%title%
Isabelle FAQ
%meta%
<style type="text/css">
<!--
.question { background-color:#C0C0C0; color:#000001; padding:3px; margin-top:12px; font-weight: bold; }
.answer { background-color:#E0E0E0; padding:3px; margin-top:3px; }
-->
</style>
%body%
<h2>General Questions</h2>
<table class="question" width="100%">
<tr>
<td>What is Isabelle?</td>
</tr>
</table>
<table class="answer" width="100%">
<tr><td>Isabelle is a popular generic theorem proving
environment developed at Cambridge University (<a
href="http://www.cl.cam.ac.uk/users/lcp/">Larry Paulson</a>)
and TU Munich (<a href="http://www.in.tum.de/~nipkow/">Tobias
Nipkow</a>). See the <a
href="http://isabelle.in.tum.de/">Isabelle homepage</a> for
more information.</td></tr>
</table>
<table class="question" width="100%">
<tr>
<td>Where can I find documentation?</td>
</tr>
</table>
<table class="answer" width="100%">
<tr><td><a href="http://isabelle.in.tum.de/docs.html">This
way, please</a>. Also have a look at the <a
href="http://isabelle.in.tum.de/library/">theory
library</a>.</td></tr>
</table>
<table class="question" width="100%">
<tr>
<td>Is it available for download?</td>
</tr>
</table> <table class="answer" width="100%"> <tr><td>Yes, it is
available from <a
href="http://isabelle.in.tum.de/dist/">several mirror
sites</a>. It should run on most recent Unix systems (Solaris,
Linux, MacOS X, etc.).</td></tr>
</table>
<h2>Syntax</h2>
<table class="question" width="100%">
<tr>
<td>There are lots of arrows in Isabelle. What's the
difference between <tt>-></tt>, <tt>=></tt>, <tt>--></tt>,
and <tt>==></tt> ?</td>
</tr>
</table>
<table class="answer" width="100%">
<tr><td>Isabelle uses the <tt>=></tt> arrow for the function
type (contrary to most functional languages which use
<tt>-></tt>). So <tt>a => b</tt> is the type of a function
that takes an element of <tt>a</tt> as input and gives you an
element of <tt>b</tt> as output. The long arrow <tt>--></tt>
and <tt>==></tt> are object and meta level
implication. Roughly speaking, the meta level implication
should only be used when stating theorems where it separates
the assumptions from the conclusion. Whenever you need an
implication inside a HOL formula, use <code>--></code>.
</td></tr>
</table>
<table class="question" width="100%"><tr><td>Where do I have to put those double quotes?</td></tr></table>
<table class="answer" width="100%"><tr><td>Isabelle distinguishes between <em>inner</em>
and <em>outer</em> syntax. The outer syntax comes from the
Isabelle framework, the inner syntax is the one in between
quotation marks and comes from the object logic (in this case HOL).
With time the distinction between the two becomes obvious, but in
the beginning the following rules of thumb may work: types should
be inside quotation marks, formulas and lemmas should be inside
quotation marks, rules and equations (e.g. for definitions) should
be inside quotation marks, commands like <tt>lemma</tt>,
<tt>consts</tt>, <tt>primrec</tt>, <tt>constdefs</tt>,
<tt>apply</tt>, <tt>done</tt> are without quotation marks, as are
the names of constants in constant definitions (<tt>consts</tt>
and <tt>constdefs</tt>)
</td></tr></table>
<table class="question" width="100%"><tr><td>What is <tt>"No such
constant: _case_syntax"</tt> supposed to tell
me?</td></tr></table>
<table class="answer" width="100%"><tr><td>You get this message if
you use a case construct on a datatype and have a typo in the
names of the constructor patterns or if the order of the
constructors in the case pattern is different from the order in
which they where defined (in the datatype definition).
</td></tr></table>
<table class="question" width="100%"><tr><td>Why doesn't Isabelle understand my
equation?</td></tr></table>
<table class="answer" width="100%"><tr><td>Isabelle's equality <tt>=</tt> binds
relatively strongly, so an equation like <tt>a = b & c</tt> might
not be what you intend. Isabelle parses it as <tt>(a = b) &
c</tt>. If you want it the other way around, you must set
explicit parentheses as in <tt>a = (b & c)</tt>. This also applies
to e.g. <tt>primrec</tt> definitions (see below).</td></tr></table>
<table class="question" width="100%"><tr><td>What does it mean "not a proper
equation"?</td></tr></table>
<table class="answer" width="100%"><tr><td>Most commonly this is an instance of the
question above. The <tt>primrec</tt> command (and others) expect
equations as input, and since equality binds strongly in Isabelle,
something like <tt>f x = b & c</tt> is not what you might expect
it to be: Isabelle parses it as <tt>(f x = b) & c</tt> (which is
indeed not a proper equation). To turn it into an equation you
must set explicit parentheses: <tt>f x = (b & c)</tt>.</td></tr></table>
<table class="question" width="100%"><tr><td>What does it mean
"<tt>Not a meta-equality (==)</tt>"?</td></tr></table>
<table class="answer" width="100%"><tr><td>This usually occurs if
you use <tt>=</tt> for <tt>constdefs</tt>. The <tt>constdefs</tt>
and <tt>defs</tt> commands expect not equations, but meta
equivalences. Just use the <tt>\<equiv></tt> or <tt>==</tt>
signs instead of <tt>=</tt>. </td></tr></table>
<h2>Proving</h2>
<table class="question" width="100%"><tr><td>What does "empty result sequence"
mean?</td></tr></table>
<table class="answer" width="100%"><tr><td>It means that the applied proof method (or
tactic) was unsuccessful. It did not transform the goal in any
way, or simply just failed to do anything. You must try another
tactic (or give the one you used more hints or lemmas to work
with)</td></tr></table>
<table class="question" width="100%"><tr><td>The Simplifier doesn't want to apply my
rule, what's wrong?</td></tr></table>
<table class="answer" width="100%"><tr><td>
Most commonly this is a typing problem. The rule you want to apply
may require a more special (or just plain different) type from
what you have in the current goal. Use the ProofGeneral menu
<tt>Isabelle/Isar -> Settings -> Show Types</tt> and the
<tt>thm</tt> command on the rule you want to apply to find out if
the types are what you expect them to be (also take a look at the
types in your goal). <tt>Show Sorts</tt>, <tt>Show Constants</tt>,
and <tt>Trace Simplifier</tt> in the same menu may also be
helpful.
</td></tr></table>
<table class="question" width="100%"><tr><td>If I do <tt>auto</tt>, it leaves me a goal
<tt>False</tt>. Is my theorem wrong?</td></tr></table>
<table class="answer" width="100%"><tr><td>Not necessarily. It just means that
<tt>auto</tt> transformed the goal into something that is not
provable any more. That could be due to <tt>auto</tt> doing
something stupid, or e.g. due to some earlier step in the proof
that lost important information. It is of course also possible
that the goal was never provable in the first place.</td></tr></table>
<table class="question" width="100%"><tr><td>Why does <tt>lemma
"1+1=2"</tt> fail?</td></tr></table>
<table class="answer" width="100%"><tr><td>Because it is not
necessarily true. Isabelle does not assume that 1 and 2 are
natural numbers. Try <tt>"(1::nat)+1=2"</tt>
instead.</td></tr></table>
<table class="question" width="100%"><tr><td>Can Isabelle find
counterexamples?</td></tr></table>
<table class="answer" width="100%"><tr><td>
<p>
For arithmetic goals, <code>arith</code> finds counterexamples.
For executable goals, <code>quickcheck</code> tries to find a
counterexample. For goals of a more logical nature (including
quantifiers, sets and inductive definitions) <code>refute</code>
searches for a countermodel.
</p>
<p>
Otherwise, negate the proposition
and instantiate (some) variables with concrete values. You may
also need additional assumptions about these values. For example,
<tt>True & False ~= True | False</tt> is a counterexample of <tt>A
& B = A | B</tt>, and <tt>A = ~B ==> A & B ~= A | B</tt> is
another one. Sometimes Isabelle can help you to find the
counterexample: just negate the proposition and do <tt>auto</tt>
or <tt>simp</tt>. If lucky, you are left with the assumptions you
need for the counterexample to work.
</p>
</td></tr></table>
<h2>Interface</h2>
<table class="question" width="100%"><tr><td>ProofGeneral appears to hang when Isabelle is started.</td></tr></table>
<table class="answer" width="100%"><tr>This may be because of UTF-8 issues e.g. in Red Hat 8.0/9.0, Suse 9.0/9.1
<p>
RedHat 8 and later has glibc 2.2 and UTF8 encoded output may be
turned on in default locale. Unfortunately Proof General relies on
8-bit characters which are UTF8 prefixes in the output of proof
assistants (inc Coq, Isabelle). These prefix characters are not
flushed to stdout individually. As a workaround we must find a way
to disable interpretation of UTF8 in the C libraries that Coq and
friends use.
<p>
Doing this inside PG/Emacs seems tricky; locale settings are
set/inherited in strange ways. One solution is to run the Emacs
process itself with an altered locale setting, for example,
starting XEmacs by typing:
<p>
<tt>$ LC_CTYPE=en_GB Isabelle &</tt>
<p>
The supplied proofgeneral script makes this setting if it sees
the string UTF in the current value of LC_CTYPE. Depending on your
distribution, this variable might also be called <tt>LANG</tt>.
<p>
Alternatively you can set LC_CTYPE or LANG inside a file ~/.i18n, which will
be read by the shell. This will affect all applications, though.
[ suggestions for a better workaround inside Emacs would be welcome ]
<p>
NB: a related issue is warnings from x-symbol: "Emacs language
environment and system locale specify different encoding, I'll
assume `iso-8859-1'". This warning appears to be mainly harmless.
Notice that the variable `buffer-file-coding-system' may determine
the format that files are saved in.<td></td></tr></table>
<table class="question" width="100%"><tr><td>X-Symbol doesn't seem to work. What can I
do?</td></tr></table>
<table class="answer" width="100%"><tr><td>The most common reason why X-Symbol doesn't
work is: it's not switched on yet. Assuming you are using
ProofGeneral and have installed the X-Symbol package, you still
need to turn X-Symbol on in ProofGeneral: select the menu items
<tt>Proof-General -> Options -> X-Symbol</tt> and (if you want to
save the setting for future sessions) select <tt>Options -> Save
Options</tt> in XEmacs.</td></tr></table>
<table class="question" width="100%"><tr><td>How do I input those X-Symbols anyway?</td></tr></table>
<table class="answer" width="100%"><tr><td>
There are lots of ways to input x-symbols. The one that always
works is writing it out in plain text (e.g. for the 'and' symbol
type <tt>\<and></tt>). For common symbols you can try to "paint
them in ASCII" and if the xsymbol package recognizes them it will
automatically convert them into their graphical
representation. Examples: <tt>--></tt> is converted into the long
single arrow, <tt>/\</tt> is converted into the 'and' symbol, the
sequence <tt>=_</tt> into the equivalence sign, <tt><_</tt> into
less-or-equal, <tt>[|</tt> into opening semantic brackets, and so
on. For greek characters, the <code>rotate</code> command works well:
to input α type <code>a</code> and then <code>C-.</code>
(control and <code>.</code>). You can also display the
grid-of-characters in the x-symbol menu to get an overview of the
available graphical representations (not all of them already have
a meaning in Isabelle, though).
</td></tr></table>
<h2>System</h2>
<table class="question" width="100%"><tr><td>I want to generate one of those flashy LaTeX
documents. How?</td></tr></table>
<table class="answer" width="100%"><tr><td>You will need to work with the
<tt>isatool</tt> command for this (in a Unix shell). The easiest
way to get to a document is the following: use <tt>isatool
mkdir</tt> to set up a new directory. The command will also create
a file called <tt>IsaMakefile</tt> in the current directory. Put
your theory file(s) into the new directory and edit the file
<tt>ROOT.ML</tt> in there (following the comments) to tell
Isabelle which of the theories to load (and in which order). Go
back to the parent directory (where the <tt>IsaMakefile</tt> is)
and type <tt>isatool make</tt>. Isabelle should then process your
theories and tell you where to find the finished document. For
more information on generating documents see the Isabelle Tutorial, Chapter 4.
</td></tr></table>
<table class="question" width="100%"><tr><td>I have a large formalization with many
theories. Must I process all of them all of the time?</td></tr></table>
<table class="answer" width="100%"><tr><td>No, you can tell Isabelle to build a so-called
heap image. This heap image can contain your preloaded
theories. To get one, set up a directory with a <tt>ROOT.ML</tt>
file (as for generating a document) and use the command
<tt>isatool usedir -b HOL MyImage</tt> in that directory to create
an image <tt>MyImage</tt> using the parent logic <tt>HOL</tt>.
You should then be able to invoke Isabelle with <tt>Isabelle -l
MyImage</tt> and have everything that is loaded in ROOT.ML
instantly available.</td></tr></table>
<table class="question" width="100%"><tr><td>Does Isabelle run on Windows?</td></tr></table>
<table class="answer" width="100%"><tr><td> After a fashion, yes,
but Isabelle is not being developed for Windows. A friendly user
(Norbert Völker) has managed to get a minimal Isabelle environment
to work on it. See the description on <a
href="http://cswww.essex.ac.uk/Research/FSS/projects/isawin/">his
website</a>. Be warned, though: emphasis is on <em>minimal</em>,
working with Windows is no fun at all. To enjoy Isabelle in its
full beauty it is recommended to get a Linux distribution (they
are inexpensive, any reasonably recent one should work, dualboot
Windows/Linux should pose no problems).
</td></tr></table>