pynucastro.autodiff.derivative_pair module#
Two independent forward-mode derivative components, stored without arrays.
A DerivativePair is the derivative payload of a Dual, not a value/derivative
pair itself. For independent variables x and y, seed x with
Dual(x, DerivativePair(1.0, 0.0)) and y with
Dual(y, DerivativePair(0.0, 1.0)). The result’s deriv.d1 and
deriv.d2 then hold derivatives with respect to x and y. Dual implements
the chain and product rules; this class adds, subtracts, and scales their
derivative components independently.
The arithmetic is written twice to support Python and Numba. Python uses the
class’s special methods (such as __add__). Numba represents instances as
native namedtuples and needs explicit @overload registrations for these
operations instead of relying on the Python methods. Functions prefixed with
_ol_ receive Numba argument types during compilation and return an
implementation that operates on values at runtime. They are not called by
users. Keep their formulas consistent with the Python methods.