Abstract

?

Author

Ion Păvăloiu
(Tiberiu Popoviciu Institute of Numerical Analysis)

Keywords

?

PDF

Cite this paper as:

I. Păvăloiu, On the monotonicity of the sequences of approximations obtained by Steffensen’s method, Mathematica, 35(58) (1993) no. 1, pp. 71-76.

About this paper

Journal

Mathematica – Revue d’analyse numérique et de théorie de l’approximation.
Mathematica

Publisher Name

Editions de l’Academie Roumaine

DOI

Not available yet.

Print ISSN

Not available yet.

Online ISSN

Not available yet.

References

[1] Balazs, M., A bilateral approximating method for finding the real roots of real equations. Rev. Anal. Numer. Theor. Approx., 21, no. 2, 111–117, (1992).

[2] Pavaloiu, I., Sur la methode de Steffensen pour la resolution des equations operationnelles non lineaires, Revue Roumaine des Mathematiques Pures et Appliquees, XIII, 6, 149–158 (1968).

[3] Pavaloiu, I., Rezolvarea ecuatiilor prin interpolare, Ed. Dacia, Cluj-Napoca (1981).

[4] Ul’m, S., Ob obobschenie metoda Steffensen dlya reshenia nelineinyh operatornyh uravnenii, Jurnal Vicisl, mat. i. mat-fiz. 4, 6, (1964).

Paper (preprint) in HTML form

On the monotonicity of the sequences of approximations obtained by Steffensen’s method

 
On the monotonicity of the sequences of approximations obtained by Steffensen’s method

Ion Păvăloiu

In a recent paper [1], M. Bálazs studied the conditions in which the sequence (xn)n0 generated by Steffensen’s method is monotonic and converges to the solution of equation

(1) f(x)=0

where f:I is a given function, and I is an interval of the real axis. Paper [1] considers the simple case of the sequence generated by the recurrence relation

(2) xn+1=xnf(xn)[xn,g(xn);f],n=0,1,,x0I,

in which g:I is given by g(x)=xf(x), and [u,v;f] is the first order divided difference of the function f.

The following theorem is proved in [1]:

Theorem 1.

[1]. Let f:I be a continuous function on I, and define g(x)=xf(x). If the following conditions:

  • (i)

    The function g:I is strictly decreasing and convex on I;

  • (ii)

    there exists a point x0I such that f(x0)<0;

  • (iii)

    I0=[x0d,x0+d]I, where d=max{|f(x0)|,|f(g(x0))|}

    hold, then all elements of the sequence (xn)n0 generated by (2) belong to I0; in addition, the following properties hold:

  • (j)

    the sequence (xn)n1 is increasing and convergent;

  • (jj)

    the sequence (g(xn))n1 is decreasing and convergent;

  • (jjj)

    limnxn=limng(xn)=x, where x is the unique solution of equation (1) in I.

We shall show further down that the properties resulting from Theorem 1 hold for more general Steffensen-type methods, while for the method (2), if hypothesis (ii) is replaced by:

  • (ii1)

    there exists x0I for which f(x0)>0 and g(x0)I, then hypothesis (iii) can be dropped, and the conclusions of the theorem remain valid putting I0=[g(x0),x0].

Consider an arbitrary function g:I whose fixed points coincide with the real roots of equation (1), and reciprocally.

The following theorem holds:

Theorem 2.

If the functions f:I and g:I are continuous on I, and if the following conditions:

  • (i2)

    the function g is strictly decreasing on I;

  • (ii2)

    the function f is strictly increasing and concave on I;

  • (iii2)

    there exists x0I such that f(x0)>0,g(x0)I and x0g(x0)>0

  • (iv2)

    the equations f(x)=0 and g(x)=x are equivalent, are fulfilled then equation (1) has a unique solution x[g(x0),x0] and the following properties hold:

  • (j2)

    the sequence (xn)n0 is decreasing and convergent;

  • (jj2)

    the sequence (g(xn))n0 is increasing and convergent;

  • (jjj2)

    limnxn=limng(xn)=x, where x is the unique solution of equation (1), therefore fixed point of g in the interval I.

Proof.

From (2), for n=1, we get

x1x0=f(x0)[x0,g(x0);f]<0,

that, is x1<x0.

Writing h(x)=xg(x), we have h(x0)>0 by hypothesis; moreover:

h(g(x0))=g(x0)g(g(x0))=[g(x0),x0;g](x0g(x0))<0

hence the equation h(x)=0 has unique solution in the interval [g(x0),x0], i.e. g has a unique fixed point x in this interval. Since the fixed points of g coincide with the roots of equation (1) and reciprocally, there follows that x is unique solution for equation (1) within the same interval, and f(g(x0))<0. Now we shall show that x1>x. First show that x1>g(x0).

It is easy to verify that the terms of the sequence (yn)n0 provided by the relations

yn+1=g(yn)f(g(yn))[yn,g(yn);f],n=0,1,,y0=x0

coincide with those of the sequence (xn)n0 generated by (2). In other words the equalities

xnf(xn)[xn,g(xn);f]=g(xn)f(g(xn))[xn,g(xn);f],n=0,1,

hold; for n=0, it follows that:

x1g(x0)=f(g(x0))[x0,g(x0);f]>0

from which it results that x1[g(x0),x0].

From Lagrange’s interpolating polynomial we get

f(x1)= f(x0)+[x0,g(x0);f](x1x0)
+[x1,x0,g(x0);f](x1x0)(x1g(x0))

and, since [x1,x0,g(x0);f]<0, taking into account (2), it follows that f(x1)>0, that is, x1>x. Since the function h is increasing, it results that h(x1)>0, i.e. x1g(x1)>0. Let us show that g(x1)[g(x0),x0]. We have g(x1)g(x0)=[x0,x1;g](x1x0)>0 that is, g(x0)<g(x1)<x<x1<x0.

Repeating the above argument, putting xk=x0,k, and supposing that the hypotheses of Theorem 2 are fulfilled for xk, we obtain:

g(xk)<g(xk+1)<x<xk+1<xk.

It results subsequently that the sequences (xn)n0 and (g(xn))n0 fulfil the properties (j2) and (jj2) of Theorem 2 and, in addition, are bounded.

Write u¯=limng(xn) and v¯=limnxn; we obtain u¯=g(v¯) and u¯xv¯. Suppose that u¯<v¯, therefore u¯v¯<0. But u¯v¯=g(v¯)v¯=h(v¯)0, since v¯x;this shows that u¯=v¯=x.

At limit (for n), equalities (2) yield f(x)=0, where the continuity of f was also taken into account. ∎

Remark 1.

If we put in Theorem 2, g(x)=xf(x), since g is decreasing, it follows that f(x)=xg(x) is increasing; since g is convex, it follows that [x,y,z;g]>0 for every x,y,zI, hence [x,y,z;f]=[x,y,z;g]<0, that is, f is concave. From f(x0)>0 it follows x0g(x0)>0, i.e. x0>g(x0). In this case, Theorem 1 in which hypotheses (ii) and (iii) are replaced by (ii2) is a consequence of Theorem 2.

The results of Theorem 2 are graphically illustrated in Figure 1.

Refer to caption
Figure 1.

In what follows we shall present, without proof. other cases in which properties of monotonicity analogous to those given by Theorem 2 hold.∎

Theorem 3.

If the functions f:I and g:I are continuous on I, and if the following conditions are fulfilled:

  • (i3)

    the function g is strictly decreasing on I;

  • (ii3)

    the function f is strictly increasing and convex on I;

  • (iii3)

    there exists x0I for which f(x0)<0,g(x0)I and x0g(x0)<0;

  • (iv3)

    the equations f(x)=0 and x=g(x) are equivalent, then the sequence (xn)n0 generated by (2) is increasing and convergent, the sequence (g(xn))n0 is decreasing and convergent, and x=limnxn=limng(xn) is the solution of equation (1).

Figures 2 plots the results of Theorem 3.

Refer to caption
Figure 2.
Refer to caption
Figure 3.
Theorem 4.

If the functions f:I and g:I are continuous on I, and if the following conditions are fulfilled:

  • (i4)

    the function g is strictly decreasing on I;

  • (ii4)

    the function f is strictly decreasing and convex on I;

  • (iii4)

    there exists x0I for which f(x0)<0,g(x0)I and x0g(x0)>0;

  • (iv4)

    the equations f(x)=0 and x=g(x) are equivalent, then the sequence (xn)n0 generated by (2) is decreasing and convergent, the sequence (g(xn))n0 is increasing and convergent, and x=limnxn=limng(xn),f(x)=0.

The results of this theorem are illustrated by Figure 3.

Theorem 5.

If the functions f:I and g:I are continuous on I, and if the following conditions are fulfilled:

  • (i5)

    the functions g is strictly decreasing on I;

  • (ii5)

    the function f is strictly decreasing and concave on I;

  • (iii5)

    there exists x0I such that f(x0)>0,g(x0)I and x0g(x0)<0;

  • (iv5)

    the equations f(x)=o and x=g(x) are equivalent then the sequence (xn)n0 generated by (2) is increasing and convergent, the sequence (g(xn))n0 is decreasing and convergent, and limnxn=limng(xn)=x,f(x)=0.

Remark 2.

The fact that the functions f and g from the above theorems are related only by the equivalence of equations f(x)=0 and x=g(x) offers large possibilities to choose these functions (i.e. to choose g when f is known, and conversely).

It is clear that if f keeps the same monotonicity and convexity on I, then we can moot the question of determining a real number λ such that g(x)=xλf(x) be a decreasing function. Under certain conditions λ can be determined, as it results from the following example:

If f is strictly increasing and strictly convex on I=[a,b], and if f is differentiable, then f is also derivable, and f(x)>f(a)>0 for every x[a,b]. Then we can put g(x)=xf(x)/f(a), and we have g(x)0 for every x[a,b], hence g is decreasing. It is clear that the equations f(x)=0 and x=g(x) have the same roots. If f(a)<0 and af(a)f(a)<b, then it is obvious that ag(a)=f(a)f(a)<0, and Theorem 3 can be applied for x0=a.

Numerical example.

Consider the equation:

f(x)=xarcsinx12(x2+1)=0,(x,1]

and the function g given by the relation:

g(x)=arcsinx12(x2+1)

Since g(x)=1x2+1 and g′′(x)=2x(x2+1)2 it follows that g is decreasing on (,1], and f is increasing and convex. One shows by direct calculation that f(2)0.75 <0 and g(2)1.25, hence f fulfils the hypotheses of Theorem 3. The table below lists the results of the calculations for x0=2.

n xn g(xn) f(xn)
0 -2.000000000000000000 -1.249045772398254430 -0.750954227601745574
1 -1.414047729532868260 -1.400933154002817630 -0.013114575530050630
2 -1.404227441155695550 -1.404222310683232820 -0.000005130472462734
3 -1.404223602392559510 -1.404223602391771120 -0.000000000000788388
4 -1.404223602391969620 -1.404223602391969620 -0.000000000000000000
Table 1.

The numerical results agree with the conclusions of Theorem 3; as one can see, after four iteration steps a solution approximation with 18 decimals is obtained (obviously, if truncation and rounding errors are neglected).∎

References

Received 18.VI.1992


Institutul de Calcul al Academiei

Filiala Cluj

3400 Cluj-Napoca

Romania

1993

Related Posts