A scaled three-term conjugate gradient method for unconstrained optimization

In this paper, we are interested in solving nonlinear large scale unconstrained optimization problems of the form where \(f:\Re^{n}\rightarrow\Re\) is an at least twice continuously differentiable function. A nonlinear conjugate gradient method is an iterative scheme that generates a sequence \(\{x_{k}\}\) of an approximation to the solution of (1), using the recurrence where \(\alpha_{k}>0\) is the steplength determined by a line search strategy which either minimizes the function or reduces it sufficiently along the search direction and \(d_{k}\) is the search direction defined by where \(g_{k}\) is the gradient of f at a point \(x_{k}\) and \(\beta_{k}\) is a scalar known as the conjugate gradient parameter. For example, Fletcher and Reeves (FR) [1], Polak-Ribiere-Polyak (PRP) [2], Liu and Storey (LS) [3], Hestenes and Stiefel (HS) [4], Dai and Yuan (DY) [5] and Fletcher (CD) [6] used an update parameter, respectively, given by where \(y_{k-1}=g_{k}-g_{k-1}\). If the objective function is quadratic, with an exact line search the performances of these methods are equivalent. For a nonlinear objective function different \(\beta_{k}\) lead to a different performance in practice. Over the years, after the practical convergence result of Al-Baali [7] and later of Gilbert and Nocedal [8] attention of researchers has been on developing on conjugate gradient methods that possesses the sufficient descent condition for some constant \(c> 0\). For instance the CG-DESCENT of Hager and Zhang [9] where and which is based on the modification of HS method. Another important class of conjugate gradient methods is the so-called three-term conjugate gradient method in which the search direction is determined as a linear combination of \(g_{k}\), \(s_{k}\), and \(y_{k}\) as where \(\tau_{1}\) and \(\tau_{2}\) are scalar. Among the generated three-term conjugate gradient methods in the literature we have the three-term conjugate methods proposed by Zhang et al. [10, 11] by considering a descent modified PRP and also a descent modified HS conjugate gradient method as and where \(s_{k}=x_{k+1}-x_{k}\). An attractive property of these methods is that at each iteration, the search direction satisfies the descent condition, namely \(g_{k}^{T} d_{k}= -c\Vert g_{k}\Vert ^{2}\) for some constant \(c> 0\). In the same manner, Andrei [12] considers the development of a three-term conjugate gradient method from the BFGS updating scheme of the inverse Hessian approximation restarted as an identity matrix at every iteration where the search direction is given by An interesting feature of this method is that both the sufficient and the conjugacy conditions are satisfied and we have global convergence for a uniformly convex function. Motivated by the good performance of the three-term conjugate gradient method, we are interested in developing a three-term conjugate gradient method which satisfies both the sufficient descent condition, the conjugacy condition, and global convergence. The remaining part of this paper is structured as follows: Section 2 deals with the derivation of the proposed method. In Section 3, we present the global convergence properties. The numerical results and discussion are reported in Section 4. Finally, a concluding remark is given in the last section.

In this section, we describe the proposed method which would satisfied both the sufficient descent and the conjugacy conditions. Let us consider the DFP method, which is a quasi-Newton method belonging to the Broyden class [13]. The search direction in the quasi-Newton methods is given by where \(H_{k}\) is the inverse Hessian approximation updated by the Broyden class. This class consists of several updating schemes, the most famous being the BFGS and the DFP; if \(H_{k}\) is updated by the DFP then such that the secant equation is satisfied. This method is also known as a variable metric method, developed by Davidon [14], Fletcher and Powell [15]. A remarkable property of this method is that it is a conjugate direction method and one of the best quasi-Newton methods that encompassed the advantage of both the Newton method and the steepest descent method, while avoiding their shortcomings [16]. Memoryless quasi-Newton methods are other techniques for solving (1), where at every step the inverse Hessian approximation is updated as an identity matrix. Thus, the search direction can be determined without requiring the storage of any matrix. It was proposed by Shanno [17] and Perry [18]. The classical conjugate gradient methods PRP [2] and FR [1] can be seen as memoryless BFGS (see Shanno [17]). We proposed our three-term conjugate gradient method by incorporating the DFP updating scheme of the inverse Hessian approximation (7), within the frame of a memoryless quasi-Newton method where at each iteration the inverse Hessian approximation is restarted as a multiple of the identity matrix with a positive scaling parameter as and thus, the search direction is given by Various strategies can be considered in deriving the scaling parameter \(\mu_{k}\); we prefer the following which is due to Wolkowicz [19]: The new search direction is then given by where and We present the algorithm of the proposed method as follows.

In this section, we present the results obtained from the numerical experiment of our proposed method in comparison with the CG-DESCENT (CG-DESC) [9], three-term Hestenes-Stiefel (TTHS) [11], three-term Polak-Ribiere-Polyak (TTPRP) [10], and TTCG [12] methods. We evaluate the performance of these methods based on iterations and function evaluations. By considering some standard unconstrained optimization test problems obtained from Andrei [25], we conducted ten numerical experiments for each test function with the size of the variable ranging from \(70\leq n \leq45\mbox{,}000\). The algorithms were implemented using Matlab subroutine programming on a PC (Intel(R) core(TM)2 Duo E8400 3.00 GHz 3 GB) 32-bit Operating system. The program terminates whenever \(\Vert g_{k}\Vert <\epsilon\) where \(\epsilon=10^{-6}\) or a method failed to converges within 2,000 iterations. The latter requirement is represented by the symbol ‘-’. An Armijo-type line search suggested by Byrd and Nocedal [26] was used for all the methods under consideration. Table 1 in the appendices gives the performance of the algorithms in terms of iterations and function evaluations. TTPRP solves 81% of the test problems, TTHS solves 88% of the test problems, CG-DESCENT solves 85% of the test problems, and STCG solves 90% of the test problems, whereas TTCG solves 85% of the test problems. The performance of STCG over TTPRP is that TTPRP needs 16% and 60% more, on average, in terms of the number of iterations and function evaluations, respectively, than STCG. The improvement of STCG over TTHS is that STCG needs 2% and 57% less, on average, in terms of number of iterations and function evaluations, respectively, than TTHS. The improvement of STCG over CG-DESCENT algorithms is that CG-DESCENT needs 10% and 70% more, on average, in terms of the number of iterations and function evaluations, respectively, than STCG. Similarly, the improvement of STCG over TTCG is that STCG needs 21% and 79% less, on average, in terms of the number of iterations and function evaluations, respectively, than TTCG. In order to further examine the performance of these methods, we employ the performance profile of Dolan and Moré [27]. Figures 1-2 give the performance profile plots of these methods in terms of iterations and function evaluations and the top curve corresponds to the method with the highest win which indicates that the performance of the proposed method is highly encouraging and substantially outperforms any of the other methods considered.