Re:Automatic Differentiation via Contour Integration
by conistonwater on Saturday January 18, 2020 @10:10AM
(#22084800)
This is a standard technique in numerical analysis for complex-differentiable functions, it's just not so common to have inputs be guaranteed to be holomorphic so people don't use it often, http://mpmath.org/doc/current/calculus/differentiation.html#... (https://github.com/fredrik-johansson/mpmath/blob/981102736a8...)
[ Parent
| on HN ]
Re:Automatic Differentiation via Contour Integration
by kkylin on Saturday January 18, 2020 @10:59AM
(#22085072)
Indeed a standard technique in numerical analysis. Among the advantages of doing things this way (when applicable, i.e., when one has holomorphic integrands) are significantly better accuracy & stability: https://people.maths.ox.ac.uk/trefethen/sirev56-3_385.pdf
(Haven't looked closely at the posted code so no idea how directly relevant this is.)