Computer Scientist (on leave)

Jean Utke

Jean Utke oversees the overall direction of the OpenAD design, determining goals and objectives, deciding how the components should be integrated, and ensuring that continuous regression testing is carried out to verify numerical correctness. He has also applied automatic differentiation to the Community Land Model, as well as on scalable solvers for sheet modeling.

Utke has devoted considerable effort to the design of OpenAD, the development of algorithms for OpenAD, their implementation, and the tool’s use in practical applications. He was one of the key developers of ADOL-C, at the time perhaps the only algorithmic differentiation tool to implement both forward and reverse modes for first- and higher-order derivatives. He also encapsulated a higher-order derivative capability in the Rhapsodia overloading library. Moreover, he also has made significant advances in the theory of AD. Deserving of mention in this regard is his work on modularity (the Open64/SL front end, enabling differentiation of multilanguage codes in Fortran/C/C++, with possible expansion to GPU-specific source code); data augmentation for numerical computation; and control flow reversal for adjoints.

Utke received his Dr. rer. Nat. (German equivalent of Ph.D.) in mathematics and scientific computing from Technical University Dresden in 1996. After a brief stay at Argonne as a visiting scholar, he worked as a consultant in industry for six years, where he applied his technical skills to such tasks as analysis and system testing, cellular communications, software design of insurance environments, and migration of a large-scale distributed C++ software system to a multithreaded architecture. In 2003, Jean joined the University of Chicago and Argonne National Laboratory as a research associate. From 2006 to 2012, he held a joint appointment as a research scientist/computer scientist.

Research Interests

  • Algorithmic (automated) differentiation (AD)
  • Software tool development
  • Theory of AD
  • Application of AD tools in real-life problems, such as ocean modeling and materials science