S S symmetry
Article A New Application of Gauss Quadrature Method for Solving Systems of Nonlinear Equations
Hari M. Srivastava 1,2,3 , Javed Iqbal 4, Muhammad Arif 4 , Alamgir Khan 4, Yusif S. Gasimov 3 and Ronnason Chinram 5,*
1 Department of Mathematics and Statistics, University of Victoria, Victoria, BC V8W 3R4, Canada; [email protected] 2 Department of Medical Research, China Medical University Hospital, China Medical University, Taichung 40402, Taiwan 3 Department of Mathematics and Informatics, Azerbaijan University, 71 Jeyhun Hajibeyli Street, Baku AZ1007, Azerbaijan; [email protected] 4 Department of Mathematics, Abdul Wali Khan University, Mardan 23200, KPK, Pakistan; [email protected] (J.I.); [email protected] (M.A.); [email protected] (A.K.) 5 Division of Computational Science, Faculty of Science, Prince of Songkla University, Hat Yai, Songkhla 90110, Thailand * Correspondence: [email protected]
Abstract: In this paper, we introduce a new three-step Newton method for solving a system of nonlin- ear equations. This new method based on Gauss quadrature rule has sixth order of convergence (with n = 3). The proposed method solves nonlinear boundary-value problems and integral equations in few iterations with good accuracy. Numerical comparison shows that the new method is remarkably effective for solving systems of nonlinear equations.
Citation: Srivastava, H.M.; Iqbal, J.; Keywords: nonlinear equations; gauss quadrature formula; ordinary differential equation (ODE); Arif, M.; Khan, A.; Gasimov, Y.S.; error equations; sixth-order convergence; numerical examples Chinram, R. A New Application of Gauss Quadrature Method for Solving Systems of Nonlinear Equations. Symmetry 2021, 13, 432. 1. Introduction https://doi.org/10.3390/sym13030432 In numerical analysis and other branches of scientific interests, solving a system of nonlinear equations by means of computational methods has always been very well Academic Editor: Sun Young Cho motivated and convincing for researchers. For a system of nonlinear equations:
Received: 26 January 2021 T P(S) = p (S) p (S) ··· p (S) = Accepted: 1 March 2021 1 , 2 , , n 0, (1) Published: 7 March 2021 T n n where S = (s1, s2, ··· , sn) and P : D ⊆ R −→ R is a nonlinear system, and pi, i = 1, 2, ··· , n : D ⊆ n −→ is a nonlinear mapping. The solution of the nonlinear system Publisher’s Note: MDPI stays neutral R R S∗ = (s∗ s∗ ··· s∗)T with regard to jurisdictional claims in of equations in (1) may be defined as the process of finding a vector 1, 2, , n ( ∗) = published maps and institutional affil- such that P S 0. The classical Newton method is one of the most commonly used iations. iterative methods: −1 S(k+1) = Sk − P0 S(k) P S(k) (k = 0, 1, 2, ··· ),
where P0(S(k)) is the Jacobian matrix of the nonlinear function P(S) in the kth iteration at Copyright: c 2021 by the authors. the point S(k) (see [1–3]). Newton’s method quadratically converges to the solution S∗ if Licensee MDPI, Basel, Switzerland. the function P is continuous and differentiable. In recent years, several methods have been This article is an open access article developed to analyze the solution of systems of nonlinear equations to improve interaction distributed under the terms and by using the quadrature formulas and fractional iterative method in the literature (see [4–12]). conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).
Symmetry 2021, 13, 432. https://doi.org/10.3390/sym13030432 https://www.mdpi.com/journal/symmetry Symmetry 2021, 13, 432 2 of 12
In particular, Codero and Torregrosa [9] developed the third-order Newton–Simpson method as follows: " ! #−1 S(k) + Y(k) S(k+1) = S(k) − 6 P0(S(k)) + 4P0 + P0(Y(k)) P(S(k)), (2) 2
and the Open Newton method:
S(k) + 3Z(k) S(k) + Z(k) S(k+1) = S(k) − 3 2P0 − P0 4 2 − (3) 3S(k) + Z(k) 1 + 2P0 P(S(k)), 4
where Z(k) represents the Newton approximation. Khirallah and Hafiz [13] suggested a cubically convergent method using the four-point Newton–Cotes formula for solving systems of nonlinear equations as follows:
2S(k) + Z(k) S(k) + 2Z(k) S(k+1) = S(k) − 8 P0(S(k)) + 3P0 + 3P0 2 2 (4) −1 + P0(Z(k)) P(S(k)).
The quadrature rule is used to approximate the definite integral of a function. The general form of a quadrature rule is given by [14]
Z b m Q = v(r)s(r)dr ' Qm = ∑ wis(ri) a i=0
where v(r) is a weight function, wi, i = 0, 1, 2, ... m are coefficients (weights). ri are points of the rule and s is a given function integrable on the interval [a, b] with the weight function v. Motivated and inspired by the research going on in this area, we have introduced a new iterative for solving nonlinear equations. Several numerical examples are considered to show the effectiveness of the proposed method. The new iterative method shows the compatibility of numerical results with the scheme’s theoretical analysis. We have solved nonlinear boundary-value problems by using the proposed method. Our method gives better results than the other methods and converges more rapidly to the solution. Section5 concludes the paper.
2. Three-Step Newton Method Let P : D ⊆ Rn −→ Rn, be s-times Fréchet differentiable function on a convex set ( ) D ⊆ Rn. Using the Mean-Value Theorem of multi-variable vectors function P(S k ) (see [1]), we have Z 1 P(S) − P(S(k)) = P0 S(k) + r(S − S(k))(S − S(k))dr. (5) 0 Using the left rectangular rule, the right-hand side of (5) can be written as
Z 1 P0 S(k) + r(S − S(k))(S − S(k))dr =∼ P0(S(k))(S − S(k)). (6) 0 From (5) and (6), we get
S = S(k) − P0(S(k))−1P(S(k))(k = 0, 1, 2, ··· ). (7)
Replacing S by S(k+1) in (7), we get the Newton method. Using (5) and different numerical integration formulas, one can obtain different iterative methods such as (2), (3), Symmetry 2021, 13, 432 3 of 12
and (4). To develop the new iterative method, we approximate the integral in (5) by the following three-point Gauss Legendre integration formula: r ! ! Z y y − x y + x 5 3 y + x f (t)dt ≈ 4 f + f (y − x) − + x 9 2 2 5 2 r ! ! (8) 5 3 y + x + f (y − x) + , 2 5 2
Thus, from (5) and (8), we have
Z 1 P0 S(k) + r(S − S(k))(S − S(k))dr 0 (k) (k) r (k) S − S 0 S + S 5 0 ( ) 3 S + S ≈ 4P + P (S − S k ) − + (9) 9 2 2 5 2 r 5 3 S + S(k) + P0 (S − S(k)) + . 2 5 2
Moreover, from (1), (5), and (9), we get ! S − S(k) S + S(k) 0 ≈ P S(k) + 4P0 9 2 r 5 3 S + S(k) + P0 S − S(k) − + (10) 2 5 2 r 5 3 S + S(k) + P0 S − S(k) + . 2 5 2
From (10), the iterative scheme is given by
S + S(k) S ≈ S(k) − 9 4P0 2 r 5 3 S + S(k) + P0 S − S(k) − + (11) 2 5 2 r −1 5 3 S + S(k) + P0 S − S(k) + P S(k). 2 5 2
Subsequently, we use the kth iteration T(k) and Z(k) of Newton’s method to replace S and S(k) respectively on the right-hand side of (11) and obtain a new iterative scheme as follows: Symmetry 2021, 13, 432 4 of 12
Algorithm 1: Three-Step Newton Method ( ) Step 1: Select an initial guess S 0 ∈ Rn and start k from 0. Step 2: Compute 5 S(k+1) = T(k) − 9 4P0(H(k)) + P0(H(k) + W(k)) 2 − 5 1 + P0(H(k) + J(k)) P(T(k)). 2 Step 3: Set for the next step: T(k) + Z(k) H(k) = , 2 r 3 J(k) = (T(k) − Z(k)) 5
and r 3 W(k) = (T(k) − Z(k)) − . 5 Step 4: Compute Z(k) = S(k) − P0(S(k))−1P(S(k))(k = 0, 1, 2, ··· ) T(k) = Z(k) − P0(Z(k))−1P(Z(k))(k = 0, 1, 2, ··· ).
Step 5: If ||S(k+1) − S(k)|| < e, then stop; otherwise. put k = k + 1 and go to Step 2.
In the next section, we discuss the convergence of the proposed method.
3. Convergence Analysis In the following theorem, we prove the convergence of the proposed method.
Theorem 1. Suppose that the function P : U ⊆ Rn −→ Rn is sufficiently Fréchet differentiable ∗ at each point of an open convex neighborhood U of the solution S ∈ Rn of (1). Assume also that P0(S) is continuous and nonsingular at S = S∗. Then, the sequence {S(k)} generated by Algorithm 5 converges to S∗ with the sixth order of convergence and the error equation is given by
5 6 7 ek+1 = 3C2ek + O(ek ), (12)
where (k) ∗ ek = S − S .
Proof. Let P : D ⊆ Rn −→ Rn be s-times Fréchet differentiable in U. Then, by using the ( ) usual notation for the mth derivative of P at v ∈ Rn, the m-linear function P m (v) : Rn × n n (m) n ∗ n · · · × R −→ R is such that P (v)(u1, u2, ··· , un) ∈ R . Suppose now that S + h ∈ R lies in the neighborhood of S∗. The Taylor polynomial for P(S∗ + h) can be of the form:
f −1 ! ∗ 0 ∗ m f P(S + h) = P (S ) h + ∑ Cmh + O(h ), (13) m=2
where 1 C = [P0(S∗)]−1P(m)(S∗)(m ≥ 2). m m! m n We observe that Cmh ∈ R , since
( ) ∗ 0 ∗ − P m (S ) ∈ L ∈ (Rn × Rn, ··· , Rn) and [P (X )] 1 ∈ L ∈ Rn. Symmetry 2021, 13, 432 5 of 12
In addition, we can express P0 as follows:
f −1 ! 0 ∗ 0 ∗ m−1 f P (S + h) = P (S ) I + ∑ mCmh + O(h ), (14) m=2
n×n m−1 n where I ∈ R is the identity matrix. We note that mCmh ∈ L ∈ R . From (13) and (14), we get
(k) 0 ∗ 2 3 4 5 6 P(S ) = P (S ) ek + C2ek + C3ek + C4ek + C5ek + C6ek + ··· . (15)
0 (k) 0 ∗ 2 3 4 5 P (S ) = P (S ) I + 2C2ek + 3C3ek + 4C4ek + 5C5ek + 6C5ek + ··· , (16) where 1 C = ( )[P0(S∗)]−1P(k)(S∗)(k = 2, 3, 4, ··· ) k k! and (k) ∗ ek = S − S . From (15), we have
0 (k) −1 2 2 P (S ) = I − 2C2ek + (4C2 − 3C3)ek + (−4C4 + 6C2C3 + 6C3C2 3 3 4 2 2 4 − 8C2)ek + (16C2 − 36C2C3 + 16C2C4 + 9C3 − 5C5)ek (17) −1 + ··· P0(S∗) .
−1 By multiplying P0 S(k) and P S(k), we obtain
0 (k) −1 (k) 2 2 3 3 P (S ) P(S ) = ek − C2ek + 2(C2 − C3)ek + (−4C2 + 4C2C3 + 3C3C2 4 4 2 2 5 − 3C4)e + (8C − 20C C3 + 6C + 10C2C4 − 4C5)e k 2 2 3 k (18) 5 3 2 2 + (−16C2 + 52C2C3 − 33C2C3 − 28C2C4 + 17C3C4+ 6 13C2C5 − 5C6)ek + ··· .
Taylor’s series expansion of P Z(k) is given by