluation relation. Dually, a {\bf coinductive definition} specifies the
1.1915 +greatest set~$R$ consistent with given rules. (Every element of~$R$ can be
1.1916 +seen as arising by applying a rule to elements of~$R$.) An important example
1.1917 +is using bisimulation relations to formalise equivalence of processes and
1.1918 +infinite data structures.
1.1919 +
1.1920 +A theory file may contain any number of inductive and coinductive
1.1921 +definitions. They may be intermixed with other declarations; in
1.1922 +particular, the (co)inductive sets {\bf must} be declared separately as
1.1923 +constants, and may have mixfix syntax or be subject to syntax translations.
1.1924 +
1.1925 +Each (co)inductive definition adds definitions to the theory and also
1.1926 +proves some theorems. Each definition creates an \ML\ structure, which is a
1.1927 +substructure of the main theory structure.
1.1928 +This package is described in detail in a separate paper,%
1.1929 +\footnote{It appeared in CADE~\cite{paulson-CADE}; a longer version is
1.1930 + distributed with Isabelle as \emph{A Fixedpoint Approach to
1.1931 + (Co)Inductive and (Co)Datatype Definitions}.} %
1.1932 +which you might refer to for background information.
1.1933 +
1.1934 +
1.1935 +\subsection{The syntax of a (co)inductive definition}
1.1936 +An inductive definition has the form
1.1937 +\begin{ttbox}
1.1938 +inductive
1.1939 + domains {\it domain declarations}
1.1940 + intrs {\it introduction rules}
1.1941 + monos {\it monotonicity theorems}
1.1942 + con_defs {\it constructor definitions}
1.1943 + type_intrs {\it introduction rules for type-checking}
1.1944 + type_elims {\it elimination rules for type-checking}
1.1945 +\end{ttbox}
1.1946 +A coinductive definition is identical, but starts with the keyword
1.1947 +{\tt coinductive}.
1.1948 +
1.1949 +The {\tt monos}, {\tt con\_defs}, {\tt type\_intrs} and {\tt type\_elims}
1.1950 +sections are optional. If present, each is specified either as a list of
1.1951 +identifiers or as a string. If the latter, then the string must be a valid
1.1952 +\textsc{ml} expression of type {\tt thm list}. The string is simply inserted
1.1953 +into the {\tt _thy.ML} file; if it is ill-formed, it will trigger \textsc{ml}
1.1954 +error messages. You can then inspect the file on the temporary directory.
1.1955 +
1.1956 +\begin{description}
1.1957 +\item[\it domain declarations] consist of one or more items of the form
1.1958 + {\it string\/}~{\tt <=}~{\it string}, associating each recursive set with
1.1959 + its domain. (The domain is some existing set that is large enough to
1.1960 + hold the new set being defined.)
1.1961 +
1.1962 +\item[\it introduction rules] specify one or more introduction rules in
1.1963 + the form {\it ident\/}~{\it string}, where the identifier gives the name of
1.1964 + the rule in the result structure.
1.1965 +
1.1966 +\item[\it monotonicity theorems] are required for each operator applied to
1.1967 + a recursive set in the introduction rules. There \textbf{must} be a theorem
1.1968 + of the form $A\subseteq B\Imp M(A)\subseteq M(B)$, for each premise $t\in M(R_i)$
1.1969 + in an introduction rule!
1.1970 +
1.1971 +\item[\it constructor definitions] contain definitions of constants
1.1972 + appearing in the introduction rules. The (co)datatype package supplies
1.1973 + the constructors' definitions here. Most (co)inductive definitions omit
1.1974 + this section; one exception is the primitive recursive functions example;
1.1975 + see theory \texttt{ex/Primrec}.
1.1976 +
1.1977 +\item[\it type\_intrs] consists of introduction rules for type-checking the
1.1978 + definition: for demonstrating that the new set is included in its domain.
1.1979 + (The proof uses depth-first search.)
1.1980 +
1.1981 +\item[\it type\_elims] consists of elimination rules for type-checking the
1.1982 + definition. They are presumed to be safe and are applied as often as
1.1983 + possible prior to the {\tt type\_intrs} search.
1.1984 +\end{description}
1.1985 +
1.1986 +The package has a few restrictions:
1.1987 +\begin{itemize}
1.1988 +\item The theory must separately declare the recursive sets as
1.1989 + constants.
1.1990 +
1.1991 +\item The names of the recursive sets must be identifiers, not infix
1.1992 +operators.
1.1993 +
1.1994 +\item Side-conditions must not be conjunctions. However, an introduction rule
1.1995 +may contain any number of side-conditions.
1.1996 +
1.1997 +\item Side-conditions of the form $x=t$, where the variable~$x$ does not
1.1998 + occur in~$t$, will be substituted through the rule \verb|mutual_induct|.
1.1999 +\end{itemize}
1.2000 +
1.2001 +
1.2002 +\subsection{Example of an inductive definition}
1.2003 +
1.2004 +Two declarations, included in a theory file, define the finite powerset
1.2005 +operator. First we declare the constant~\texttt{Fin}. Then we declare it
1.2006 +inductively, with two introduction rules:
1.2007 +\begin{ttbox}
1.2008 +consts Fin :: i=>i
1.2009 +
1.2010 +inductive
1.2011 + domains "Fin(A)" <= "Pow(A)"
1.2012 + intrs
1.2013 + emptyI "0 : Fin(A)"
1.2014 + consI "[| a: A; b: Fin(A) |] ==> cons(a,b) : Fin(A)"
1.2015 + type_intrs empty_subsetI, cons_subsetI, PowI
1.2016 + type_elims "[make_elim PowD]"
1.2017 +\end{ttbox}
1.2018 +The resulting theory structure contains a substructure, called~\texttt{Fin}.
1.2019 +It contains the \texttt{Fin}$~A$ introduction rules as the list
1.2020 +\texttt{Fin.intrs}, and also individually as \texttt{Fin.emptyI} and
1.2021 +\texttt{Fin.consI}. The induction rule is \texttt{Fin.induct}.
1.2022 +
1.2023 +The chief problem with making (co)inductive definitions involves type-checking
1.2024 +the rules. Sometimes, additional theorems need to be supplied under
1.2025 +\texttt{type_intrs} or \texttt{type_elims}. If the package fails when trying
1.2026 +to prove your introduction rules, then set the flag \ttindexbold{trace_induct}
1.2027 +to \texttt{true} and try again. (See the manual \emph{A Fixedpoint Approach
1.2028 + \ldots} for more discussion of type-checking.)
1.2029 +
1.2030 +In the example above, $\texttt{Pow}(A)$ is given as the domain of
1.2031 +$\texttt{Fin}(A)$, for obviously every finite subset of~$A$ is a subset
1.2032 +of~$A$. However, the inductive definition package can only prove that given a
1.2033 +few hints.
1.2034 +Here is the output that results (with the flag set) when the
1.2035 +\texttt{type_intrs} and \texttt{type_elims} are omitted from the inductive
1.2036 +definition above:
1.2037 +\begin{ttbox}
1.2038 +Inductive definition Finite.Fin
1.2039 +Fin(A) ==
1.2040 +lfp(Pow(A),
1.2041 + \%X. {z: Pow(A) . z = 0 | (EX a b. z = cons(a, b) & a : A & b : X)})
1.2042 + Proving monotonicity...
1.2043 +\ttbreak
1.2044 + Proving the introduction rules...
1.2045 +The typechecking subgoal:
1.2046 +0 : Fin(A)
1.2047 + 1. 0 : Pow(A)
1.2048 +\ttbreak
1.2049 +The subgoal after monos, type_elims:
1.2050 +0 : Fin(A)
1.2051 + 1. 0 : Pow(A)
1.2052 +*** prove_goal: tactic failed
1.2053 +\end{ttbox}
1.2054 +We see the need to supply theorems to let the package prove
1.2055 +$\emptyset\in\texttt{Pow}(A)$. Restoring the \texttt{type_intrs} but not the
1.2056 +\texttt{type_elims}, we again get an error message:
1.2057 +\begin{ttbox}
1.2058 +The typechecking subgoal:
1.2059 +0 : Fin(A)
1.2060 + 1. 0 : Pow(A)
1.2061 +\ttbreak
1.2062 +The subgoal after monos, type_elims:
1.2063 +0 : Fin(A)
1.2064 + 1. 0 : Pow(A)
1.2065 +\ttbreak
1.2066 +The typechecking subgoal:
1.2067 +cons(a, b) : Fin(A)
1.2068 + 1. [| a : A; b : Fin(A) |] ==> cons(a, b) : Pow(A)
1.2069 +\ttbreak
1.2070 +The subgoal after monos, type_elims:
1.2071 +cons(a, b) : Fin(A)
1.2072 + 1. [| a : A; b : Pow(A) |] ==> cons(a, b) : Pow(A)
1.2073 +*** prove_goal: tactic failed
1.2074 +\end{ttbox}
1.2075 +The first rule has been type-checked, but the second one has failed. The
1.2076 +simplest solution to such problems is to prove the failed subgoal separately
1.2077 +and to supply it under \texttt{type_intrs}. The solution actually used is
1.2078 +to supply, under \texttt{type_elims}, a rule that changes
1.2079 +$b\in\texttt{Pow}(A)$ to $b\subseteq A$; together with \texttt{cons_subsetI}
1.2080 +and \texttt{PowI}, it is enough to complete the type-checking.
1.2081 +
1.2082 +
1.2083 +
1.2084 +\subsection{Further examples}
1.2085 +
1.2086 +An inductive definition may involve arbitrary monotonic operators. Here is a
1.2087 +standard example: the accessible part of a relation. Note the use
1.2088 +of~\texttt{Pow} in the introduction rule and the corresponding mention of the
1.2089 +rule \verb|Pow_mono| in the \texttt{monos} list. If the desired rule has a
1.2090 +universally quantified premise, usually the effect can be obtained using
1.2091 +\texttt{Pow}.
1.2092 +\begin{ttbox}
1.2093 +consts acc :: i=>i
1.2094 +inductive
1.2095 + domains "acc(r)" <= "field(r)"
1.2096 + intrs
1.2097 + vimage "[| r-``{a}: Pow(acc(r)); a: field(r) |] ==> a: acc(r)"
1.2098 + monos Pow_mono
1.2099 +\end{ttbox}
1.2100 +
1.2101 +Finally, here is a coinductive definition. It captures (as a bisimulation)
1.2102 +the notion of equality on lazy lists, which are first defined as a codatatype:
1.2103 +\begin{ttbox}
1.2104 +consts llist :: i=>i
1.2105 +codatatype "llist(A)" = LNil | LCons ("a: A", "l: llist(A)")
1.2106 +\ttbreak
1.2107 +
1.2108 +consts lleq :: i=>i
1.2109 +coinductive
1.2110 + domains "lleq(A)" <= "llist(A) * llist(A)"
1.2111 + intrs
1.2112 + LNil "<LNil, LNil> : lleq(A)"
1.2113 + LCons "[| a:A; <l,l'>: lleq(A) |]
1.2114 + ==> <LCons(a,l), LCons(a,l')>: lleq(A)"
1.2115 + type_intrs "llist.intrs"
1.2116 +\end{ttbox}
1.2117 +This use of \texttt{type_intrs} is typical: the relation concerns the
1.2118 +codatatype \texttt{llist}, so naturally the introduction rules for that
1.2119 +codatatype will be required for type-checking the rules.
1.2120 +
1.2121 +The Isabelle distribution contains many other inductive definitions. Simple
1.2122 +examples are collected on subdirectory \texttt{ZF/ex}. The directory
1.2123 +\texttt{Coind} and the theory \texttt{ZF/ex/LList} contain coinductive
1.2124 +definitions. Larger examples may be found on other subdirectories of
1.2125 +\texttt{ZF}, such as \texttt{IMP}, and \texttt{Resid}.
1.2126 +
1.2127 +
1.2128 +\subsection{The result structure}
1.2129 +
1.2130 +Each (co)inductive set defined in a theory file generates an \ML\ substructure
1.2131 +having the same name. The the substructure contains the following elements:
1.2132 +
1.2133 +\begin{ttbox}
1.2134 +val intrs : thm list \textrm{the introduction rules}
1.2135 +val elim : thm \textrm{the elimination (case analysis) rule}
1.2136 +val mk_cases : thm list -> string -> thm \textrm{case analysis, see below}
1.2137 +val induct : thm \textrm{the standard induction rule}
1.2138 +val mutual_induct : thm \textrm{the mutual induction rule, or \texttt{True}}
1.2139 +val defs : thm list \textrm{definitions of operators}
1.2140 +val bnd_mono : thm list \textrm{monotonicity property}
1.2141 +val dom_subset : thm list \textrm{inclusion in `bounding set'}
1.2142 +\end{ttbox}
1.2143 +Furthermore there is the theorem $C$\texttt{_I} for every constructor~$C$; for
1.2144 +example, the \texttt{list} datatype's introduction rules are bound to the
1.2145 +identifiers \texttt{Nil_I} and \texttt{Cons_I}.
1.2146 +
1.2147 +For a codatatype, the component \texttt{coinduct} is the coinduction rule,
1.2148 +replacing the \texttt{induct} component.
1.2149 +
1.2150 +Recall that \ttindex{mk_cases} generates simplified instances of the
1.2151 +elimination (case analysis) rule. It is as useful for inductive definitions
1.2152 +as it is for datatypes. There are many examples in the theory
1.2153 +\texttt{ex/Comb}, which is discussed at length
1.2154 +elsewhere~\cite{paulson-generic}. The theory first defines the datatype
1.2155 +\texttt{comb} of combinators:
1.2156 +\begin{ttbox}
1.2157 +consts comb :: i
1.2158 +datatype "comb" = K
1.2159 + | S
1.2160 + | "#" ("p: comb", "q: comb") (infixl 90)
1.2161 +\end{ttbox}
1.2162 +The theory goes on to define contraction and parallel contraction
1.2163 +inductively. Then the file \texttt{ex/Comb.ML} defines special cases of
1.2164 +contraction using \texttt{mk_cases}:
1.2165 +\begin{ttbox}
1.2166 +val K_contractE = contract.mk_cases comb.con_defs "K -1-> r";
1.2167 +{\out val K_contractE = "K -1-> ?r ==> ?Q" : thm}
1.2168 +\end{ttbox}
1.2169 +We can read this as saying that the combinator \texttt{K} cannot reduce to
1.2170 +anything. Similar elimination rules for \texttt{S} and application are also
1.2171 +generated and are supplied to the classical reasoner. Note that
1.2172 +\texttt{comb.con_defs} is given to \texttt{mk_cases} to allow freeness
1.2173 +reasoning on datatype \texttt{comb}.
1.2174 +
1.2175 +\index{*coinductive|)} \index{*inductive|)}
1.2176 +
1.2177 +
1.2178 +
1.2179 +
1.2180 +\section{The outer reaches of set theory}
1.2181 +
1.2182 +The constructions of the natural numbers and lists use a suite of
1.2183 +operators for handling recursive function definitions. I have described
1.2184 +the developments in detail elsewhere~\cite{paulson-set-II}. Here is a brief
1.2185 +summary:
1.2186 +\begin{itemize}
1.2187 + \item Theory \texttt{Trancl} defines the transitive closure of a relation
1.2188 + (as a least fixedpoint).
1.2189 +
1.2190 + \item Theory \texttt{WF} proves the Well-Founded Recursion Theorem, using an
1.2191 + elegant approach of Tobias Nipkow. This theorem permits general
1.2192 + recursive definitions within set theory.
1.2193 +
1.2194 + \item Theory \texttt{Ord} defines the notions of transitive set and ordinal
1.2195 + number. It derives transfinite induction. A key definition is {\bf
1.2196 + less than}: $i<j$ if and only if $i$ and $j$ are both ordinals and
1.2197 + $i\in j$. As a special case, it includes less than on the natural
1.2198 + numbers.
1.2199 +
1.2200 + \item Theory \texttt{Epsilon} derives $\varepsilon$-induction and
1.2201 + $\varepsilon$-recursion, which are generalisations of transfinite
1.2202 + induction and recursion. It also defines \cdx{rank}$(x)$, which
1.2203 + is the least ordinal $\alpha$ such that $x$ is constructed at
1.2204 + stage $\alpha$ of the cumulative hierarchy (thus $x\in
1.2205 + V@{\alpha+1}$).
1.2206 +\end{itemize}
1.2207 +
1.2208 +Other important theories lead to a theory of cardinal numbers. They have
1.2209 +not yet been written up anywhere. Here is a summary:
1.2210 +\begin{itemize}
1.2211 +\item Theory \texttt{Rel} defines the basic properties of relations, such as
1.2212 + (ir)reflexivity, (a)symmetry, and transitivity.
1.2213 +
1.2214 +\item Theory \texttt{EquivClass} develops a theory of equivalence
1.2215 + classes, not using the Axiom of Choice.
1.2216 +
1.2217 +\item Theory \texttt{Order} defines partial orderings, total orderings and
1.2218 + wellorderings.
1.2219 +
1.2220 +\item Theory \texttt{OrderArith} defines orderings on sum and product sets.
1.2221 + These can be used to define ordinal arithmetic and have applications to
1.2222 + cardinal arithmetic.
1.2223 +
1.2224 +\item Theory \texttt{OrderType} defines order types. Every wellordering is
1.2225 + equivalent to a unique ordinal, which is its order type.
1.2226 +
1.2227 +\item Theory \texttt{Cardinal} defines equipollence and cardinal numbers.
1.2228 +
1.2229 +\item Theory \texttt{CardinalArith} defines cardinal addition and
1.2230 + multiplication, and proves their elementary laws. It proves that there
1.2231 + is no greatest cardinal. It also proves a deep result, namely
1.2232 + $\kappa\otimes\kappa=\kappa$ for every infinite cardinal~$\kappa$; see
1.2233 + Kunen~\cite[page 29]{kunen80}. None of these results assume the Axiom of
1.2234 + Choice, which complicates their proofs considerably.
1.2235 +\end{itemize}
1.2236 +
1.2237 +The following developments involve the Axiom of Choice (AC):
1.2238 +\begin{itemize}
1.2239 +\item Theory \texttt{AC} asserts the Axiom of Choice and proves some simple
1.2240 + equivalent forms.
1.2241 +
1.2242 +\item Theory \texttt{Zorn} proves Hausdorff's Maximal Principle, Zorn's Lemma
1.2243 + and the Wellordering Theorem, following Abrial and
1.2244 + Laffitte~\cite{abrial93}.
1.2245 +
1.2246 +\item Theory \verb|Cardinal_AC| uses AC to prove simplified theorems about
1.2247 + the cardinals. It also proves a theorem needed to justify
1.2248 + infinitely branching datatype declarations: if $\kappa$ is an infinite
1.2249 + cardinal and $|X(\alpha)| \le \kappa$ for all $\alpha<\kappa$ then
1.2250 + $|\union\sb{\alpha<\kappa} X(\alpha)| \le \kappa$.
1.2251 +
1.2252 +\item Theory \texttt{InfDatatype} proves theorems to justify infinitely
1.2253 + branching datatypes. Arbitrary index sets are allowed, provided their
1.2254 + cardinalities have an upper bound. The theory also justifies some
1.2255 + unusual cases of finite branching, involving the finite powerset operator
1.2256 + and the finite function space operator.
1.2257 +\end{itemize}
1.2258 +
1.2259 +
1.2260 +
1.2261 +\section{The examples directories}
1.2262 +Directory \texttt{HOL/IMP} contains a mechanised version of a semantic
1.2263 +equivalence proof taken from Winskel~\cite{winskel93}. It formalises the
1.2264 +denotational and operational semantics of a simple while-language, then
1.2265 +proves the two equivalent. It contains several datatype and inductive
1.2266 +definitions, and demonstrates their use.
1.2267 +
1.2268 +The directory \texttt{ZF/ex} contains further developments in {\ZF} set
1.2269 +theory. Here is an overview; see the files themselves for more details. I
1.2270 +describe much of this material in other
1.2271 +publications~\cite{paulson-set-I,paulson-set-II,paulson-CADE}.
1.2272 +\begin{itemize}
1.2273 +\item File \texttt{misc.ML} contains miscellaneous examples such as
1.2274 + Cantor's Theorem, the Schr\"oder-Bernstein Theorem and the `Composition
1.2275 + of homomorphisms' challenge~\cite{boyer86}.
1.2276 +
1.2277 +\item Theory \texttt{Ramsey} proves the finite exponent 2 version of
1.2278 + Ramsey's Theorem, following Basin and Kaufmann's
1.2279 + presentation~\cite{basin91}.
1.2280 +
1.2281 +\item Theory \texttt{Integ} develops a theory of the integers as
1.2282 + equivalence classes of pairs of natural numbers.
1.2283 +
1.2284 +\item Theory \texttt{Primrec} develops some computation theory. It
1.2285 + inductively defines the set of primitive recursive functions and presents a
1.2286 + proof that Ackermann's function is not primitive recursive.
1.2287 +
1.2288 +\item Theory \texttt{Primes} defines the Greatest Common Divisor of two
1.2289 + natural numbers and and the ``divides'' relation.
1.2290 +
1.2291 +\item Theory \texttt{Bin} defines a datatype for two's complement binary
1.2292 + integers, then proves rewrite rules to perform binary arithmetic. For
1.2293 + instance, $1359\times {-}2468 = {-}3354012$ takes under 14 seconds.
1.2294 +
1.2295 +\item Theory \texttt{BT} defines the recursive data structure ${\tt
1.2296 + bt}(A)$, labelled binary trees.
1.2297 +
1.2298 +\item Theory \texttt{Term} defines a recursive data structure for terms
1.2299 + and term lists. These are simply finite branching trees.
1.2300 +
1.2301 +\item Theory \texttt{TF} defines primitives for solving mutually
1.2302 + recursive equations over sets. It constructs sets of trees and forests
1.2303 + as an example, including induction and recursion rules that handle the
1.2304 + mutual recursion.
1.2305 +
1.2306 +\item Theory \texttt{Prop} proves soundness and completeness of
1.2307 + propositional logic~\cite{paulson-set-II}. This illustrates datatype
1.2308 + definitions, inductive definitions, structural induction and rule
1.2309 + induction.
1.2310 +
1.2311 +\item Theory \texttt{ListN} inductively defines the lists of $n$
1.2312 + elements~\cite{paulin92}.
1.2313 +
1.2314 +\item Theory \texttt{Acc} inductively defines the accessible part of a
1.2315 + relation~\cite{paulin92}.
1.2316 +
1.2317 +\item Theory \texttt{Comb} defines the datatype of combinators and
1.2318 + inductively defines contraction and parallel contraction. It goes on to
1.2319 + prove the Church-Rosser Theorem. This case study follows Camilleri and
1.2320 + Melham~\cite{camilleri92}.
1.2321 +
1.2322 +\item Theory \texttt{LList} defines lazy lists and a coinduction
1.2323 + principle for proving equations between them.
1.2324 +\end{itemize}
1.2325 +
1.2326 +
1.2327 +\section{A proof about powersets}\label{sec:ZF-pow-example}
1.2328 +To demonstrate high-level reasoning about subsets, let us prove the
1.2329 +equation ${{\tt Pow}(A)\cap {\tt Pow}(B)}= {\tt Pow}(A\cap B)$. Compared
1.2330 +with first-order logic, set theory involves a maze of rules, and theorems
1.2331 +have many different proofs. Attempting other proofs of the theorem might
1.2332 +be instructive. This proof exploits the lattice properties of
1.2333 +intersection. It also uses the monotonicity of the powerset operation,
1.2334 +from \texttt{ZF/mono.ML}:
1.2335 +\begin{ttbox}
1.2336 +\tdx{Pow_mono} A<=B ==> Pow(A) <= Pow(B)
1.2337 +\end{ttbox}
1.2338 +We enter the goal and make the first step, which breaks the equation into
1.2339 +two inclusions by extensionality:\index{*equalityI theorem}
1.2340 +\begin{ttbox}
1.2341 +Goal "Pow(A Int B) = Pow(A) Int Pow(B)";
1.2342 +{\out Level 0}
1.2343 +{\out Pow(A Int B) = Pow(A) Int Pow(B)}
1.2344 +{\out 1. Pow(A Int B) = Pow(A) Int Pow(B)}
1.2345 +\ttbreak
1.2346 +by (resolve_tac [equalityI] 1);
1.2347 +{\out Level 1}
1.2348 +{\out Pow(A Int B) = Pow(A) Int Pow(B)}
1.2349 +{\out 1. Pow(A Int B) <= Pow(A) Int Pow(B)}
1.2350 +{\out 2. Pow(A) Int Pow(B) <= Pow(A Int B)}
1.2351 +\end{ttbox}
1.2352 +Both inclusions could be tackled straightforwardly using \texttt{subsetI}.
1.2353 +A shorter proof results from noting that intersection forms the greatest
1.2354 +lower bound:\index{*Int_greatest theorem}
1.2355 +\begin{ttbox}
1.2356 +by (resolve_tac [Int_greatest] 1);
1.2357 +{\out Level 2}
1.2358 +{\out Pow(A Int B) = Pow(A) Int Pow(B)}
1.2359 +{\out 1. Pow(A Int B) <= Pow(A)}
1.2360 +{\out 2. Pow(A Int B) <= Pow(B)}
1.2361 +{\out 3. Pow(A) Int Pow(B) <= Pow(A Int B)}
1.2362 +\end{ttbox}
1.2363 +Subgoal~1 follows by applying the monotonicity of \texttt{Pow} to $A\int
1.2364 +B\subseteq A$; subgoal~2 follows similarly:
1.2365 +\index{*Int_lower1 theorem}\index{*Int_lower2 theorem}
1.2366 +\begin{ttbox}
1.2367 +by (resolve_tac [Int_lower1 RS Pow_mono] 1);
1.2368 +{\out Level 3}
1.2369 +{\out Pow(A Int B) = Pow(A) Int Pow(B)}
1.2370 +{\out 1. Pow(A Int B) <= Pow(B)}
1.2371 +{\out 2. Pow(A) Int Pow(B) <= Pow(A Int B)}
1.2372 +\ttbreak
1.2373 +by (resolve_tac [Int_lower2 RS Pow_mono] 1);
1.2374 +{\out Level 4}
1.2375 +{\out Pow(A Int B) = Pow(A) Int Pow(B)}
1.2376 +{\out 1. Pow(A) Int Pow(B) <= Pow(A Int B)}
1.2377 +\end{ttbox}
1.2378 +We are left with the opposite inclusion, which we tackle in the
1.2379 +straightforward way:\index{*subsetI theorem}
1.2380 +\begin{ttbox}
1.2381 +by (resolve_tac [subsetI] 1);
1.2382 +{\out Level 5}
1.2383 +{\out Pow(A Int B) = Pow(A) Int Pow(B)}
1.2384 +{\out 1. !!x. x : Pow(A) Int Pow(B) ==> x : Pow(A Int B)}
1.2385 +\end{ttbox}
1.2386 +The subgoal is to show $x\in {\tt Pow}(A\cap B)$ assuming $x\in{\tt
1.2387 +Pow}(A)\cap {\tt Pow}(B)$; eliminating this assumption produces two
1.2388 +subgoals. The rule \tdx{IntE} treats the intersection like a conjunction
1.2389 +instead of unfolding its definition.
1.2390 +\begin{ttbox}
1.2391 +by (eresolve_tac [IntE] 1);
1.2392 +{\out Level 6}
1.2393 +{\out Pow(A Int B) = Pow(A) Int Pow(B)}
1.2394 +{\out 1. !!x. [| x : Pow(A); x : Pow(B) |] ==> x : Pow(A Int B)}
1.2395 +\end{ttbox}
1.2396 +The next step replaces the \texttt{Pow} by the subset
1.2397 +relation~($\subseteq$).\index{*PowI theorem}
1.2398 +\begin{ttbox}
1.2399 +by (resolve_tac [PowI] 1);
1.2400 +{\out Level 7}
1.2401 +{\out Pow(A Int B) = Pow(A) Int Pow(B)}
1.2402 +{\out 1. !!x. [| x : Pow(A); x : Pow(B) |] ==> x <= A Int B}
1.2403 +\end{ttbox}
1.2404 +We perform the same replacement in the assumptions. This is a good
1.2405 +demonstration of the tactic \ttindex{dresolve_tac}:\index{*PowD theorem}
1.2406 +\begin{ttbox}
1.2407 +by (REPEAT (dresolve_tac [PowD] 1));
1.2408 +{\out Level 8}
1.2409 +{\out Pow(A Int B) = Pow(A) Int Pow(B)}
1.2410 +{\out 1. !!x. [| x <= A; x <= B |] ==> x <= A Int B}
1.2411 +\end{ttbox}
1.2412 +The assumptions are that $x$ is a lower bound of both $A$ and~$B$, but
1.2413 +$A\int B$ is the greatest lower bound:\index{*Int_greatest theorem}
1.2414 +\begin{ttbox}
1.2415 +by (resolve_tac [Int_greatest] 1);
1.2416 +{\out Level 9}
1.2417 +{\out Pow(A Int B) = Pow(A) Int Pow(B)}
1.2418 +{\out 1. !!x. [| x <= A; x <= B |] ==> x <= A}
1.2419 +{\out 2. !!x. [| x <= A; x <= B |] ==> x <= B}
1.2420 +\end{ttbox}
1.2421 +To conclude the proof, we clear up the trivial subgoals:
1.2422 +\begin{ttbox}
1.2423 +by (REPEAT (assume_tac 1));
1.2424 +{\out Level 10}
1.2425 +{\out Pow(A Int B) = Pow(A) Int Pow(B)}
1.2426 +{\out No subgoals!}
1.2427 +\end{ttbox}
1.2428 +\medskip
1.2429 +We could have performed this proof in one step by applying
1.2430 +\ttindex{Blast_tac}. Let us
1.2431 +go back to the start:
1.2432 +\begin{ttbox}
1.2433 +choplev 0;
1.2434 +{\out Level 0}
1.2435 +{\out Pow(A Int B) = Pow(A) Int Pow(B)}
1.2436 +{\out 1. Pow(A Int B) = Pow(A) Int Pow(B)}
1.2437 +by (Blast_tac 1);
1.2438 +{\out Depth = 0}
1.2439 +{\out Depth = 1}
1.2440 +{\out Depth = 2}
1.2441 +{\out Depth = 3}
1.2442 +{\out Level 1}
1.2443 +{\out Pow(A Int B) = Pow(A) Int Pow(B)}
1.2444 +{\out No subgoals!}
1.2445 +\end{ttbox}
1.2446 +Past researchers regarded this as a difficult proof, as indeed it is if all
1.2447 +the symbols are replaced by their definitions.
1.2448 +\goodbreak
1.2449 +
1.2450 +\section{Monotonicity of the union operator}
1.2451 +For another example, we prove that general union is monotonic:
1.2452 +${C\subseteq D}$ implies $\bigcup(C)\subseteq \bigcup(D)$. To begin, we
1.2453 +tackle the inclusion using \tdx{subsetI}:
1.2454 +\begin{ttbox}
1.2455 +Goal "C<=D ==> Union(C) <= Union(D)";
1.2456 +{\out Level 0}
1.2457 +{\out C <= D ==> Union(C) <= Union(D)}
1.2458 +{\out 1. C <= D ==> Union(C) <= Union(D)}
1.2459 +\ttbreak
1.2460 +by (resolve_tac [subsetI] 1);
1.2461 +{\out Level 1}
1.2462 +{\out C <= D ==> Union(C) <= Union(D)}
1.2463 +{\out 1. !!x. [| C <= D; x : Union(C) |] ==> x : Union(D)}
1.2464 +\end{ttbox}
1.2465 +Big union is like an existential quantifier --- the occurrence in the
1.2466 +assumptions must be eliminated early, since it creates parameters.
1.2467 +\index{*UnionE theorem}
1.2468 +\begin{ttbox}
1.2469 +by (eresolve_tac [UnionE] 1);
1.2470 +{\out Level 2}
1.2471 +{\out C <= D ==> Union(C) <= Union(D)}
1.2472 +{\out 1. !!x B. [| C <= D; x : B; B : C |] ==> x : Union(D)}
1.2473 +\end{ttbox}
1.2474 +Now we may apply \tdx{UnionI}, which creates an unknown involving the
1.2475 +parameters. To show $x\in \bigcup(D)$ it suffices to show that $x$ belongs
1.2476 +to some element, say~$\Var{B2}(x,B)$, of~$D$.
1.2477 +\begin{ttbox}
1.2478 +by (resolve_tac [UnionI] 1);
1.2479 +{\out Level 3}
1.2480 +{\out C <= D ==> Union(C) <= Union(D)}
1.2481 +{\out 1. !!x B. [| C <= D; x : B; B : C |] ==> ?B2(x,B) : D}
1.2482 +{\out 2. !!x B. [| C <= D; x : B; B : C |] ==> x : ?B2(x,B)}
1.2483 +\end{ttbox}
1.2484 +Combining \tdx{subsetD} with the assumption $C\subseteq D$ yields
1.2485 +$\Var{a}\in C \Imp \Var{a}\in D$, which reduces subgoal~1. Note that
1.2486 +\texttt{eresolve_tac} has removed that assumption.
1.2487 +\begin{ttbox}
1.2488 +by (eresolve_tac [subsetD] 1);
1.2489 +{\out Level 4}
1.2490 +{\out C <= D ==> Union(C) <= Union(D)}
1.2491 +{\out 1. !!x B. [| x : B; B : C |] ==> ?B2(x,B) : C}
1.2492 +{\out 2. !!x B. [| C <= D; x : B; B : C |] ==> x : ?B2(x,B)}
1.2493 +\end{ttbox}
1.2494 +The rest is routine. Observe how~$\Var{B2}(x,B)$ is instantiated.
1.2495 +\begin{ttbox}
1.2496 +by (assume_tac 1);
1.2497 +{\out Level 5}
1.2498 +{\out C <= D ==> Union(C) <= Union(D)}
1.2499 +{\out 1. !!x B. [| C <= D; x : B; B : C |] ==> x : B}
1.2500 +by (assume_tac 1);
1.2501 +{\out Level 6}
1.2502 +{\out C <= D ==> Union(C) <= Union(D)}
1.2503 +{\out No subgoals!}
1.2504 +\end{ttbox}
1.2505 +Again, \ttindex{Blast_tac} can prove the theorem in one step.
1.2506 +\begin{ttbox}
1.2507 +by (Blast_tac 1);
1.2508 +{\out Depth = 0}
1.2509 +{\out Depth = 1}
1.2510 +{\out Depth = 2}
1.2511 +{\out Level 1}
1.2512 +{\out C <= D ==> Union(C) <= Union(D)}
1.2513 +{\out No subgoals!}
1.2514 +\end{ttbox}
1.2515 +
1.2516 +The file \texttt{ZF/equalities.ML} has many similar proofs. Reasoning about
1.2517 +general intersection can be difficult because of its anomalous behaviour on
1.2518 +the empty set. However, \ttindex{Blast_tac} copes well with these. Here is
1.2519 +a typical example, borrowed from Devlin~\cite[page 12]{devlin79}:
1.2520 +\begin{ttbox}
1.2521 +a:C ==> (INT x:C. A(x) Int B(x)) = (INT x:C. A(x)) Int (INT x:C. B(x))
1.2522 +\end{ttbox}
1.2523 +In traditional notation this is
1.2524 +\[ a\in C \,\Imp\, \inter@{x\in C} \Bigl(A(x) \int B(x)\Bigr) =
1.2525 + \Bigl(\inter@{x\in C} A(x)\Bigr) \int
1.2526 + \Bigl(\inter@{x\in C} B(x)\Bigr) \]
1.2527 +
1.2528 +\section{Low-level reasoning about functions}
1.2529 +The derived rules \texttt{lamI}, \texttt{lamE}, \texttt{lam_type}, \texttt{beta}
1.2530 +and \texttt{eta} support reasoning about functions in a
1.2531 +$\lambda$-calculus style. This is generally easier than regarding
1.2532 +functions as sets of ordered pairs. But sometimes we must look at the
1.2533 +underlying representation, as in the following proof
1.2534 +of~\tdx{fun_disjoint_apply1}. This states that if $f$ and~$g$ are
1.2535 +functions with disjoint domains~$A$ and~$C$, and if $a\in A$, then
1.2536 +$(f\un g)`a = f`a$:
1.2537 +\begin{ttbox}
1.2538 +Goal "[| a:A; f: A->B; g: C->D; A Int C = 0 |] ==> \ttback
1.2539 +\ttback (f Un g)`a = f`a";
1.2540 +{\out Level 0}
1.2541 +{\out [| a : A; f : A -> B; g : C -> D; A Int C = 0 |]}
1.2542 +{\out ==> (f Un g) ` a = f ` a}
1.2543 +{\out 1. [| a : A; f : A -> B; g : C -> D; A Int C = 0 |]}
1.2544 +{\out ==> (f Un g) ` a = f ` a}
1.2545 +\end{ttbox}
1.2546 +Using \tdx{apply_equality}, we reduce the equality to reasoning about
1.2547 +ordered pairs. The second subgoal is to verify that $f\un g$ is a function.
1.2548 +To save space, the assumptions will be abbreviated below.
1.2549 +\begin{ttbox}
1.2550 +by (resolve_tac [apply_equality] 1);
1.2551 +{\out Level 1}
1.2552 +{\out [| \ldots |] ==> (f Un g) ` a = f ` a}
1.2553 +{\out 1. [| \ldots |] ==> <a,f ` a> : f Un g}
1.2554 +{\out 2. [| \ldots |] ==> f Un g : (PROD x:?A. ?B(x))}
1.2555 +\end{ttbox}
1.2556 +We must show that the pair belongs to~$f$ or~$g$; by~\tdx{UnI1} we
1.2557 +choose~$f$:
1.2558 +\begin{ttbox}
1.2559 +by (resolve_tac [UnI1] 1);
1.2560 +{\out Level 2}
1.2561 +{\out [| \ldots |] ==> (f Un g) ` a = f ` a}
1.2562 +{\out 1. [| \ldots |] ==> <a,f ` a> : f}
1.2563 +{\out 2. [| \ldots |] ==> f Un g : (PROD x:?A. ?B(x))}
1.2564 +\end{ttbox}
1.2565 +To show $\pair{a,f`a}\in f$ we use \tdx{apply_Pair}, which is
1.2566 +essentially the converse of \tdx{apply_equality}:
1.2567 +\begin{ttbox}
1.2568 +by (resolve_tac [apply_Pair] 1);
1.2569 +{\out Level 3}
1.2570 +{\out [| \ldots |] ==> (f Un g) ` a = f ` a}
1.2571 +{\out 1. [| \ldots |] ==> f : (PROD x:?A2. ?B2(x))}
1.2572 +{\out 2. [| \ldots |] ==> a : ?A2}
1.2573 +{\out 3. [| \ldots |] ==> f Un g : (PROD x:?A. ?B(x))}
1.2574 +\end{ttbox}
1.2575 +Using the assumptions $f\in A\to B$ and $a\in A$, we solve the two subgoals
1.2576 +from \tdx{apply_Pair}. Recall that a $\Pi$-set is merely a generalized
1.2577 +function space, and observe that~{\tt?A2} is instantiated to~\texttt{A}.
1.2578 +\begin{ttbox}
1.2579 +by (assume_tac 1);
1.2580 +{\out Level 4}
1.2581 +{\out [| \ldots |] ==> (f Un g) ` a = f ` a}
1.2582 +{\out 1. [| \ldots |] ==> a : A}
1.2583 +{\out 2. [| \ldots |] ==> f Un g : (PROD x:?A. ?B(x))}
1.2584 +by (assume_tac 1);
1.2585 +{\out Level 5}
1.2586 +{\out [| \ldots |] ==> (f Un g) ` a = f ` a}
1.2587 +{\out 1. [| \ldots |] ==> f Un g : (PROD x:?A. ?B(x))}
1.2588 +\end{ttbox}
1.2589 +To construct functions of the form $f\un g$, we apply
1.2590 +\tdx{fun_disjoint_Un}:
1.2591 +\begin{ttbox}
1.2592 +by (resolve_tac [fun_disjoint_Un] 1);
1.2593 +{\out Level 6}
1.2594 +{\out [| \ldots |] ==> (f Un g) ` a = f ` a}
1.2595 +{\out 1. [| \ldots |] ==> f : ?A3 -> ?B3}
1.2596 +{\out 2. [| \ldots |] ==> g : ?C3 -> ?D3}
1.2597 +{\out 3. [| \ldots |] ==> ?A3 Int ?C3 = 0}
1.2598 +\end{ttbox}
1.2599 +The remaining subgoals are instances of the assumptions. Again, observe how
1.2600 +unknowns are instantiated:
1.2601 +\begin{ttbox}
1.2602 +by (assume_tac 1);
1.2603 +{\out Level 7}
1.2604 +{\out [| \ldots |] ==> (f Un g) ` a = f ` a}
1.2605 +{\out 1. [| \ldots |] ==> g : ?C3 -> ?D3}
1.2606 +{\out 2. [| \ldots |] ==> A Int ?C3 = 0}
1.2607 +by (assume_tac 1);
1.2608 +{\out Level 8}
1.2609 +{\out [| \ldots |] ==> (f Un g) ` a = f ` a}
1.2610 +{\out 1. [| \ldots |] ==> A Int C = 0}
1.2611 +by (assume_tac 1);
1.2612 +{\out Level 9}
1.2613 +{\out [| \ldots |] ==> (f Un g) ` a = f ` a}
1.2614 +{\out No subgoals!}
1.2615 +\end{ttbox}
1.2616 +See the files \texttt{ZF/func.ML} and \texttt{ZF/WF.ML} for more
1.2617 +examples of reasoning about functions.
1.2618 +
1.2619 +\index{set theory|)}