Class BrentOptimizer


  • public class BrentOptimizer
    extends UnivariateOptimizer
    For a function defined on some interval (lo, hi), this class finds an approximation x to the point at which the function attains its minimum. It implements Richard Brent's algorithm (from his book "Algorithms for Minimization without Derivatives", p. 79) for finding minima of real univariate functions.
    This code is an adaptation, partly based on the Python code from SciPy (module "optimize.py" v0.5); the original algorithm is also modified
    • to use an initial guess provided by the user,
    • to ensure that the best point encountered is the one returned.
    Since:
    2.0
    • Field Detail

      • GOLDEN_SECTION

        private static final double GOLDEN_SECTION
        Golden section.
      • MIN_RELATIVE_TOLERANCE

        private static final double MIN_RELATIVE_TOLERANCE
        Minimum relative tolerance.
      • relativeThreshold

        private final double relativeThreshold
        Relative threshold.
      • absoluteThreshold

        private final double absoluteThreshold
        Absolute threshold.
    • Constructor Detail

      • BrentOptimizer

        public BrentOptimizer​(double rel,
                              double abs,
                              ConvergenceChecker<UnivariatePointValuePair> checker)
        The arguments are used implement the original stopping criterion of Brent's algorithm. abs and rel define a tolerance tol = rel |x| + abs. rel should be no smaller than 2 macheps and preferably not much less than sqrt(macheps), where macheps is the relative machine precision. abs must be positive.
        Parameters:
        rel - Relative threshold.
        abs - Absolute threshold.
        checker - Additional, user-defined, convergence checking procedure.
        Throws:
        NotStrictlyPositiveException - if abs <= 0.
        NumberIsTooSmallException - if rel < 2 * Math.ulp(1d).
      • BrentOptimizer

        public BrentOptimizer​(double rel,
                              double abs)
        The arguments are used for implementing the original stopping criterion of Brent's algorithm. abs and rel define a tolerance tol = rel |x| + abs. rel should be no smaller than 2 macheps and preferably not much less than sqrt(macheps), where macheps is the relative machine precision. abs must be positive.
        Parameters:
        rel - Relative threshold.
        abs - Absolute threshold.
        Throws:
        NotStrictlyPositiveException - if abs <= 0.
        NumberIsTooSmallException - if rel < 2 * Math.ulp(1d).