Is auto differential module is suitable for the fortran-stdlib
dengbangjie opened this issue · 2 comments
Motivation
I am willing to contribute a basic differential programming module using an operator overloading approach (as opposed to a tape-based method). Differential modeling is quite interesting because it simultaneously provides the value of an object and its gradient. From my personal experience, this capability significantly reduces the time required for nonlinear optimization tasks. However, I am unsure whether this domain is appropriate for the Fortran-stdlib library. And, I am unsure whether the library prefers arbitrary-order differential objects [x_i, [dx_i/dx_j], ...] or just dual number calculus [x, dx].
Prior Art
No response
Additional Information
No response
Hi,
This looks like a promising contribution. Do you have a stand-alone repo that showcases the contribution you wish to carry out? that would help the discussion. Some details about implementation, capabilities, limitations, performance would help to push the discussion.
Duplicate of #531. @dengbangjie, you are very welcome to pick this up. I think a forward-mode operator-overloading based auto-diff would be fully in the scope of stdlib.