mathematics
and
computer
science

LANS

LANS Research

Sensitivity of flow throu

OpenAD/F: Automatic Differentiation of Fortran Codes

PIs:
Paul Hovland, Boyana Norris, Jean Utke

Additional People:
Patrick Heimbach , Chris Hill , Carl Wunsch

Other Institutes:
MIT

Funding:
NSF , NASA

[project website]

The OpenAD/F project seeks to develop a modular, open-source tool for the automatic generation of adjoint code from Fortran 95 source code. Discrete adjoint computations are used for sensitivity analysis and to provide the gradients used in geophysical state estimation. Because derivatives are needed with respect to millions or billions of independent variables, finite different approximations are impractical: a gradient computation that would take minutes or hours using an adjoint computation would take months or years using finite differences.