-- File: BFGS.cdl -- Created: Tue May 14 16:01:21 1991 -- Author: Laurent PAINNOT -- ---Copyright: Matra Datavision 1991, 1992 class BFGS from math ---Purpose: -- This class implements the Broyden-Fletcher-Goldfarb-Shanno variant of -- Davidson-Fletcher-Powell minimization algorithm of a function of -- multiple variables.Knowledge of the function's gradient is required. uses Vector from math, Matrix from math, MultipleVarFunctionWithGradient from math, Status from math, OStream from Standard raises NotDone from StdFail, DimensionError from Standard is Create(F: in out MultipleVarFunctionWithGradient; StartingPoint: Vector; Tolerance: Real=1.0e-8; NbIterations: Integer=200; ZEPS: Real=1.0e-12) ---Purpose: -- Given the starting point StartingPoint, -- the Broyden-Fletcher-Goldfarb-Shanno variant of Davidson-Fletcher-Powell -- minimization is done on the function F. -- The tolerance required on F is given by Tolerance. -- The solution F = Fi is found when : -- 2.0 * abs(Fi - Fi-1) <= Tolerance * (abs(Fi) + abs(Fi-1) + ZEPS). -- The maximum number of iterations allowed is given by NbIterations. returns BFGS; Create(F: in out MultipleVarFunctionWithGradient; Tolerance: Real = 1.0e-8; NbIterations: Integer = 200; ZEPS: Real = 1.0e-12) ---Purpose: Initializes the computation of the minimum of F. -- Warning -- A call to the Perform method must be made after this -- initialization to effectively compute the minimum of the -- function F. returns BFGS; Delete(me:out) is virtual; ---C++: alias "Standard_EXPORT virtual ~math_BFGS(){Delete() ; }" Perform(me: in out; F: in out MultipleVarFunctionWithGradient; StartingPoint: Vector) ---Purpose: Is used internally by the constructors. is static; IsSolutionReached(me; F: in out MultipleVarFunctionWithGradient) ---Purpose: -- This method is called at the end of each iteration to check if the -- solution is found. -- It can be redefined in a sub-class to implement a specific test to -- stop the iterations. returns Boolean is virtual; IsDone(me) ---Purpose: Returns true if the computations are successful, otherwise returns false. ---C++: inline returns Boolean is static; Location(me) ---Purpose: returns the location vector of the minimum. -- Exception NotDone is raised if the minimum was not found. ---C++: inline ---C++: return const& returns Vector raises NotDone is static; Location(me; Loc: out Vector) ---Purpose: outputs the location vector of the minimum in Loc. -- Exception NotDone is raised if the minimum was not found. -- Exception DimensionError is raised if the range of Loc is not -- equal to the range of the StartingPoint. ---C++: inline raises DimensionError, NotDone is static; Minimum(me) ---Purpose: returns the value of the minimum. -- Exception NotDone is raised if the minimum was not found. ---C++: inline returns Real raises NotDone is static; Gradient(me) ---Purpose: Returns the gradient vector at the minimum. -- Exception NotDone is raised if the minimum was not found. ---C++: inline ---C++: return const& returns Vector raises NotDone is static; Gradient(me; Grad: out Vector) ---Purpose: Returns the value of the gradient vector at the minimum in Grad. -- Exception NotDone is raised if the minimum was not found. -- Exception DimensionError is raised if the range of Grad is not -- equal to the range of the StartingPoint. ---C++: inline raises DimensionError, NotDone is static; NbIterations(me) ---Purpose: Returns the number of iterations really done in the -- calculation of the minimum. -- The exception NotDone is raised if the minimum was not found. ---C++: inline returns Integer raises NotDone is static; Dump(me; o: in out OStream) ---Purpose: Prints on the stream o information on the current state -- of the object. -- Is used to redefine the operator <<. is static; fields Done: Boolean; TheStatus: Status is protected; TheLocation: Vector is protected; TheGradient: Vector is protected; PreviousMinimum: Real is protected; TheMinimum: Real is protected; XTol: Real is protected; EPSZ: Real is protected; nbiter: Integer is protected; Itermax: Integer; end BFGS;