JaCoP package org.jacop.floats offers also experimental extensions. They include a method to generate constraints that define the derivative of a function, represented by constraints, and a interval Newton method.

JaCoP can do symbolic (partial) derivatives of functions, that is using the defined constraints for a given function and a variable defining the result of the function, JaCoP will generate constraints that define a function for a derivative of the original function. There is a complication in this process since constraints are not functions and sometime it is difficult or impossible to find out which constraint defines a function for a given variable. In such situations JaCoP uses a heuristic to find out this but it is also possible to define this explicitly. The following examples explains the process of computing derivatives of constrains.

Set<FloatVar> vars = new HashSet<FloatVar>();

vars.add(x1);

vars.add(x2);

Derivative.init(store);

// Derivative.defineConstraint(x1x1, c0);

// Derivative.defineConstraint(x2x2, c1);

// Derivative.defineConstraint(x1x1x1x1, c2);

// Derivative.defineConstraint(x2x2x2x2, c3);

// ================== df/dx1 =================

FloatVar fx1 = Derivative.getDerivative(store, f, vars, x1);

// ================== df/dx2 =================

FloatVar fx2 = Derivative.getDerivative(store, f, vars, x2);

vars.add(x1);

vars.add(x2);

Derivative.init(store);

// Derivative.defineConstraint(x1x1, c0);

// Derivative.defineConstraint(x2x2, c1);

// Derivative.defineConstraint(x1x1x1x1, c2);

// Derivative.defineConstraint(x2x2x2x2, c3);

// ================== df/dx1 =================

FloatVar fx1 = Derivative.getDerivative(store, f, vars, x1);

// ================== df/dx2 =================

FloatVar fx2 = Derivative.getDerivative(store, f, vars, x2);

In the above example, JaCoP will generate two partial derivatives and . Set vars is used to keep all variables of function defined by variable f. The function defined by variable f is usually defined by several constraints using temporary variables, such as x1x1, x2x2, x1x1x1x1 and x2x2x2x2. In such situations JaCoP tries, using heuristics, to find out what constraint defines a given temporary variable. Sometimes it is not possible and the programmer needs to define it, as indicated by the commented code and methods Derivative.defineConstraint. For example, constraint c0 defines variable x1x1. Method getDerivative of class Derivative generates a variable that defines the result of the derivative.

Derivatives, can be used for setting additional constraints on solutions. For example, when looking for minimum or maximum of a function or in the interval Newton method.

Interval Newton methods combine classical Newton method with interval analysis [9] and are used to provide better bounds on variables.

Traditionally, univariate Newton method computes a solution for equation f(x) = 0 using the following approximation.

This can be extended by using vector x and a Jacobian matrix instead of f′(x). Furthermore one can extend it to interval arithmetic. This is implemented in JaCoP using “constraint” EquationSystem. This constraint takes as parameters a vector of functions and variables of these functions. The constraints computes then all partial derivatives and form a set of equations that are later solved, if possible, by interval Gauss-Seidel method.

This can be formalized as system of linear equations that, when solved, provides bounds
on values (x_{n+1} - x_{n}) of variables of equations.

Below, constraint eqs specifies a set of two functions f0 and f1 of two variables x1 and x2. Constraint eqs will try to prune domains of variables x1 and x2.