Christian H. Bischof, George F. Corliss, Larry Green, Andreas Griewank, Ken Haigler, and Perry Newman. Automatic differentiation of advanced CFD codes for multidisciplinary design. Journal on Computing Systems in Engineering, 3:625--638, 1992.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....method Consistent, efficient, and versatile SA algorithm Not automatic Application of these techniques to advanced CFD codes has been proposed and discussed in [1] 5] In [1] and [2] the incremental iterative method (IIM) for calculation of sensitivity derivatives (SD s) is discussed. In [3], the application of automatic differentiation (AD) to obtain SD s from a 3 D thin layer Navier Stokes code is demonstrated. References [4] and [5] are recent summaries of the studies of this work; the combination of AD and IIM to efficiently obtain consistent discrete SD s from a ....

C. Bischof, G. Corliss, L. Green, A. Griewank, K. Haigler, and P. Newman, Automatic differentiation of advanced CFD codes for multidisciplinary design, Computing Systems in Engineering, 3 (6), (1992), pp. 625--637.

....of the adjoint seems to have taken almost a decade and has started before AD tools were well enough developed to tackle this challenge. Code for evaluation of second order derivatives, as a consequence of its even larger degree of complexity, has not been hand written for large scale applications [4]. Recently a number of AD tools are being developed that are capable of generating adjoint code (Odyssee [15] GRESS [12] TAMC [7] see also other contributions to this document) Other tools operating in reverse mode are employing operator overloading capabilities of C or Fortran 90 ....

Christian H. Bischof, George F. Corliss, Larry Green, Andreas Griewank, Ken Haigler, and Perry Newman. Automatic differentiation of advanced CFD codes for multidisciplinary design. Journal on Computing Systems in Engineering, 3:625--638, 1992.

....Rice or direct collaborations with the developers of the system. Even with such a limited distribution, it was successfully employed in many different areas of science and engineering, including aeronautical multidisciplinary design optimization [2, 31] aeronautical computational fluid dynamics [6, 7, 15, 19, 25, 26], weather modeling [12, 14, 28, 29] groundwater contaminant transport [11, 32] aquifer modeling [17, 21] structural engineering [16] statistics [13] mechanical system design [20] power networks [23] reaction modeling [27] and large scale numerical optimization [1, 8, 30] In July 1995, ....

....the product of the JacobJan by a vector. The seed matrix mechanism allows for flexible use of the code generated by ADIFORTWO. For example, it can be employed to compute compressed versions of large sparse Jacobians [1] to chain derivatives generated by programs running on different platforms [6, 15], or to decrease turnaround time for derivative computations through a parallel stripmining approach [7] The benefit from proper initialization of the seed matrix is substantial, since the cost of derivative computation is more or less proportional to the number p of directional derivatives ....

C. Bischof, G. Corliss, L. Green, A. Griewank, K. Haigler, and P. Newman, Automatic differentiation of advanced CFD codes for multidisciplinary design, Journal on Computing Systems in Engineering, 3 (1992), pp. 625-638.

....or the complexity of the program. Hence, general techniques that rely on the output of computer simulation models, such as optimal design and Automatic Differentiation of FCAP 2 sensitivity or reliability analysis, can all benefit from using automatic differentiation. See, for example, the work in [4, 6, 8]. In this paper, we describe how we applied our automatic differentiation tool, called ADIC (Automatic Differentiation in C) to generate derivative codes for FCAP2 and FCAP3 programs. We also present runtime performance results of the derivative code for FCAP3. The results show that ....

C. Bischof, G. Corliss, L. Green, A. Griewank, K. Haigler, and P. Newman, "Automatic Differentiation of Advanced CFD Codes for Multidisciplinary Design," Journal on Computing Systems in Engineering, 3(5):625-638, 1993.

....x 0 = x 0 (t) provided that the spectral radius of Phi(x; t) x is less than one in the vicinity of (x ; t) This fundamental result has removed some serious doubts regarding the general applicability of automatic differentiation. It has been verified on several large codes, including cases [3] where the assumptions of Gilbert s theorem do not appear to be satisfied. Therefore, we wish to relax the hypothesis and avoid derivatives that are not needed either from a theoretical or from a practical point of view. We will also establish rates of convergence, provide a practical stopping ....

....argue that if a P k with contractive D k = I Gamma P k F x is known, the linear system F x x 0 = GammaF t can be solved iteratively after the solution x has been computed with satisfactory accuracy. This approach has long been used by engineers, as evidenced for example in some references of [3]. It certainly may be advantageous to start the derivative recurrences (7) or (10) with an initial x 0 k = 0 only when the underlying iteration has reached the vicinity of the solution point. If for some weight vector w, one actually wants to calculate the adjoint sensitivity w T x 0 k = ....

Christian Bischof, George Corliss, Larry Green, Andreas Griewank, K. Haigler, and Perry Newman. Automatic differentiation of advanced CFD codes for multidisciplinary design. Preprint MCS--P339-- 1192, Mathematics and Computer Science Division, Argonne National Laboratory, Argonne, Ill., January 1993.

No context found.

Bischof, C., Corliss, G., Green, L., Griewank, A., Haigler, K., and Newman, P., "Automatic Differentiation of Advanced CFD Codes for Multidisciplinary Design," Journal on Computing Systems in Engineering, Vol.3, No. 6, 1993, pp. 625--637.

....code that allows the computation of the derivatives of the dependent variables with respect to the independent ones. 1 2. Interface Contraction. Automatic differentiation tools such as ADIFOR produce derivative code that typically outperforms divided difference approximations (see, for example, [1, 4, 7, 9, 23]) but, not surprisingly, is usually much less efficient than optimal hand derived code. We describe a technique using what we call interface contraction, that can reduce dramatically the runtime and storage requirements for computing derivatives via automatic differentiation. This technique ....

C. Bischof, G. Corliss, L. Green, A. Griewank, K. Haigler, and P. Newman, Automatic differentiation of advanced CFD codes for multidisciplinary design, Journal on Computing Systems in Engineering, 3 (1992), pp. 625--638.

....development time efficacy of AD surpass those of hand coding approaches. For example, the AD tool ADIFOR (Automatic DIfferentiation of FORtran) 4, 6, 7] has been used to generate derivative codes for many applications in areas such as large scale optimization [2, 13] computational fluid dynamics [8, 9, 14], weather modeling [22] and groundwater modeling [11] Other examples of available AD tools are Odyss ee [23] and GRESS [20] for Fortran, and ADOL C [18] and ADIC [10] for C programs. The forward and reverse modes are the two basic modes of AD, and are distinguished by the manner in which the ....

Christian Bischof, George Corliss, Larry Green, Andreas Griewank, Kara Haigler, and Perry Newman. Automatic differentiation of advanced CFD codes for multidisciplinary design. Journal on Computing Systems in Engineering, 3(6):625--638, 1992.

....77 code that allows the computation of the derivatives of the dependent variables with respect to the independent ones. 3 Interface Contraction Automatic differentiation tools such as ADIFOR produce derivative code that typically outperforms divided difference approximations (see, for example, [1, 3, 5, 6, 18]) but, not surprisingly, is usually much less efficient than a hand derived code probably could be. We introduce a technique, called interface contraction, that can dramatically reduce the runtime and storage requirements for computing derivatives via automatic differentiation. This ....

Christian Bischof, George Corliss, Larry Green, Andreas Griewank, Kara Haigler, and Perry Newman. Automatic differentiation of advanced CFD codes for multidisciplinary design. Journal on Computing Systems in Engineering, 3(6):625--638, 1992.

....development time efficacy of AD surpass those of hand coding approaches. For example, the AD tool ADIFOR (Automatic DIfferentiation of FORtran) 5, 7, 8] has been used to generate derivative codes for many applications in areas such as large scale optimization [2, 14] computational fluid dynamics [9, 10, 15], weather modeling [26] and groundwater modeling [11] Other examples of available AD tools are Odyss ee [27] and GRESS [23] for Fortran, and ADOL C [21] and ADIC [13] for C programs. The forward and reverse modes are the two basic modes of AD, and are distinguished by the manner in which the ....

Christian Bischof, George Corliss, Larry Green, Andreas Griewank, Kara Haigler, and Perry Newman. Automatic differentiation of advanced CFD codes for multidisciplinary design. Journal on Computing Systems in Engineering, 3(6):625--638, 1992.

....thin layer Navier Stokes equations with a finite volume formulation. The code employs grid sequencing, multigrid, and local time stepping to accelerate convergence and efficiently obtain steady state high Reynolds number turbulent flow solutions. Experiences with ADIFOR on TLNS3D are described in [4, 18]. dF dx S S = e 13, e 17] S = e 7, e 12] S = e 1, e 6] dF dx S dF dx S g(1:6) g(7:12) g(13:17) Figure 8: An Example of parallel sensitivity computation # Processors Average Time Std. Deviation 1 34,160 (est. 0 5 6,832 53.1 15 3,757 11.5 Table 3: Results of parallel ....

.... useful and informative [32] 14] More important, Gilbert has shown recently that in the case of contractive fixed point iterations, the corresponding derivative values also converge to their correct values [17] We have confirmed the validity of this result on a transsonic fluid dynamics code [4], where the previously used semi analytic approximation to the derivative of the lift coefficient with respect to the Mach number turned out to be off by 50 . We have extended Gilbert s result to a wider class of iterative schemes, including Broyden s method, and other quasi Newton schemes. Our ....

Christian Bischof, George Corliss, Larry Green, Andreas Griewank, Kara Haigler, and Perry Newman. Automatic differentiation of advanced CFD codes for multidisciplinary design. Journal on Computing Systems in Engineering, 3(6):625--638, 1992.

No context found.

Bischof, C., G. Corliss, L. Green, A. Griewank, K. Haigler, and P. Newman. Automatic differentiation of advanced CFD codes for multidisciplinary design. Journal on Computing Systems in Engineering, 3(6):625--638, 1992c.

....Two major revisions of the system were subsequently completed; the June 1993 version is called ADIFOR 1.0. ADIFOR 1. 0 was successfully employed in very different areas of science of engineering: aeronautical multidisciplinary design optimization [5, 68] aeronautical computational fluid dynamics [9, 13, 27, 41, 56, 57], weather modeling [19, 25, 61, 62] groundwater contaminant transport [18, 69] aquifer modeling [30, 50] structural engineering [29] statistics [23] mechanical system design [49] power networks [52] reactor modeling [60] and large scale numerical optimization [4, 14, 64] The largest of ....

....affect the dependent variables (see Subsection 3.1) The seed matrix mechanism allows for flexible use of ADIFOR generated code. For example, it can be employed to compute compressed versions of large sparse Jacobians [4] to chain derivatives generated by programs running on different platforms [9, 27], or to decrease turnaround time for derivative computations through a parallel stripmining approach [13] 3 The ADIFOR 2.0 System The ADIFOR 2.0 system has three major components: ADIFOR 2.0 preprocessor: The ADIFOR 2.0 preprocessor parses the code, performs certain code normalizations, ....

Christian Bischof, George Corliss, Larry Green, Andreas Griewank, Kara Haigler, and Perry Newman. Automatic differentiation of advanced CFD codes for multidisciplinary design. Journal on Computing Systems in Engineering, 3(6):625--638, 1992.

....arrive at a code that also computes the derivatives. Hand coding of derivatives for a large code is a tedious and error prone process, in particular as real codes are often not well documented. In fact, the effort can take months or years, and in some cases may even be considered prohibitive [6]. However, depending on the skill of the implementer, hand coding may lead to the most efficient code possible. Divided Differences: One can approximate the derivative of F with respect to the ith component of x at a particular point x 0 by differencing, for example by a one sided difference, ....

Christian Bischof, George Corliss, Larry Green, Andreas Griewank, Kara Haigler, and Perry Newman. Automatic differentiation of advanced CFD codes for multidisciplinary design. Journal on Computing Systems in Engineering, 3(6):625--638, 1992.

....Rice or direct collaborations with the developers of the system. Even with such a limited distribution, it was successfully employed in many different areas of science and engineering, including aeronautical multidisciplinary design optimization [2, 31] aeronautical computational fluid dynamics [6, 7, 15, 19, 25, 26], weather modeling [12, 14, 28, 29] groundwater contaminant transport [11, 32] aquifer modeling [17, 21] structural engineering [16] statistics [13] mechanical system design [20] power networks [23] reaction modeling [27] and large scale numerical optimization [1, 8, 30] In July 1995, the ....

....the product of the Jacobian by a vector. The seed matrix mechanism allows for flexible use of the code generated by ADIFORTWO. For example, it can be employed to compute compressed versions of large sparse Jacobians [1] to chain derivatives generated by programs running on different platforms [6, 15], or to decrease turnaround time for derivative computations through a parallel stripmining approach [7] The benefit from proper initialization of the seed matrix is substantial, since the cost of derivative computation is more or less proportional to the number p of directional derivatives ....

C. Bischof, G. Corliss, L. Green, A. Griewank, K. Haigler, and P. Newman, Automatic differentiation of advanced CFD codes for multidisciplinary design, Journal on Computing Systems in Engineering, 3 (1992), pp. 625--638.

No context found.

Christian H. Bischof, George F. Corliss, Larry Green, Andreas Griewank, Ken Haigler, and Perry Newman. Automatic differentiation of advanced CFD codes for multidisciplinary design. Journal on Computing Systems in Engineering, 3:625--638, 1992.

No context found.

Bischof, C., et. al., "Automatic Differentiation of Advanced CFD Codes for Multidisciplinary Design," Computer Systems in Engineering No. 3, pp 625-637, 1993.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC