Derivative free Davidon-Fletcher-Powell (DFP) for solving symmetric systems of nonlinear equations

• Main Authors: Mamat, M., Dauda, M.K., Mohamed, M.A., Waziri, M.Y., Mohamad, F.S., Abdullah, H.
• Format: Conference or Workshop Item
• Language: English; English
• Published: 2018
• Subjects: QA Mathematics; QA75 Electronic computers. Computer science
• Online Access: http://eprints.unisza.edu.my/1723/1/FH03-FIK-18-13683.jpg; http://eprints.unisza.edu.my/1723/2/FH03-FIK-19-23937.pdf; http://eprints.unisza.edu.my/1723/

Research from the work of engineers, economist, modelling, industry, computing, and scientist are mostly nonlinear equations in nature. Numerical solution to such systems is widely applied in those areas of mathematics. Over the years, there has been significant theoretical study to develop methods for solving such systems, despite these efforts, unfortunately the methods developed do have deficiency. In a contribution to solve systems of the form F (x) = 0, x is an element of R-n, a derivative free method via the classical Davidon-Fletcher-Powell (DFP) update is presented. This is achieved by simply approximating the inverse Hessian matrix with Q(k+1)(-1) to theta I-k. The modified method satisfied the descent condition and possess local superlinear convergence properties. Interestingly, without computing any derivative, the proposed method never fail to converge throughout the numerical experiments. The output is based on number of iterations and CPU time, different initial starting points were used on a solve 40 benchmark test problems. With the aid of the squared norm merit function and derivative-free line search technique, the approach yield a method of solving symmetric systems of nonlinear equations that is capable of significantly reducing the CPU time and number of iteration, as compared to its counterparts. A comparison between the proposed method and classical DFP update were made and found that the proposed methodis the top performer and outperformed the existing method in almost all the cases. In terms of number of iterations, out of the 40 problems solved, the proposed method solved 38 successfully, (95%) while classical DFP solved 2 problems (i.e. 05%). In terms of CPU time, the proposed method solved 29 out of the 40 problems given, (i.e. 72.5%) successfully whereas classical DFP solves 11 (27.5%). The method is valid in terms of derivation, reliable in terms of number of iterations and accurate in terms of CPU time. Thus, suitable and achived the objective.