#+TITLE: =randltl=
#+EMAIL spot@lrde.epita.fr
#+OPTIONS: H:2 num:nil toc:t
#+LINK_UP: file:tools.html
This tool generates random formulas. By default, it will generate one
random LTL formula using atomic propositions supplied on the
command-line. It can be instructed to generate random Boolean or PSL
formulas instead, but let us first focus on LTL generation.
For instance to obtain one random LTL formula over the propositions
=a=, =b=, or =c=, use:
#+BEGIN_SRC sh :results verbatim :exports both
randltl a b c
#+END_SRC
#+RESULTS:
: G(Gb W (Gb M c))
Note that the result does not always use all atomic propositions.
The syntax of the formula output can be changed using the
[[file:ioltl.org][common output options]]:
#+BEGIN_SRC sh :results verbatim :exports results
randltl --help | sed -n '/Output options:/,/^$/p' | sed '1d;$d'
#+END_SRC
#+RESULTS:
: -8, --utf8 output using UTF-8 characters
: -l, --lbt output in LBT's syntax
: -p, --full-parentheses output fully-parenthesized formulas
: -s, --spin output in Spin's syntax
: --spot output in Spot's syntax (default)
: --wring output in Wring's syntax
When you select Spin's or Wring's syntax, operators =W= and =M= are
automatically rewritten using =U= and =R= (written =V= for Spin).
When you select LBT's syntax, you should name you atomic propositions
like =p0=, =p1=, etc... (Atomic proposition named differently will be
output by Spot in double-quotes, but this is not supported by LBT.)
#+BEGIN_SRC sh :results verbatim :exports both
randltl -l p1 p2 p3
randltl -8 p1 p2 p3
randltl -s p1 p2 p3
randltl --wring p1 p2 p3
#+END_SRC
#+RESULTS:
: G W G p2 M G p2 p3
: □(□p2 W (□p2 M p3))
: []((p3 U (p3 && []p2)) V ([]p2 || (p3 U (p3 && []p2))))
: (G(((p3=1) U ((p3=1) * (G(p2=1)))) R ((G(p2=1)) + ((p3=1) U ((p3=1) * (G(p2=1)))))))
As you might guess from the above result, for a given set of atomic
propositions (and on the same computer) the generated formula will
always be the same. This is because each time =randltl= starts, it
initialize the seed of the random number generator to =0=. This seed
can be changed using the =--seed= option. For instance the following
three commands:
#+BEGIN_SRC sh :results verbatim :exports both
randltl a b c
randltl --seed=123 a b c
randltl --seed=0 a b c
#+END_SRC
Will give three formulas in which the first and last are identical:
#+RESULTS:
: G(Gb W (Gb M c))
: X((c <-> F(a M Xc)) -> a)
: G(Gb W (Gb M c))
When generating random formulas, we usually want large quantity of
them. Rather than running =randltl= several times with different
seeds, we can use the =-n= option to specify a number of formulas to
produce.
#+BEGIN_SRC sh :results verbatim :exports both
randltl -n 5 a b c
#+END_SRC
#+RESULTS:
: G(Gb W (Gb M c))
: !(GFb -> Fa)
: (((c xor Xc) R c) R Gc) & !c
: X(1 U Xb) M Fb
: !XFb U !(Xa W 0)
By default =randltl= will never output the same formula twice (this
can be changed with the =--allow-dups= option), so it may generate
more formulas internally than it eventually prints. To ensure
termination, for each output formula the number of ignored (because
duplicated) random formulas that are generated is limited to 100000.
Therefore in some situations, most likely when generating small
formulas, with few atomic proposition, you may see =randltl= stop
before the requested number of formulas has been output with an error
message.
If the integer passed to =-n= is negative, =randltl= will attempt to
generate as many formulas as it can. This is most useful when
=randltl= is piped to =ltlfilt= to select random formulas matching a
certain criterion, as we shall see later.
Besides the list of atomic propositions (=a b c= in our example) and
the seed, several other parameters control the generation of the
random formulas.
Initially, the random generator selects a tree size for the formula.
The default size is =15=, but it can be changed using the =--tree-size=
option. For instance in the following, for each formula the tree size
will be chosen randomly in the range =22..30=.
#+BEGIN_SRC sh :results verbatim :exports both
randltl -n 5 a b c --tree-size=22..30
#+END_SRC
#+RESULTS:
: ((Fc W 0) M 1) & ((a | XXc) M (b W Fb))
: (!(c | (a R Xb)) <-> (Xc R (a & c))) -> !a
: 0
: X(Xc R GX(1 U Xb)) & GX(b M Xc)
: (!b -> ((c xor Fa) W b)) W (!Fb U a)
The tree size is just the number of nodes in the syntax tree of the
formula during its construction. However because Spot automatically
applies some /trivial simplifications/ during the construction of its
formulas (e.g., =F(F(a)= is reduced to =F(a)=, =a&0= to =0=, etc.),
the actual size of the formula output may be smaller than the
tree size specified.
It is pretty common to obtain the formulas =0= or =1= among the first
formulas output, since many random formulas trivially simplify to
these. However because duplicate formulas are suppressed by default,
they shall only occur once.
Stronger simplifications may be requested using the =-r= option, that
implements many rewritings that helps Spot translators algorithms (so
beware that using =-r= reduces the randomness of the output).
#+BEGIN_SRC sh :results verbatim :exports both
randltl -n 5 a b c --tree-size=22..30 -r
#+END_SRC
#+RESULTS:
: GFc & ((a | XXc) M Fb)
: !a | (!c & (!a U X!b)) | (a & c & Xc)
: 0
: XG(b M Xc)
: (((!c & Fa) | (c & G!a)) W b) W (G!b U a)
The generator build the syntax tree recursively from its root, by
considering all operators that could be used for a given tree size (for
example a tree-size of 2 disables binary operators). A /priority/ is
associated to each operator, and the probability of this operator
being selected is this priority over the sum of the priorities of all
considered operators. The default priorities for each operator can
be seen with =--dump-priorities=:
#+BEGIN_SRC sh :results verbatim :exports both
randltl a b c --dump-priorities
#+END_SRC
#+RESULTS:
#+begin_example
Use --ltl-priorities to set the following LTL priorities:
ap 3
false 1
true 1
not 1
F 1
G 1
X 1
equiv 1
implies 1
xor 1
R 1
U 1
W 1
M 1
and 1
or 1
#+end_example
Where =ap= stands for /atomic propositions/ (=a=, =b=, =c=). In this
example, when the generator builds a leaf of the syntax tree (i.e., a
subformula of tree-size 1), it must ignore all operators, and chose
between =ap=, =false=, or =true=, and the odds of choosing =ap= is
3/(3+1+1).
As indicated at the top of the output, these priorities can be changed
using the =--ltl-priorities= options. The most common scenario is to
disable some of the operators by giving them a null priority. The
following example disables 6 operators, and augments the priority of
=U= to 3 to favor its occurrence.
#+BEGIN_SRC sh :results verbatim :exports both
randltl -n 5 a b c --ltl-priorities 'xor=0,implies=0,equiv=0,W=0,M=0,X=0,U=3'
#+END_SRC
#+RESULTS:
: F(Fb U (c & Gb))
: !(Fb U Fa)
: ((Gc U c) U c) U Fc
: 0
: 1
When using =-r= to simplify generated formulas, beware that these
rewritings may use operators that you disabled during the initial
random generation: you may obtain a formula that uses =W= even if =W=
has a null priority. (You can use =ltlfilt= to rewrite these
operators if that is a problem.)
If the =--weak-fairness= option is used, for each random formula
generated, a weak-fairness formula of the form =GF(a) & GF(b) & GF(c)=
is generated for a subset of the atomic propositions and "ANDed" to
the random formula. The =--tree-size= option has no influence on the
weak-fairness formula appended.
#+BEGIN_SRC sh :results verbatim :exports both
randltl -n 5 a b c --weak-fairness
#+END_SRC
#+RESULTS:
: G(Gb W (Gb M c)) & GFb
: GFb & !(GFb -> Fa) & GFa
: GFb & (((c xor Xc) R c) R Gc) & !c
: GFb & GFa & (X(1 U Xb) M Fb)
: GFb & (!XFb U !(Xa W 0)) & GFc
Boolean formulas may be output with the =-B= option:
#+BEGIN_SRC sh :results verbatim :exports both
randltl -B -n 5 a b c
#+END_SRC
#+RESULTS:
: !a -> !b
: !(!(a -> (b xor c)) -> !a)
: !c | (!c xor (c xor (c xor !c)))
: !b
: a xor !(!b <-> (!a -> c))
In that case, priorities should be set with =--boolean-priorities=.
Finally, PSL formulas may be output using the =-P= option. However
keep in mind that since LTL formulas are PSL formulas, generating
random PSL formula may produce many LTL formulas that do not use any
PSL operator (this is even more so the case when simplifications are
enabled with =-r=).
#+BEGIN_SRC sh :results verbatim :exports both
randltl -P -n 5 a b c
#+END_SRC
#+RESULTS:
: G(Gb M ((c & Xb) | (!a M 1)))
: !(Fa xor X({{c | {a xor !b}}[*]}[]-> Fb))
: c & Gc
: 0
: 1
As shown with the =--dump-priorities= output below, tweaking the
priorities used to generated PSL formulas requires three different
options:
#+BEGIN_SRC sh :results verbatim :exports both
randltl -P a b c --dump-priorities
#+END_SRC
#+RESULTS:
#+begin_example
Use --ltl-priorities to set the following LTL priorities:
ap 3
false 1
true 1
not 1
F 1
G 1
X 1
Closure 1
equiv 1
implies 1
xor 1
R 1
U 1
W 1
M 1
and 1
or 1
EConcat 1
UConcat 1
Use --sere-priorities to set the following SERE priorities:
eword 1
boolform 1
star 1
star_b 1
and 1
andNLM 1
or 1
concat 1
fusion 1
Use --boolean-priorities to set the following Boolean formula priorities:
ap 3
false 1
true 1
not 1
equiv 1
implies 1
xor 1
and 1
or 1
#+end_example
The =--ltl-priorities= option we have seen previously now recognize
some new PSL-specific operators: =Closure= is the ={sere}= operator,
=EConcat= is the ={sere}<>->f= operator, and =UConcat= is the
={sere}[]->f= operator. When these operator are selected, they
require a SERE argument which is generated according to the priorities
set by =--sere-priorities=: =eword= is the empty word, =boolform= is a
Boolean formula (generated using the priorities set by
=--boolean-priorities=), =star= is the unbounded Kleen star, while
=star_b= is the bounded version, and =andNLM= is the
non-length-matching variant of the =and= operator.
# LocalWords: randltl num toc LTL PSL SRC Gb sed utf UTF lbt LBT's
# LocalWords: Xc GFb Gc Xb Fb XFb Xa dups ltlfilt Fc XXc GX GFc XG
# LocalWords: rewritings ltl ap GF ANDed GFa boolean EConcat eword
# LocalWords: UConcat boolform andNLM concat Kleen eval setenv setq
# LocalWords: getenv