hi yaron and markus,

thanks for the info! i have tried out the code in the old blog post and
for my simple needs it seems to work ok at the moment -- although it is
twice slower than pre-computed explicit derivatives.

n.