Code Differentiation for Hydrodynamic Model Optimization
Use of a hydrodynamics code for experimental data fitting purposes (an optimization problem) requires information about how a computed result changes when the model parameters change. These so-called sensitivities provide the gradient that determines the search direction for modifying the parameters to find an optimal result. Here, the authors apply code-based automatic differentiation (AD) techniques applied in the forward and adjoint modes to two problems with 12 parameters to obtain these gradients and compare the computational efficiency and accuracy of the various methods. They fit the pressure trace from a one-dimensional flyer-plate experiment and examine the accuracy for a two-dimensional jet-formation problem. For the flyer-plate experiment, the adjoint mode requires similar or less computer time than the forward methods. Additional parameters will not change the adjoint mode run time appreciably, which is a distinct advantage for this method. Obtaining ''accurate'' sensitivities for the j et problem parameters remains problematic.
- Research Organization:
- Los Alamos National Lab. (LANL), Los Alamos, NM (United States)
- Sponsoring Organization:
- USDOE Office of Defense Programs (DP) (US)
- DOE Contract Number:
- W-7405-ENG-36
- OSTI ID:
- 759173
- Report Number(s):
- LA-UR-99-3075; TRN: US0004312
- Resource Relation:
- Conference: American Physical Society Conference on Shock Compression of Condensed Matter, Snowbird, UT (US), 06/27/1999--07/02/1999; Other Information: PBD: 27 Jun 1999
- Country of Publication:
- United States
- Language:
- English
Similar Records
Differential sensitivity theory applied to the MESA code
Differential Sensitivity Theory applied to the MESA code for high pressure interactions