integrate_1d_adjoint()

template<typename F , typename T_a , typename T_b , typename Integrator , typename... Args>

return_type_t< T_a, T_b, Args... > stan::math::internal::integrate_1d_adjoint ( const char *  function, const F &  f, const T_a &  a, const T_b &  b, Integrator &&  integrator, std::ostream *  msgs, const Args &...  args  )

inline

Build the reverse-mode result of a one-dimensional adaptive quadrature.

The quadrature routine is supplied as a callable integrator so the three integrate_1d* variants (plain, double-exponential, Gauss-Kronrod) can share this body, differing only in which Boost routine and tolerances they bind. integrator is invoked as integrator(g) where g is an integrand with signature g(x, xc); it returns the scalar integral of g over [a, b].

The integral value is computed with an all-double integrand. Adjoints are then accumulated into the original var inputs:

• endpoints contribute -f(a) and f(b) (skipped when infinite),
• each var argument component contributes the integral of d f / d arg, obtained with nested reverse-mode autodiff.

Gradients of f that come back NaN at a point where f == 0 are set to zero. Any other NaN propagates into the component integral and is reported as a domain_error naming the (flattened) parameter index.

Template Parameters

F	Type of f
T_a	type of first limit
T_b	type of second limit
Integrator	callable (integrand) -> double
Args	types of parameter pack arguments

Parameters

	function	name of the calling function, used in error messages
	f	the functor to integrate
	a	lower limit of integration
	b	upper limit of integration
	integrator	binds the quadrature routine and its tolerances
[in,out]	msgs	the print stream for warning messages
	args	additional arguments to pass to f

Returns

numeric integral of function f

Definition at line 57 of file integrate_1d_adjoint.hpp.