wenzelm@30184
|
1 |
|
nipkow@3950
|
2 |
\chapter{Simplification}
|
nipkow@3950
|
3 |
\label{chap:simplification}
|
lcp@104
|
4 |
\index{simplification|(}
|
lcp@104
|
5 |
|
wenzelm@9695
|
6 |
This chapter describes Isabelle's generic simplification package. It performs
|
wenzelm@9695
|
7 |
conditional and unconditional rewriting and uses contextual information
|
wenzelm@9695
|
8 |
(`local assumptions'). It provides several general hooks, which can provide
|
wenzelm@9695
|
9 |
automatic case splits during rewriting, for example. The simplifier is
|
wenzelm@9695
|
10 |
already set up for many of Isabelle's logics: FOL, ZF, HOL, HOLCF.
|
nipkow@1860
|
11 |
|
wenzelm@4395
|
12 |
The first section is a quick introduction to the simplifier that
|
wenzelm@4395
|
13 |
should be sufficient to get started. The later sections explain more
|
wenzelm@4395
|
14 |
advanced features.
|
wenzelm@4395
|
15 |
|
nipkow@1860
|
16 |
|
nipkow@1860
|
17 |
\section{Simplification for dummies}
|
nipkow@1860
|
18 |
\label{sec:simp-for-dummies}
|
nipkow@1860
|
19 |
|
wenzelm@4395
|
20 |
Basic use of the simplifier is particularly easy because each theory
|
wenzelm@4557
|
21 |
is equipped with sensible default information controlling the rewrite
|
wenzelm@4557
|
22 |
process --- namely the implicit {\em current
|
wenzelm@4557
|
23 |
simpset}\index{simpset!current}. A suite of simple commands is
|
wenzelm@4557
|
24 |
provided that refer to the implicit simpset of the current theory
|
wenzelm@4557
|
25 |
context.
|
wenzelm@4395
|
26 |
|
wenzelm@4395
|
27 |
\begin{warn}
|
wenzelm@4395
|
28 |
Make sure that you are working within the correct theory context.
|
wenzelm@4395
|
29 |
Executing proofs interactively, or loading them from ML files
|
wenzelm@4395
|
30 |
without associated theories may require setting the current theory
|
wenzelm@4395
|
31 |
manually via the \ttindex{context} command.
|
wenzelm@4395
|
32 |
\end{warn}
|
wenzelm@4395
|
33 |
|
wenzelm@4395
|
34 |
\subsection{Simplification tactics} \label{sec:simp-for-dummies-tacs}
|
nipkow@1860
|
35 |
\begin{ttbox}
|
wenzelm@4395
|
36 |
Simp_tac : int -> tactic
|
wenzelm@4395
|
37 |
Asm_simp_tac : int -> tactic
|
wenzelm@4395
|
38 |
Full_simp_tac : int -> tactic
|
wenzelm@4395
|
39 |
Asm_full_simp_tac : int -> tactic
|
wenzelm@4395
|
40 |
trace_simp : bool ref \hfill{\bf initially false}
|
wenzelm@7920
|
41 |
debug_simp : bool ref \hfill{\bf initially false}
|
wenzelm@4395
|
42 |
\end{ttbox}
|
wenzelm@4395
|
43 |
|
wenzelm@4395
|
44 |
\begin{ttdescription}
|
wenzelm@4395
|
45 |
\item[\ttindexbold{Simp_tac} $i$] simplifies subgoal~$i$ using the
|
wenzelm@4395
|
46 |
current simpset. It may solve the subgoal completely if it has
|
wenzelm@4395
|
47 |
become trivial, using the simpset's solver tactic.
|
wenzelm@4395
|
48 |
|
wenzelm@4395
|
49 |
\item[\ttindexbold{Asm_simp_tac}]\index{assumptions!in simplification}
|
wenzelm@4395
|
50 |
is like \verb$Simp_tac$, but extracts additional rewrite rules from
|
wenzelm@4395
|
51 |
the local assumptions.
|
wenzelm@4395
|
52 |
|
wenzelm@4395
|
53 |
\item[\ttindexbold{Full_simp_tac}] is like \verb$Simp_tac$, but also
|
wenzelm@4395
|
54 |
simplifies the assumptions (without using the assumptions to
|
wenzelm@4395
|
55 |
simplify each other or the actual goal).
|
wenzelm@4395
|
56 |
|
wenzelm@4395
|
57 |
\item[\ttindexbold{Asm_full_simp_tac}] is like \verb$Asm_simp_tac$,
|
nipkow@4889
|
58 |
but also simplifies the assumptions. In particular, assumptions can
|
nipkow@4889
|
59 |
simplify each other.
|
nipkow@4889
|
60 |
\footnote{\texttt{Asm_full_simp_tac} used to process the assumptions from
|
nipkow@4889
|
61 |
left to right. For backwards compatibilty reasons only there is now
|
nipkow@4889
|
62 |
\texttt{Asm_lr_simp_tac} that behaves like the old \texttt{Asm_full_simp_tac}.}
|
wenzelm@7920
|
63 |
\item[set \ttindexbold{trace_simp};] makes the simplifier output internal
|
wenzelm@7920
|
64 |
operations. This includes rewrite steps, but also bookkeeping like
|
wenzelm@7920
|
65 |
modifications of the simpset.
|
wenzelm@7920
|
66 |
\item[set \ttindexbold{debug_simp};] makes the simplifier output some extra
|
wenzelm@7920
|
67 |
information about internal operations. This includes any attempted
|
wenzelm@7920
|
68 |
invocation of simplification procedures.
|
wenzelm@4395
|
69 |
\end{ttdescription}
|
wenzelm@4395
|
70 |
|
wenzelm@4395
|
71 |
\medskip
|
wenzelm@4395
|
72 |
|
wenzelm@9695
|
73 |
As an example, consider the theory of arithmetic in HOL. The (rather trivial)
|
wenzelm@9695
|
74 |
goal $0 + (x + 0) = x + 0 + 0$ can be solved by a single call of
|
wenzelm@9695
|
75 |
\texttt{Simp_tac} as follows:
|
wenzelm@4395
|
76 |
\begin{ttbox}
|
wenzelm@4395
|
77 |
context Arith.thy;
|
paulson@5205
|
78 |
Goal "0 + (x + 0) = x + 0 + 0";
|
nipkow@1860
|
79 |
{\out 1. 0 + (x + 0) = x + 0 + 0}
|
wenzelm@4395
|
80 |
by (Simp_tac 1);
|
wenzelm@4395
|
81 |
{\out Level 1}
|
wenzelm@4395
|
82 |
{\out 0 + (x + 0) = x + 0 + 0}
|
wenzelm@4395
|
83 |
{\out No subgoals!}
|
nipkow@1860
|
84 |
\end{ttbox}
|
wenzelm@4395
|
85 |
|
wenzelm@4395
|
86 |
The simplifier uses the current simpset of \texttt{Arith.thy}, which
|
wenzelm@4395
|
87 |
contains suitable theorems like $\Var{n}+0 = \Var{n}$ and $0+\Var{n} =
|
nipkow@1860
|
88 |
\Var{n}$.
|
nipkow@1860
|
89 |
|
wenzelm@4395
|
90 |
\medskip In many cases, assumptions of a subgoal are also needed in
|
wenzelm@4395
|
91 |
the simplification process. For example, \texttt{x = 0 ==> x + x = 0}
|
wenzelm@4395
|
92 |
is solved by \texttt{Asm_simp_tac} as follows:
|
nipkow@1860
|
93 |
\begin{ttbox}
|
nipkow@1860
|
94 |
{\out 1. x = 0 ==> x + x = 0}
|
paulson@2479
|
95 |
by (Asm_simp_tac 1);
|
nipkow@1860
|
96 |
\end{ttbox}
|
nipkow@1860
|
97 |
|
paulson@4597
|
98 |
\medskip \texttt{Asm_full_simp_tac} is the most powerful of this quartet
|
wenzelm@4395
|
99 |
of tactics but may also loop where some of the others terminate. For
|
wenzelm@4395
|
100 |
example,
|
nipkow@1860
|
101 |
\begin{ttbox}
|
wenzelm@4395
|
102 |
{\out 1. ALL x. f x = g (f (g x)) ==> f 0 = f 0 + 0}
|
nipkow@1860
|
103 |
\end{ttbox}
|
paulson@4597
|
104 |
is solved by \texttt{Simp_tac}, but \texttt{Asm_simp_tac} and {\tt
|
berghofe@13616
|
105 |
Asm_full_simp_tac} loop because the rewrite rule $f\,\Var{x} =
|
wenzelm@4395
|
106 |
g\,(f\,(g\,\Var{x}))$ extracted from the assumption does not
|
wenzelm@4395
|
107 |
terminate. Isabelle notices certain simple forms of nontermination,
|
nipkow@4889
|
108 |
but not this one. Because assumptions may simplify each other, there can be
|
berghofe@13616
|
109 |
very subtle cases of nontermination. For example, invoking
|
berghofe@13616
|
110 |
{\tt Asm_full_simp_tac} on
|
nipkow@1860
|
111 |
\begin{ttbox}
|
berghofe@13616
|
112 |
{\out 1. [| P (f x); y = x; f x = f y |] ==> Q}
|
nipkow@1860
|
113 |
\end{ttbox}
|
berghofe@13616
|
114 |
gives rise to the infinite reduction sequence
|
berghofe@13616
|
115 |
\[
|
berghofe@13693
|
116 |
P\,(f\,x) \stackrel{f\,x = f\,y}{\longmapsto} P\,(f\,y) \stackrel{y = x}{\longmapsto}
|
berghofe@13693
|
117 |
P\,(f\,x) \stackrel{f\,x = f\,y}{\longmapsto} \cdots
|
berghofe@13616
|
118 |
\]
|
berghofe@13616
|
119 |
whereas applying the same tactic to
|
berghofe@13616
|
120 |
\begin{ttbox}
|
berghofe@13616
|
121 |
{\out 1. [| y = x; f x = f y; P (f x) |] ==> Q}
|
berghofe@13616
|
122 |
\end{ttbox}
|
berghofe@13616
|
123 |
terminates.
|
nipkow@1860
|
124 |
|
wenzelm@4395
|
125 |
\medskip
|
wenzelm@4395
|
126 |
|
wenzelm@3108
|
127 |
Using the simplifier effectively may take a bit of experimentation.
|
wenzelm@4395
|
128 |
Set the \verb$trace_simp$\index{tracing!of simplification} flag to get
|
wenzelm@4395
|
129 |
a better idea of what is going on. The resulting output can be
|
wenzelm@4395
|
130 |
enormous, especially since invocations of the simplifier are often
|
wenzelm@4395
|
131 |
nested (e.g.\ when solving conditions of rewrite rules).
|
nipkow@1860
|
132 |
|
wenzelm@4395
|
133 |
|
wenzelm@4395
|
134 |
\subsection{Modifying the current simpset}
|
nipkow@1860
|
135 |
\begin{ttbox}
|
wenzelm@4395
|
136 |
Addsimps : thm list -> unit
|
wenzelm@4395
|
137 |
Delsimps : thm list -> unit
|
wenzelm@4395
|
138 |
Addsimprocs : simproc list -> unit
|
wenzelm@4395
|
139 |
Delsimprocs : simproc list -> unit
|
wenzelm@4395
|
140 |
Addcongs : thm list -> unit
|
wenzelm@4395
|
141 |
Delcongs : thm list -> unit
|
oheimb@5549
|
142 |
Addsplits : thm list -> unit
|
oheimb@5549
|
143 |
Delsplits : thm list -> unit
|
wenzelm@4395
|
144 |
\end{ttbox}
|
wenzelm@4395
|
145 |
|
wenzelm@4395
|
146 |
Depending on the theory context, the \texttt{Add} and \texttt{Del}
|
wenzelm@4395
|
147 |
functions manipulate basic components of the associated current
|
wenzelm@4395
|
148 |
simpset. Internally, all rewrite rules have to be expressed as
|
wenzelm@4395
|
149 |
(conditional) meta-equalities. This form is derived automatically
|
wenzelm@4395
|
150 |
from object-level equations that are supplied by the user. Another
|
wenzelm@4395
|
151 |
source of rewrite rules are \emph{simplification procedures}, that is
|
wenzelm@4395
|
152 |
\ML\ functions that produce suitable theorems on demand, depending on
|
wenzelm@4395
|
153 |
the current redex. Congruences are a more advanced feature; see
|
oheimb@11181
|
154 |
{\S}\ref{sec:simp-congs}.
|
wenzelm@4395
|
155 |
|
wenzelm@4395
|
156 |
\begin{ttdescription}
|
wenzelm@4395
|
157 |
|
wenzelm@4395
|
158 |
\item[\ttindexbold{Addsimps} $thms$;] adds rewrite rules derived from
|
wenzelm@4395
|
159 |
$thms$ to the current simpset.
|
wenzelm@4395
|
160 |
|
wenzelm@4395
|
161 |
\item[\ttindexbold{Delsimps} $thms$;] deletes rewrite rules derived
|
wenzelm@4395
|
162 |
from $thms$ from the current simpset.
|
wenzelm@4395
|
163 |
|
wenzelm@4395
|
164 |
\item[\ttindexbold{Addsimprocs} $procs$;] adds simplification
|
wenzelm@4395
|
165 |
procedures $procs$ to the current simpset.
|
wenzelm@4395
|
166 |
|
wenzelm@4395
|
167 |
\item[\ttindexbold{Delsimprocs} $procs$;] deletes simplification
|
wenzelm@4395
|
168 |
procedures $procs$ from the current simpset.
|
wenzelm@4395
|
169 |
|
wenzelm@4395
|
170 |
\item[\ttindexbold{Addcongs} $thms$;] adds congruence rules to the
|
wenzelm@4395
|
171 |
current simpset.
|
wenzelm@4395
|
172 |
|
oheimb@5549
|
173 |
\item[\ttindexbold{Delcongs} $thms$;] deletes congruence rules from the
|
oheimb@5549
|
174 |
current simpset.
|
oheimb@5549
|
175 |
|
oheimb@5549
|
176 |
\item[\ttindexbold{Addsplits} $thms$;] adds splitting rules to the
|
oheimb@5549
|
177 |
current simpset.
|
oheimb@5549
|
178 |
|
oheimb@5549
|
179 |
\item[\ttindexbold{Delsplits} $thms$;] deletes splitting rules from the
|
wenzelm@4395
|
180 |
current simpset.
|
wenzelm@4395
|
181 |
|
wenzelm@4395
|
182 |
\end{ttdescription}
|
wenzelm@4395
|
183 |
|
wenzelm@9695
|
184 |
When a new theory is built, its implicit simpset is initialized by the union
|
wenzelm@9695
|
185 |
of the respective simpsets of its parent theories. In addition, certain
|
wenzelm@9695
|
186 |
theory definition constructs (e.g.\ \ttindex{datatype} and \ttindex{primrec}
|
wenzelm@9695
|
187 |
in HOL) implicitly augment the current simpset. Ordinary definitions are not
|
wenzelm@9695
|
188 |
added automatically!
|
wenzelm@4395
|
189 |
|
wenzelm@4395
|
190 |
It is up the user to manipulate the current simpset further by
|
wenzelm@4395
|
191 |
explicitly adding or deleting theorems and simplification procedures.
|
wenzelm@4395
|
192 |
|
wenzelm@4395
|
193 |
\medskip
|
wenzelm@4395
|
194 |
|
paulson@5205
|
195 |
Good simpsets are hard to design. Rules that obviously simplify,
|
paulson@5205
|
196 |
like $\Var{n}+0 = \Var{n}$, should be added to the current simpset right after
|
paulson@5205
|
197 |
they have been proved. More specific ones (such as distributive laws, which
|
paulson@5205
|
198 |
duplicate subterms) should be added only for specific proofs and deleted
|
paulson@5205
|
199 |
afterwards. Conversely, sometimes a rule needs
|
paulson@5205
|
200 |
to be removed for a certain proof and restored afterwards. The need of
|
paulson@5205
|
201 |
frequent additions or deletions may indicate a badly designed
|
paulson@5205
|
202 |
simpset.
|
wenzelm@4395
|
203 |
|
wenzelm@4395
|
204 |
\begin{warn}
|
wenzelm@4395
|
205 |
The union of the parent simpsets (as described above) is not always
|
wenzelm@4395
|
206 |
a good starting point for the new theory. If some ancestors have
|
wenzelm@4395
|
207 |
deleted simplification rules because they are no longer wanted,
|
wenzelm@4395
|
208 |
while others have left those rules in, then the union will contain
|
paulson@5205
|
209 |
the unwanted rules. After this union is formed, changes to
|
paulson@5205
|
210 |
a parent simpset have no effect on the child simpset.
|
wenzelm@4395
|
211 |
\end{warn}
|
wenzelm@4395
|
212 |
|
wenzelm@4395
|
213 |
|
wenzelm@4395
|
214 |
\section{Simplification sets}\index{simplification sets}
|
wenzelm@4395
|
215 |
|
wenzelm@4395
|
216 |
The simplifier is controlled by information contained in {\bf
|
wenzelm@4395
|
217 |
simpsets}. These consist of several components, including rewrite
|
wenzelm@4395
|
218 |
rules, simplification procedures, congruence rules, and the subgoaler,
|
wenzelm@4395
|
219 |
solver and looper tactics. The simplifier should be set up with
|
wenzelm@4395
|
220 |
sensible defaults so that most simplifier calls specify only rewrite
|
wenzelm@4395
|
221 |
rules or simplification procedures. Experienced users can exploit the
|
wenzelm@4395
|
222 |
other components to streamline proofs in more sophisticated manners.
|
wenzelm@4395
|
223 |
|
wenzelm@4395
|
224 |
\subsection{Inspecting simpsets}
|
wenzelm@4395
|
225 |
\begin{ttbox}
|
wenzelm@4395
|
226 |
print_ss : simpset -> unit
|
nipkow@4889
|
227 |
rep_ss : simpset -> \{mss : meta_simpset,
|
oheimb@4664
|
228 |
subgoal_tac: simpset -> int -> tactic,
|
nipkow@7620
|
229 |
loop_tacs : (string * (int -> tactic))list,
|
nipkow@7620
|
230 |
finish_tac : solver list,
|
nipkow@7620
|
231 |
unsafe_finish_tac : solver list\}
|
nipkow@1860
|
232 |
\end{ttbox}
|
nipkow@1860
|
233 |
\begin{ttdescription}
|
wenzelm@4395
|
234 |
|
wenzelm@4395
|
235 |
\item[\ttindexbold{print_ss} $ss$;] displays the printable contents of
|
wenzelm@4395
|
236 |
simpset $ss$. This includes the rewrite rules and congruences in
|
wenzelm@4395
|
237 |
their internal form expressed as meta-equalities. The names of the
|
wenzelm@4395
|
238 |
simplification procedures and the patterns they are invoked on are
|
wenzelm@4395
|
239 |
also shown. The other parts, functions and tactics, are
|
wenzelm@4395
|
240 |
non-printable.
|
wenzelm@4395
|
241 |
|
oheimb@4664
|
242 |
\item[\ttindexbold{rep_ss} $ss$;] decomposes $ss$ as a record of its internal
|
oheimb@4664
|
243 |
components, namely the meta_simpset, the subgoaler, the loop, and the safe
|
oheimb@4664
|
244 |
and unsafe solvers.
|
oheimb@4664
|
245 |
|
nipkow@1860
|
246 |
\end{ttdescription}
|
nipkow@1860
|
247 |
|
nipkow@1860
|
248 |
|
wenzelm@4395
|
249 |
\subsection{Building simpsets}
|
wenzelm@4395
|
250 |
\begin{ttbox}
|
wenzelm@4395
|
251 |
empty_ss : simpset
|
wenzelm@4395
|
252 |
merge_ss : simpset * simpset -> simpset
|
wenzelm@4395
|
253 |
\end{ttbox}
|
wenzelm@4395
|
254 |
\begin{ttdescription}
|
wenzelm@4395
|
255 |
|
wenzelm@9695
|
256 |
\item[\ttindexbold{empty_ss}] is the empty simpset. This is not very useful
|
wenzelm@9695
|
257 |
under normal circumstances because it doesn't contain suitable tactics
|
wenzelm@9695
|
258 |
(subgoaler etc.). When setting up the simplifier for a particular
|
wenzelm@9695
|
259 |
object-logic, one will typically define a more appropriate ``almost empty''
|
wenzelm@9695
|
260 |
simpset. For example, in HOL this is called \ttindexbold{HOL_basic_ss}.
|
wenzelm@4395
|
261 |
|
wenzelm@4395
|
262 |
\item[\ttindexbold{merge_ss} ($ss@1$, $ss@2$)] merges simpsets $ss@1$
|
wenzelm@4395
|
263 |
and $ss@2$ by building the union of their respective rewrite rules,
|
wenzelm@4395
|
264 |
simplification procedures and congruences. The other components
|
wenzelm@4557
|
265 |
(tactics etc.) cannot be merged, though; they are taken from either
|
wenzelm@4557
|
266 |
simpset\footnote{Actually from $ss@1$, but it would unwise to count
|
wenzelm@4557
|
267 |
on that.}.
|
paulson@2479
|
268 |
|
wenzelm@4395
|
269 |
\end{ttdescription}
|
lcp@104
|
270 |
|
lcp@104
|
271 |
|
wenzelm@4395
|
272 |
\subsection{Accessing the current simpset}
|
oheimb@5575
|
273 |
\label{sec:access-current-simpset}
|
lcp@323
|
274 |
|
wenzelm@4395
|
275 |
\begin{ttbox}
|
oheimb@5575
|
276 |
simpset : unit -> simpset
|
oheimb@5575
|
277 |
simpset_ref : unit -> simpset ref
|
wenzelm@4395
|
278 |
simpset_of : theory -> simpset
|
wenzelm@4395
|
279 |
simpset_ref_of : theory -> simpset ref
|
wenzelm@4395
|
280 |
print_simpset : theory -> unit
|
oheimb@5575
|
281 |
SIMPSET :(simpset -> tactic) -> tactic
|
oheimb@5575
|
282 |
SIMPSET' :(simpset -> 'a -> tactic) -> 'a -> tactic
|
wenzelm@4395
|
283 |
\end{ttbox}
|
wenzelm@4395
|
284 |
|
wenzelm@4395
|
285 |
Each theory contains a current simpset\index{simpset!current} stored
|
wenzelm@4395
|
286 |
within a private ML reference variable. This can be retrieved and
|
wenzelm@4395
|
287 |
modified as follows.
|
wenzelm@4395
|
288 |
|
wenzelm@4395
|
289 |
\begin{ttdescription}
|
wenzelm@4395
|
290 |
|
wenzelm@4395
|
291 |
\item[\ttindexbold{simpset}();] retrieves the simpset value from the
|
wenzelm@4395
|
292 |
current theory context.
|
wenzelm@4395
|
293 |
|
wenzelm@4395
|
294 |
\item[\ttindexbold{simpset_ref}();] retrieves the simpset reference
|
wenzelm@4395
|
295 |
variable from the current theory context. This can be assigned to
|
wenzelm@4395
|
296 |
by using \texttt{:=} in ML.
|
wenzelm@4395
|
297 |
|
wenzelm@4395
|
298 |
\item[\ttindexbold{simpset_of} $thy$;] retrieves the simpset value
|
wenzelm@4395
|
299 |
from theory $thy$.
|
wenzelm@4395
|
300 |
|
wenzelm@4395
|
301 |
\item[\ttindexbold{simpset_ref_of} $thy$;] retrieves the simpset
|
wenzelm@4395
|
302 |
reference variable from theory $thy$.
|
wenzelm@4395
|
303 |
|
oheimb@5575
|
304 |
\item[\ttindexbold{print_simpset} $thy$;] prints the current simpset
|
oheimb@5575
|
305 |
of theory $thy$ in the same way as \texttt{print_ss}.
|
oheimb@5575
|
306 |
|
wenzelm@5574
|
307 |
\item[\ttindexbold{SIMPSET} $tacf$, \ttindexbold{SIMPSET'} $tacf'$]
|
wenzelm@5574
|
308 |
are tacticals that make a tactic depend on the implicit current
|
wenzelm@5574
|
309 |
simpset of the theory associated with the proof state they are
|
wenzelm@5574
|
310 |
applied on.
|
wenzelm@5574
|
311 |
|
wenzelm@4395
|
312 |
\end{ttdescription}
|
wenzelm@4395
|
313 |
|
wenzelm@5574
|
314 |
\begin{warn}
|
paulson@8136
|
315 |
There is a small difference between \texttt{(SIMPSET'~$tacf$)} and
|
paulson@8136
|
316 |
\texttt{($tacf\,$(simpset()))}. For example \texttt{(SIMPSET'
|
wenzelm@5574
|
317 |
simp_tac)} would depend on the theory of the proof state it is
|
wenzelm@5574
|
318 |
applied to, while \texttt{(simp_tac (simpset()))} implicitly refers
|
wenzelm@5574
|
319 |
to the current theory context. Both are usually the same in proof
|
wenzelm@5574
|
320 |
scripts, provided that goals are only stated within the current
|
wenzelm@5574
|
321 |
theory. Robust programs would not count on that, of course.
|
wenzelm@5574
|
322 |
\end{warn}
|
wenzelm@5574
|
323 |
|
lcp@104
|
324 |
|
lcp@332
|
325 |
\subsection{Rewrite rules}
|
wenzelm@4395
|
326 |
\begin{ttbox}
|
wenzelm@4395
|
327 |
addsimps : simpset * thm list -> simpset \hfill{\bf infix 4}
|
wenzelm@4395
|
328 |
delsimps : simpset * thm list -> simpset \hfill{\bf infix 4}
|
wenzelm@4395
|
329 |
\end{ttbox}
|
wenzelm@4395
|
330 |
|
wenzelm@4395
|
331 |
\index{rewrite rules|(} Rewrite rules are theorems expressing some
|
wenzelm@4395
|
332 |
form of equality, for example:
|
lcp@323
|
333 |
\begin{eqnarray*}
|
lcp@323
|
334 |
Suc(\Var{m}) + \Var{n} &=& \Var{m} + Suc(\Var{n}) \\
|
lcp@323
|
335 |
\Var{P}\conj\Var{P} &\bimp& \Var{P} \\
|
nipkow@714
|
336 |
\Var{A} \un \Var{B} &\equiv& \{x.x\in \Var{A} \disj x\in \Var{B}\}
|
lcp@323
|
337 |
\end{eqnarray*}
|
nipkow@1860
|
338 |
Conditional rewrites such as $\Var{m}<\Var{n} \Imp \Var{m}/\Var{n} =
|
wenzelm@4395
|
339 |
0$ are also permitted; the conditions can be arbitrary formulas.
|
lcp@104
|
340 |
|
wenzelm@4395
|
341 |
Internally, all rewrite rules are translated into meta-equalities,
|
wenzelm@4395
|
342 |
theorems with conclusion $lhs \equiv rhs$. Each simpset contains a
|
wenzelm@4395
|
343 |
function for extracting equalities from arbitrary theorems. For
|
oheimb@11181
|
344 |
example, $\neg(\Var{x}\in \{\})$ could be turned into $\Var{x}\in \{\}
|
wenzelm@4395
|
345 |
\equiv False$. This function can be installed using
|
wenzelm@4395
|
346 |
\ttindex{setmksimps} but only the definer of a logic should need to do
|
oheimb@11181
|
347 |
this; see {\S}\ref{sec:setmksimps}. The function processes theorems
|
wenzelm@4395
|
348 |
added by \texttt{addsimps} as well as local assumptions.
|
lcp@104
|
349 |
|
wenzelm@4395
|
350 |
\begin{ttdescription}
|
wenzelm@4395
|
351 |
|
wenzelm@4395
|
352 |
\item[$ss$ \ttindexbold{addsimps} $thms$] adds rewrite rules derived
|
wenzelm@4395
|
353 |
from $thms$ to the simpset $ss$.
|
wenzelm@4395
|
354 |
|
wenzelm@4395
|
355 |
\item[$ss$ \ttindexbold{delsimps} $thms$] deletes rewrite rules
|
wenzelm@4395
|
356 |
derived from $thms$ from the simpset $ss$.
|
wenzelm@4395
|
357 |
|
wenzelm@4395
|
358 |
\end{ttdescription}
|
lcp@104
|
359 |
|
lcp@332
|
360 |
\begin{warn}
|
wenzelm@4395
|
361 |
The simplifier will accept all standard rewrite rules: those where
|
wenzelm@4395
|
362 |
all unknowns are of base type. Hence ${\Var{i}+(\Var{j}+\Var{k})} =
|
wenzelm@4395
|
363 |
{(\Var{i}+\Var{j})+\Var{k}}$ is OK.
|
wenzelm@4395
|
364 |
|
wenzelm@4395
|
365 |
It will also deal gracefully with all rules whose left-hand sides
|
wenzelm@4395
|
366 |
are so-called {\em higher-order patterns}~\cite{nipkow-patterns}.
|
wenzelm@4395
|
367 |
\indexbold{higher-order pattern}\indexbold{pattern, higher-order}
|
wenzelm@4395
|
368 |
These are terms in $\beta$-normal form (this will always be the case
|
wenzelm@4395
|
369 |
unless you have done something strange) where each occurrence of an
|
wenzelm@4395
|
370 |
unknown is of the form $\Var{F}(x@1,\dots,x@n)$, where the $x@i$ are
|
wenzelm@4395
|
371 |
distinct bound variables. Hence $(\forall x.\Var{P}(x) \land
|
wenzelm@4395
|
372 |
\Var{Q}(x)) \bimp (\forall x.\Var{P}(x)) \land (\forall
|
wenzelm@4395
|
373 |
x.\Var{Q}(x))$ is also OK, in both directions.
|
wenzelm@4395
|
374 |
|
wenzelm@4395
|
375 |
In some rare cases the rewriter will even deal with quite general
|
wenzelm@4395
|
376 |
rules: for example ${\Var{f}(\Var{x})\in range(\Var{f})} = True$
|
wenzelm@4395
|
377 |
rewrites $g(a) \in range(g)$ to $True$, but will fail to match
|
wenzelm@4395
|
378 |
$g(h(b)) \in range(\lambda x.g(h(x)))$. However, you can replace
|
wenzelm@4395
|
379 |
the offending subterms (in our case $\Var{f}(\Var{x})$, which is not
|
wenzelm@4395
|
380 |
a pattern) by adding new variables and conditions: $\Var{y} =
|
wenzelm@4395
|
381 |
\Var{f}(\Var{x}) \Imp \Var{y}\in range(\Var{f}) = True$ is
|
wenzelm@4395
|
382 |
acceptable as a conditional rewrite rule since conditions can be
|
wenzelm@4395
|
383 |
arbitrary terms.
|
wenzelm@4395
|
384 |
|
wenzelm@4395
|
385 |
There is basically no restriction on the form of the right-hand
|
wenzelm@4395
|
386 |
sides. They may not contain extraneous term or type variables,
|
wenzelm@4395
|
387 |
though.
|
lcp@104
|
388 |
\end{warn}
|
lcp@332
|
389 |
\index{rewrite rules|)}
|
lcp@332
|
390 |
|
wenzelm@4395
|
391 |
|
nipkow@4947
|
392 |
\subsection{*Simplification procedures}
|
wenzelm@4395
|
393 |
\begin{ttbox}
|
wenzelm@4395
|
394 |
addsimprocs : simpset * simproc list -> simpset
|
wenzelm@4395
|
395 |
delsimprocs : simpset * simproc list -> simpset
|
wenzelm@4395
|
396 |
\end{ttbox}
|
wenzelm@4395
|
397 |
|
wenzelm@4557
|
398 |
Simplification procedures are {\ML} objects of abstract type
|
wenzelm@4557
|
399 |
\texttt{simproc}. Basically they are just functions that may produce
|
wenzelm@4395
|
400 |
\emph{proven} rewrite rules on demand. They are associated with
|
wenzelm@4395
|
401 |
certain patterns that conceptually represent left-hand sides of
|
wenzelm@4395
|
402 |
equations; these are shown by \texttt{print_ss}. During its
|
wenzelm@4395
|
403 |
operation, the simplifier may offer a simplification procedure the
|
wenzelm@4395
|
404 |
current redex and ask for a suitable rewrite rule. Thus rules may be
|
wenzelm@4395
|
405 |
specifically fashioned for particular situations, resulting in a more
|
wenzelm@4395
|
406 |
powerful mechanism than term rewriting by a fixed set of rules.
|
wenzelm@4395
|
407 |
|
wenzelm@4395
|
408 |
|
wenzelm@4395
|
409 |
\begin{ttdescription}
|
wenzelm@4395
|
410 |
|
paulson@4597
|
411 |
\item[$ss$ \ttindexbold{addsimprocs} $procs$] adds the simplification
|
wenzelm@4395
|
412 |
procedures $procs$ to the current simpset.
|
wenzelm@4395
|
413 |
|
paulson@4597
|
414 |
\item[$ss$ \ttindexbold{delsimprocs} $procs$] deletes the simplification
|
wenzelm@4395
|
415 |
procedures $procs$ from the current simpset.
|
wenzelm@4395
|
416 |
|
wenzelm@4395
|
417 |
\end{ttdescription}
|
wenzelm@4395
|
418 |
|
wenzelm@4557
|
419 |
For example, simplification procedures \ttindexbold{nat_cancel} of
|
wenzelm@4557
|
420 |
\texttt{HOL/Arith} cancel common summands and constant factors out of
|
wenzelm@4557
|
421 |
several relations of sums over natural numbers.
|
wenzelm@4557
|
422 |
|
wenzelm@4557
|
423 |
Consider the following goal, which after cancelling $a$ on both sides
|
wenzelm@4557
|
424 |
contains a factor of $2$. Simplifying with the simpset of
|
wenzelm@4557
|
425 |
\texttt{Arith.thy} will do the cancellation automatically:
|
wenzelm@4557
|
426 |
\begin{ttbox}
|
wenzelm@4557
|
427 |
{\out 1. x + a + x < y + y + 2 + a + a + a + a + a}
|
wenzelm@4557
|
428 |
by (Simp_tac 1);
|
wenzelm@4557
|
429 |
{\out 1. x < Suc (a + (a + y))}
|
wenzelm@4557
|
430 |
\end{ttbox}
|
wenzelm@4557
|
431 |
|
wenzelm@4395
|
432 |
|
wenzelm@4395
|
433 |
\subsection{*Congruence rules}\index{congruence rules}\label{sec:simp-congs}
|
wenzelm@4395
|
434 |
\begin{ttbox}
|
wenzelm@4395
|
435 |
addcongs : simpset * thm list -> simpset \hfill{\bf infix 4}
|
wenzelm@4395
|
436 |
delcongs : simpset * thm list -> simpset \hfill{\bf infix 4}
|
wenzelm@4395
|
437 |
addeqcongs : simpset * thm list -> simpset \hfill{\bf infix 4}
|
wenzelm@4395
|
438 |
deleqcongs : simpset * thm list -> simpset \hfill{\bf infix 4}
|
wenzelm@4395
|
439 |
\end{ttbox}
|
wenzelm@4395
|
440 |
|
lcp@104
|
441 |
Congruence rules are meta-equalities of the form
|
wenzelm@3108
|
442 |
\[ \dots \Imp
|
lcp@104
|
443 |
f(\Var{x@1},\ldots,\Var{x@n}) \equiv f(\Var{y@1},\ldots,\Var{y@n}).
|
lcp@104
|
444 |
\]
|
lcp@323
|
445 |
This governs the simplification of the arguments of~$f$. For
|
lcp@104
|
446 |
example, some arguments can be simplified under additional assumptions:
|
lcp@104
|
447 |
\[ \List{\Var{P@1} \bimp \Var{Q@1};\; \Var{Q@1} \Imp \Var{P@2} \bimp \Var{Q@2}}
|
lcp@104
|
448 |
\Imp (\Var{P@1} \imp \Var{P@2}) \equiv (\Var{Q@1} \imp \Var{Q@2})
|
lcp@104
|
449 |
\]
|
wenzelm@4395
|
450 |
Given this rule, the simplifier assumes $Q@1$ and extracts rewrite
|
wenzelm@4395
|
451 |
rules from it when simplifying~$P@2$. Such local assumptions are
|
wenzelm@4395
|
452 |
effective for rewriting formulae such as $x=0\imp y+x=y$. The local
|
wenzelm@4395
|
453 |
assumptions are also provided as theorems to the solver; see
|
oheimb@11181
|
454 |
{\S}~\ref{sec:simp-solver} below.
|
lcp@698
|
455 |
|
wenzelm@4395
|
456 |
\begin{ttdescription}
|
wenzelm@4395
|
457 |
|
wenzelm@4395
|
458 |
\item[$ss$ \ttindexbold{addcongs} $thms$] adds congruence rules to the
|
wenzelm@4395
|
459 |
simpset $ss$. These are derived from $thms$ in an appropriate way,
|
wenzelm@4395
|
460 |
depending on the underlying object-logic.
|
wenzelm@4395
|
461 |
|
wenzelm@4395
|
462 |
\item[$ss$ \ttindexbold{delcongs} $thms$] deletes congruence rules
|
wenzelm@4395
|
463 |
derived from $thms$.
|
wenzelm@4395
|
464 |
|
wenzelm@4395
|
465 |
\item[$ss$ \ttindexbold{addeqcongs} $thms$] adds congruence rules in
|
wenzelm@4395
|
466 |
their internal form (conclusions using meta-equality) to simpset
|
wenzelm@4395
|
467 |
$ss$. This is the basic mechanism that \texttt{addcongs} is built
|
wenzelm@4395
|
468 |
on. It should be rarely used directly.
|
wenzelm@4395
|
469 |
|
wenzelm@4395
|
470 |
\item[$ss$ \ttindexbold{deleqcongs} $thms$] deletes congruence rules
|
wenzelm@4395
|
471 |
in internal form from simpset $ss$.
|
wenzelm@4395
|
472 |
|
wenzelm@4395
|
473 |
\end{ttdescription}
|
wenzelm@4395
|
474 |
|
wenzelm@4395
|
475 |
\medskip
|
wenzelm@4395
|
476 |
|
wenzelm@4395
|
477 |
Here are some more examples. The congruence rule for bounded
|
wenzelm@4395
|
478 |
quantifiers also supplies contextual information, this time about the
|
wenzelm@4395
|
479 |
bound variable:
|
lcp@286
|
480 |
\begin{eqnarray*}
|
lcp@286
|
481 |
&&\List{\Var{A}=\Var{B};\;
|
lcp@286
|
482 |
\Forall x. x\in \Var{B} \Imp \Var{P}(x) = \Var{Q}(x)} \Imp{} \\
|
lcp@286
|
483 |
&&\qquad\qquad
|
lcp@286
|
484 |
(\forall x\in \Var{A}.\Var{P}(x)) = (\forall x\in \Var{B}.\Var{Q}(x))
|
lcp@286
|
485 |
\end{eqnarray*}
|
lcp@323
|
486 |
The congruence rule for conditional expressions can supply contextual
|
lcp@323
|
487 |
information for simplifying the arms:
|
lcp@104
|
488 |
\[ \List{\Var{p}=\Var{q};~ \Var{q} \Imp \Var{a}=\Var{c};~
|
oheimb@11181
|
489 |
\neg\Var{q} \Imp \Var{b}=\Var{d}} \Imp
|
lcp@104
|
490 |
if(\Var{p},\Var{a},\Var{b}) \equiv if(\Var{q},\Var{c},\Var{d})
|
lcp@104
|
491 |
\]
|
lcp@698
|
492 |
A congruence rule can also {\em prevent\/} simplification of some arguments.
|
lcp@104
|
493 |
Here is an alternative congruence rule for conditional expressions:
|
lcp@104
|
494 |
\[ \Var{p}=\Var{q} \Imp
|
lcp@104
|
495 |
if(\Var{p},\Var{a},\Var{b}) \equiv if(\Var{q},\Var{a},\Var{b})
|
lcp@104
|
496 |
\]
|
lcp@104
|
497 |
Only the first argument is simplified; the others remain unchanged.
|
lcp@104
|
498 |
This can make simplification much faster, but may require an extra case split
|
lcp@104
|
499 |
to prove the goal.
|
lcp@104
|
500 |
|
lcp@104
|
501 |
|
wenzelm@4395
|
502 |
\subsection{*The subgoaler}\label{sec:simp-subgoaler}
|
wenzelm@4395
|
503 |
\begin{ttbox}
|
wenzelm@7990
|
504 |
setsubgoaler :
|
wenzelm@7990
|
505 |
simpset * (simpset -> int -> tactic) -> simpset \hfill{\bf infix 4}
|
wenzelm@4395
|
506 |
prems_of_ss : simpset -> thm list
|
wenzelm@4395
|
507 |
\end{ttbox}
|
wenzelm@4395
|
508 |
|
lcp@104
|
509 |
The subgoaler is the tactic used to solve subgoals arising out of
|
lcp@104
|
510 |
conditional rewrite rules or congruence rules. The default should be
|
wenzelm@4395
|
511 |
simplification itself. Occasionally this strategy needs to be
|
wenzelm@4395
|
512 |
changed. For example, if the premise of a conditional rule is an
|
wenzelm@4395
|
513 |
instance of its conclusion, as in $Suc(\Var{m}) < \Var{n} \Imp \Var{m}
|
wenzelm@4395
|
514 |
< \Var{n}$, the default strategy could loop.
|
lcp@104
|
515 |
|
wenzelm@4395
|
516 |
\begin{ttdescription}
|
wenzelm@4395
|
517 |
|
wenzelm@4395
|
518 |
\item[$ss$ \ttindexbold{setsubgoaler} $tacf$] sets the subgoaler of
|
wenzelm@4395
|
519 |
$ss$ to $tacf$. The function $tacf$ will be applied to the current
|
wenzelm@4395
|
520 |
simplifier context expressed as a simpset.
|
wenzelm@4395
|
521 |
|
wenzelm@4395
|
522 |
\item[\ttindexbold{prems_of_ss} $ss$] retrieves the current set of
|
wenzelm@4395
|
523 |
premises from simplifier context $ss$. This may be non-empty only
|
wenzelm@4395
|
524 |
if the simplifier has been told to utilize local assumptions in the
|
wenzelm@4395
|
525 |
first place, e.g.\ if invoked via \texttt{asm_simp_tac}.
|
wenzelm@4395
|
526 |
|
wenzelm@4395
|
527 |
\end{ttdescription}
|
wenzelm@4395
|
528 |
|
wenzelm@4395
|
529 |
As an example, consider the following subgoaler:
|
lcp@104
|
530 |
\begin{ttbox}
|
wenzelm@4395
|
531 |
fun subgoaler ss =
|
wenzelm@4395
|
532 |
assume_tac ORELSE'
|
wenzelm@4395
|
533 |
resolve_tac (prems_of_ss ss) ORELSE'
|
wenzelm@4395
|
534 |
asm_simp_tac ss;
|
lcp@104
|
535 |
\end{ttbox}
|
wenzelm@4395
|
536 |
This tactic first tries to solve the subgoal by assumption or by
|
wenzelm@4395
|
537 |
resolving with with one of the premises, calling simplification only
|
wenzelm@4395
|
538 |
if that fails.
|
wenzelm@4395
|
539 |
|
lcp@104
|
540 |
|
lcp@698
|
541 |
\subsection{*The solver}\label{sec:simp-solver}
|
wenzelm@4395
|
542 |
\begin{ttbox}
|
nipkow@7620
|
543 |
mk_solver : string -> (thm list -> int -> tactic) -> solver
|
nipkow@7620
|
544 |
setSolver : simpset * solver -> simpset \hfill{\bf infix 4}
|
nipkow@7620
|
545 |
addSolver : simpset * solver -> simpset \hfill{\bf infix 4}
|
nipkow@7620
|
546 |
setSSolver : simpset * solver -> simpset \hfill{\bf infix 4}
|
nipkow@7620
|
547 |
addSSolver : simpset * solver -> simpset \hfill{\bf infix 4}
|
wenzelm@4395
|
548 |
\end{ttbox}
|
wenzelm@4395
|
549 |
|
nipkow@7620
|
550 |
A solver is a tactic that attempts to solve a subgoal after
|
wenzelm@4395
|
551 |
simplification. Typically it just proves trivial subgoals such as
|
paulson@4597
|
552 |
\texttt{True} and $t=t$. It could use sophisticated means such as {\tt
|
wenzelm@4395
|
553 |
blast_tac}, though that could make simplification expensive.
|
nipkow@7620
|
554 |
To keep things more abstract, solvers are packaged up in type
|
nipkow@7620
|
555 |
\texttt{solver}. The only way to create a solver is via \texttt{mk_solver}.
|
lcp@286
|
556 |
|
wenzelm@3108
|
557 |
Rewriting does not instantiate unknowns. For example, rewriting
|
wenzelm@3108
|
558 |
cannot prove $a\in \Var{A}$ since this requires
|
wenzelm@3108
|
559 |
instantiating~$\Var{A}$. The solver, however, is an arbitrary tactic
|
wenzelm@3108
|
560 |
and may instantiate unknowns as it pleases. This is the only way the
|
wenzelm@3108
|
561 |
simplifier can handle a conditional rewrite rule whose condition
|
paulson@3485
|
562 |
contains extra variables. When a simplification tactic is to be
|
wenzelm@3108
|
563 |
combined with other provers, especially with the classical reasoner,
|
wenzelm@4395
|
564 |
it is important whether it can be considered safe or not. For this
|
nipkow@7620
|
565 |
reason a simpset contains two solvers, a safe and an unsafe one.
|
oheimb@2628
|
566 |
|
wenzelm@3108
|
567 |
The standard simplification strategy solely uses the unsafe solver,
|
wenzelm@4395
|
568 |
which is appropriate in most cases. For special applications where
|
wenzelm@3108
|
569 |
the simplification process is not allowed to instantiate unknowns
|
wenzelm@4395
|
570 |
within the goal, simplification starts with the safe solver, but may
|
wenzelm@4395
|
571 |
still apply the ordinary unsafe one in nested simplifications for
|
oheimb@9398
|
572 |
conditional rules or congruences. Note that in this way the overall
|
oheimb@9398
|
573 |
tactic is not totally safe: it may instantiate unknowns that appear also
|
oheimb@9398
|
574 |
in other subgoals.
|
lcp@323
|
575 |
|
wenzelm@4395
|
576 |
\begin{ttdescription}
|
nipkow@7620
|
577 |
\item[\ttindexbold{mk_solver} $s$ $tacf$] converts $tacf$ into a new solver;
|
nipkow@7620
|
578 |
the string $s$ is only attached as a comment and has no other significance.
|
nipkow@7620
|
579 |
|
wenzelm@4395
|
580 |
\item[$ss$ \ttindexbold{setSSolver} $tacf$] installs $tacf$ as the
|
wenzelm@4395
|
581 |
\emph{safe} solver of $ss$.
|
wenzelm@4395
|
582 |
|
wenzelm@4395
|
583 |
\item[$ss$ \ttindexbold{addSSolver} $tacf$] adds $tacf$ as an
|
wenzelm@4395
|
584 |
additional \emph{safe} solver; it will be tried after the solvers
|
wenzelm@4395
|
585 |
which had already been present in $ss$.
|
wenzelm@4395
|
586 |
|
wenzelm@4395
|
587 |
\item[$ss$ \ttindexbold{setSolver} $tacf$] installs $tacf$ as the
|
wenzelm@4395
|
588 |
unsafe solver of $ss$.
|
wenzelm@4395
|
589 |
|
wenzelm@4395
|
590 |
\item[$ss$ \ttindexbold{addSolver} $tacf$] adds $tacf$ as an
|
wenzelm@4395
|
591 |
additional unsafe solver; it will be tried after the solvers which
|
wenzelm@4395
|
592 |
had already been present in $ss$.
|
lcp@104
|
593 |
|
wenzelm@4395
|
594 |
\end{ttdescription}
|
wenzelm@4395
|
595 |
|
wenzelm@4395
|
596 |
\medskip
|
wenzelm@4395
|
597 |
|
wenzelm@4395
|
598 |
\index{assumptions!in simplification} The solver tactic is invoked
|
wenzelm@4395
|
599 |
with a list of theorems, namely assumptions that hold in the local
|
wenzelm@4395
|
600 |
context. This may be non-empty only if the simplifier has been told
|
wenzelm@4395
|
601 |
to utilize local assumptions in the first place, e.g.\ if invoked via
|
wenzelm@4395
|
602 |
\texttt{asm_simp_tac}. The solver is also presented the full goal
|
wenzelm@4395
|
603 |
including its assumptions in any case. Thus it can use these (e.g.\
|
wenzelm@4395
|
604 |
by calling \texttt{assume_tac}), even if the list of premises is not
|
wenzelm@4395
|
605 |
passed.
|
wenzelm@4395
|
606 |
|
wenzelm@4395
|
607 |
\medskip
|
wenzelm@4395
|
608 |
|
oheimb@11181
|
609 |
As explained in {\S}\ref{sec:simp-subgoaler}, the subgoaler is also used
|
wenzelm@4395
|
610 |
to solve the premises of congruence rules. These are usually of the
|
wenzelm@4395
|
611 |
form $s = \Var{x}$, where $s$ needs to be simplified and $\Var{x}$
|
wenzelm@4395
|
612 |
needs to be instantiated with the result. Typically, the subgoaler
|
wenzelm@4395
|
613 |
will invoke the simplifier at some point, which will eventually call
|
wenzelm@4395
|
614 |
the solver. For this reason, solver tactics must be prepared to solve
|
wenzelm@4395
|
615 |
goals of the form $t = \Var{x}$, usually by reflexivity. In
|
wenzelm@4395
|
616 |
particular, reflexivity should be tried before any of the fancy
|
paulson@4597
|
617 |
tactics like \texttt{blast_tac}.
|
lcp@323
|
618 |
|
wenzelm@3108
|
619 |
It may even happen that due to simplification the subgoal is no longer
|
wenzelm@3108
|
620 |
an equality. For example $False \bimp \Var{Q}$ could be rewritten to
|
oheimb@11181
|
621 |
$\neg\Var{Q}$. To cover this case, the solver could try resolving
|
oheimb@11181
|
622 |
with the theorem $\neg False$.
|
lcp@323
|
623 |
|
wenzelm@4395
|
624 |
\medskip
|
wenzelm@4395
|
625 |
|
lcp@104
|
626 |
\begin{warn}
|
ballarin@13938
|
627 |
If a premise of a congruence rule cannot be proved, then the
|
ballarin@13938
|
628 |
congruence is ignored. This should only happen if the rule is
|
ballarin@13938
|
629 |
\emph{conditional} --- that is, contains premises not of the form $t
|
ballarin@13938
|
630 |
= \Var{x}$; otherwise it indicates that some congruence rule, or
|
ballarin@13938
|
631 |
possibly the subgoaler or solver, is faulty.
|
lcp@104
|
632 |
\end{warn}
|
lcp@104
|
633 |
|
lcp@104
|
634 |
|
wenzelm@4395
|
635 |
\subsection{*The looper}\label{sec:simp-looper}
|
wenzelm@4395
|
636 |
\begin{ttbox}
|
oheimb@5549
|
637 |
setloop : simpset * (int -> tactic) -> simpset \hfill{\bf infix 4}
|
oheimb@5549
|
638 |
addloop : simpset * (string * (int -> tactic)) -> simpset \hfill{\bf infix 4}
|
oheimb@5549
|
639 |
delloop : simpset * string -> simpset \hfill{\bf infix 4}
|
wenzelm@4395
|
640 |
addsplits : simpset * thm list -> simpset \hfill{\bf infix 4}
|
oheimb@5549
|
641 |
delsplits : simpset * thm list -> simpset \hfill{\bf infix 4}
|
wenzelm@4395
|
642 |
\end{ttbox}
|
lcp@104
|
643 |
|
oheimb@5549
|
644 |
The looper is a list of tactics that are applied after simplification, in case
|
wenzelm@4395
|
645 |
the solver failed to solve the simplified goal. If the looper
|
wenzelm@4395
|
646 |
succeeds, the simplification process is started all over again. Each
|
wenzelm@4395
|
647 |
of the subgoals generated by the looper is attacked in turn, in
|
wenzelm@4395
|
648 |
reverse order.
|
lcp@104
|
649 |
|
oheimb@9398
|
650 |
A typical looper is \index{case splitting}: the expansion of a conditional.
|
wenzelm@4395
|
651 |
Another possibility is to apply an elimination rule on the
|
wenzelm@4395
|
652 |
assumptions. More adventurous loopers could start an induction.
|
oheimb@2567
|
653 |
|
wenzelm@4395
|
654 |
\begin{ttdescription}
|
wenzelm@4395
|
655 |
|
oheimb@5549
|
656 |
\item[$ss$ \ttindexbold{setloop} $tacf$] installs $tacf$ as the only looper
|
oheimb@5549
|
657 |
tactic of $ss$.
|
wenzelm@4395
|
658 |
|
oheimb@5549
|
659 |
\item[$ss$ \ttindexbold{addloop} $(name,tacf)$] adds $tacf$ as an additional
|
oheimb@5549
|
660 |
looper tactic with name $name$; it will be tried after the looper tactics
|
oheimb@5549
|
661 |
that had already been present in $ss$.
|
oheimb@5549
|
662 |
|
oheimb@5549
|
663 |
\item[$ss$ \ttindexbold{delloop} $name$] deletes the looper tactic $name$
|
oheimb@5549
|
664 |
from $ss$.
|
wenzelm@4395
|
665 |
|
wenzelm@4395
|
666 |
\item[$ss$ \ttindexbold{addsplits} $thms$] adds
|
oheimb@5549
|
667 |
split tactics for $thms$ as additional looper tactics of $ss$.
|
oheimb@5549
|
668 |
|
oheimb@5549
|
669 |
\item[$ss$ \ttindexbold{addsplits} $thms$] deletes the
|
oheimb@5549
|
670 |
split tactics for $thms$ from the looper tactics of $ss$.
|
wenzelm@4395
|
671 |
|
wenzelm@4395
|
672 |
\end{ttdescription}
|
wenzelm@4395
|
673 |
|
oheimb@5549
|
674 |
The splitter replaces applications of a given function; the right-hand side
|
oheimb@5549
|
675 |
of the replacement can be anything. For example, here is a splitting rule
|
oheimb@5549
|
676 |
for conditional expressions:
|
oheimb@5549
|
677 |
\[ \Var{P}(if(\Var{Q},\Var{x},\Var{y})) \bimp (\Var{Q} \imp \Var{P}(\Var{x}))
|
oheimb@11181
|
678 |
\conj (\neg\Var{Q} \imp \Var{P}(\Var{y}))
|
oheimb@5549
|
679 |
\]
|
paulson@8136
|
680 |
Another example is the elimination operator for Cartesian products (which
|
paulson@8136
|
681 |
happens to be called~$split$):
|
oheimb@5549
|
682 |
\[ \Var{P}(split(\Var{f},\Var{p})) \bimp (\forall a~b. \Var{p} =
|
oheimb@5549
|
683 |
\langle a,b\rangle \imp \Var{P}(\Var{f}(a,b)))
|
oheimb@5549
|
684 |
\]
|
oheimb@5549
|
685 |
|
oheimb@5549
|
686 |
For technical reasons, there is a distinction between case splitting in the
|
oheimb@5549
|
687 |
conclusion and in the premises of a subgoal. The former is done by
|
oheimb@9398
|
688 |
\texttt{split_tac} with rules like \texttt{split_if} or \texttt{option.split},
|
oheimb@9398
|
689 |
which do not split the subgoal, while the latter is done by
|
oheimb@9398
|
690 |
\texttt{split_asm_tac} with rules like \texttt{split_if_asm} or
|
oheimb@9398
|
691 |
\texttt{option.split_asm}, which split the subgoal.
|
oheimb@5549
|
692 |
The operator \texttt{addsplits} automatically takes care of which tactic to
|
oheimb@5549
|
693 |
call, analyzing the form of the rules given as argument.
|
oheimb@5549
|
694 |
\begin{warn}
|
oheimb@5549
|
695 |
Due to \texttt{split_asm_tac}, the simplifier may split subgoals!
|
oheimb@5549
|
696 |
\end{warn}
|
oheimb@5549
|
697 |
|
oheimb@5549
|
698 |
Case splits should be allowed only when necessary; they are expensive
|
oheimb@5549
|
699 |
and hard to control. Here is an example of use, where \texttt{split_if}
|
oheimb@5549
|
700 |
is the first rule above:
|
oheimb@5549
|
701 |
\begin{ttbox}
|
paulson@8136
|
702 |
by (simp_tac (simpset()
|
paulson@8136
|
703 |
addloop ("split if", split_tac [split_if])) 1);
|
oheimb@5549
|
704 |
\end{ttbox}
|
wenzelm@5776
|
705 |
Users would usually prefer the following shortcut using \texttt{addsplits}:
|
oheimb@5549
|
706 |
\begin{ttbox}
|
oheimb@5549
|
707 |
by (simp_tac (simpset() addsplits [split_if]) 1);
|
oheimb@5549
|
708 |
\end{ttbox}
|
paulson@8136
|
709 |
Case-splitting on conditional expressions is usually beneficial, so it is
|
paulson@8136
|
710 |
enabled by default in the object-logics \texttt{HOL} and \texttt{FOL}.
|
wenzelm@4395
|
711 |
|
wenzelm@4395
|
712 |
|
wenzelm@4395
|
713 |
\section{The simplification tactics}\label{simp-tactics}
|
wenzelm@4395
|
714 |
\index{simplification!tactics}\index{tactics!simplification}
|
lcp@104
|
715 |
\begin{ttbox}
|
oheimb@9398
|
716 |
generic_simp_tac : bool -> bool * bool * bool ->
|
oheimb@9398
|
717 |
simpset -> int -> tactic
|
wenzelm@4395
|
718 |
simp_tac : simpset -> int -> tactic
|
wenzelm@4395
|
719 |
asm_simp_tac : simpset -> int -> tactic
|
wenzelm@4395
|
720 |
full_simp_tac : simpset -> int -> tactic
|
wenzelm@4395
|
721 |
asm_full_simp_tac : simpset -> int -> tactic
|
wenzelm@4395
|
722 |
safe_asm_full_simp_tac : simpset -> int -> tactic
|
wenzelm@4395
|
723 |
\end{ttbox}
|
lcp@104
|
724 |
|
oheimb@9398
|
725 |
\texttt{generic_simp_tac} is the basic tactic that is underlying any actual
|
oheimb@9398
|
726 |
simplification work. The others are just instantiations of it. The rewriting
|
oheimb@9398
|
727 |
strategy is always strictly bottom up, except for congruence rules,
|
oheimb@9398
|
728 |
which are applied while descending into a term. Conditions in conditional
|
oheimb@9398
|
729 |
rewrite rules are solved recursively before the rewrite rule is applied.
|
lcp@104
|
730 |
|
wenzelm@4395
|
731 |
\begin{ttdescription}
|
wenzelm@4395
|
732 |
|
oheimb@9398
|
733 |
\item[\ttindexbold{generic_simp_tac} $safe$ ($simp\_asm$, $use\_asm$, $mutual$)]
|
oheimb@9398
|
734 |
gives direct access to the various simplification modes:
|
oheimb@9398
|
735 |
\begin{itemize}
|
oheimb@9398
|
736 |
\item if $safe$ is {\tt true}, the safe solver is used as explained in
|
oheimb@11181
|
737 |
{\S}\ref{sec:simp-solver},
|
oheimb@9398
|
738 |
\item $simp\_asm$ determines whether the local assumptions are simplified,
|
oheimb@9398
|
739 |
\item $use\_asm$ determines whether the assumptions are used as local rewrite
|
oheimb@9398
|
740 |
rules, and
|
oheimb@9398
|
741 |
\item $mutual$ determines whether assumptions can simplify each other rather
|
oheimb@9398
|
742 |
than being processed from left to right.
|
oheimb@9398
|
743 |
\end{itemize}
|
oheimb@9398
|
744 |
This generic interface is intended
|
oheimb@9398
|
745 |
for building special tools, e.g.\ for combining the simplifier with the
|
oheimb@9398
|
746 |
classical reasoner. It is rarely used directly.
|
oheimb@9398
|
747 |
|
wenzelm@4395
|
748 |
\item[\ttindexbold{simp_tac}, \ttindexbold{asm_simp_tac},
|
wenzelm@4395
|
749 |
\ttindexbold{full_simp_tac}, \ttindexbold{asm_full_simp_tac}] are
|
wenzelm@4395
|
750 |
the basic simplification tactics that work exactly like their
|
oheimb@11181
|
751 |
namesakes in {\S}\ref{sec:simp-for-dummies}, except that they are
|
wenzelm@4395
|
752 |
explicitly supplied with a simpset.
|
wenzelm@4395
|
753 |
|
wenzelm@4395
|
754 |
\end{ttdescription}
|
lcp@104
|
755 |
|
wenzelm@4395
|
756 |
\medskip
|
lcp@104
|
757 |
|
wenzelm@4395
|
758 |
Local modifications of simpsets within a proof are often much cleaner
|
wenzelm@4395
|
759 |
by using above tactics in conjunction with explicit simpsets, rather
|
wenzelm@4395
|
760 |
than their capitalized counterparts. For example
|
nipkow@1860
|
761 |
\begin{ttbox}
|
nipkow@1860
|
762 |
Addsimps \(thms\);
|
paulson@2479
|
763 |
by (Simp_tac \(i\));
|
nipkow@1860
|
764 |
Delsimps \(thms\);
|
nipkow@1860
|
765 |
\end{ttbox}
|
wenzelm@4395
|
766 |
can be expressed more appropriately as
|
nipkow@1860
|
767 |
\begin{ttbox}
|
wenzelm@4395
|
768 |
by (simp_tac (simpset() addsimps \(thms\)) \(i\));
|
nipkow@1860
|
769 |
\end{ttbox}
|
nipkow@1860
|
770 |
|
wenzelm@4395
|
771 |
\medskip
|
nipkow@1860
|
772 |
|
wenzelm@4395
|
773 |
Also note that functions depending implicitly on the current theory
|
wenzelm@4395
|
774 |
context (like capital \texttt{Simp_tac} and the other commands of
|
oheimb@11181
|
775 |
{\S}\ref{sec:simp-for-dummies}) should be considered harmful outside of
|
wenzelm@4395
|
776 |
actual proof scripts. In particular, ML programs like theory
|
wenzelm@4395
|
777 |
definition packages or special tactics should refer to simpsets only
|
wenzelm@4395
|
778 |
explicitly, via the above tactics used in conjunction with
|
wenzelm@4395
|
779 |
\texttt{simpset_of} or the \texttt{SIMPSET} tacticals.
|
lcp@104
|
780 |
|
wenzelm@4395
|
781 |
|
wenzelm@5370
|
782 |
\section{Forward rules and conversions}
|
wenzelm@5370
|
783 |
\index{simplification!forward rules}\index{simplification!conversions}
|
wenzelm@5370
|
784 |
\begin{ttbox}\index{*simplify}\index{*asm_simplify}\index{*full_simplify}\index{*asm_full_simplify}\index{*Simplifier.rewrite}\index{*Simplifier.asm_rewrite}\index{*Simplifier.full_rewrite}\index{*Simplifier.asm_full_rewrite}
|
wenzelm@4395
|
785 |
simplify : simpset -> thm -> thm
|
wenzelm@4395
|
786 |
asm_simplify : simpset -> thm -> thm
|
wenzelm@4395
|
787 |
full_simplify : simpset -> thm -> thm
|
wenzelm@5370
|
788 |
asm_full_simplify : simpset -> thm -> thm\medskip
|
wenzelm@5370
|
789 |
Simplifier.rewrite : simpset -> cterm -> thm
|
wenzelm@5370
|
790 |
Simplifier.asm_rewrite : simpset -> cterm -> thm
|
wenzelm@5370
|
791 |
Simplifier.full_rewrite : simpset -> cterm -> thm
|
wenzelm@5370
|
792 |
Simplifier.asm_full_rewrite : simpset -> cterm -> thm
|
wenzelm@4395
|
793 |
\end{ttbox}
|
wenzelm@4395
|
794 |
|
wenzelm@5370
|
795 |
The first four of these functions provide \emph{forward} rules for
|
wenzelm@5370
|
796 |
simplification. Their effect is analogous to the corresponding
|
oheimb@11181
|
797 |
tactics described in {\S}\ref{simp-tactics}, but affect the whole
|
wenzelm@5370
|
798 |
theorem instead of just a certain subgoal. Also note that the
|
oheimb@11181
|
799 |
looper~/ solver process as described in {\S}\ref{sec:simp-looper} and
|
oheimb@11181
|
800 |
{\S}\ref{sec:simp-solver} is omitted in forward simplification.
|
wenzelm@5370
|
801 |
|
wenzelm@5370
|
802 |
The latter four are \emph{conversions}, establishing proven equations
|
wenzelm@5370
|
803 |
of the form $t \equiv u$ where the l.h.s.\ $t$ has been given as
|
wenzelm@5370
|
804 |
argument.
|
wenzelm@4395
|
805 |
|
wenzelm@4395
|
806 |
\begin{warn}
|
wenzelm@5370
|
807 |
Forward simplification rules and conversions should be used rarely
|
wenzelm@5370
|
808 |
in ordinary proof scripts. The main intention is to provide an
|
wenzelm@5370
|
809 |
internal interface to the simplifier for special utilities.
|
wenzelm@4395
|
810 |
\end{warn}
|
wenzelm@4395
|
811 |
|
wenzelm@4395
|
812 |
|
lcp@332
|
813 |
\section{Permutative rewrite rules}
|
lcp@323
|
814 |
\index{rewrite rules!permutative|(}
|
lcp@323
|
815 |
|
lcp@323
|
816 |
A rewrite rule is {\bf permutative} if the left-hand side and right-hand
|
lcp@323
|
817 |
side are the same up to renaming of variables. The most common permutative
|
lcp@323
|
818 |
rule is commutativity: $x+y = y+x$. Other examples include $(x-y)-z =
|
lcp@323
|
819 |
(x-z)-y$ in arithmetic and $insert(x,insert(y,A)) = insert(y,insert(x,A))$
|
lcp@323
|
820 |
for sets. Such rules are common enough to merit special attention.
|
lcp@323
|
821 |
|
wenzelm@4395
|
822 |
Because ordinary rewriting loops given such rules, the simplifier
|
wenzelm@4395
|
823 |
employs a special strategy, called {\bf ordered
|
wenzelm@4395
|
824 |
rewriting}\index{rewriting!ordered}. There is a standard
|
wenzelm@4395
|
825 |
lexicographic ordering on terms. This should be perfectly OK in most
|
wenzelm@4395
|
826 |
cases, but can be changed for special applications.
|
lcp@323
|
827 |
|
nipkow@4947
|
828 |
\begin{ttbox}
|
nipkow@4947
|
829 |
settermless : simpset * (term * term -> bool) -> simpset \hfill{\bf infix 4}
|
nipkow@4947
|
830 |
\end{ttbox}
|
wenzelm@4395
|
831 |
\begin{ttdescription}
|
wenzelm@4395
|
832 |
|
wenzelm@4395
|
833 |
\item[$ss$ \ttindexbold{settermless} $rel$] installs relation $rel$ as
|
wenzelm@4395
|
834 |
term order in simpset $ss$.
|
wenzelm@4395
|
835 |
|
wenzelm@4395
|
836 |
\end{ttdescription}
|
wenzelm@4395
|
837 |
|
wenzelm@4395
|
838 |
\medskip
|
wenzelm@4395
|
839 |
|
wenzelm@4395
|
840 |
A permutative rewrite rule is applied only if it decreases the given
|
wenzelm@4395
|
841 |
term with respect to this ordering. For example, commutativity
|
wenzelm@4395
|
842 |
rewrites~$b+a$ to $a+b$, but then stops because $a+b$ is strictly less
|
wenzelm@4395
|
843 |
than $b+a$. The Boyer-Moore theorem prover~\cite{bm88book} also
|
wenzelm@4395
|
844 |
employs ordered rewriting.
|
wenzelm@4395
|
845 |
|
wenzelm@4395
|
846 |
Permutative rewrite rules are added to simpsets just like other
|
wenzelm@4395
|
847 |
rewrite rules; the simplifier recognizes their special status
|
wenzelm@4395
|
848 |
automatically. They are most effective in the case of
|
wenzelm@4395
|
849 |
associative-commutative operators. (Associativity by itself is not
|
wenzelm@4395
|
850 |
permutative.) When dealing with an AC-operator~$f$, keep the
|
wenzelm@4395
|
851 |
following points in mind:
|
lcp@323
|
852 |
\begin{itemize}\index{associative-commutative operators}
|
wenzelm@4395
|
853 |
|
wenzelm@4395
|
854 |
\item The associative law must always be oriented from left to right,
|
wenzelm@4395
|
855 |
namely $f(f(x,y),z) = f(x,f(y,z))$. The opposite orientation, if
|
wenzelm@4395
|
856 |
used with commutativity, leads to looping in conjunction with the
|
wenzelm@4395
|
857 |
standard term order.
|
lcp@323
|
858 |
|
lcp@323
|
859 |
\item To complete your set of rewrite rules, you must add not just
|
lcp@323
|
860 |
associativity~(A) and commutativity~(C) but also a derived rule, {\bf
|
paulson@4597
|
861 |
left-com\-mut\-ativ\-ity} (LC): $f(x,f(y,z)) = f(y,f(x,z))$.
|
lcp@323
|
862 |
\end{itemize}
|
lcp@323
|
863 |
Ordered rewriting with the combination of A, C, and~LC sorts a term
|
lcp@323
|
864 |
lexicographically:
|
lcp@323
|
865 |
\[\def\maps#1{\stackrel{#1}{\longmapsto}}
|
lcp@323
|
866 |
(b+c)+a \maps{A} b+(c+a) \maps{C} b+(a+c) \maps{LC} a+(b+c) \]
|
lcp@323
|
867 |
Martin and Nipkow~\cite{martin-nipkow} discuss the theory and give many
|
lcp@323
|
868 |
examples; other algebraic structures are amenable to ordered rewriting,
|
lcp@323
|
869 |
such as boolean rings.
|
lcp@323
|
870 |
|
wenzelm@3108
|
871 |
\subsection{Example: sums of natural numbers}
|
wenzelm@4395
|
872 |
|
wenzelm@9695
|
873 |
This example is again set in HOL (see \texttt{HOL/ex/NatSum}). Theory
|
wenzelm@9695
|
874 |
\thydx{Arith} contains natural numbers arithmetic. Its associated simpset
|
wenzelm@9695
|
875 |
contains many arithmetic laws including distributivity of~$\times$ over~$+$,
|
wenzelm@9695
|
876 |
while \texttt{add_ac} is a list consisting of the A, C and LC laws for~$+$ on
|
wenzelm@9695
|
877 |
type \texttt{nat}. Let us prove the theorem
|
lcp@323
|
878 |
\[ \sum@{i=1}^n i = n\times(n+1)/2. \]
|
lcp@323
|
879 |
%
|
paulson@4597
|
880 |
A functional~\texttt{sum} represents the summation operator under the
|
paulson@4597
|
881 |
interpretation $\texttt{sum} \, f \, (n + 1) = \sum@{i=0}^n f\,i$. We
|
paulson@4597
|
882 |
extend \texttt{Arith} as follows:
|
lcp@323
|
883 |
\begin{ttbox}
|
lcp@323
|
884 |
NatSum = Arith +
|
clasohm@1387
|
885 |
consts sum :: [nat=>nat, nat] => nat
|
berghofe@9445
|
886 |
primrec
|
paulson@4245
|
887 |
"sum f 0 = 0"
|
paulson@4245
|
888 |
"sum f (Suc n) = f(n) + sum f n"
|
lcp@323
|
889 |
end
|
lcp@323
|
890 |
\end{ttbox}
|
paulson@4245
|
891 |
The \texttt{primrec} declaration automatically adds rewrite rules for
|
wenzelm@4557
|
892 |
\texttt{sum} to the default simpset. We now remove the
|
wenzelm@4557
|
893 |
\texttt{nat_cancel} simplification procedures (in order not to spoil
|
wenzelm@4557
|
894 |
the example) and insert the AC-rules for~$+$:
|
lcp@323
|
895 |
\begin{ttbox}
|
wenzelm@4557
|
896 |
Delsimprocs nat_cancel;
|
paulson@4245
|
897 |
Addsimps add_ac;
|
lcp@323
|
898 |
\end{ttbox}
|
paulson@4597
|
899 |
Our desired theorem now reads $\texttt{sum} \, (\lambda i.i) \, (n+1) =
|
lcp@323
|
900 |
n\times(n+1)/2$. The Isabelle goal has both sides multiplied by~$2$:
|
lcp@323
|
901 |
\begin{ttbox}
|
paulson@5205
|
902 |
Goal "2 * sum (\%i.i) (Suc n) = n * Suc n";
|
lcp@323
|
903 |
{\out Level 0}
|
wenzelm@3108
|
904 |
{\out 2 * sum (\%i. i) (Suc n) = n * Suc n}
|
wenzelm@3108
|
905 |
{\out 1. 2 * sum (\%i. i) (Suc n) = n * Suc n}
|
lcp@323
|
906 |
\end{ttbox}
|
wenzelm@3108
|
907 |
Induction should not be applied until the goal is in the simplest
|
wenzelm@3108
|
908 |
form:
|
lcp@323
|
909 |
\begin{ttbox}
|
paulson@4245
|
910 |
by (Simp_tac 1);
|
lcp@323
|
911 |
{\out Level 1}
|
wenzelm@3108
|
912 |
{\out 2 * sum (\%i. i) (Suc n) = n * Suc n}
|
wenzelm@3108
|
913 |
{\out 1. n + (sum (\%i. i) n + sum (\%i. i) n) = n * n}
|
lcp@323
|
914 |
\end{ttbox}
|
wenzelm@3108
|
915 |
Ordered rewriting has sorted the terms in the left-hand side. The
|
wenzelm@3108
|
916 |
subgoal is now ready for induction:
|
lcp@323
|
917 |
\begin{ttbox}
|
paulson@4245
|
918 |
by (induct_tac "n" 1);
|
lcp@323
|
919 |
{\out Level 2}
|
wenzelm@3108
|
920 |
{\out 2 * sum (\%i. i) (Suc n) = n * Suc n}
|
wenzelm@3108
|
921 |
{\out 1. 0 + (sum (\%i. i) 0 + sum (\%i. i) 0) = 0 * 0}
|
lcp@323
|
922 |
\ttbreak
|
paulson@4245
|
923 |
{\out 2. !!n. n + (sum (\%i. i) n + sum (\%i. i) n) = n * n}
|
paulson@8136
|
924 |
{\out ==> Suc n + (sum (\%i. i) (Suc n) + sum (\%i.\,i) (Suc n)) =}
|
paulson@4245
|
925 |
{\out Suc n * Suc n}
|
lcp@323
|
926 |
\end{ttbox}
|
lcp@323
|
927 |
Simplification proves both subgoals immediately:\index{*ALLGOALS}
|
lcp@323
|
928 |
\begin{ttbox}
|
paulson@4245
|
929 |
by (ALLGOALS Asm_simp_tac);
|
lcp@323
|
930 |
{\out Level 3}
|
wenzelm@3108
|
931 |
{\out 2 * sum (\%i. i) (Suc n) = n * Suc n}
|
lcp@323
|
932 |
{\out No subgoals!}
|
lcp@323
|
933 |
\end{ttbox}
|
paulson@4597
|
934 |
Simplification cannot prove the induction step if we omit \texttt{add_ac} from
|
paulson@4245
|
935 |
the simpset. Observe that like terms have not been collected:
|
lcp@323
|
936 |
\begin{ttbox}
|
paulson@4245
|
937 |
{\out Level 3}
|
paulson@4245
|
938 |
{\out 2 * sum (\%i. i) (Suc n) = n * Suc n}
|
paulson@4245
|
939 |
{\out 1. !!n. n + sum (\%i. i) n + (n + sum (\%i. i) n) = n + n * n}
|
paulson@8136
|
940 |
{\out ==> n + (n + sum (\%i. i) n) + (n + (n + sum (\%i.\,i) n)) =}
|
paulson@4245
|
941 |
{\out n + (n + (n + n * n))}
|
lcp@323
|
942 |
\end{ttbox}
|
lcp@323
|
943 |
Ordered rewriting proves this by sorting the left-hand side. Proving
|
lcp@323
|
944 |
arithmetic theorems without ordered rewriting requires explicit use of
|
lcp@323
|
945 |
commutativity. This is tedious; try it and see!
|
lcp@323
|
946 |
|
lcp@323
|
947 |
Ordered rewriting is equally successful in proving
|
lcp@323
|
948 |
$\sum@{i=1}^n i^3 = n^2\times(n+1)^2/4$.
|
lcp@323
|
949 |
|
lcp@323
|
950 |
|
lcp@323
|
951 |
\subsection{Re-orienting equalities}
|
paulson@4597
|
952 |
Ordered rewriting with the derived rule \texttt{symmetry} can reverse
|
wenzelm@4557
|
953 |
equations:
|
lcp@323
|
954 |
\begin{ttbox}
|
lcp@323
|
955 |
val symmetry = prove_goal HOL.thy "(x=y) = (y=x)"
|
paulson@3128
|
956 |
(fn _ => [Blast_tac 1]);
|
lcp@323
|
957 |
\end{ttbox}
|
lcp@323
|
958 |
This is frequently useful. Assumptions of the form $s=t$, where $t$ occurs
|
lcp@323
|
959 |
in the conclusion but not~$s$, can often be brought into the right form.
|
paulson@4597
|
960 |
For example, ordered rewriting with \texttt{symmetry} can prove the goal
|
lcp@323
|
961 |
\[ f(a)=b \conj f(a)=c \imp b=c. \]
|
paulson@4597
|
962 |
Here \texttt{symmetry} reverses both $f(a)=b$ and $f(a)=c$
|
lcp@323
|
963 |
because $f(a)$ is lexicographically greater than $b$ and~$c$. These
|
lcp@323
|
964 |
re-oriented equations, as rewrite rules, replace $b$ and~$c$ in the
|
lcp@323
|
965 |
conclusion by~$f(a)$.
|
lcp@323
|
966 |
|
oheimb@11181
|
967 |
Another example is the goal $\neg(t=u) \imp \neg(u=t)$.
|
lcp@323
|
968 |
The differing orientations make this appear difficult to prove. Ordered
|
paulson@4597
|
969 |
rewriting with \texttt{symmetry} makes the equalities agree. (Without
|
lcp@323
|
970 |
knowing more about~$t$ and~$u$ we cannot say whether they both go to $t=u$
|
lcp@323
|
971 |
or~$u=t$.) Then the simplifier can prove the goal outright.
|
lcp@323
|
972 |
|
lcp@323
|
973 |
\index{rewrite rules!permutative|)}
|
lcp@323
|
974 |
|
lcp@323
|
975 |
|
wenzelm@4395
|
976 |
\section{*Coding simplification procedures}
|
wenzelm@4395
|
977 |
\begin{ttbox}
|
wenzelm@13474
|
978 |
val Simplifier.simproc: Sign.sg -> string -> string list
|
wenzelm@15027
|
979 |
-> (Sign.sg -> simpset -> term -> thm option) -> simproc
|
wenzelm@13474
|
980 |
val Simplifier.simproc_i: Sign.sg -> string -> term list
|
wenzelm@15027
|
981 |
-> (Sign.sg -> simpset -> term -> thm option) -> simproc
|
wenzelm@4395
|
982 |
\end{ttbox}
|
wenzelm@4395
|
983 |
|
wenzelm@4395
|
984 |
\begin{ttdescription}
|
wenzelm@13477
|
985 |
\item[\ttindexbold{Simplifier.simproc}~$sign$~$name$~$lhss$~$proc$] makes
|
wenzelm@13477
|
986 |
$proc$ a simplification procedure for left-hand side patterns $lhss$. The
|
wenzelm@13477
|
987 |
name just serves as a comment. The function $proc$ may be invoked by the
|
wenzelm@13477
|
988 |
simplifier for redex positions matched by one of $lhss$ as described below
|
wenzelm@13477
|
989 |
(which are be specified as strings to be read as terms).
|
wenzelm@13477
|
990 |
|
wenzelm@13477
|
991 |
\item[\ttindexbold{Simplifier.simproc_i}] is similar to
|
wenzelm@13477
|
992 |
\verb,Simplifier.simproc,, but takes well-typed terms as pattern argument.
|
wenzelm@4395
|
993 |
\end{ttdescription}
|
wenzelm@4395
|
994 |
|
wenzelm@4395
|
995 |
Simplification procedures are applied in a two-stage process as
|
wenzelm@4395
|
996 |
follows: The simplifier tries to match the current redex position
|
wenzelm@4395
|
997 |
against any one of the $lhs$ patterns of any simplification procedure.
|
wenzelm@4395
|
998 |
If this succeeds, it invokes the corresponding {\ML} function, passing
|
wenzelm@4395
|
999 |
with the current signature, local assumptions and the (potential)
|
wenzelm@4395
|
1000 |
redex. The result may be either \texttt{None} (indicating failure) or
|
wenzelm@4395
|
1001 |
\texttt{Some~$thm$}.
|
wenzelm@4395
|
1002 |
|
wenzelm@4395
|
1003 |
Any successful result is supposed to be a (possibly conditional)
|
wenzelm@4395
|
1004 |
rewrite rule $t \equiv u$ that is applicable to the current redex.
|
wenzelm@4395
|
1005 |
The rule will be applied just as any ordinary rewrite rule. It is
|
wenzelm@4395
|
1006 |
expected to be already in \emph{internal form}, though, bypassing the
|
wenzelm@4395
|
1007 |
automatic preprocessing of object-level equivalences.
|
wenzelm@4395
|
1008 |
|
wenzelm@4395
|
1009 |
\medskip
|
wenzelm@4395
|
1010 |
|
wenzelm@4395
|
1011 |
As an example of how to write your own simplification procedures,
|
wenzelm@4395
|
1012 |
consider eta-expansion of pair abstraction (see also
|
wenzelm@4395
|
1013 |
\texttt{HOL/Modelcheck/MCSyn} where this is used to provide external
|
wenzelm@4395
|
1014 |
model checker syntax).
|
wenzelm@4395
|
1015 |
|
wenzelm@9695
|
1016 |
The HOL theory of tuples (see \texttt{HOL/Prod}) provides an operator
|
wenzelm@9695
|
1017 |
\texttt{split} together with some concrete syntax supporting
|
wenzelm@9695
|
1018 |
$\lambda\,(x,y).b$ abstractions. Assume that we would like to offer a tactic
|
wenzelm@9695
|
1019 |
that rewrites any function $\lambda\,p.f\,p$ (where $p$ is of some pair type)
|
wenzelm@9695
|
1020 |
to $\lambda\,(x,y).f\,(x,y)$. The corresponding rule is:
|
wenzelm@4395
|
1021 |
\begin{ttbox}
|
wenzelm@4395
|
1022 |
pair_eta_expand: (f::'a*'b=>'c) = (\%(x, y). f (x, y))
|
wenzelm@4395
|
1023 |
\end{ttbox}
|
wenzelm@4395
|
1024 |
Unfortunately, term rewriting using this rule directly would not
|
wenzelm@4395
|
1025 |
terminate! We now use the simplification procedure mechanism in order
|
wenzelm@4395
|
1026 |
to stop the simplifier from applying this rule over and over again,
|
wenzelm@4395
|
1027 |
making it rewrite only actual abstractions. The simplification
|
wenzelm@4395
|
1028 |
procedure \texttt{pair_eta_expand_proc} is defined as follows:
|
wenzelm@4395
|
1029 |
\begin{ttbox}
|
wenzelm@13474
|
1030 |
val pair_eta_expand_proc =
|
wenzelm@13477
|
1031 |
Simplifier.simproc (Theory.sign_of (the_context ()))
|
wenzelm@13477
|
1032 |
"pair_eta_expand" ["f::'a*'b=>'c"]
|
wenzelm@13477
|
1033 |
(fn _ => fn _ => fn t =>
|
wenzelm@13477
|
1034 |
case t of Abs _ => Some (mk_meta_eq pair_eta_expand)
|
wenzelm@13477
|
1035 |
| _ => None);
|
wenzelm@4395
|
1036 |
\end{ttbox}
|
wenzelm@4395
|
1037 |
This is an example of using \texttt{pair_eta_expand_proc}:
|
wenzelm@4395
|
1038 |
\begin{ttbox}
|
wenzelm@4395
|
1039 |
{\out 1. P (\%p::'a * 'a. fst p + snd p + z)}
|
wenzelm@4395
|
1040 |
by (simp_tac (simpset() addsimprocs [pair_eta_expand_proc]) 1);
|
wenzelm@4395
|
1041 |
{\out 1. P (\%(x::'a,y::'a). x + y + z)}
|
wenzelm@4395
|
1042 |
\end{ttbox}
|
wenzelm@4395
|
1043 |
|
wenzelm@4395
|
1044 |
\medskip
|
wenzelm@4395
|
1045 |
|
wenzelm@4395
|
1046 |
In the above example the simplification procedure just did fine
|
wenzelm@4395
|
1047 |
grained control over rule application, beyond higher-order pattern
|
wenzelm@4395
|
1048 |
matching. Usually, procedures would do some more work, in particular
|
wenzelm@4395
|
1049 |
prove particular theorems depending on the current redex.
|
wenzelm@4395
|
1050 |
|
wenzelm@4395
|
1051 |
|
wenzelm@7990
|
1052 |
\section{*Setting up the Simplifier}\label{sec:setting-up-simp}
|
lcp@323
|
1053 |
\index{simplification!setting up}
|
lcp@286
|
1054 |
|
wenzelm@9712
|
1055 |
Setting up the simplifier for new logics is complicated in the general case.
|
wenzelm@9712
|
1056 |
This section describes how the simplifier is installed for intuitionistic
|
wenzelm@9712
|
1057 |
first-order logic; the code is largely taken from {\tt FOL/simpdata.ML} of the
|
wenzelm@9712
|
1058 |
Isabelle sources.
|
lcp@286
|
1059 |
|
wenzelm@16019
|
1060 |
The case splitting tactic, which resides on a separate files, is not part of
|
wenzelm@16019
|
1061 |
Pure Isabelle. It needs to be loaded explicitly by the object-logic as
|
wenzelm@16019
|
1062 |
follows (below \texttt{\~\relax\~\relax} refers to \texttt{\$ISABELLE_HOME}):
|
lcp@286
|
1063 |
\begin{ttbox}
|
wenzelm@6569
|
1064 |
use "\~\relax\~\relax/src/Provers/splitter.ML";
|
lcp@286
|
1065 |
\end{ttbox}
|
lcp@286
|
1066 |
|
paulson@4597
|
1067 |
Simplification requires converting object-equalities to meta-level rewrite
|
paulson@4597
|
1068 |
rules. This demands rules stating that equal terms and equivalent formulae
|
paulson@4597
|
1069 |
are also equal at the meta-level. The rule declaration part of the file
|
paulson@4597
|
1070 |
\texttt{FOL/IFOL.thy} contains the two lines
|
lcp@323
|
1071 |
\begin{ttbox}\index{*eq_reflection theorem}\index{*iff_reflection theorem}
|
lcp@286
|
1072 |
eq_reflection "(x=y) ==> (x==y)"
|
lcp@286
|
1073 |
iff_reflection "(P<->Q) ==> (P==Q)"
|
lcp@286
|
1074 |
\end{ttbox}
|
lcp@323
|
1075 |
Of course, you should only assert such rules if they are true for your
|
lcp@286
|
1076 |
particular logic. In Constructive Type Theory, equality is a ternary
|
wenzelm@4395
|
1077 |
relation of the form $a=b\in A$; the type~$A$ determines the meaning
|
wenzelm@4395
|
1078 |
of the equality essentially as a partial equivalence relation. The
|
paulson@4597
|
1079 |
present simplifier cannot be used. Rewriting in \texttt{CTT} uses
|
wenzelm@4395
|
1080 |
another simplifier, which resides in the file {\tt
|
wenzelm@4395
|
1081 |
Provers/typedsimp.ML} and is not documented. Even this does not
|
wenzelm@4395
|
1082 |
work for later variants of Constructive Type Theory that use
|
lcp@323
|
1083 |
intensional equality~\cite{nordstrom90}.
|
lcp@286
|
1084 |
|
lcp@286
|
1085 |
|
lcp@286
|
1086 |
\subsection{A collection of standard rewrite rules}
|
wenzelm@4557
|
1087 |
|
wenzelm@4557
|
1088 |
We first prove lots of standard rewrite rules about the logical
|
wenzelm@4557
|
1089 |
connectives. These include cancellation and associative laws. We
|
wenzelm@4557
|
1090 |
define a function that echoes the desired law and then supplies it the
|
wenzelm@9695
|
1091 |
prover for intuitionistic FOL:
|
lcp@286
|
1092 |
\begin{ttbox}
|
lcp@286
|
1093 |
fun int_prove_fun s =
|
lcp@286
|
1094 |
(writeln s;
|
lcp@286
|
1095 |
prove_goal IFOL.thy s
|
lcp@286
|
1096 |
(fn prems => [ (cut_facts_tac prems 1),
|
wenzelm@4395
|
1097 |
(IntPr.fast_tac 1) ]));
|
lcp@286
|
1098 |
\end{ttbox}
|
lcp@286
|
1099 |
The following rewrite rules about conjunction are a selection of those
|
paulson@4597
|
1100 |
proved on \texttt{FOL/simpdata.ML}. Later, these will be supplied to the
|
lcp@286
|
1101 |
standard simpset.
|
lcp@286
|
1102 |
\begin{ttbox}
|
wenzelm@4395
|
1103 |
val conj_simps = map int_prove_fun
|
lcp@286
|
1104 |
["P & True <-> P", "True & P <-> P",
|
lcp@286
|
1105 |
"P & False <-> False", "False & P <-> False",
|
lcp@286
|
1106 |
"P & P <-> P",
|
lcp@286
|
1107 |
"P & ~P <-> False", "~P & P <-> False",
|
lcp@286
|
1108 |
"(P & Q) & R <-> P & (Q & R)"];
|
lcp@286
|
1109 |
\end{ttbox}
|
lcp@286
|
1110 |
The file also proves some distributive laws. As they can cause exponential
|
lcp@286
|
1111 |
blowup, they will not be included in the standard simpset. Instead they
|
lcp@323
|
1112 |
are merely bound to an \ML{} identifier, for user reference.
|
lcp@286
|
1113 |
\begin{ttbox}
|
wenzelm@4395
|
1114 |
val distrib_simps = map int_prove_fun
|
lcp@286
|
1115 |
["P & (Q | R) <-> P&Q | P&R",
|
lcp@286
|
1116 |
"(Q | R) & P <-> Q&P | R&P",
|
lcp@286
|
1117 |
"(P | Q --> R) <-> (P --> R) & (Q --> R)"];
|
lcp@286
|
1118 |
\end{ttbox}
|
lcp@286
|
1119 |
|
lcp@286
|
1120 |
|
lcp@286
|
1121 |
\subsection{Functions for preprocessing the rewrite rules}
|
lcp@323
|
1122 |
\label{sec:setmksimps}
|
wenzelm@4395
|
1123 |
\begin{ttbox}\indexbold{*setmksimps}
|
wenzelm@4395
|
1124 |
setmksimps : simpset * (thm -> thm list) -> simpset \hfill{\bf infix 4}
|
wenzelm@4395
|
1125 |
\end{ttbox}
|
lcp@286
|
1126 |
The next step is to define the function for preprocessing rewrite rules.
|
paulson@4597
|
1127 |
This will be installed by calling \texttt{setmksimps} below. Preprocessing
|
lcp@286
|
1128 |
occurs whenever rewrite rules are added, whether by user command or
|
lcp@286
|
1129 |
automatically. Preprocessing involves extracting atomic rewrites at the
|
lcp@286
|
1130 |
object-level, then reflecting them to the meta-level.
|
lcp@286
|
1131 |
|
wenzelm@12725
|
1132 |
To start, the function \texttt{gen_all} strips any meta-level
|
wenzelm@12717
|
1133 |
quantifiers from the front of the given theorem.
|
oheimb@5549
|
1134 |
|
paulson@4597
|
1135 |
The function \texttt{atomize} analyses a theorem in order to extract
|
lcp@286
|
1136 |
atomic rewrite rules. The head of all the patterns, matched by the
|
paulson@4597
|
1137 |
wildcard~\texttt{_}, is the coercion function \texttt{Trueprop}.
|
lcp@286
|
1138 |
\begin{ttbox}
|
lcp@286
|
1139 |
fun atomize th = case concl_of th of
|
lcp@286
|
1140 |
_ $ (Const("op &",_) $ _ $ _) => atomize(th RS conjunct1) \at
|
lcp@286
|
1141 |
atomize(th RS conjunct2)
|
lcp@286
|
1142 |
| _ $ (Const("op -->",_) $ _ $ _) => atomize(th RS mp)
|
lcp@286
|
1143 |
| _ $ (Const("All",_) $ _) => atomize(th RS spec)
|
lcp@286
|
1144 |
| _ $ (Const("True",_)) => []
|
lcp@286
|
1145 |
| _ $ (Const("False",_)) => []
|
lcp@286
|
1146 |
| _ => [th];
|
lcp@286
|
1147 |
\end{ttbox}
|
lcp@286
|
1148 |
There are several cases, depending upon the form of the conclusion:
|
lcp@286
|
1149 |
\begin{itemize}
|
lcp@286
|
1150 |
\item Conjunction: extract rewrites from both conjuncts.
|
lcp@286
|
1151 |
\item Implication: convert $P\imp Q$ to the meta-implication $P\Imp Q$ and
|
lcp@286
|
1152 |
extract rewrites from~$Q$; these will be conditional rewrites with the
|
lcp@286
|
1153 |
condition~$P$.
|
lcp@286
|
1154 |
\item Universal quantification: remove the quantifier, replacing the bound
|
lcp@286
|
1155 |
variable by a schematic variable, and extract rewrites from the body.
|
paulson@4597
|
1156 |
\item \texttt{True} and \texttt{False} contain no useful rewrites.
|
lcp@286
|
1157 |
\item Anything else: return the theorem in a singleton list.
|
lcp@286
|
1158 |
\end{itemize}
|
lcp@286
|
1159 |
The resulting theorems are not literally atomic --- they could be
|
oheimb@5549
|
1160 |
disjunctive, for example --- but are broken down as much as possible.
|
oheimb@5549
|
1161 |
See the file \texttt{ZF/simpdata.ML} for a sophisticated translation of
|
oheimb@5549
|
1162 |
set-theoretic formulae into rewrite rules.
|
oheimb@5549
|
1163 |
|
oheimb@5549
|
1164 |
For standard situations like the above,
|
oheimb@5549
|
1165 |
there is a generic auxiliary function \ttindexbold{mk_atomize} that takes a
|
oheimb@5549
|
1166 |
list of pairs $(name, thms)$, where $name$ is an operator name and
|
oheimb@5549
|
1167 |
$thms$ is a list of theorems to resolve with in case the pattern matches,
|
oheimb@5549
|
1168 |
and returns a suitable \texttt{atomize} function.
|
oheimb@5549
|
1169 |
|
lcp@286
|
1170 |
|
lcp@286
|
1171 |
The simplified rewrites must now be converted into meta-equalities. The
|
paulson@4597
|
1172 |
rule \texttt{eq_reflection} converts equality rewrites, while {\tt
|
lcp@286
|
1173 |
iff_reflection} converts if-and-only-if rewrites. The latter possibility
|
oheimb@11181
|
1174 |
can arise in two other ways: the negative theorem~$\neg P$ is converted to
|
paulson@4597
|
1175 |
$P\equiv\texttt{False}$, and any other theorem~$P$ is converted to
|
paulson@4597
|
1176 |
$P\equiv\texttt{True}$. The rules \texttt{iff_reflection_F} and {\tt
|
lcp@286
|
1177 |
iff_reflection_T} accomplish this conversion.
|
lcp@286
|
1178 |
\begin{ttbox}
|
lcp@286
|
1179 |
val P_iff_F = int_prove_fun "~P ==> (P <-> False)";
|
lcp@286
|
1180 |
val iff_reflection_F = P_iff_F RS iff_reflection;
|
lcp@286
|
1181 |
\ttbreak
|
lcp@286
|
1182 |
val P_iff_T = int_prove_fun "P ==> (P <-> True)";
|
lcp@286
|
1183 |
val iff_reflection_T = P_iff_T RS iff_reflection;
|
lcp@286
|
1184 |
\end{ttbox}
|
oheimb@5549
|
1185 |
The function \texttt{mk_eq} converts a theorem to a meta-equality
|
lcp@286
|
1186 |
using the case analysis described above.
|
lcp@286
|
1187 |
\begin{ttbox}
|
oheimb@5549
|
1188 |
fun mk_eq th = case concl_of th of
|
lcp@286
|
1189 |
_ $ (Const("op =",_)$_$_) => th RS eq_reflection
|
lcp@286
|
1190 |
| _ $ (Const("op <->",_)$_$_) => th RS iff_reflection
|
lcp@286
|
1191 |
| _ $ (Const("Not",_)$_) => th RS iff_reflection_F
|
lcp@286
|
1192 |
| _ => th RS iff_reflection_T;
|
lcp@286
|
1193 |
\end{ttbox}
|
oheimb@11162
|
1194 |
The
|
wenzelm@12725
|
1195 |
three functions \texttt{gen_all}, \texttt{atomize} and \texttt{mk_eq}
|
oheimb@5549
|
1196 |
will be composed together and supplied below to \texttt{setmksimps}.
|
lcp@286
|
1197 |
|
lcp@286
|
1198 |
|
lcp@286
|
1199 |
\subsection{Making the initial simpset}
|
wenzelm@4395
|
1200 |
|
wenzelm@9712
|
1201 |
It is time to assemble these items. The list \texttt{IFOL_simps} contains the
|
wenzelm@9712
|
1202 |
default rewrite rules for intuitionistic first-order logic. The first of
|
wenzelm@9712
|
1203 |
these is the reflexive law expressed as the equivalence
|
wenzelm@9712
|
1204 |
$(a=a)\bimp\texttt{True}$; the rewrite rule $a=a$ is clearly useless.
|
wenzelm@4395
|
1205 |
\begin{ttbox}
|
wenzelm@4395
|
1206 |
val IFOL_simps =
|
wenzelm@4395
|
1207 |
[refl RS P_iff_T] \at conj_simps \at disj_simps \at not_simps \at
|
wenzelm@4395
|
1208 |
imp_simps \at iff_simps \at quant_simps;
|
lcp@286
|
1209 |
\end{ttbox}
|
paulson@4597
|
1210 |
The list \texttt{triv_rls} contains trivial theorems for the solver. Any
|
lcp@286
|
1211 |
subgoal that is simplified to one of these will be removed.
|
lcp@286
|
1212 |
\begin{ttbox}
|
lcp@286
|
1213 |
val notFalseI = int_prove_fun "~False";
|
lcp@286
|
1214 |
val triv_rls = [TrueI,refl,iff_refl,notFalseI];
|
lcp@286
|
1215 |
\end{ttbox}
|
wenzelm@9712
|
1216 |
We also define the function \ttindex{mk_meta_cong} to convert the conclusion
|
wenzelm@9712
|
1217 |
of congruence rules into meta-equalities.
|
wenzelm@9712
|
1218 |
\begin{ttbox}
|
wenzelm@9712
|
1219 |
fun mk_meta_cong rl = standard (mk_meta_eq (mk_meta_prems rl));
|
wenzelm@9712
|
1220 |
\end{ttbox}
|
lcp@323
|
1221 |
%
|
wenzelm@9695
|
1222 |
The basic simpset for intuitionistic FOL is \ttindexbold{FOL_basic_ss}. It
|
oheimb@11162
|
1223 |
preprocess rewrites using
|
wenzelm@12725
|
1224 |
{\tt gen_all}, \texttt{atomize} and \texttt{mk_eq}.
|
wenzelm@9695
|
1225 |
It solves simplified subgoals using \texttt{triv_rls} and assumptions, and by
|
wenzelm@9695
|
1226 |
detecting contradictions. It uses \ttindex{asm_simp_tac} to tackle subgoals
|
wenzelm@9695
|
1227 |
of conditional rewrites.
|
wenzelm@4395
|
1228 |
|
paulson@4597
|
1229 |
Other simpsets built from \texttt{FOL_basic_ss} will inherit these items.
|
wenzelm@4395
|
1230 |
In particular, \ttindexbold{IFOL_ss}, which introduces {\tt
|
wenzelm@4395
|
1231 |
IFOL_simps} as rewrite rules. \ttindexbold{FOL_ss} will later
|
oheimb@11181
|
1232 |
extend \texttt{IFOL_ss} with classical rewrite rules such as $\neg\neg
|
wenzelm@4395
|
1233 |
P\bimp P$.
|
oheimb@2628
|
1234 |
\index{*setmksimps}\index{*setSSolver}\index{*setSolver}\index{*setsubgoaler}
|
lcp@286
|
1235 |
\index{*addsimps}\index{*addcongs}
|
lcp@286
|
1236 |
\begin{ttbox}
|
wenzelm@4395
|
1237 |
fun unsafe_solver prems = FIRST'[resolve_tac (triv_rls {\at} prems),
|
oheimb@2628
|
1238 |
atac, etac FalseE];
|
wenzelm@4395
|
1239 |
|
paulson@8136
|
1240 |
fun safe_solver prems = FIRST'[match_tac (triv_rls {\at} prems),
|
paulson@8136
|
1241 |
eq_assume_tac, ematch_tac [FalseE]];
|
wenzelm@4395
|
1242 |
|
wenzelm@9712
|
1243 |
val FOL_basic_ss =
|
paulson@8136
|
1244 |
empty_ss setsubgoaler asm_simp_tac
|
paulson@8136
|
1245 |
addsimprocs [defALL_regroup, defEX_regroup]
|
paulson@8136
|
1246 |
setSSolver safe_solver
|
paulson@8136
|
1247 |
setSolver unsafe_solver
|
wenzelm@12725
|
1248 |
setmksimps (map mk_eq o atomize o gen_all)
|
wenzelm@9712
|
1249 |
setmkcong mk_meta_cong;
|
wenzelm@4395
|
1250 |
|
paulson@8136
|
1251 |
val IFOL_ss =
|
paulson@8136
|
1252 |
FOL_basic_ss addsimps (IFOL_simps {\at}
|
paulson@8136
|
1253 |
int_ex_simps {\at} int_all_simps)
|
paulson@8136
|
1254 |
addcongs [imp_cong];
|
lcp@286
|
1255 |
\end{ttbox}
|
paulson@4597
|
1256 |
This simpset takes \texttt{imp_cong} as a congruence rule in order to use
|
lcp@286
|
1257 |
contextual information to simplify the conclusions of implications:
|
lcp@286
|
1258 |
\[ \List{\Var{P}\bimp\Var{P'};\; \Var{P'} \Imp \Var{Q}\bimp\Var{Q'}} \Imp
|
lcp@286
|
1259 |
(\Var{P}\imp\Var{Q}) \bimp (\Var{P'}\imp\Var{Q'})
|
lcp@286
|
1260 |
\]
|
paulson@4597
|
1261 |
By adding the congruence rule \texttt{conj_cong}, we could obtain a similar
|
lcp@286
|
1262 |
effect for conjunctions.
|
lcp@286
|
1263 |
|
lcp@286
|
1264 |
|
oheimb@5549
|
1265 |
\subsection{Splitter setup}\index{simplification!setting up the splitter}
|
wenzelm@4557
|
1266 |
|
oheimb@5549
|
1267 |
To set up case splitting, we have to call the \ML{} functor \ttindex{
|
oheimb@5549
|
1268 |
SplitterFun}, which takes the argument signature \texttt{SPLITTER_DATA}.
|
oheimb@5549
|
1269 |
So we prove the theorem \texttt{meta_eq_to_iff} below and store it, together
|
oheimb@5549
|
1270 |
with the \texttt{mk_eq} function described above and several standard
|
oheimb@5549
|
1271 |
theorems, in the structure \texttt{SplitterData}. Calling the functor with
|
oheimb@5549
|
1272 |
this data yields a new instantiation of the splitter for our logic.
|
lcp@286
|
1273 |
\begin{ttbox}
|
oheimb@5549
|
1274 |
val meta_eq_to_iff = prove_goal IFOL.thy "x==y ==> x<->y"
|
oheimb@5549
|
1275 |
(fn [prem] => [rewtac prem, rtac iffI 1, atac 1, atac 1]);
|
lcp@286
|
1276 |
\ttbreak
|
oheimb@5549
|
1277 |
structure SplitterData =
|
oheimb@5549
|
1278 |
struct
|
oheimb@5549
|
1279 |
structure Simplifier = Simplifier
|
oheimb@5549
|
1280 |
val mk_eq = mk_eq
|
oheimb@5549
|
1281 |
val meta_eq_to_iff = meta_eq_to_iff
|
oheimb@5549
|
1282 |
val iffD = iffD2
|
oheimb@5549
|
1283 |
val disjE = disjE
|
oheimb@5549
|
1284 |
val conjE = conjE
|
oheimb@5549
|
1285 |
val exE = exE
|
oheimb@5549
|
1286 |
val contrapos = contrapos
|
oheimb@5549
|
1287 |
val contrapos2 = contrapos2
|
oheimb@5549
|
1288 |
val notnotD = notnotD
|
oheimb@5549
|
1289 |
end;
|
oheimb@5549
|
1290 |
\ttbreak
|
oheimb@5549
|
1291 |
structure Splitter = SplitterFun(SplitterData);
|
lcp@286
|
1292 |
\end{ttbox}
|
lcp@286
|
1293 |
|
lcp@104
|
1294 |
|
lcp@104
|
1295 |
\index{simplification|)}
|
wenzelm@5370
|
1296 |
|
wenzelm@5370
|
1297 |
|
wenzelm@5370
|
1298 |
%%% Local Variables:
|
wenzelm@5370
|
1299 |
%%% mode: latex
|
wenzelm@5370
|
1300 |
%%% TeX-master: "ref"
|
wenzelm@5370
|
1301 |
%%% End:
|