scotty.derivatives module#

scotty.derivatives.STENCILS: Dict[str, Dict[Tuple, float]] = {'d1_CFD2': {(-1,): -0.5, (1,): 0.5}, 'd1_CFD4': {(-2,): 0.08333333333333333, (-1,): -0.6666666666666666, (1,): 0.6666666666666666, (2,): -0.08333333333333333}, 'd1_FFD2': {(0,): -1.5, (1,): 2, (2,): -0.5}, 'd1d1_CFD_CFD2': {(-1, -1): 0.25, (-1, 1): -0.25, (1, -1): -0.25, (1, 1): 0.25}, 'd1d1_FFD_CFD2': {(0, -1): 0.25, (0, 0): 1, (0, 1): -1.25, (1, 0): -2, (1, 1): 2, (2, -1): -0.25, (2, 0): 1, (2, 1): -0.75}, 'd1d1_FFD_FFD2': {(0, 0): 2.25, (0, 1): -3, (0, 2): 0.75, (1, 0): -3, (1, 1): 4, (1, 2): -1, (2, 0): 0.75, (2, 1): -1, (2, 2): 0.25}, 'd2_CFD2': {(-1,): 1, (0,): -2, (1,): 1}, 'd2_FFD2': {(0,): 2, (1,): -5, (2,): 4, (3,): -1}}#

Finite difference stencils for derivative.

The second-derivative stencils are optimised for minimal function evaluations.

The naming scheme here is <order>_<kind><error> with:

  • <order> is d<n> for the ``n``th derivative, and this is repeated for multiple dimensions

  • <kind> is either FFD for forward finite difference or CFD for central finite difference, again repeated for each dimension

  • <error> is the order of the error scaling

So d1d1_FFD_CFD2 is the first mixed-derivative of two dimensions, using forward differences for the first dimension and central for the second, with second-order error overall

scotty.derivatives.cache(func)#

Simple caching of a function compatible with numpy array arguments

scotty.derivatives.derivative(func, dims, args, spacings=1e-08, stencil=None, use_cache=True)#

Partial derivative of a function along one or more of its arguments.

Currently this can take partial derivatives in one or two arguments, given by dims and evaluated at args. The arguments to func must be exactly the keys of args, and func must be able to take them as keyword arguments.

The function will be evaluated at points around args given by the integer offsets in the stencil multiplied by the step size in each dimension given in spacings.

By default, calls to func with a given set of arguments will be cached, which enables re-use of evaluation points across multiple derivatives. This is particularly useful when needing to take partial derivatives in many directions. However, if func has side effects, such as printing variables, these will not work, and you should set use_cache = False.

Parameters

  • func (Callable) – Function to take the derivative of

  • dims (Union[str, Tuple[str, ...]]) – The name(s) of the dimension(s) to take the derivative along. dims must appear in the keys of args. For second derivatives in one dimension, pass a tuple with the argument name repeated, for example ("x", "x") for the second derivative in x

  • args (Dict[str, Union[float, numpy.ndarray[Any, numpy.dtype[numpy.float64]]]]) – Arguments to func at the point to evaluate the derivative

  • spacings (Union[float, Dict[str, float]]) – Step size for derivative. If a single float, used for all dimensions, otherwise if a mapping of dimension names to step size then the keys must be identical to those of dims

  • stencil (Optional[str]) – Stencil name (see STENCILS for supported stencils). Defaults to central differences

  • use_cache (bool) – If true, then wraps func with the decorator cache

Returns

Derivative of func evaluated at location

Return type

ArrayLike

Examples

>>> derivative(lambda x: np.sin(x), "x", {"x": 0.0})
1.0

>>> derivative(lambda x, y: np.sin(x) * y**2, ("x", "y"), {"x": 0.0, "y": 2.0})
4.0