A New Kind of Pre-Mean-Type Mappings and Its Gauss Iteration

Abstract

A function which is reflexive is called by pre-mean, a more generalized definition of a mean. In this paper, we define a new pre-mean and study its properties, and then using the given invariant curve we consider the problem of convergence of Gauss iteration of a kind of pre-mean type mappings generated by the exponential and logarithmic functions.

Share and Cite:

Ni, Y. and Fan, Y. (2024) A New Kind of Pre-Mean-Type Mappings and Its Gauss Iteration. Journal of Applied Mathematics and Physics, 12, 2574-2581. doi: 10.4236/jamp.2024.127152.

1. Introduction

Mean value is an important fundamental concept in mathematics. For thousands of years, many forms of mean have been proposed, such as the arithmetic mean A: p :

A( x 1 ,, x p )= x 1 ++ x p p ,

the geometric mean G: ( 0,+ ) p ( 0, ) :

G( x 1 ,, x p )= x 1 x p p ,

the harmonize mean H: ( 0,+ ) p ( 0, ) :

H( x 1 ,, x p )= p 1 x 1 ++ 1 x p .

Although there are various mean forms, they all meet the following definitions ([1]):

Let I be an open interval. If the function M: I p I satisfies

min{ x 1 ,, x p }M( x,y )max{ x 1 ,, x p }, x 1 ,, x p I, (1)

then M is called a mean on the interval I. When the above inequalities strictly hold, M is called strict mean.

Theorem 1. ([2] [3]) If the mean M and N of interval I are continuous, that is, any x,yI , xy , we have

max( M( x,y ),N( x,y ) )min( M( x,y ),N( x,y ) )<max( x,y )min( x,y ),

then

1) For any n , Gauss iteration ( M,N ) n of the mapping ( M,N ) is the mean type map on I 2 .

2) There is a continuous mean K: I 2 I , so that ( ( M,N ) n ) n=0 converges to ( K,K ) .

3) K is ( M,N ) -invariant mean, and is the unique ( M,N ) -invariant mean.

4) If ( M,N ) is a strict mean mapping, that is, M and N are a strict mean, then K is also a strict mean.

5) If M and N are (strictly) increasing for each variable, then K is also (strictly) increasing for each variable.

The above theorem provides the relationship between invariant equations and convergence of iteration of mean-type mappings. By studying the problem of invariant equations, we can study the dynamic behavior of mean-type mappings. There are many works to study kinds of invariant equations with different means ([4]-[8]). For the pre-mean-type mapping, the results have been still few. Next we will introduce some definitions for the pre-mean.

Obviously, it can be seen from the inequalities (1) that any mean satisfies the reflexivity, i.e.

M( x,,x )=x,xI.

However, not all the functions that satisfy the reflexivity are means. In other words, there are functions in the interval I that satisfy the reflexivity but not be means. In 2006, Matkowski ([9]) gave the definition of pre-mean as follows

If the function M: I p I satisfies reflexivity, M is called a pre-mean.

This indicates that a mean must be a pre-mean, but a pre-mean is not necessarily a mean, and the pre-mean is a generalized function than the mean. For example, the function M: 2 is defined as

M( x,y )=ln( 1 3 e 2x + 2 3 e 2y 1 3 e x + 2 3 e y ),x,y,

which satisfies reflexivity, but M( ln 1 2 ,0 )=ln 27 24 >max{ ln 1 2 ,0 } does not satisfy the definition of mean, so this function is a pre-mean but not a mean. But if a pre-mean is (strictly) increasing with respect to each variable, then it must be a (strict) mean ([9] [10]).

In this paper, we will consider the convergence of iterates of pre-mean type mappings of the form ( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] ) with λ,μ( 0,1 ) , p,q , pq , where E λ,μ [ p,q ] : 2 satisfies

E λ,μ [ p,q ] ( x,y )=ln ( λ e px +( 1λ ) e py μ e qx +( 1μ ) e qy ) 1 pq ,x,y.

2. Some Definitions and Auxiliary Results

Lemma 1 Suppose p,q , pq and λ,μ( 0,1 ) , then

1) E λ,μ [ p,q ] is the pre-mean.

2) E λ,μ [ p,q ] is the mean if and only if λ=μ or pq0 .

Proof. 1) If x , make x=y , then E λ,μ [ p,q ] ( x,x )=ln ( e px e qx ) 1 pq =x , so it is a pre-mean.

2) Firstly, for the case λ=μ , E λ,μ [ p,q ] ( x,x )=ln ( λ e px +( 1λ ) e py λ e qx +( 1λ ) e qy ) 1 pq . Let x<y , p>q , then ( pq )x<( pq )y and

e px+qy < e qx+py .

It can be obtained after doing a simple calculation on the above formula

( λ e px +( 1λ ) e py ) e qy <( λ e qx +( 1λ ) e qy ) e py ,

that is

ln ( λ e px +( 1λ ) e py λ e qx +( 1λ ) e qy ) 1 pq <y.

Similarly, in the same way, we get ln ( λ e px +( 1λ ) e py λ e qx +( 1λ ) e qy ) 1 pq >x, i.e.

x<ln ( λ e px +( 1λ ) e py λ e qx +( 1λ ) e qy ) 1 pq <y.

For the case of p<q , it can be proved similarly. From the definition of the mean value, the ln ( λ e px +( 1λ ) e py λ e qx +( 1λ ) e qy ) 1 pq ( x,y ), E λ,μ [ p,q ] is a mean at λ=μ .

Secondly, for the case pq0 , it is obvious that the function E λ,μ [ p,q ] is increasing with respect to each of the variables, so it is a mean. The proof of the “if” part has been finished. To prove the “only if” result, assume that λ=μ or pq0 and consider two possible cases.

Case p>0 and q>0 .

If E λ,μ [ p,q ] were a mean, there is

E λ,μ [ p,q ] ( ln( x ),0 )<0and E λ,μ [ p,q ] ( 0,ln( x ) )<0,x( 0,1 )

that is

( λ x p +( 1λ ) μ x q +( 1μ ) ) 1 pq <1and ( λ+( 1λ ) x p μ+( 1μ ) x q ) 1 pq <1,x( 0,1 ).

Therefore, if pq>0 , then

λ x p +( 1λ ) μ x q +( 1μ ) <1and λ+( 1λ ) x p μ+( 1μ ) x q <1,x( 0,1 ).

Let x0 , then

λ 1μ 1and λ μ 1. (2)

If pq<0 , then

λ x p +( 1λ ) μ x q +( 1μ ) >1and λ+( 1λ ) x p μ+( 1μ ) x q >1,x( 0,1 ).

Let x0 , then

λ 1μ 1and λ μ 1. (3)

So, if and only if λ=μ , Equations (2) and (3) hold.

Case p<0 and q<0 .

Since

E λ,μ [ p,q ] ( x,y )+ E 1λ,1μ [ p,q ] ( x,y ) =ln ( λ e px +( 1λ ) e py μ e qx +( 1μ ) e qy ( 1μ ) e qx +μ e qy ( 1λ ) e px +μ e py ) 1 pq =ln ( λ e px +( 1λ ) e py μ e qx +( 1μ ) e qy e p( x+y ) e q( x+y ) ( 1μ ) e qy +μ e qx ( 1λ ) e py +λ e px ) 1 pq = 1 pq ln e ( pq )( x+y ) =x+y,

then E 1λ,1μ [ p,q ] ( x,y )=x+y E λ,μ [ p,q ] ( x,y ) . Therefore, if E λ,μ [ p,q ] were a mean, that is, for all x<y we have

x E λ,μ [ p,q ] ( x,y )y,

then by E 1λ,1μ [ p,q ] ( x,y )=x+y E λ,μ [ p,q ] ( x,y ) , we have

x E 1λ,1μ [ p,q ] ( x,y )y,

that is, E 1λ,1μ [ p,q ] would be a mean, which contradicts the previous case.

Lemma 2 Let p,q , pq , λ,μ( 0,1 ) , then the arithmetic mean A is invariant with respect to the pre-mean type mapping ( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] ) , that is

A( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] )=A.

Proof. For all x,y , we have

A( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] )( x,y ) = ln ( λ e px +( 1λ ) e py μ e qx +( 1μ ) e qy ( 1μ ) e qx +μ e qy ( 1λ ) e px +μ e py ) 1 pq 2 = ln ( λ e px +( 1λ ) e py μ e qx +( 1μ ) e qy e p( x+y ) e q( x+y ) ( 1μ ) e qy +μ e qx ( 1λ ) e py +λ e px ) 1 pq 2 = 1 2( pq ) ln e ( pq )( x+y ) = x+y 2 =A( x,y ).

To consider the convergence of iterates of the pre-mean but not mean mappings, we now give the definition of invariant curves.

Let M,N: I 2 I , K: I 2 be some functions, let JI , f:JI be a function on J. If there is

f( M( x,f( x ) ) )=N( x,f( x ) ),xJ,

we say that the graph of the function f is an invariant curve with respect to the map ( M,N ) , briefly, ( M,N ) -invariant curve ([2]).

To explore the connection between the pre-mean map and its invariant curve, Lemma 3 is introduced.

Lemma 3 Let p,q , pq , λ,μ( 0,1 ) , then

(1) For every a , the graph of the function f( x )=2ax , x , that is,

H a ={ ( x,2ax )|x },

is ( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] ) -invariant curve, and in particular ( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] )( H a ) H a .

(2) The point ( a,a ) is the only fixed point of the mapping ( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] ) in the set H a .

(3) The family of sets { H a :a>0 } forms a partition of 2 , that is H a H b = for all a,b and ab , and a H a = 2 .

Proof. Fix a , we have

A( x,f( x ) )= x+f( x ) 2 =a,

and using Lemma 2, we get

A( E λ,μ [ p,q ] ( x,f( x ) ), E 1λ,1μ [ p,q ] ( x,f( x ) ) )=A( x,f( x ) )=a,

that is

E 1λ,1μ [ p,q ] ( x,f( x ) )+ E λ,μ [ p,q ] ( x,f( x ) ) 2 = x+f( x ) 2 =a,

E 1λ,1μ [ p,q ] ( x,f( x ) )=2a E λ,μ [ p,q ] ( x,f( x ) ),

therefore,

f( E λ,μ [ p,q ] ( x,f( x ) ) )= E 1λ,1μ [ p,q ] ( x,f( x ) ),

According to the definition of invariant curves, the function f( x )=2ax is the invariant curve of ( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] ) . Therefore, the first result is correct and the other two parts are obvious.

3. Convergence of Iteration of the Pre-Mean Mappings

Using Theorem 1 and Lemma 2, we can get the following:

Theorem 2 If λ=μ or pq0 , the iterative sequence ( ( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] ) n ) n=0 is pointwise convergent on 2 and

lim n ( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] ) n ( x,y )=( x+y 2 , x+y 2 ),x,y.

It is difficult to consider the convergence problem for the case λμ and pq>0 . Lemma 3 tells us that, H a is the invariant curve of ( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] ) . We will study the convergence problem of the restrictions of this map to the invariant sets H a ,a .

Theorem 3 Let p,q , pq , λ,μ( 0,1 ) . For every a and n , we have for all x it holds

( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] ) n ( x,2ax )=( ( f [ λ,μ ];a [ p,q ] ) n ( x ),2a ( f [ λ,μ ];a [ p,q ] ) n ( x ) ), (4)

where the function f [ λ,μ ];a [ p,q ] is defined as

f [ λ,μ ];a [ p,q ] ( x )= E λ,μ [ p,q ] ( x,2ax ),x,

that is

f [ λ,μ ];a [ p,q ] ( x )=ln ( λ e px +( 1λ ) e p( 2ax ) μ e qx +( 1μ ) e q( 2ax ) ) 1 pq ,x.

Proof. For the convenience of writing, write E= E λ,μ [ p,q ] , f a = f [ λ,μ ];a [ p,q ] . Because

E λ,μ [ p,q ] ( x,y )+ E 1λ,1μ [ p,q ] =x+y,

there is,

( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] )( x,y )=( E( x,y ),x+yE( x,y ) ),x,y.

When n=1 , the Equation (4) is correct, in fact

( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] )( x,2ax )=( E( x,2ax ),2aE( x,2ax ) ) =( f a ( x ),2a f a ( x ) ).

Suppose (4) holds for some n=k , that is

( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] ) k ( x,2ax )=( f a k ( x ),2a f a k ( x ) ),

then for all x

( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] ) k+1 ( x,2ax ) = ( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] ) k ( E( x,2ax ),2aE( x,2ax ) ) =( f a k ( E( x,2ax ) ),2a f a k ( E( x,2ax ) ) ) =( f a k ( f a ( x ) ),2a f a k ( f a ( x ) ) )=( f a k+1 ( x ),2a f a k+1 ( x ) ),

that is to say, when n=k+1 , (4) is still established, the certificate is finished.

To further study the theorem 3, we obtain the following

Corollary 1 Let p,q , pq , λ,μ( 0,1 ) , if

| ( 2λ1 )p( 2μ1 )q pq |<1,

then there is an open set U 2 containing the diagonal Δ=( x,x )|x such that

lim n ( E λ,μ [ p,q ] , E 1λ,1μ [ p,q ] ) n ( x,y )=( x+y 2 , x+y 2 ),( x,y )U.

Proof. According to theorem 3, we can obtain that

f [ λ,μ ];a [ p,q ] ( a )=E( a,a )=a,

as well as

( f [ λ,μ ];a [ p,q ] )( x )=[ ln ( λ e px +( 1λ ) e 2ax μ e qx +( 1μ ) e 2ax ) 1 pq ] = 1 pq [ p 2p( 1λ ) e p( 2ax ) λ e px +( 1λ ) e p( 2ax ) q+ 2q( 1μ ) e q( 2ax ) μ e qx +( 1μ ) e q( 2ax ) ].

So,

( f [ λ,μ ];a [ p,q ] )( a )= ( 2λ1 )p( 2μ1 )q pq .

It can be seen that any real number a, a are the fixed point of f [ λ,μ ];a [ p,q ] and ( f [ λ,μ ];a [ p,q ] )( a ) is independent of the value of a.

Since ( f [ λ,μ ];a [ p,q ] )( a )<1 , according to the compression mapping principle,

lim n ( f [ λ,μ ];a [ p,q ] ) n ( x )= f [ λ,μ ];a [ p,q ] ( a )=a= x+y 2 ,

which complete the proof.

Conflicts of Interest

The authors declare no conflicts of interest regarding the publication of this paper.

References

[1] Cauchy, A.L. (1821) Cours d’analyse de l’École Royale Polytechnique, Ire partie: Analyse Algébrique. Imprimérie Royale.
[2] Matkowski, J. (2011) Mean-Type Mappings and Invariant Curves. Journal of Mathematical Analysis and Applications, 384, 431-438.
https://doi.org/10.1016/j.jmaa.2011.05.078
[3] Matkowski, J. (1999) Iterations of Mean-Type Mappings and Invariant Means. Annales Mathematicae Silesianae, 13, 211-226.
https://api.semanticscholar.org/CorpusID:124866044
[4] Jarczyk, J. and Jarczyk, W. (2018) Invariance of Means. Aequationes Mathematicae, 92, 801-872.
https://doi.org/10.1007/s00010-018-0564-5
[5] Daróczy, Z. and Páles, Z. (2002) Gauss-Composition of Means and the Solution of the Matkowski-Sutô Problem. Publicationes Mathematicae Debrecen, 61, 157-218.
https://doi.org/10.5486/pmd.2002.2713
[6] Zhang, Q. and Li, L. (2023) On the Invariance of Generalized Quasiarithmetic Means. Journal of Applied Analysis & Computation, 13, 1581-1596.
https://doi.org/10.11948/20220380
[7] Zhang, Q. and Xu, B. (2011) An Invariance of Geometric Mean with Respect to Generalized Quasi-Arithmetic Means. Journal of Mathematical Analysis and Applications, 379, 65-74.
https://doi.org/10.1016/j.jmaa.2010.12.025
[8] Zhang, Q. (2024) Means and the Equations of Means. Advances in Mathematics (China), 53, 468-498.
[9] Matkowski, J. (2006) On Iterations of Means and Functional Equations. Iteration Theory (ECIT’04), Vol. 350, 184-197.
https://www.researchgate.net/publication/265979149
[10] Matkowski, J. (2014) Convergence of Iterates of Pre-Mean-Type Mappings. ESAIM: Proceedings and Surveys, 46, 196-212.
https://doi.org/10.1051/proc/201446016

Copyright © 2025 by authors and Scientific Research Publishing Inc.

Creative Commons License

This work and the related PDF file are licensed under a Creative Commons Attribution 4.0 International License.