Z3 and DIMACS output - z3

Z3 currently supports the DIMACS format for input. Is there any way to output the DIMACS format for the problem before solution? I mean converting the problem to a system CNFs and output it in a DIMACS format.
If not, any ideas towards this direction would be more than helpful.

The DIMACS format is very primitive, it supports only Boolean variables. Z3 does not reduce every problem into SAT. Some problems are solved using a propositional SAT solver, but this is not the rule. This usually only happens if the input contains only Boolean and/or Bit-vector variables. Moreover, even if the input problem contains only Boolean and Bit-vector variables, there is no guarantee that Z3 will use a pure SAT solver to solve it.
That being said, you can use the tactic framework to control Z3. For example, for Bit-vector problems, the following tactic will convert it into a propositional formula in CNF format. It should be straightforward to convert it into DIMACS. Here is the example. You can try it online at: http://rise4fun.com/Z3Py/E1s
x, y, z = BitVecs('x y z', 16)
g = Goal()
g.add(x == y, z > If(x < 0, x, -x))
print g
# t is a tactic that reduces a Bit-vector problem into propositional CNF
t = Then('simplify', 'bit-blast', 'tseitin-cnf')
subgoal = t(g)
assert len(subgoal) == 1
# Traverse each clause of the first subgoal
for c in subgoal[0]:
print c

Thanks to Leonardo's answer I came up with this code that will do what you want:
private static void Output(Context ctx,Solver slv)
{
var goal = ctx.MkGoal();
goal.Add(slv.Assertions);
var applyResult = ctx.Then(ctx.MkTactic("simplify"),
ctx.MkTactic("bit-blast"),
ctx.MkTactic("tseitin-cnf")).Apply(goal);
Debug.Assert(applyResult.Subgoals.Length==1);
var map = new Dictionary<BoolExpr,int>();
foreach (var f in applyResult.Subgoals[0].Formulas)
{
Debug.Assert(f.IsOr);
foreach (var e in f.Args)
if (e.IsNot)
{
Debug.Assert(e.Args.Length==1);
Debug.Assert(e.Args[0].IsConst);
map[(BoolExpr)e.Args[0]] = 0;
}
else
{
Debug.Assert(e.IsConst);
map[(BoolExpr)e] = 0;
}
}
var id = 1;
foreach (var key in map.Keys.ToArray())
map[key] = id++;
using (var fos = File.CreateText("problem.cnf"))
{
fos.WriteLine("c DIMACS file format");
fos.WriteLine($"p cnf {map.Count} {applyResult.Subgoals[0].Formulas.Length}");
foreach(var f in applyResult.Subgoals[0].Formulas)
{
foreach (var e in f.Args)
if (e.IsNot)
fos.Write($"{map[(BoolExpr)e.Args[0]]} ");
else
fos.Write($"-{map[(BoolExpr)e]} ");
fos.WriteLine("0");
}
}
}
For it to work you should add all your constraints to the solver directly, by calling slv.Assert(...).

Related

Autodiff for Jacobian derivative with respect to individual joint angles

I am trying to compute $\partial{J}{q_i}$ in drake C++ for manipulator and as per my search, the best approach seems to be using autodiff function. I was not able to fully understand autodiff approach from the resources that I found, so I apologize if my approach is not clear enough. I have used my understanding from some already asked questions mentioned on the forum regarding auto diff as well as https://drake.mit.edu/doxygen_cxx/classdrake_1_1multibody_1_1_multibody_plant.html as reference.
As I want to calculate $\partial{J}{q_i}$, the return type will be a tensor i.e. 3 * 7 * 7(or 6 * 7 * 7 depending on the spatial jacobian). I can think of using std::vectorEigen::MatrixXd to allocate the output or alternatively just doing one $q_i$ at a time and computing the respective jacobian for the auto diff. In either case, I was struggling to pass it in the initializing the jacobian function.
I did the following to initialize autodiff
std::unique_ptr<multibody::MultibodyPlant<AutoDiffXd>> mplant_autodiff = systems::System<double>::ToAutoDiffXd(mplant);
std::unique_ptr<systems::Context<AutoDiffXd>> mContext_autodiff = mplant_autodiff->CreateDefaultContext();
mContext_autodiff->SetTimeStateAndParametersFrom(*mContext);
const multibody::Frame<AutoDiffXd>* mFrame_EE_autodiff = &mplant_autodiff->GetBodyByName(mEE_link).body_frame();
const multibody::Frame<AutoDiffXd>* mWorld_Frame_autodiff = &(mplant_autodiff->world_frame());
//Initialize the q as autodiff vector
drake::AutoDiffVecXd q_autodiff = drake::math::InitializeAutoDiff(mq_robot);
MatrixX<AutoDiffXd> mJacobian_autodiff; // Linear Jacobian matrix.
mplant_autodiff->SetPositions(context_autodiff.get(), q_autodiff);
mplant_autodiff->CalcJacobianTranslationalVelocity(*mContext_autodiff,
multibody::JacobianWrtVariable::kQDot,
*mFrame_EE_autodiff,
Eigen::Vector3d::Zero(),
*mWorld_Frame_autodiff,
*mWorld_Frame_autodiff,
&mJacobian_autodiff
);
However, as far as I understand, InitializeAutoDiff initializes to the identity matrix, whereas I want to $\partial{J}{q_i}$, so is there is a better way to do it. In addition, I get error messages when I try to call the jacobian matrix. Is there a way to address this problem both for $\partial{J}{q_i}$ for each q_i and changing q_i in a for loop or directly getting the result in a tensor. My apologies if I am doing something total tangent to the correct approach. I thank you in anticipation.
However, as far as I understand, InitializeAutoDiff initializes to the identity matrix, whereas I want to $\partial{J}{q_i}$, so is there is a better way to do it
That is correct. When you call InitializeAutoDiff and compute mJacobian_autodiff, you get a matrix of AutoDiffXd. Each AutoDiffXd has a value() function that stores the double value, and a derivatives() storing the gradient as an Eigen::VectorXd. We have
mJacobian(i, j).value() = J(i, j)
mJacobian_autodiff(i, j).derivatives()(k) = ∂J(i, j)/∂q(k)
So if you want to create a std::vecot<Eigen::MatrixXd> such that the k'th entry of this vector stores the matrix ∂J/∂q(k), then here is a code
std::vector<Eigen::MatrixXd> dJdq(q_autodiff.rows());
for (int i = 0; i < q_autodiff.rows(); ++i) {
dJdq[i].resize(mJacobian_autodiff.rows(), mJacobian_autodiff.cols());
}
for (int i = 0; i < q_autodiff.rows(); ++i) {
// dJidq stores the gradient of the ∂J.col(i)/∂q, namely dJidq(j, k) = ∂J(j, i)/∂q(k)
auto dJidq = ExtractGradient(mJacobian_autodiff.col(i));
for (int j = 0; j < static_cast<int>(dJdq.size()); ++j) {
dJdq[j].col(i) = dJidq.col(j);
}
}
Compute ∂J/∂q(i) for a single i
If you do not want to compute ∂J/∂q(i) for all i, but only for one specific i, you can change the initialization of q_autodiff from InitializeAutoDiff to this
AutoDiffVecXd q_autodiff(q.rows());
for (int k = 0; k < q_autodiff.rows(); ++k) {
q_autodiff(k).value() = q(k)
q_autodiff(k).derivatives() = Vector1d::Zero();
if (k == i) {
q_autodiff(k).derivatives()(0) = 1;
}
}
namely q_autodiff stores the gradient ∂q/∂q(i), which is 0 for all k != i and 1 when k == i. And then you can compute mJacobian_autodiff using your current code. Now mJacobian_autodiff(m, n).derivatives() store the gradient of ∂J(m, m)/∂q(i) for that specific i. You can extract this gradient as
Eigen::Matrix dJdqi(mJacobian_autodiff.rows(), mJacobian_autodiff.cols());
for (int m = 0; m < dJdqi.rows(); ++m) {
for (int n = 0; n < dJdqi.cols(); ++n) {
dJdqi(m, n) = mJacobian_autodiff(m, n).derivatives()(0);
}
}

using a string in a math equation in Dart

I store various formulas in Postgres and I want to use those formulas in my code. It would look something like this:
var amount = 100;
var formula = '5/105'; // normally something I would fetch from Postgres
var total = amount * formula; // should return 4.76
Is there a way to evaluate the string in this manner?
As far as I'm aware, there isn't a formula solver package developed for Dart yet. (If one exists or gets created after this post, we can edit it into the answer.)
EDIT: Mattia in the comments points out the math_expressions package, which looks pretty robust and easy to use.
There is a way to execute arbitrary Dart code as a string, but it has several problems. A] It's very roundabout and convoluted; B] it becomes a massive security issue; and C] it only works if the Dart is compiled in JIT mode (so in Flutter this means it will only work in debug builds, not release builds).
So the answer is that unfortunately, you will have to implement it yourself. The good news is that, for simple 4-function arithmetic, this is pretty straight-forward, and you can follow a tutorial on writing a calculator app like this one to see how it's done.
Of course, if all your formulas only contain two terms with an operator between them like in your example snippet, it becomes even easier. You can do the whole thing in just a few lines of code:
void main() {
final amount = 100;
final formula = '5/105';
final pattern = RegExp(r'(\d+)([\/+*-])(\d+)');
final match = pattern.firstMatch(formula);
final value = process(num.parse(match[1]), match[2], num.parse(match[3]));
final total = amount * value;
print(total); // Prints: 4.761904761904762
}
num process(num a, String operator, num b) {
switch (operator) {
case '+': return a + b;
case '-': return a - b;
case '*': return a * b;
case '/': return a / b;
}
throw ArgumentError(operator);
}
There are a few packages that can be used to accomplish this:
pub.dev/packages/function_tree
pub.dev/packages/math_expressions
pub.dev/packages/expressions
I used function_tree as follows:
double amount = 100.55;
String formula = '5/105*.5'; // From Postgres
final tax = amount * formula.interpret();
I haven't tried it, but using math_expressions it should look like this:
double amount = 100.55;
String formula = '5/105*.5'; // From Postgres
Parser p = Parser();
// Context is used to evaluate variables, can be empty in this case.
ContextModel cm = ContextModel();
Expression exp = p.parse(formula) * p.parse(amount.toString());
// or..
//Expression exp = p.parse(formula) * Number(amount);
double result = exp.evaluate(EvaluationType.REAL, cm);
// Result: 2.394047619047619
print('Result: ${result}');
Thanks to fkleon for the math_expressions help.

Z3 - How to extract variables from a given formula?

I'm using Z3 C++ API (Version 4.3.1) and I want to extract variables of a formula (An object of type expr). I've found a similar question but it is in Z3py. I am wonder if there is a method in Z3 C/C++ API to extract variables from expr object. Thanks!
For example (some details omitted):
expr fs = implies(x + y == 0, z * x < 15);
std::vector<expr> varlist = get_vars(fs);
Then varlist should contain x,y,z.
The C++ example in the distribution (examples/c++/example.cpp) shows a sample visitor pattern.
It is very simplistic, but will give the idea.
I repeat it here below:
void visit(expr const & e) {
if (e.is_app()) {
unsigned num = e.num_args();
for (unsigned i = 0; i < num; i++) {
visit(e.arg(i));
}
// do something
// Example: print the visited expression
func_decl f = e.decl();
std::cout << "application of " << f.name() << ": " << e << "\n";
}
else if (e.is_quantifier()) {
visit(e.body());
// do something
}
else {
assert(e.is_var());
// do something
}
}
The visitor function can be improved by using
a cache of previously visited expressions because in general Z3 uses shared sub-expressions.
This is similar to the Python example.
Hope this helps

Finding follow sets - infinite recursion

While finding follow sets, rules such as
A->aA can lead to infinite recursion. Is there any coding technique to avoid it?
Note that the above example is just an example, in practice such a recursion could happen indirectly as well.
Here is my sample C code for finding follow sets. The grammar is stored as an array of linked lists. Please tell me if the code is unclear at any point.
set findFollowSet(char nonTerminal[], Grammar G, hashTable2 h) //later assume that all first sets are already in the hashtable.
{
LINK temp1 = find2(h, nonTerminal);
set s= createEmptySet();
set temp = createEmptySet();
char lhs[80] = "\0";
int i;
//special case
if(temp1->numRightSideOf==0) //its not on right side of any grammar rule
return insert(s, "$");
for(i=0;i<temp1->numRightSideOf;i++)
{
link l = G.rules[temp1->rightSideOf[i]];
strcpy(lhs, l->symbol); //storing the lhs just in case the nonTerm appears on the rightmost end of the rule.
printf("!!!!! %s\n", lhs);
sleep(1);
//finding nonTerminal in G
while(l!=NULL)
{
if(strcmp(l->symbol, nonTerminal) == 0)
break;
l=l->next;
}
//found the nonTerminal in G
if(l->next!=NULL)
{
temp = findFirstSet(l->next, G, h);
temp = removeElement(temp, "EPSILON");
}
else //its on the rightmost end of the rule
temp = findFollowSet(lhs, G, h);
s = setUnion(s, temp); destroySet(temp);
}
return s;
}
FIRST and FOLLOW sets are defined recursively, so you need to find the recursive closure. What this mean in practice is that you don't find the FOLLOW set for a single non-terminal -- you find all the FOLLOW sets for all the terminals simultaneously, by starting with all sets empty and going over the grammar adding symbols to different sets, until no more symbols can be added to any set. So you end up with something like:
FOLLOW[*] = {}; // all follow sets start empty
done = false;
while (!done)
done = true;
for (R : each rule in the grammar)
A = RHS[R];
tmp = FOLLOW[A];
for (S : each symbol in LHS[R] from right to left)
if (S is terminal)
tmp = {S};
else
if (!(FOLLOW[S] contains tmp))
done = false
FOLLOW[S] |= tmp
if (epsilon in FIRST[S])
tmp |= FIRST[S] - epsilon
else
tmp = FIRST[S]
Ok I got the answer but its inefficient.
So if anyone wants to suggest some more efficient answer, please feel welcomed.
Just store the recursion stack explicitly and at each recursive call, check if the entry already exists in the stack.
Mind you, you need to check the entire stack not just the top of it.

An error appears when running exist quantifier and fixedpoint Z3 in C#

I tried to use ctx.mkExist in the fixedpoint, howwever, it occurs error said "contains recursive predicate", I don't know why? and How to use ctx.MkExists in fixedpoint?For example:
exist (lamda real) that lamb>=0 AND inv(c,i) AND phi(c+lamb,i) => phi(c,i)
using (Context ctx = new Context())
{
var s = ctx.MkFixedpoint();
IntSort B = ctx.IntSort;
BoolSort T = ctx.BoolSort;
RealSort R = ctx.RealSort;
FuncDecl phi = ctx.MkFuncDecl("phi", new Sort[] { R,B }, T);
s.RegisterRelation(phi);
FuncDecl Inv = ctx.MkFuncDecl("inv", new Sort[] { R, B }, T);
s.RegisterRelation(Inv);
RealExpr c= (RealExpr)ctx.MkBound(0, R);
IntExpr i = (IntExpr) ctx.MkBound(1, B);
Expr[] InvArg=new Expr[2];
InvArg[0] = ctx.MkConst("inv0" , Inv.Domain[0]);
InvArg[1] = ctx.MkConst("inv1", Inv.Domain[1]);
Expr invExpr = ctx.MkImplies(ctx.MkOr(
ctx.MkAnd(ctx.MkEq(InvArg[1], ctx.MkInt(0)), ctx.MkGe((RealExpr)InvArg[0], ctx.MkReal(0))),
ctx.MkAnd(ctx.MkEq(InvArg[1], ctx.MkInt(1)), ctx.MkGe((RealExpr)InvArg[0], ctx.MkReal(2)))
),
(BoolExpr)Inv[InvArg]);
Quantifier invQ = ctx.MkForall(InvArg, invExpr, 1);
s.AddRule(invQ);
RealExpr[] lamb = new RealExpr[1];
lamb[0] = ctx.MkRealConst("lamb");
Expr existExpr = ctx.MkAnd(
(BoolExpr)Inv[c,i],
(BoolExpr)phi[ctx.MkAdd(c,lamb[0]),i],
ctx.MkGe(lamb[0], ctx.MkReal(0)));
BoolExpr t= ctx.MkExists(lamb, existExpr, 1);
s.AddRule(ctx.MkImplies(t,(BoolExpr)phi[c,i]));
}
sometimes, there is an error said "AccessViolationException was unhandlered,Attempted to read or write protected memory. This is often an indication that other memory is corrupt." when running to ctx.MkExists()
The fixedpoint solver only supports universal quantifiers at the top-level.
You should rewrite the rule as follows:
s.AddRule(ctx.MkForall(lamb,
ctx.MkImplies((BoolExpr)existExpr,(BoolExpr)phi[c,i])));
Z3 should ideally not result in any access violation. This is typically indicating a bug.
I would really appreciate repros for such bugs when/if you encounter them.

Resources