ADOL-C

A Package for Automatic Differentiation

of Algorithms Written in C/C++

Current versions:

For UNIX / LINUX:

For Windows / DOS:

ADOL-C 1.9.0 (May 2004)
ADOL-C 1.9.0 (Please contact us for detailed information!)

go to: Synopsis - Functionality - Availability - Sources and Articles - Applications - Contact

Synopsis

The package ADOL-C facilitates the evaluation of first and higher derivatives of vector functions that are defined by computer programs written in C or C++. The resulting derivative evaluation routines may be called from C/C++, Fortran, or any other language that can be linked with C.

The numerical values of derivative vectors are obtained free of truncation errors at a small multiple of the run time and randomly accessed memory of the given function evaluation program.

Functionality

ADOL-C uses the operator overloading concept to compute in forward and reverse mode of automatic differentiation:

derivatives of any order
and one-sided derivatives in non-smooth cases (e.g. evaluation of fabs)

for a function that is given as a C/C++ code. The user has to make a few modifications to the evaluation code:

redeclare type of all variables that are involved in the evaluation code (active variables) to the new type adouble,
mark the evaluation section,
specify independent and dependent variables and pass their values,
recompile the code and link the ADOL-C library.

ADOL-C provides:

convenient drivers for common differentiation tasks,
application oriented drivers, e.g.

for the computation of sparsity patterns
for ordinary differential equations
for the calculation of full higher-derivative tensors

active vector and matrix classes.

The operator overloading is used to produce the tape, an internal representation of the evaluation section marked. For each operation on the tape

truncated Taylor series are propagated in the forward mode
the corresponding adjoint operations are performed on the adjoint variables in the reverse mode.

Instead of calling the routines for the forward and reverse mode directly the user may apply appropriate drivers to get the desired derivatives for a scalar-valued function

. Some of these covenient drivers are listed in the following table:

function(tag,m,n,x,y)
gradient(tag,n,x,g)
hessian(tag,n,x,H)	$H =\,$ lower triangle of
hess_vec(tag,n,x,v,z)
lagra_hess_vec(tag,m,n,x,v,u,h)

For vector-valued functions ADOL-C provides also easy to use drivers. Again some of them are listed below:

function(tag,m,n,x,y)
jacobian(tag,m,n,x,J)	$f^\prime(x) = J(x) = \left[\frac{\partial f_i}{\partial x_j}\right] \left. \right\vert _x$
jac_vec(tag,m,n,x,v,j)	$j = J(x)\,v$
vec_jac(tag,m,n,repeat,x,u,j)	$j = u^T\,J(x)$

Drivers for ODE's and routines to get full higher order derivative tensors are available. Storage and runtime requirements are clearly predictable. The tape is transferred to external mass storage devices and only strictly sequentially accessed. The much smaller randomly accessed memory can be precalculated using information on the tape.

Availability

ADOL-C is available free of charge provided under the terms of the Common Public License. Check the ftp sites for the source, the manual and examples. ADOL-C was tested on several platforms including Sun, LinuX, and SGI.

Source and Articles

Software:

Version for UNIX / LINUX:

Documentation:

User's guide
Short reference (english or german)

Recent Papers:

K. Röbenack:

N. Arora and L. T. Biegler:

G. Haase, U. Langer, E. Lindner, and W. Mühlhuber:

Corliss et. al.,

M. J. Huiskes:

W. Klein, A. Griewank, A. Walther:

K. Röbenack, K. J. Reinschke:

Olaf Vogel :

A. Griewank.

Evaluating Derivatives, Principles and Techniques of Algorithmic Differentiation

A. Griewank, D. Juedes, H. Mitev, J. Utke, O. Vogel, and A. Walther:

ADOL-C: A Package for the Automatic Differentiation of Algorithms Written in C/C++

this is the updated version of the paper published in

A. Griewank, J. Utke, and A. Walther

Evaluating higher derivative tensors by forward propagation of univariate Taylor series,

T. Coleman and G. Jonsson,

The Efficient Computation of Structured Gradients using Automatic Differentiation

Introduction to AD: Short Course at SIAM Optimization Conference in Toronto

Applications

DYNOPC, Department of Chemical Engineering, Carnegie Mellon University

   (Solution of dynamic optimization problems)

MUSCOD-II , Interdisciplinary Center for Scientific Computing (IWR), University of Heidelberg

   (Numerical Solution of Optimal Control Problems involving Differential-Algebraic Equations )

NEOS Server , Argonne National Laboratory

   (Network Enabled Optimization System for solving optimization problems remotely over the Internet)

Omuses, Institut für Automatisierungs- und Systemtechnik, TU Ilmenau

   (Optimization of Multistage Systems)

Whom to Contact

Andrea Walther
Institut für Wissenschaftliches Rechnen
Technische Universität Dresden
Mommsenstr. 13
01062 Dresden, Germany
phone: +49 (0)351 463 35018
FAX: +49 (0)351 463 37096
email

Andreas Griewank
Humboldt-Universität zu Berlin
Institut für Mathematik
Unter den Linden 6
D-10099 Berlin
phone: +49 (0)30 2093 - 5820
FAX: +49 (0)30 2093 - 8959
email

Jean Utke
Mathematics and Computer Science
Argonne National Laboratory
9700 S Cass Ave
Argonne, IL
60439-4844, USA
phone: +1 630 252 4552
FAX: +1 630 252 5986
email

A comprehensive collection of AD related news, papers and tools can be found at www.autodiff.org

email the page owner
(May/10/2004)