complex_step.py

"""Complex Step derivative approximations."""

from collections import defaultdict

import numpy as np

from openmdao.utils.om_warnings import issue_warning, DerivativesWarning
from openmdao.approximation_schemes.approximation_scheme import ApproximationScheme


class ComplexStep(ApproximationScheme):
    r"""
    Approximation scheme using complex step to calculate derivatives.

    For example, using  a step size of 'h' will approximate the derivative in
    the following way:

    .. math::

        f'(x) = \Im{\frac{f(x+ih)}{h}}.

    Attributes
    ----------
    _fd : <FiniteDifference>
        When nested complex step is detected, we switch to Finite Difference.
    """

    DEFAULT_OPTIONS = {
        'step': 1e-40,
        'directional': False,
    }

    def __init__(self):
        """
        Initialize the ApproximationScheme.
        """
        super().__init__()

        # Only used when nested under complex step.
        self._fd = None

    def add_approximation(self, abs_key, system, kwargs, vector=None):
        """
        Use this approximation scheme to approximate the derivative d(of)/d(wrt).

        Parameters
        ----------
        abs_key : tuple(str,str)
            Absolute name pairing of (of, wrt) for the derivative.
        system : System
            Containing System.
        kwargs : dict
            Additional keyword arguments, to be interpreted by sub-classes.
        vector : ndarray or None
            Direction for difference when using directional derivatives.
        """
        options = self.DEFAULT_OPTIONS.copy()
        options.update(kwargs)
        options['vector'] = vector

        wrt = abs_key[1]
        if wrt in self._wrt_meta:
            self._wrt_meta[wrt].update(options)
        else:
            self._wrt_meta[wrt] = options
        self._reset()  # force later regen of approx_groups

    def _get_approx_data(self, system, wrt, meta):
        """
        Given approximation metadata, compute necessary delta for complex step.

        Parameters
        ----------
        system : System
            System whose derivatives are being approximated.
        wrt : str
            Name of wrt variable.
        meta : dict
            Metadata dict.

        Returns
        -------
        float
            Delta needed for complex step perturbation.
        """
        step = meta['step']
        step *= 1j
        return step

    def compute_approx_col_iter(self, system, under_cs=False):
        """
        Execute the system to compute the approximate sub-Jacobians.

        Parameters
        ----------
        system : System
            System on which the execution is run.
        under_cs : bool
            True if we're currently under complex step at a higher level.

        Yields
        ------
        int
            column index
        ndarray
            solution array corresponding to the jacobian column at the given column index
        """
        if not self._wrt_meta:
            return

        if system.under_complex_step:

            # If we are nested under another complex step, then warn and swap to FD.
            if not self._fd:
                from openmdao.approximation_schemes.finite_difference import FiniteDifference

                issue_warning("Nested complex step detected. Finite difference will be used.",
                              prefix=system.pathname, category=DerivativesWarning)

                fd = self._fd = FiniteDifference()
                empty = {}
                for wrt in self._wrt_meta:
                    fd.add_approximation(wrt, system, empty)

            yield from self._fd.compute_approx_col_iter(system)
            return

        saved_inputs = system._inputs._get_data().copy()
        system._inputs._data.imag[:] = 0.0
        saved_outputs = system._outputs.asarray(copy=True)
        system._outputs._data.imag[:] = 0.0
        saved_resids = system._residuals.asarray(copy=True)
        system._residuals._data.imag[:] = 0.0

        # Turn on complex step.
        system._set_complex_step_mode(True)

        try:
            for tup in self._compute_approx_col_iter(system, under_cs=True):
                yield tup
                # this was needed after adding relevance to the NL solve in order to clean
                # out old results left over in the output array from a previous solve.
                system._outputs.set_val(saved_outputs)
        finally:
            # Turn off complex step.
            system._set_complex_step_mode(False)

        system._inputs.set_val(saved_inputs)
        system._outputs.set_val(saved_outputs)
        system._residuals.set_val(saved_resids)

    def _get_multiplier(self, delta):
        """
        Return a multiplier to be applied to the jacobian.

        Parameters
        ----------
        delta : complex
            Complex number used to compute the multiplier.

        Returns
        -------
        float
            multiplier to apply to the jacobian.
        """
        return (1.0 / delta * 1j).real

    def _transform_result(self, array):
        """
        Return the imaginary part of the given array.

        Parameters
        ----------
        array : ndarray of complex
            Result array after doing a complex step.

        Returns
        -------
        ndarray
            Imaginary part of the result array.
        """
        return array.imag

    def _run_point(self, system, idx_info, delta, result_array, total, idx_start=0):
        """
        Perturb the system inputs with a complex step, run, and return the results.

        Parameters
        ----------
        system : System
            The system having its derivs approximated.
        idx_info : tuple of (Vector, ndarray of int)
            Tuple of wrt indices and corresponding data vector to perturb.
        delta : complex
            Perturbation amount.
        result_array : ndarray
            An array used to store the results.
        total : bool
            If True total derivatives are being approximated, else partials.
        idx_start : int
            Vector index of the first element of this wrt variable.

        Returns
        -------
        ndarray
            Copy of the outputs or residuals array after running the perturbed system.
        """
        for vec, idxs in idx_info:
            if vec is not None and idxs is not None:
                vec.iadd(delta, idxs)

        if total:
            system.run_solve_nonlinear()
            result_array[:] = system._outputs.asarray()
        else:
            system.run_apply_nonlinear()
            result_array[:] = system._residuals.asarray()

        for vec, idxs in idx_info:
            if vec is not None and idxs is not None:
                vec.isub(delta, idxs)

        return result_array

    def apply_directional(self, data, direction):
        """
        Apply stepsize to direction and embed into approximation data.

        Parameters
        ----------
        data : float
            Step size for complex step.
        direction : ndarray
            Vector containing derivative direction.

        Returns
        -------
        ndarray
            New step direction.
        """
        return data * direction