Publication: Adjoint code development and optimization using automatic differentiation
Introduction
Applications
Tools
Research Groups
Workshops
Publications
   List Publications
   Advanced Search
   Info
   Add Publications
My Account
About

Adjoint code development and optimization using automatic differentiation

- Technical report -
 

Area
Computational Fluid Dynamics, Engineering, Optimization

Author(s)
C. Praveen

Institution
India National Aerospace Laboratories

Year
2006

Publisher
NAL

Abstract
Adjoint code for 1-D and 2-D Euler equations are developed using automatic differentiation tool called TAPENADE. A piecemeal approach is used in which the subroutines in the fow solver are differentiated individually and used in an adjoint iterative solver. This approach is useful for problems requiring iterative solution procedures since it leads to enormous savings in memory and time. For 2-D case, the adjoint solver requires about 38% more memory compared to the flow solver. The time per adjoint iteration is about twice that of the flow solver. The adjoint code is used to solve pressure matching problem for quasi 1-D flow through a duct. A smoothing procedure based on an elliptic equation is developed for this purpose. In 2-D, a second order vertex-centroid scheme on triangular grids is used to develop an adjoint solver. Both the flow and adjoint solvers are accelerated using LUSGS scheme with spectral radius approximation for flux jacobians. The adjoint code is validated by computing the slope of the Cl-alpha curve.

AD Tools
TAPENADE

AD Theory and Techniques
Adjoint, Code Optimization, Implementation Strategies, Iteration, Reverse Mode

Related Applications
- Adjoint CFD Solver Development Using TAPENADE

BibTeX
@TECHREPORT{
         Praveen2006Acd,
       title = "Adjoint code development and optimization using automatic differentiation",
       author = "C. Praveen",
       publisher = "NAL",
       year = "2006",
       number = "PD CF 0604",
       institution = "India National Aerospace Laboratories",
       abstract = "Adjoint code for 1-D and 2-D Euler equations are developed using automatic
         differentiation tool called Tapenade. A piecemeal approach is used in which the subroutines in the
         fow solver are differentiated individually and used in an adjoint iterative solver. This approach is
         useful for problems requiring iterative solution procedures since it leads to enormous savings in
         memory and time. For 2-D case, the adjoint solver requires about 38% more memory compared to the
         flow solver. The time per adjoint iteration is about twice that of the flow solver. The adjoint code
         is used to solve pressure matching problem for quasi 1-D flow through a duct. A smoothing procedure
         based on an elliptic equation is developed for this purpose. In 2-D, a second order vertex-centroid
         scheme on triangular grids is used to develop an adjoint solver. Both the flow and adjoint solvers
         are accelerated using LUSGS scheme with spectral radius approximation for flux jacobians. The
         adjoint code is validated by computing the slope of the Cl-alpha curve.",
       ad_area = "Computational Fluid Dynamics, Engineering, Optimization",
       ad_tools = "TAPENADE",
       ad_theotech = "Adjoint, Code Optimization, Implementation Strategies, Iteration, Reverse
         Mode"
}


back
  

LinkedIn:    Contact:
autodiff.org
Username:
Password:
(lost password)