Optimization methods

The available optimization methods are listed in the table below.

Table 18.1. Available methods

Method Description Function type Constraints
BCBarrierNewtonMethod - bound constrained barrier Newton method The Newton method in which linear constraints are modeled by a logarithmic barrier NLF.NLF2required
BCBarrierQuasiNewtonMethod - bound constrained barrier quasi Newton method The quasi Newton method in which linear constraints are modeled by a logarithmic barrier NLF.NLF1required
BCEllipsoidMethod - bound constrained ellipsoid method An algorithm with polynomial complexity, an additional parameter - the radius of the initial ellipsoid can be set with the setInitialEllipsoid(double radius) method NLF.NLF1required
BCFDNewtonMethod - bound constrained finite differences Newton method A modification of the finite difference Newton method in which constrains are added NLF.NLF1required
BCNewtonMethod - bound constrained Newton method A modification of the Newton method in which constraints are added NLF.NLF2required
BCQuasiNewtonMethod - bound constrained quasi Newton method A modification of the quasi Newton method in which constraints are added NLF.NLF1required
ConjugateGradientMethod - conjugate gradient method The conjugate gradient method has relative small memory complexity (which is linear in the domain dimension as compared to square complexity of other algorithms). Low memory requirment. It is used for for symmetric and positive defined functions. NLF.NLF1not required
FDNewtonMethod - finite difference Newton method A modification of the Newton-Raphson method in which the Hessian is approximated by finite differences. Very important is the proper setting of the initial point. Fast first-order derivatives calculation is also required NLF.NLF1not required
NewtonMethod - classical Newton-Raphson method An efficient, classical Newton-Raphson algorithm based on the local approximation of the function with a polynomial of order two (Taylor series expansion). Very important is the proper setting of the initial point NLF.NLF2not required
QuasiNewtonMethod - quasi Newton method A modification of the Newton-Raphson method in which the Hessian is approximated by finite differences of the first order derivatives at the consecutive points. Very important is the proper setting of the initial point. It is sometimes slower than FDNewtonMethod due to approximation errors. NLF.NLF1not required