I published a paper on arXiv on hyper-dual numbers where I explain how
to use the above code to compute the first and second order derivatives
of Matlab functions by means of automatic differentiation. Notice that
the derivatives are computed in tuple forward mode. More efficient in
terms of the number of function calls would be a mixed forward-backward
call, i.e.: the gradient is evaluated by adjoint differentiation, and
the Hessian is computed from the gradient in tangent-linear mode. An
adjoint number class may follow in the future and will then be
available for download on this page.
For students: I am happy for contributions or expresses of interest or
students who search a thesis topic related to numerical optimization,
optimal control and automatic differentiation. I can suggest lots of
promising topics in these directions that could substantially support
my research.
Contact via e-mail: MartinNeuenhofen@googlemail.com