# A Modified Nonlinear Conjugate Gradient Method for Engineering Computation.

1. Introduction

Unconstrained optimization methods are widely used in the fields of nonlinear dynamic systems and engineering computation to obtain the numerical solution of the optimal control problem [1-4]. In this paper, we consider the unconstrained optimization problem:

min f (x), x [member of] [R.sup.n], (1)

where f : [R.sup.n] [right arrow] R is a continuously differentiable function. The nonlinear conjugate gradient (CG) method is highly useful for solving this kind of problems because of its simplicity and its very low memory requirement [1]. The iterative formula of the CG methods is given by

[x.sub.k+1] = [x.sub.k] + [[alpha].sub.k] [d.sub.k] (2)

where [[alpha].sub.k] > 0 is step length which is obtained by carrying out some linear search, such as exact or inexact line search. In practical computation, exact line search is consumption time and the workload is very large, so we usually take the following inexact line search (see [5-7]). Usually, a major inexact line search is the strong Wolfe-Powell line search. The strong Wolfe-Powell line search is to find the step-length [[alpha].sub.k] in (2) satisfying

[mathematical expression not reproducible], (3)

[mathematical expression not reproducible], (4)

where 0 < [delta] < 1/2 and [delta] < [sigma] <1. In this paper, the following modified Wolfe-Powell line search is to find the step-length [[alpha].sub.k] in (2) satisfying (3) and the following:

[mathematical expression not reproducible], (5)

and [.sub.d]k is the search direction defined by

[mathematical expression not reproducible], (6)

where [g.sub.k] denotes the gradient [nabla] f([x.sub.k]), [[beta].sub.k] is scalar, and [[beta].sub.k] is chosen so that [d.sub.k] becomes the kth conjugate direction. There have been many well-known formulae for the scalar [[beta].sub.k], for example,

[[beta].sup.FR.sub.k] = [parallel][g.sub.k][[parallel].sup.2]/[parallel][g.sub.k-1][[parallel].sup.2] (7)

(Fletcher-Reeves [8], 1964),

[[beta].sup.PRP.sub.k] = [g.sup.t.sub.k] [y.sup.k-1]/[parallel][g.sub.k- 1][[parallel].sup.2]] (8)

(Polak-Ribiere-Polyak [9], 1969),

[[beta].sup.DY.sub.k] = [parallel][g.sub.k][[parallel].sup.2]/[g.sup.T.sub.k] [y.sup.k-1] (9)

(Dai-Yuan [10], 1999), and other formulae (e.g., [11-13]), where [parallel] * [parallel] is the Euclidean norm of vectors, [y.sub.k-1] = [g.sub.k] - [g.sub.k-1], and "T" stand for the transpose. These methods are generally regarded as very efficient conjugate gradient methods in practical computation.

In recent decades, in order to obtain the CG method which has not only good convergence property but also excellent computation, many researchers have studied the CG method extensively and obtained some improved methods with good properties [14-20]. Li and Feng [21] gave the modified CG method which generates a sufficient descent direction and showed its global convergence property under the strong Wolfe-Powell conditions. Dai and Wen [22] gave a scaled conjugate gradient method. They proved its global convergence property under the strong Wolfe-Powell conditions. Al-Baali [23] proved that the FR method satisfies the sufficient descent condition and converges globally for general objective functions if the strong Wolfe-Powell line search is used. Dai and Yuan [24] also introduced a formula for [[beta].sub.k]

[mathematical expression not reproducible], (10)

where [mathematical expression not reproducible]. Because

[mathematical expression not reproducible], (11)

we can rewrite (10) as

[mathematical expression not reproducible]. (12)

This formula includes the above three classes of CG method as an extreme case, and global convergence of three parameters of CG method was proved under strong Wolfe-Powell line search. If [[omega].sub.k] = 0, then the family reduces to the two- parameter family of conjugate gradient methods in [25]. Further, if [[lambda].sub.k] = 0, [[mu].sub.k] = [mu], and [[omega].sub.k] = 0, then the family reduces to the one-parameter family in [26]. Therefore, the three-parameter family has the one-parameter family in [26] and the two-parameter family in [25] as its subfamilies. In addition, some hybrid methods can also be regarded as special cases of the three-parameter family [24]. Above many modified CG methods, global convergence was obtained under strong Wolfe-Powell line search; however, in this paper, we further study the CG method, and our main aim is to improve the numerical performance of the CG method while keeping its global convergence with modified Wolfe-Powell line search.

This paper is organized as follows. We first present a criterion for the global convergence of CG method in the next section. In Section 3, we propose a new modified three-parameter conjugate gradient method and establish global convergence results for relative algorithm under modified Wolfe-Powell line search. The preliminary numerical results are contained in Section 4. One engineering example is analyzed for illustration in Section 5. Finally, conclusions appear in Section 6.

2. A Criterion for the Global Convergence of CG Method

In this section, first, we adopt the following assumption used commonly in the research literatures.

Assumption 1. The function f is LC in a neighborhood N of the level set [OMEGA] := {x [member of] [R.sup.n] | f(x) [less than or equal to] f([x.sub.1])} and [OMEGA] is bounded. Here, by LC, we mean that the gradient [nabla]f ([x.sub.k]) is Lipschitz continuous with modulus L; that is, there exists L > 0 such that

[mathematical expression not reproducible]. (13)

Lemma 2 (Zoutendijk condition [27]). Suppose that Assumption 1 holds, [x.sub.k] is given by (2) and (6), and [[alpha].sub.k] is obtained by the modified Wolfe-Powell line search ((3), (5)), while the direction [d.sub.k] satisfies [g.sup.T.sub.k] [d.sub.k] < 0. Then,

[mathematical expression not reproducible]. (14)

Lemma 3 (see [28]). Suppose that [[lambda].sub.k] (>0) and C are constants; if {[a.sub.i]} satisfy [mathematical expression not reproducible].

Theorem 4. Suppose that the objective function satisfies Assumption 1 and that [x.sub.k] is given by (2) and (6), where [[alpha].sub.k] satisfies the modified Wolfe-Powell (3) and (5), and [absolute value of [mathematical expression not reproducible] then, either [g.sub.k] = 0 holds for certain k or

[mathematical expression not reproducible]. (15)

Proof. Suppose, by contradiction, that the stated conclusion is not true. Then, in view of [parallel][g.sub.k][parallel].sup.2] > 0, there exits a constant [epsilon] > 0, such that

[parallel][g.sub.k][[parallel].sup.2] [greater than or equal to] [epsilon], k = 1, 2, ... (16)

From (6), we have

[d.sub.k] + [g.sub.k] = [[beta].sub.k] [d.sub.k-1], k = 2, 3, ..., (17)

By multiplying [d.sub.k] + [g.sub.k] on both sides of (17), then we have

[mathematical expression not reproducible]. (18)

Let [mathematical expression not reproducible]; then,

[mathematical expression not reproducible]. (19)

Thus, from (19) and [mathematical expression not reproducible], we get

[mathematical expression not reproducible]. (20)

Note that [t.sub.1] = 1/[parallel][g.sub.k][[parallel].sup.2] and [r.sub.1] = 1; then, it follows from (20) that

[mathematical expression not reproducible]. (21)

From Assumption 1, it follows that there exists constant M (>0), such that

[parallel]g (x)[[parallel].sup.2] [less than or equal to] M, [for all]x [member of] [OMEGA], (22)

and from (16), (21), and (22), we get

[mathematical expression not reproducible]. (23)

From the above, it is obvious that

[mathematical expression not reproducible]. (24)

From the other side, for [t.sub.k] [greater than or equal to] 0 from (23), it follows that

[k.summation over (i=1)] [absolute value of [r.sub.i]] [greater than or equal to] k[epsilon]/2M. (25)

From (24) and (25) and Lemma 3, we have

[mathematical expression not reproducible], (26)

what contradicts Lemma 2. Therefore, the global convergence is proved.

3. The Global Convergence for the New Formula and Algorithm Frame

3.1. The New Formula and the Corresponding Properties. Based on formula (10), we put forward a new formula of [[beta].sub.k]:

[mathematical expression not reproducible], (27)

where 1/2 [mathematical expression not reproducible] Because of possible negative values of

min {(1 - [[lambda].sub.k]) [parallel][g.sub.k][[parallel].sup.2], [[lambda].sub.k], [g.sup.T.sub.k] ([g.sub.k-1] - [d.sub.k-1] (28)

we use the maximum function to truncate zero and

min {(1 - [[lambda].sub.k]) [parallel][g.sub.k][[parallel].sup.2], [[lambda].sub.k], [g.sup.T.sub.k] ([g.sub.k-1] - [d.sub.k-1] (29)

Using the equality

[g.sup.T.sub.k-1] = - [parallel][g.sub.k-1][[parallel].sup.2] + [[beta].sub.k- 1]] [g.sup.T.sub.k-1] [d.sub.k-2], (30)

we can rewrite the denominator of (27) as

[mathematical expression not reproducible]. (31)

When

[mathematical expression not reproducible], (32)

then the denominator of [[beta].sup.new.sub.k] given by (27) reduces to the denominator of [[beta].sup.*.sub.k]. On the other hand, when

0 < [[lambda].sub.k ] [g.sup.T.sub.k] ([g.sub.k-1] - [d.sub.k-1]) < (1 - [[lambda].sub.k]) [parallel][g.sub.k][[parallel].sup.2], (33)

the numerator of (27) reduces to

[[lambda].sub.k] [g.sup.T.sub.k] ([g.sub.k-1] - [d.sub.k-1]). (34)

When

[d.sub.k-1] = 1/[[lambda].sub.k] [g.sub.k], (35)

then

[mathematical expression not reproducible]. (36)

Now, the numerator of [[beta].sup.new.sub.k] (27) reduces to the numerator of [[beta].sup.*.sub.k]. From the above analysis, we can see that (27) indeed is an extension of (10). Due to the existence of the parameters [[lambda].sub.k], [[mu].sub.k], and [w.sub.k]., it would be more flexible to call methods (2), (6), and (27) by this paper of conjugate gradient methods. Numerical experiments results in Section 4 demonstrate the influence of these parameters versus formula (27).

Lemma 5. Suppose that Assumption 1 holds and that xk is given by (2) and (6), where [[alpha].sub.k] satisfies the modified Wolfe-Powell conditions (3) and (5), while [[beta].sub.k] is computed by (27). Then, one has

[g.sup.T.sub.k] [d.sub.k]/[parallel][g.sub.k] [[parallel].sup.2] < 0 [for all]k [less than or equal to] 1. (37)

Proof. When k = 1, we have

[mathematical expression not reproducible]. (38)

Suppose [g.sup.T.sub.k-1] [d.sub.k-1] < 0 hold, [[beta].sub.k] = [[beta].sup.new.sub.k] in formula (27), and the conclusion holds.

If [[beta].sub.k] = [[beta].sup.new.sub.k], we have

[mathematical expression not reproducible], (39)

where [l.sub.k] = max{0, min((1 - [[lambda].sub.k]) [parallel][g.sub.k][[parallel].sup.2], [[lambda].sub.k] [g.sup.T.sub.k] ([g.sup.k-1] - [d.sub.k-1])}}; by formulas (3) and (5), we have [g.sup.T.sub.k] [d.sub.dk-1] < 0. Hence,

[mathematical expression not reproducible]. (40)

When [mathematical expression not reproducible], we obtain

[mathematical expression not reproducible]. (41)

Due to (1-[[mu].sub.k] - [w.sub.k]) [parallel] [g.sup.T.sub.k-1] [[parallel].sup.2] > 0 and [l.sub.k] > 0, through the above analysis, we have

[mathematical expression not reproducible]. (42)

Hence, [g.sup.T.sub.k] [d.sub.k]/[parallel][g.sub.k] [[parallel].sup.2] < 0.

The result shows that the search direction satisfies descent condition ([g.sup.T.sub.k] [d.sub.k] < 0); this condition may be crucial for convergence analysis of any conjugate gradient method.

Lemma 6. Suppose that Assumption 1 holds and that {[x.sub.k]} is given by (2) and (6), where [[alpha].sub.k] satisfies the modified WolfePowell conditions (3) and (5), while [[beta].sub.k] is computed by (27). Then, one has

[absolute value of [[beta].sub.k]] [less than or equal to] [parallel][g.sub.k][[parallel].sup.2]/[parallel] [g.sup.T.sub.k-1] [[parallel].sup.2]. (43)

Proof. Let [mathematical expression not reproducible]

By Lemma 5, then [mathematical expression not reproducible]

To sum up, [mathematical expression not reproducible], and [mathematical expression not reproducible], and hence [absolute value of [[beta].sub.k]] [less than or equal to] [parallel][g.sub.k][[parallel].sup.2]/[parallel] [g.sup.T.sub.k-1] [[parallel].sup.2].

Theorem 7. Suppose the objective function f(x) satisfies Assumption 1; consider methods (2) and (6), where [[beta].sub.k] is given by (27) and [[alpha].sub.k] satisfies the modified Wolfe-Powell line search condition. Then, either [g.sub.k] = 0 holds for certain k or

[mathematical expression not reproducible]. (44)

Proof. By Theorem 4 and Lemma 6, Theorem 7 is proved.

The result shows that the proposed algorithm with the modified Wolfe-Powell line search possesses global convergence.

3.2. Algorithm A. Based on the discussed above, now we can describe the algorithm frame for solving the unconstrained optimization problems (1) as follows.

Step 0. Choose an initial point [x.sub.0] [member of] [R.sup.n], given constants [[epsilon].sub.0] > 0, [delta] [member of] (0,1), [sigma] [member of] ([delta], 1) and [[lambda].sub.k] [member of] (1/2,1], [[mu].sub.k] [member of] [0,1], and [w.sub.k] [member of] [0,1], subject to [[lambda].sub.k] [less than or equal to] [[mu].sub.k] + [w.sub.k], set [d.sub.0] = [g.sub.0], and let k := 0.

Step 1. If a stopping criterion [parallel][g.sub.k][parallel] < [[epsilon].sub.0] is satisfied, then stop; otherwise, go to Step 2.

Step 2. Determine a step size [[alpha].sub.k] by line searches (3) and (5).

Step 3. Let [x.sub.k+1] = [x.sub.k] + [[alpha].sub.k] [d.sub.k]; compute [[beta].sub.k] and [d.sub.k] by (27) and (6).

Step 4. Set k := k+1 and go to Step 1.

4. Numerical Experiments and Results

In this section, in order to show the performance of the given algorithm, we test our proposed algorithm (algorithm A) and DY algorithm (given by formula (10)) via unconstrained optimization problems from Andrei [29] as follow. These testing functions are often used in engineering field:

(1) Sphere function [mathematical expression not reproducible]

(2) Rastrigin function [mathematical expression not reproducible]

(3) Freudenstein and Roth function (Froth) [mathematical expression not reproducible]

(4) Perturbed Quadratic diagonal function (Pqd) [mathematical expression not reproducible]

(5) Extended White & Holst function (Ewh) [mathematical expression not reproducible]

(6) Raydan1 function [f.sub.Ray1(x)] = [[summation].sup.2.sub.i=1] (1/10)(exp([x.sub.i]) - [x.sub.i]), x* =(0,0), [f.sub.Ray1](x*) = 0.3.

(7) Raydan 2 function [f.sub.Ray2(x)] = [[summation].sup.500.sub.i=1] (exp([x.sub.i]) - [x.sub.i]), x* = (0,0), [f.sub.Ray2](x*) = 500.

(8) Extended Trigonometric function [mathematical expression not reproducible]

(9) Extended Powell function [mathematical expression not reproducible]

(10) Wood function [mathematical expression not reproducible]

(11) Extended Wood function (Ewood) [mathematical expression not reproducible]

(12) Perturbed Quadratic function (Perq) [mathematical expression not reproducible]

(13) Extended Tridiagonal 1 function (Etri1) [mathematical expression not reproducible].

(14) Extended Miele & Cantrell function (Emic) [mathematical expression not reproducible]

(15) Extended Rosenbrock function (Erosen) [mathematical expression not reproducible]

(16) Generalized Rosenbrock function (Grosen) [mathematical expression not reproducible]

(17) QUARTC function [f.sub.QUAR] (x) = [[summation].sup.20.sub.i=1] [([x.sub.i] -1).sup.4], [x.sup.*] = (1, 1, ..., 1, 1), [f.sub.Grosen]([x.sup.*])

(18) LIARWHD function [mathematical expression not reproducible]

(19) Staircase 1 function [mathematical expression not reproducible]

(20) Staircase 2 function [mathematical expression not reproducible]

(21) POWER function [f.sub.[POWER] (x) = [[summation].sup.1000.sub.i=1] [([ix.sub.i]).sup.2], [x.sup.*] = (0, 0, ..., 0, 0), [f.sub.POWER]([x.sup.*]) = 0

(22) Diagonal 4 function [mathematical expression not reproducible]

(23) Extended BD1 function [mathematical expression not reproducible]

(24) CUBE function [mathematical expression not reproducible].

Here, [x.sup.*] and f([x.sup.*]) are the optimal solution and the function value at the optimal solution, respectively. For each algorithm, the parameters are chosen as [delta] = 0.04 and [sigma] = 0.5. All codes were written in MATLAB 7.5 and run on Lenovo with 1.90 GHz CPU processor, 2.43 GB RAM memory, and Windows XP operating system. The stop criterion of the iteration is one of the following conditions: (1) [parallel][g.sub.k][parallel] [less than or equal to] [[epsilon].sub.0] = [10.sup.-4] and (2) the number of iterations Itr > 5000. If condition (2) occurs, the method is deemed to fail for solving the corresponding test problem, and denote it by "F." For the first three test problems, we present experimental results to observe the behavior of the proposed and DY (given by formula (10)) conjugate gradient algorithm for different [[lambda].sub.k], different [[mu].sub.k], and different [[omega].sub.k]. Details of the schemes for parameters set are given in Table 1. Numerical results of test problems are listed in Tables 2, 3, 4, 5, 6, 7, and 8, respectively. Table 9 shows numerical results of other test problems. Here, [x.sub.0] denotes the initial point of the test problems and x and f([bar.x]) are iteration value and the function value at the final iteration, respectively.

Based on Table 1's sixteen kinds of scheme (different parameters set), we compared algorithm A with DY (given by formula (10)) conjugate gradient algorithm based on different initial point for three test problems. It is easy to see that the two algorithms based on different scheme (different parameters set) are successful for the first and the second test problems listed in Tables 2, 3, and 4. From Tables 5, 6, 7, 8, and 9, we can see that algorithm A is more successful than DY (given by formula (10)) conjugate gradient algorithm. For example, for the first three test problems based on different scheme (different parameters set), algorithm A based on different initial point all achieved satisfied iteration value and the function value at the final iteration. Nevertheless, under some scheme (parameters set), DY (given by formula (10)) conjugate gradient algorithm cannot search satisfied iteration solution and the function value at the final iteration. From Tables 5-9, we can also see that DY (given by formula (10)) conjugate gradient algorithm sometimes is failed based on some scheme (parameters set); however, our algorithm is failed only one time. These indicate that the influence of parameters value's changing in formula (27) on the algorithm is not big. We presented the Dolan and More [30] performance profiles for the algorithm A and DY method. Note that the performance ratio p([tau]) is the probability for a solver s for the tested problems with the factor [tau] of the smallest cost. As we can see from Figure 1, algorithm A is superior to DY method for the absolute errors of f([bar.x]) versus f([x.sup.*]). Hence, compared with the DY (given by formula (10)) conjugate gradient algorithm, algorithm A has higher stability and adaptability. Therefore, algorithm A yields a better numerical performance than the DY (given by formula (10)) conjugate gradient algorithm. From the above analysis, we can conclude that algorithm A is competitive for solving unconstrained optimization problems.

5. Application to Engineering

In this section, we present a real example to illustrate application of the algorithm proposed in this article. The example is the results of tests on endurance of deep groove ball bearings. For illustrating the purposes, we applied the real dataset of 23 observed failure times that was initially reported in Lieblein and Zelen [31] and later by a number of authors including Abouammoh and Alshingiti [32] and Krishna and Kumar [33]. The following dataset represents the number of millions of revolutions before failure for each of the 23 ball bearings in a life test: 17.88, 28.92, 33.0, 41.52, 42.12, 45.60, 48.40, 51.84, 51.96, 54.12, 55.56, 67.80, 68.64, 68.64, 68.88, 84.12, 93.12, 98.64, 105.12, 105.84, 127.92, 128.04, and 173.4. Dey and Pradhan [34] indicated that Weibull distribution fits this dataset better than the exponential, inverted exponential, and gamma distribution. A random variable X follows the Weibull distribution with probability density function (pdf) being that

[mathematical expression not reproducible], (45)

where [alpha] > 0 and [lambda] > 0 are the shape and scale parameters, respectively. Let n denote the number of observed failures and [t.sub.1], ..., [t.sub.n] denote the complete sample; the logarithm likelihood function is

[mathematical expression not reproducible]. (46)

The corresponding differential equations are

[mathematical expression not reproducible]. (47)

From (47), a closed-form solution of [alpha] and [lambda] does not exist, so a numerical technique (minimization - log L) must be used to find the maximum likelihood estimation (MLE) of [alpha] and [lambda] for any given dataset. By using algorithm A, We obtain [??] = 3.183499 and [??] = 7.0363e - 7. Dey and Pradhan [34] obtained the MLE of the parameters as follows: ([mathematical expression not reproducible]) = (3.1835,1.4329e - 6). From the numerical results, we can see that our algorithm is alternative for the above real unconstrained optimization problem.

6. Conclusion

In this article, by modifying the scalar [[beta].sub.k], we have proposed a three-parameter family of conjugate gradient method for solving large-scale unconstrained optimization problems. Global convergence of the proposed methods under modified Wolfe-Powell line search and general criterion are established, respectively. Numerical results show that our algorithm is competitive for solving unconstrained optimization problems. So, the proposed method is an alternative method used in the reliability data.

http://dx.doi.org/10.1155/2017/1425857

Competing Interests

The authors declare that they have no competing interests.

Acknowledgments

The research work are supported by National Natural Science Foundation of China (11361036; 11461051), the Joint Specialized Research Fund for the Doctoral Program of Higher Education, Inner Mongolia Education Department (20131514110005), Higher School Science Research Project of Inner Mongolia (NJZY16394), Natural Science Foundation of Inner Mongolia (2014MS0112), and Science Research Project of Inner Mongolia University of Technology (ZD201409).

References

[1] B. Balaram, M. D. Narayanan, and P. K. Rajendrakumar, "Optimal design of multi-parametric nonlinear systems using a parametric continuation based genetic algorithm approach," Nonlinear Dynamics, vol. 67, no. 4, pp. 2759-2777, 2012.

[2] L. Zhang, H. Gao, Z. Chen, Q. Sun, and X. Zhang, "Multiobjective global optimal parafoil homing trajectory optimization via Gauss pseudospectral method," Nonlinear Dynamics, vol. 72, no. 1-2, pp. 1-8, 2013.

[3] X.-z. Jiang and J.-b. Jian, "A sufficient descent Dai-Yuan type nonlinear conjugate gradient method for unconstrained optimization problems," Nonlinear Dynamics, vol. 72, no. 1-2, pp. 101-112, 2013.

[4] L. Zhu, G. Wang, and H. Chen, "Estimating steady multivariables inverse heat conduction problem by using conjugate gradient method," Proceedings of the Chinese Society of Electrical Engineering, vol. 31, no. 8, pp. 58-61, 2011.

[5] Z. Wei, G. Li, and L. Qi, "New nonlinear conjugate gradient formulas for large-scale unconstrained optimization problems," Applied Mathematics and Computation, vol. 179, no. 2, pp. 407430, 2006.

[6] Z. X. Wei, G. Y. Li, and L. Q. Qi, "Global convergence of the Polak-Ribiere-Polyak conjugate gradient method with an Armijo-type inexact line search for nonconvex unconstrained optimization problems," Mathematics of Computation, vol. 77, no. 264, pp. 2173-2193, 2008.

[7] G. Yuan, Z. Wei, and Q. Zhao, "A modified Polak-Ribiere-Polyak conjugate gradient algorithm for large-scale optimization problems," IIE Transactions, vol. 46, no. 4, pp. 397-413, 2014.

[8] R. Fletcher and C. M. Reeves, "Function minimization by conjugate gradients," The Computer Journal, vol. 7, pp. 149-154, 1964.

[9] B. T. Polyak, "The conjugate gradient method in extremal problems," USSR Computational Mathematics and Mathematical Physics, vol. 9, no. 4, pp. 94-112, 1969.

[10] Y. H. Dai and Y. Yuan, "A nonlinear conjugate gradient method with a strong global convergence property," SIAM Journal on Optimization, vol. 10, no. 1, pp. 177-182, 1999.

[11] M. R. Hestenes and E. Stiefel, "Method of conjugate gradient for solving linear equations," Journal of Research of the National Bureau of Standards, vol. 49, pp. 409-436, 1952.

[12] R. Fletcher, Practical Methods of Optimization. Vol. I, A Wiley- Interscience Publication, John Wiley & Sons, Chichester, UK, 2nd edition, 1987

[13] Y. Liu and C. Storey, "Efficient generalized conjugate gradient algorithms, part 1: theory," Journal of Optimization Theory and Applications, vol. 69, no. 1, pp. 129-137, 1991.

[14] H. Schramm and J. Zowe, "A version of the bundle idea for minimizing a nonsmooth function: conceptual idea, convergence analysis, numerical results," SIAM Journal on Optimization, vol. 2, no. 1, pp. 121-152, 1992.

[15] L. Luksan and J. Vlcek, "A bundle-Newton method for nonsmooth unconstrained minimization," Mathematical Programming, vol. 83, no. 3, pp. 373-391,1998.

[16] G. Yuan, Z. Wei, and G. Li, "A modified Polak-Ribiere-Polyak conjugate gradient algorithm for nonsmooth convex programs," Journal of Computational and Applied Mathematics, vol. 255, pp. 86-96, 2014.

[17] Z.-F. Dai, "Two modified HS type conjugate gradient methods for unconstrained optimization problems," Nonlinear Analysis: Theory, Methods & Applications, vol. 74, no. 3, pp. 927-936, 2011.

[18] Z. Dai and F. Wen, "Another improved Wei-Yao-Liu nonlinear conjugate gradient method with sufficient descent property," Applied Mathematics and Computation, vol. 218, no. 14, pp. 7421-7430, 2012.

[19] X. Z. Jiang, L. Han, and J. B. Jian, "A globally convergent mixed conjugate gradient method with Wolfe-Powell line search," Mathematica Numerica Sinica, vol. 34, no. 1, pp. 103-112, 2012.

[20] K. Sugiki, Y. Narushima, and H. Yabe, "Globally convergent three-term conjugate gradient methods that use secant conditions and generate descent search directions for unconstrained optimization," Journal of Optimization Theory and Applications, vol. 153, no. 3, pp. 733-757, 2012.

[21] M. Li and H. Feng, "A sufficient descent LS conjugate gradient method for unconstrained optimization problems," Applied Mathematics and Computation, vol. 218, no. 5, pp. 1577-1586, 2011.

[22] Z. Dai and F. Wen, "A modified CG-DESCENT method for unconstrained optimization," Journal of Computational and Applied Mathematics, vol. 235, no. 11, pp. 3332-3341, 2011.

[23] M. Al-Baali, "Descent property and global convergence of the Fletcher-Reeves method with inexact line search," IMA Journal of Numerical Analysis, vol. 5, no. 1, pp. 121-124, 1985.

[24] Y. H. Dai and Y. Yuan, "A three-parameter family of nonlinear conjugate gradient methods," Mathematics of Computation, vol. 70, no. 235, pp. 1155-1167, 2001.

[25] L. Nazareth, "Conjugate-gradient methods," in Encyclopedia of Optimization, C. Floudas and P. Pardalos, Eds., Kluwer Academic Publishers, Boston, Mass, USA, 1999.

[26] Y. H. Dai and Y. Yuan, "A class of globally convergent conjugate gradient methods," Research Report ICM-98-030, Institute of Computational Mathematics and Scientific/Engineering Computing, Chinese Academy of Sciences, 1998.

[27] G. Zoutendijk, "Nonlinear programming, computational methods," in Integer and Nonlinear Programming, pp. 37-86, North-Holland, Amsterdam, The Netherlands, 1970.

[28] Y. H. Dai and Y. X. Yuan, Nonlinear Conjugate Gradient Method, Science Press of Shanghai, Shanghai, China, 2000.

[29] N. Andrei, "An unconstrained optimization test functions collection," Advanced Modeling and Optimization, vol. 10, no. 1, pp. 147-161, 2008.

[30] E. D. Dolan and J. J. More, "Benchmarking optimization software with performance profiles," Mathematical Programming, vol. 91, no. 2, pp. 201-213, 2002.

[31] J. Lieblein and M. Zelen, "Statistical investigation of the fatigue life of deep-groove ball bearings," Journal of Research of the National Bureau of Standards, vol. 57, no. 5, pp. 273-316, 1956.

[32] A. M. Abouammoh and A. M. Alshingiti, "Reliability estimation of generalized inverted exponential distribution," Journal of Statistical Computation and Simulation, vol. 79, no. 11-12, pp. 1301-1315, 2009.

[33] H. Krishna and K. Kumar, "Reliability estimation in generalized inverted exponential distribution with progressively type-II censored sample," Journal of Statistical Computation and Simulation, vol. 83, no. 6, pp. 1007-1019, 2013.

[34] S. Dey and B. Pradhan, "Generalized inverted exponential distribution under hybrid censoring," Statistical Methodology, vol. 18, pp. 101-114, 2014.

Tiefeng Zhu, (1,2) Zaizai Yan, (1) and Xiuyun Peng (1)

(1) Science College, Inner Mongolia University of Technology, Hohhot 010051, China

(2) Department of Information Engineering, College of Youth Politics, Inner Mongolia Normal University, Hohhot 010051, China

Correspondence should be addressed to Zaizai Yan; zz.yan@163.com

Received 6 July 2016; Revised 6 November 2016; Accepted 8 December 2016; Published 11 January 2017

Caption: Figure 1: Performance profile on the absolute errors of f([bar.x]) versus f([x.sup.*]) (algorithm A versus DY).
```Table 1: Several schemes for the parameters set.

Parameters set                                       Scheme
number

[[lambda].sub.k] = 1.0, [[mu].sub.k] = 0.1,          [1]
[[omega].sub.k] = 0.1

[[lambda].sub.k] = 0.8, [[mu].sub.k] = 0.1,          [2]
[[omega].sub.k] = 0.1

[[lambda].sub.k] = 0.7, [[mu].sub.k] = 0.1,          [3]
[[omega].sub.k] = 0.1

[[lambda].sub.k] = 0.6, [[mu].sub.k] = 0.1,          [4]
[[omega].sub.k] = 0.1

[[lambda].sub.k] = 1.0, [[mu].sub.k] = 0.1,          [5]
[[omega].sub.k] = 0.8

[[lambda].sub.k] = 1.0, [[mu].sub.k] = 0.1,          [6]
[[omega].sub.k] = 0.6

[[lambda].sub.k] = 1.0, [[mu].sub.k] = 0.1,          [7]
[[omega].sub.k] = 0.4

[[lambda].sub.k] = 1.0, [[mu].sub.k] = 0.1,          [8]
[[omega].sub.k] = 0.2

[[lambda].sub.k] = 0.9, [[mu].sub.k] = 0.3,          [9]
[[omega].sub.k] = 0.5

[[lambda].sub.k] = 0.9, [[mu].sub.k] = 0.3,          [10]
[[omega].sub.k] = 0.4

[[lambda].sub.k] = 0.9, [[mu].sub.k] = 0.3,          [11]
[[omega].sub.k] = 0.3

[[lambda].sub.k] = 0.9, [[mu].sub.k] = 0.3,          [12]
[[omega].sub.k] = 0.1

[[lambda].sub.k] = 0.7, [[mu].sub.k] = 0.2,          [13]
[[omega].sub.k] = 0.1

[[lambda].sub.k] = 0.7, [[mu].sub.k] = 0.3,          [14]
[[omega].sub.k] = 0.1

[[lambda].sub.k] = 0.7, [[mu].sub.k] = 0.4,          [15]
[[omega].sub.k] = 0.1

[[lambda].sub.k] = 0.7, [[mu].sub.k] = 0.5,          [16]
[[omega].sub.k] = 0.1

Table 2: The numerical results of sphere function for different scheme.

Algorithm A            DY (given by formula
(10)) algorithm
[bar.x]/f)([bar.x])    [bar.x]/f)([bar.x])

[x.sub.0]    (-1,-2,-3, -4, -5,     (-1, -2, -3, -4,
1, 2, 3, 4, 5,         -5, 1, 2, 3, 4,
-1, -2, ..., 1, 2,     5, -1, -2, ...,
3, 4, 5)              1, 2, 3, 4, 5)

1        [bar.x]/2.9098e-11     [bar.x]/2.9098e-11
2        [bar.x]/2.9098e-11     [bar.x]/2.9098e-11
Sphere         3        [bar.x]/2.9098e-11     [bar.x]/2.9098e-11
4        [bar.x]/2.9098e-11     [bar.x]/2.9098e-11
5        [bar.x]/2.9098e-11     [bar.x]/2.9098e-11

6        [bar.x]/2.9098e-11     [bar.x]/2.9098e-11
7        [bar.x]/2.9098e-11     [bar.x]/2.9098e-11
8        [bar.x]/2.9098e-11     [bar.x]/2.9098e-11
Scheme         9        [bar.x]/2.9098e-11     [bar.x]/2.9098e-11
10       [bar.x]/2.9098e-11     [bar.x]/2.9098e-11
11       [bar.x]/2.9098e-11     [bar.x]/2.9098e-11
12       [bar.x]/2.9098e-11     [bar.x]/2.9098e-11

Note: [bar.x] = 1.0e -5 * (0.0512, 0.1024, 0.1536, 0.2048, 0.2560,
-0.0512, -0.1024, -0.1536, -0.2048, -0.2560, 0.0512, 0.1024, 0.1536,
0.2048, 0.2560, ..., -0.0512, -0.1024, -0.1536, -0.2048, -0.2560).

Table 3: The numerical results of sphere function for different scheme.

Algorithm A
[bar.x]/([bar.x])

[x.sub.0]                      (3, 3, ..., 3)

1      1.0e-5 * (-0.1536, -0.1536,..., -0.1536)/2.5952e-11
2      1.0e-5 * (-0.1536, -0.1536,..., -0.1536)/2.5952e-11
Sphere      3      1.0e-5 * (-0.1536, -0.1536,..., -0.1536)/2.5952e-11
4      1.0e-5 * (-0.1536, -0.1536,..., -0.1536)/2.5952e-11
5      1.0e-5 * (-0.1536, -0.1536,..., -0.1536)/2.5952e-11
6      1.0e-5 * (-0.1536, -0.1536,..., -0.1536)/2.5952e-11

7      1.0e-5 * (-0.1536, -0.1536,..., -0.1536)/2.5952e-11
8      1.0e-5 * (-0.1536, -0.1536,..., -0.1536)/2.5952e-11
Scheme      9      1.0e-5 * (-0.1536, -0.1536,..., -0.1536)/2.5952e-11
10      1.0e-5 * (-0.1536, -0.1536,..., -0.1536)/2.5952e-11
11      1.0e-5 * (-0.1536, -0.1536,..., -0.1536)/2.5952e-11
12      1.0e-5 * (-0.1536, -0.1536,..., -0.1536)/2.5952e-11

DY (given by formula (10)) algorithm
[bar.x]/([bar.x])

[x.sub.0]                      (3, 3, ..., 3)

1      1.0e-5 * (0.2556,0.2556,..., 0.2556)/7.1845e-11
2      1.0e-5 * (0.2556,0.2556,..., 0.2556)/7.1845e-11
Sphere      3      1.0e-5 * (0.2556,0.2556,..., 0.2556)/7.1845e-11
4      1.0e-5 * (0.2556,0.2556,..., 0.2556)/7.1845e-11
5      1.0e-5 * (0.2556,0.2556,..., 0.2556)/7.1845e-11
6      1.0e-5 * (0.2556,0.2556,..., 0.2556)/7.1845e-11

7      1.0e-5 * (0.2556,0.2556,..., 0.2556)/7.1845e-11
8      1.0e-5 * (0.2556,0.2556,..., 0.2556)/7.1845e-11
Scheme      9      1.0e-5 * (0.2556,0.2556,..., 0.2556)/7.1845e-11
10      1.0e-5 * (0.2556,0.2556,..., 0.2556)/7.1845e-11
11      1.0e-5 * (0.2556,0.2556,..., 0.2556)/7.1845e-11
12      1.0e-5 * (0.2556,0.2556,..., 0.2556)/7.1845e-11

Table 4: The numerical results of Rastrigin function for different
scheme.

Algorithm A
[bar.x]/f([bar.x])

[x.sub.0]                 (0.5, 0.5, ..., 0.5)

1      1.0e-7 * (0.6520,0.6520,..., 0.6520)/2.5260e-12
2      1.0e-7 * (0.6520,0.6520,..., 0.6520)/2.5260e-12
Rastrigin      3      1.0e-7 * (0.3504,0.3504,..., 0.3504)/7.3008e-13
4      -1.0e-7 * (0.6856,0.6856,..., 0.6856)/2.7978e-12
5      1.0e-7 * (0.6520,0.6520,..., 0.6520)/2.5260e-12
6      1.0e-7 * (0.6520,0.6520,..., 0.6520)/2.5260e-12

7      1.0e-7 * (0.6520,0.6520,..., 0.6520)/2.5260e-12
8      1.0e-7 * (0.6520,0.6520,..., 0.6520)/2.5260e-12
Scheme        13      -1.0e-7 * (0.7804,0.7804,..., 0.7804)/3.6238e-12
14      1.0e-7 * (0.6697,0.6697,..., 0.6697)/2.6699e-12
15      -1.0e-7 * (0.7222,0.7222,..., 0.7222)/3.1015e-12
16      1.0e-7 * (0.5194,0.5194,..., 0.5194)/1.6094e-12

DY (given by formula (10)) algorithm
[bar.x]/f([bar.x])

[x.sub.0]                 (0.5, 0.5, ..., 0.5)

1      1.0e-7 * (0.6520,0.6520,..., 0.6520)/2.5260e-12
2      1.0e-7 * (0.6520,0.6520,..., 0.6520)/2.5260e-12
Rastrigin      3      1.0e-7 * (0.3504,0.3504,...,0.3504)/7.3008e-13
4      -1.0e-7 * (0.6856,0.6856,...,0.6856)/2.7978e-12
5      1.0e-7 * (0.6520,0.6520,..., 0.6520)/2.5260e-12
6      1.0e-7 * (0.6520,0.6520,..., 0.6520)/2.5260e-12

7      1.0e-7 * (0.6520,0.6520,..., 0.6520)/2.5260e-12
8      1.0e-7 * (0.6520,0.6520,..., 0.6520)/2.5260e-12
Scheme        13      -1.0e-7 * (0.7804,0.7804,...,0.7804)/3.6238e-12
14      1.0e-7 * (0.6697,0.6697,..., 0.6697)/2.6699e-12
15      -1.0e-7 * (0.7222,0.7222,...,0.7222)/3.1015e-12
16      1.0e-7 * (0.5194,0.5194,...,0.5194)/1.6094e-12

Table 5: The numerical results of Rastrigin function for different
scheme.

Algorithm A
[bar.x]/f ([bar.x])

[x.sub.0]   (0.5, 0.5, 0.5, 0.5, 0.5,
0.1, 0.1, 0.1, 0.1, 0.1)

1           1.0e-7 * (0.0658, ..., 0.0658, 0.1700, ...,
0.1700)/2.6645e-14

2           1.0e-7 * (0.2044, ..., 0.2044, 0.5798, ...,
0.5798)/2.4514e-13

Rastrigin   3           1.0e-7 * (0.2786, ..., 0.2786, 0.8842,...,
0.8842)/4.5830e-13

4           1.0e-6 * (0.0288, ..., 0.0288, 0.1034, ...,
0.1034)/4.9027e-13

5           1.0e-7 * (0.0658, ..., 0.0658, 0.1700, ...,
0.1700)/2.6645e-14

6           1.0e-7 * (0.0658, ..., 0.0658, 0.1700, ...,
0.1700)/2.6645e-14

7           1.0e-7 * (0.0658, ..., 0.0658, 0.1700, ...,
0.1700)/2.6645e-14

8           1.0e-7 * (0.0658, ..., 0.0658, 0.1700, ...,
0.1700)/2.6645e-14

13          1.0e-7 * (0.2687, ..., 0.2687, 0.8076, ...,
Scheme                  0.8076)/4.2633e-13

14          1.0e-7 * (0.2558, ..., 0.2558, 0.7716, ...,
0.7716)/3.8369e-13

15          1.0e-7 * (0.2430, ..., 0.2430, 0.7357, ...,
0.7357)/3.5172e-13

16          1.0e-7 * (0.2303,..., 0.2303,0.7000,...,
0.7000)/3.1442e-13

DY (given by formula (10)) algorithm
[bar.x]/f ([bar.x])

[x.sub.0]   (0.5, 0.5, 0.5, 0.5, 0.5,
0.1, 0.1, 0.1, 0.1, 0.1)

1           1.0e-7 * (0.0658, ..., 0.0658, 0.1700, ...,
0.1700)/2.6645e-14

2           1.0e-7 * (0.2044, ..., 0.2044, 0.5798, ...,
0.5798)/2.4514e-13

Rastrigin   3           1.0e-7 * (0.2786, ..., 0.2786, 0.8842, ...,
0.8842)/4.5830e-13

4           1.0e-6 * (0.0288, ..., 0.0288, 0.1034, ...,
0.1034)/4.9027e-13

5           (-0.9950,..., -0.9950, 13.9269, ...,
13.9269)/2.9849

6           (0.0000,..., 0.0000, -8.9540,...,
-8.9540)/4.6096e-12

7           (0.0000, ..., 0.0000,1 9.8909, ...,
19.8909)/1.6094e-12

8           (0.9550, ..., 0.9550, -20.8843, ...,
-20.8843)/2.9849

13                               F/F
Scheme
14                               F/F

15                               F/F

16                               F/F

Table 6: The numerical results of Freudenstein and Roth function for
different scheme.

Algorithm A        DY (given by formula (10))
algorithm
[bar.x]/f/([bar.x])        [bar.x]/f/([bar.x])
[x.sub.0]        (0.5, -2)                  (0.5, -2)

1      (4.9992, 4)/8.6900e- 7  (5.0000, 4.0000)/4.0440e-10
2      (4.9992, 4)/8.6900e- 7              F/F
Froth       3      (4.9992, 4)/8.6900e- 7  (5.0000, 4.0000)/4.0440e-10
4      (4.9992, 4)/8.6900e- 7  (11.4128, -0.8968)/48.9843
5      (4.9992, 4)/8.6900e- 7  (5.0000, 4.0000)/4.0440e-10
6      (4.9992, 4)/8.6900e- 7  (5.0000, 4.0000)/4.0440e-10

7      (4.9992, 4)/8.6900e- 7              F/F
8      (4.9992, 4)/8.6900e- 7  (11.4128, -0.8968)/48.9843
Scheme      9      (4.9992, 4)/8.6900e- 7  (11.4128, -0.8968)/48.9843
10      (4.9992, 4)/8.6900e- 7  (11.4128, -0.8968)/48.9843
11      (4.9992, 4)/8.6900e- 7  (11.4128, -0.8968)/48.9843
12      (4.9992, 4)/8.6900e- 7              F/F

Table 7: The numerical results of Freudenstein and Roth function for
different scheme.

Algorithm A       DY (given by formula (10))
algorithm
[bar.x]/f([bar.x])        [bar.x]/f([bar.x])

[x.sub.0]  (-0.5,2)                        (-0.5,2)
1      (4.9986,4)/3.0289e-6              F/F
2      (4.9986,4)/3.0289e-6              F/F
Froth        3      (4.9986,4)/3.0289e-6   (11.4128, -0.8968)/48.9843
4      (4.9986,4)/3.0289e-6  (4.9999, -4.0000)/1.0003e-8
5      (4.9986,4)/3.0289e-6              F/F
6      (4.9986,4)/3.0289e-6              F/F

7      (4.9986,4)/3.0289e-6              F/F
8      (4.9986,4)/3.0289e-6              F/F
Scheme       9      (4.9986,4)/3.0289e-6              F/F
10      (4.9986,4)/3.0289e-6              F/F
11      (4.9986,4)/3.0289e-6              F/F
12      (4.9986,4)/3.0289e-6              F/F

Table 8: The numerical results of Freudenstein and Roth function for
different scheme.

Algorithm A
[bar.x]/f ([bar.x])
[x.sub.0]                   (-0.5, -2)

1            (4.9989,4.0000)/1.7488e-6
2            (4.9989,4.0000)/1.7488e-6
Froth     3            (4.9989,4.0000)/1.7488e-6
4            (4.9989,4.0000)/1.7488e-6
5            (4.9989,4.0000)/1.7488e-6
6            (4.9989,4.0000)/1.7488e-6

7            (4.9989,4.0000)/1.7488e-6
8            (4.9989,4.0000)/1.7488e-6
Scheme    13           (4.9989,4.0000)/1.7488e-6
14           (4.9989,4.0000)/1.7488e-6
15           (4.9989,4.0000)/1.7488e-6
16           (4.9989,4.0000)/1.7488e-6

DY (given by formula (10)) algorithm
[bar.x]/f ([bar.x])
[x.sub.0]                   (-0.5, -2)

1            (5.0000,4.0000)/1.7660e-10
2            (11.4128, -0.8968)/48.9843
Froth     3            F/F
4            F/F
5            F/F
6            (5.0000,4.0000)/1.7404e-10

7            (11.4128, -0.8968)/48.9843
8            F/F
Scheme    13           F/F
14           (5.0000, 4.0000)/2.6700e-10
15           (11.4128, -0.8968)/48.9843
16           (11.4128, -0.8968)/48.9843

Table 9: The numerical results of different function for
[[lambda].sub.k] = 0.9,  [[mu].sub.k] = 0.3, and [[omega].sub.k = 0.1.

Functions      [x.sub.0]                Algorithm
[bar.x]/f([bar.x])

Pqd            (0.5, -0.5, 0.5,         (0.0081, 0.0081,...,
0.8, 0.9)                -0.0182)/2.2226e-4

Ewh            (1, 2, 1, 2,...,         (1.0028, 1.0083,..., 1.0028,
1, 2, 1, 2)              1.0083)/2.3050e--5

Raydan1        (1, 1)                   1.0e-3 * (0.9370, 0.0005)/
0.3000

Raydan2        (1, 1, ..., 1, 1)        (0, 0,..., 0, 0)/500

Etri           (1, 1, ..., 1, 1)        a

Epow           (3, -1, 0, 1)            (2.0000, 0, 0, 2.0000)/
5.8875e-11

Wood           (-3, -1, -3, -1)         (1.0002, 1.0004, 0.9998,
0.9996)/1.6386e--7

Ewood          (-3, 1.2, -3, 1.2)       (1.0001, 1.0002, 0.9999,
0.9998)/2.0820e-8

Perq           (1, 2, 3)                1.0e-4 * (0.0973, -0.0020,
-0.1014)/4.0320e-10

Etri1          (2, 2, ..., 2)           F/F

Emic           (2, 2,..., 2)            c

Erosen         (-1.2, 1, ..., -1.2, 1)  (0.9999,0.9999,..., 0.9999)/
6.3141e-9

Grosen         (2, 2, ..., 2)           (1, 1, ..., 1)/5.8136e-10

QUARTC         (2, 2, ... ,2)           (0.6338, 0.6338,..., 0.6338)/
0.0899

LIARWHD        (4, 4, ..., 4)           (1, 1, ..., 1)/4.2649e-10

Staircase1     (2, 2, ..., 2)           1.0e-4 * (0.0448, -0.1380,
-0.0297, -0.1576)/2.7012e-10

Staircase2     (0, 0, ..., 0)           (1, 1, ..., 1)/5.1016e-10

POWER          (0, 1, ..., 0, 1)        -1.0e--4 * (0.1081, 0, 0,
..., 0, 0.0214)/1.9507e--9

Diagonal4      (2, 2, ..., 2)           1.0e--4 * (-0.2394, 0.0062,
-0.2394, 0.0062)/6.1191e-10

EBD1           (0, 1, ..., 0, 1)        (1, 1, ..., 1)/3.1321e--10

CUBE           (-1.2, 1, ..., -1.2, 1)  (1, 1.0001, ..., 1, 1.0001)/
1.4082e-9

Functions      DY (given by formula (10)) algorithm
[bar.x]/f([bar.x])

Pqd            (-0.0345, -0.0345, -0.0346,
-0.0347, -0.0612)/0.0401

Ewh            (1, 1, ..., 1, 1)/5.1174e-11

Raydan1        1.0e-3 * (0.9917, 0.0008)/
0.3000

Raydan2        (0, 0, ..., 0, 0)/500

Etri           b

Epow           F/F

Wood           F/F

Ewood          F/F

Perq           1.0e--4 * (0.0480, 0.0011,
0.1524)/7.2434e--10

Etri1          (1.2, 1.2, ..., 1.2)/6.5680e- 11

Emic           F/F

Erosen         (1, 1, ..., 1)/3.2274e- 12

Grosen         (1, 1, ..., 1)/5.8136e--10

QUARTC         (0.5278, 0.5278, ..., 0.5278)/
0.2487

LIARWHD        (1, 1, ..., 1)/7.5304e--10

Staircase1     1.0e-4 * (-0.2478, 0.7435,
-0.8469, 0.5991)/4.9189e--9

Staircase2     (1, 0.9999, ..., 1,0.9999)/
4.1476e-9

POWER          -1.0e -4 * (0.1081, 0 ,0, ...,0,
0.0214)/1.9507e--9

Diagonal4      1.0e-4 * (0.4352, 0.0055, 0.4352,
0.0055)/1.9247e-9

EBD1           (1, 1, ..., 1)/3.1321e-10

CUBE           (0.9693, 0.9108, ..., 0.9693, 0.9108)/
9.3983e-4

a = 1.0e3 * (0.0009, 0.0258, -0.0314, 0.0067, 0.0817, 1.0242, 0.1885,
-0.5213, -0.9109, -0.4524)/8.0025e - 11.

b = 1.0e4 * (0.0522, -0.1413, 0.0641, 0.0603, -0.2218, -0.5115,
0.2249, 0.7358, -1.3182, 0.3192)/4.0124e - 10.

c = (0.2140, 1.2386, 1.2810, 1.3032, -0.5016, 0.6837, 0.8063,
0.7930)/0.0105.
```