No label defined (Q4028401)

[For the entire collection see Zbl 0747.00030.] Making automatic differentiation easily accessible is greatly facilitated by such things as operator and function overloading and the ability to create user-defined functions. The natural milieu for doing these is an extensible, object-oriented language -- such as C++, Objective-C, or ADA -- in which we can create new types of variables that behave, in all respects, like fully functional elements of the language. We shall briefly introduce two implementations of (forward mode) automatic differentiation written in C++. The first, ``nstd'', a trivial implementation, was hacked in a few days and possesses fundamental limitations arising from storing derivatives in arrays. The second, ``MXYZPTLK'', which uses dynamically allocated lists as its fundamental structure, involved considerably more effort and is, consequently, more practical.