forked from SymbolicPathFinder/jpf-symbc
-
Notifications
You must be signed in to change notification settings - Fork 5
/
dreal.readme
49 lines (41 loc) · 3.73 KB
/
dreal.readme
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
In this file we explain how to use dReal solver with Symbolic Path Finder (SPF). dReal is an SMT solver
that support first-order logic formulas over the real numbers. The tool can handle formulas with nonlinear
real functions in the framework of delta-complete decision procedures (for more information please look
at https://dreal.github.io/).
In order to install SPF that supports dReal one needs to 1) download SPF version that contains support
for dReal, and 2) install dReal itself. "http://babelfish.arc.nasa.gov/trac/jpf/wiki/projects/jpf-symbc"
contains address of the repository. Any version committed on or after "Thu, 30 Jul 2015 12:51:29 -0700"
should be fine for our purpose. After obtaining a copy of SPF, one also need to install dReal. Required
information regarding downloading and installing this tool can be found in "https://dreal.github.io/".
dReal problem solver of SPF has been tested with version 2 of dReal.
After downloading and installing both SPF and dReal, we need to configure properties required for SPF.
As with the other properties of SPF and JPF (Java Path Finder), properties related to dReal solver are
specified in "jpf.properties" file. Currently, there are three properties specific to the dReal solver
that should be specified in this file: "dreal.timeBound", "dreal.precision", and "dreal.mode".
1) dreal.timeBound, an integer value with default -1 (in milliseconds),
SPF constructs mathematical constraints and give them to dReal solver to check for their satisfiability.
Since these constraints can become arbitrary large, even if their satisfiability remains decidable, it is
possible that for some formula, dReal cannot find an answer in a reasonable amount of time. Setting this
parameter will force dReal to stop and return "unknown" if it did not find an answer after the given
number of milliseconds. Non-positive values mean dReal will wait as much as required to find an answer.
2) dreal.precision, a positive real (double) value with default "0.00001",
dReal, looks for "delta" satisfiability of the input formula. The delta relaxation is specified with this
parameter and will be directly passed to the dReal.
3) dreal.mode, "strictified" or "relaxed" with default "strictified".
As we mentioned before, dReal looks for satisfiability of "relaxed" formula. As a result, if a formula is
found to be delta-sat, there is no guarantee that actual formula will also be satisfiable. In other words,
dReal decides between "unsat" and "delta-sat". This is the behavior one can expect with "relaxed" value
parameter. On the other hand, one can also ask dReal to decide between "sat" and "delta-unsat". In order
to do this, "strictified" should be used as the value of this parameter.
Note that, as we are writing this document, strictification is not directly supported by dReal. What
actually happens is that each constraint will be strictified first, and the result will be given to dReal
next. As a consequence, all equalities will be replaced by 'false'. Similarly, in the other scenario,
'true' will take place of all inequalities.
Finally, note that even if a formula is found to be satisfiable, there is no guarantee that the generated
model will actually satisfy constraints in the implementation. The reason for this is that dReal works
with Type II computable functions which in principle are very different from floating point numbers and
functions used by java.
Last but not least, one should also tell were to find the dReal executable file. The current interface to
dReal is completely independent of SPF and JPF. As a result, jpf.properties file is not used to specify
the path to dReal. Instead, the path is specified by setting the property "dreal-bin-path" through command
line (e.g. "-Ddreal-bin-path=/path/to/dreal/bin").