BibTeX
@ARTICLE{
         Griewank2003AJa,
       abstract = "The chain rule - fundamental to any kind of analytical differentiation - can be
         applied in various ways to computational graphs representing vector functions. These variants result
         in different operations counts for the calculation of the corresponding Jacobian matrices. The
         minimization of the number of arithmetic operations required for the calculation of the complete
         Jacobian leads to a hard combinatorial optimization problem. We will describe an approach to the
         solution of this problem that builds on the idea of optimizing chained matrix products using dynamic
         programming techniques. Reductions by a factor of 3 and more are possible regarding the operations
         count for the Jacobian accumulation. After discussing the mathematical basics of Automatic
         Differentiation we will show how to compute Jacobians by chained sparse matrix products. These
         matrix chains can be reordered, must be pruned, and are finally subject to a dynamic programming
         algorithm to reduce the number of scalar multiplications performed.",
       ad_theotech = "Code Optimization, General",
       author = "Andreas Griewank and Uwe Naumann",
       title = "Accumulating {J}acobians as Chained Sparse Matrix Products",
       journal = "Math. Prog.",
       number = "95",
       volume = "3",
       pages = "555--571",
       year = "2003",
       doi = "10.1007/s10107-002-0329-7",
       publisher = "Springer"
}