Ulam stability of a linear difference equation in locally convex spaces

Abstract

We obtain a characterization of Ulam stability for the linear difference equation with constant coefficients \(x_{n+p}=a_₁x_{n+p-1}+…+a_{p}x_{n}\) in locally convex spaces. Moreover, for the first order linear difference equation we determine the best Ulam constant.

Authors

Adela Novac
Technical University of Cluj-Napoca, Romania


Technical University of Cluj-Napoca, Romania
Tiberiu Popoviciu Institute of Numerical Analysis, Romanian Academy

Dorian Popa
Technical University of Cluj-Napoca, Romania

Keywords

Ulam stability; difference equations; best constant; locally convex spaces

PDF

soon

Cite this paper as:

A. Novac, D. Otrocol, D. Popa, Ulam stability of a linear difference equation in locally convex spaces, Results Math., 76, 33 (2021). https://doi.org/10.1007/s00025-021-01344-2

About this paper

Journal

Results in Mathematics

Publisher Name

Springer

Print ISSN

Not available yet.

Online ISSN

Not available yet.

Google Scholar Profile

[1] Anderson, D.R., Onitsuka, M., Hyers–Ulam stability for a discrete time scale with two step sizes. Appl. Math. Comput. 344–345, 128–140 (2019),MathSciNet MATH Google Scholar
[2] Baias, A.R., Blaga, F., Popa, D., Best Ulam constant for a linear difference equation. Carpatian J. Math. 35(1), 13–21 (2019), MathSciNet MATH Google Scholar
[3] Brillouët-Belluot, N., Brzdȩk, J., Ciepliński, K., On Some Recent Developments in Ulam’s Type Stability, Abstract and Applied Analysis Volume 2012, Article ID 716936, 41p
[4] Brzdȩk, J., Popa, D., Raşa, I., Xu, B., Ulam Stability of Operators. Academic Press, Cambridge (2018), MATH Google Scholar
[5] Brzdȩk, J., Wojcek, P., On approximate solutions of some difference equation. Bull. Aust. Math. Soc. 95(3), 476–481 (2017), MathSciNet Article Google Scholar
[6] Brzdȩk, J., Popa, D., Xu, B., The Hyers–Ulam stability of the nonlinear recurrence. J. Math. Anal. Appl. 335, 443–449 (2007), MathSciNet Article Google Scholar
[7] Brzdȩk, J., Popa, D., Xu, B., Note on the nonstability of the nonlinear recurrence. Abh. Math. Sem. Univ. Hamburg 76, 183–189 (2006), MathSciNet Article Google Scholar
[8] Brzdȩk, J., Popa, D., Xu, B., On nonstability of the linear recurrence of order one. J. Math. Anal. Appl. 367, 146–153 (2010), MathSciNet Article Google Scholar
[9] Hyers, D.H., On the stability of the linear functional equations. In: Proc. Nat, Acad. Sci., USA, 27, pp. 222–224 (1941)
[10] Jameson, G.J.O., Convex series. Proc. Camb. Phil. Soc. 37–47, (1972)
[11] Moslehian, M.S., Popa, D., On the stability of the first-order linear recurrence in topological vector spaces. Nonlinear Anal. 73, 2792–2799 (2010), MathSciNet Article Google Scholar
[12] Onitsuka, M., Influence of the stepsize on Hyers-Ulam stability of first order homogeneous linear difference equations. Int. J. Differ. Equ. 12(2), 281–302 (2017), MathSciNet Google Scholar
[13] Polya, G., Szegö, G., Aufgaben und Lehrsatze aus der Analysis I. Julius Springer, Berlin (1925), Book Google Scholar
[14] Popa, D., Hyers–Ulam stability of the linear recurrence with constant coefficient. Adv. Differ. Equ.-NY, 101–107 (2005)
[15] Popa, D., Hyers–Ulam–Rassias stability of the linear recurrence. J. Math. Anal. Appl. 309, 591–597 (2005), MathSciNet Article Google Scholar
[16] Popa, D., Raşa, I., Best constant in stability of some positive linear operators. Aequ. Math. 90, 719–726 (2016), MathSciNet Article Google Scholar
[17] Rudin, W., Functional analysis. In: International Series in Pure and Applied Mathematics, 2nd ed. McGraw-Hill Inc., New York (1991)
[18] Ulam, S.M., Problems in Modern Mathematics. Wiley, New York (1964), MATH Google Scholar

Ulam stability of a linear difference equation in locally convex spaces

Adela Novac Technical University of Cluj-Napoca, Department of Mathematics, 28 Memorandumului Street, 400114 Cluj-Napoca, Romania adela.chis@math.utcluj.ro , Diana Otrocol Technical University of Cluj-Napoca, Department of Mathematics, 28 Memorandumului Street, 400114 Cluj-Napoca, Tiberiu Popoviciu Institute of Numerical Analysis, Romanian Academy, P.O.Box. 68-1, 400110 Cluj-Napoca, Romania Diana.Otrocol@math.utcluj.ro and Dorian Popa Technical University of Cluj-Napoca, Department of Mathematics, 28 Memorandumului Street, 400114 Cluj-Napoca, Romania Popa.Dorian@math.utcluj.ro
Abstract.

We obtain a characterization of Ulam stability for the linear difference equation with constant coefficients xn+p=a1xn+p1++apxnx_{n+p}=a_{1}x_{n+p-1}+\ldots+a_{p}x_{n} in locally convex spaces. Moreover, for the first order linear difference equation we determine the best Ulam constant.
Keywords: Ulam stability; difference equations; best constant; locally convex spaces.
MSC: 39B82; 39A10; 39B72.

1. Introduction

It seems that the first result on the stability of functional equations appeared in the famous book of Pólya and Szegö [13] and concerns the Cauchy functional equation on the set of positive integers. But the stability problem of functional equations was originally raised by S.M. Ulam [18] in 1940 who formulated a question concerning the perturbation of homomorphisms of metric groups. In 1941, this problem was solved by D. H. Hyers [9] for the Cauchy functional equation in Banach spaces. After Hyers’ result many papers dedicated to this topic have been published (see e.g. [3], [6], [7], [11], [14]).

Recall the result on stability for the Cauchy’s functional equation (see [4], [18]):

”Let (G,+)(G,+) be an abelian group, XX a Banach space and ε>0\varepsilon>0. Then for every f:GXf:G\rightarrow X satisfying

(1.1) f(x+y)f(x)f(y)ε,x,yG,\left\|f(x+y)-f(x)-f(y)\right\|\leq\varepsilon,\ \ x,y\in G,

there exists a unique additive function a:GXa:G\rightarrow X such that

(1.2) f(x)a(x)ε,xG.\left\|f(x)-a(x)\right\|\leq\varepsilon,\ \ x\in G."

Remark that conditions (1.1) and (1.2) can be rewritten in the form

f(x+y)f(x)f(y)εB(0,1),x,yG,f(x+y)-f(x)-f(y)\in\varepsilon B(0,1),\ x,y\in G,

respectively

f(x)a(x)εB(0,1),xG,f(x)-a(x)\in\varepsilon B(0,1),\ x\in G,

where B(0,1)B(0,1) is the unit ball of the Banach space XX.

This new form of the relations (1.1) and (1.2) suggests the possibility to formulate the problem of Ulam stability in topological vector spaces for various equations. The goal of this paper is to give a result on Ulam stability for the linear difference equation with constant coefficients in locally convex spaces and to obtain in some cases the best Ulam constant.

Throughout this paper, it will be assumed that XX is a sequentially complete locally convex space over the field \mathbb{C} of complex numbers.

Let AA be a nonempty subset of XX.

The following relation holds:

(λ+μ)AλA+μA.(\lambda+\mu)A\subseteq\lambda A+\mu A.

If AA is a convex set and λ,μ\lambda,\mu are positive numbers, then

(λ+μ)A=λA+μA.(\lambda+\mu)A=\lambda A+\mu A.

For more details and terminology, see [17].

By a convex series of elements of AA we mean a series of the form n=1λnan\sum\limits_{n=1}^{\infty}\lambda_{n}a_{n}, where anAa_{n}\in A and λn0,n1\lambda_{n}\geq 0,\ n\geq 1, and n=1λn=1\sum\limits_{n=1}^{\infty}\lambda_{n}=1. The set AA is called:

  • (i)

    CS-closed if the sum of every convergent series of elements of AA belongs to AA;

  • (ii)

    CS-compact if every convex series of elements of AA is convergent to an element of AA.

It can be proved that every CS-compact set is a CS-closed set and bounded, and every CS-closed set is a convex set (for more details, see [10]).

As a consequence of the previous assertions it follows that if n=1pn\sum\limits_{n=1}^{\infty}p_{n} is a convergent series of positive terms, AA is a CS-compact set and anA,n1a_{n}\in A,\ n\geq 1, then

n=1pnan(n=1pn)A\sum\limits_{n=1}^{\infty}p_{n}a_{n}\in\left(\sum\limits_{n=1}^{\infty}p_{n}\right)A

All over this paper by VV we denote a CS-compact and balanced neighbourhood of 0X0_{X}.

We consider the linear difference equation of order pp

(1.3) xn+p=a1xn+p1++apxn,n0x_{n+p}=a_{1}x_{n+p-1}+\ldots+a_{p}x_{n},\ n\geq 0

where a1,a2,,apa_{1},a_{2},\ldots,a_{p}\in\mathbb{C}, x0,x1,,xp1Xx_{0},x_{1},\ldots,x_{p-1}\in X, and p={1,2,}.p\in\mathbb{N}=\{1,2,\ldots\}.

The equation (1.3) is called Ulam stable (or VV-Ulam stable) if there exists a positive constant KK such that for every ε>0\varepsilon>0 and every (xn)n0(x_{n})_{n\geq 0} satisfying

(1.4) xn+pa1xn+p1apxnεV,n0,x_{n+p}-a_{1}x_{n+p-1}-\ldots-a_{p}x_{n}\in\varepsilon V,\ \forall n\geq 0,

there exists a sequence (yn)n0(y_{n})_{n\geq 0} in XX such that

yn+p=a1yn+p1++apyn,y_{n+p}=a_{1}y_{n+p-1}+\ldots+a_{p}y_{n},
(1.5) xnynKεV,n0.x_{n}-y_{n}\in K\varepsilon V,\ n\geq 0.

A positive number KK satisfying (1.5) is called an Ulam constant of (1.3). Denote by KRK_{R} the infimum of all Ulam constants of (1.3). Generally KRK_{R} is not necessary an Ulam constant of (1.3) (see [16]), but if it is an Ulam constant we call it the best Ulam constant or the Ulam constant of (1.3). If in (1.4) the number ε\varepsilon is replaced by a sequence of positive numbers (εn)n0(\varepsilon_{n})_{n\geq 0} and KεK\varepsilon in (1.5) by a sequence of positive numbers (δn)n0(\delta_{n})_{n\geq 0} depending on (εn)n0(\varepsilon_{n})_{n\geq 0} then the linear difference equation is called generalized Ulam stable.

Let

(1.6) rp=a1rp1++apr^{p}=a_{1}r^{p-1}+\ldots+a_{p}

be the characteristic equation of (1.4) and denote by r1,,rpr_{1},\ldots,r_{p} its complex roots.

A first result on Ulam stability for the equation (1.3) in Banach spaces was given by D. Popa in [14] who proved that if all the roots of (1.6) are not on the unit circle then the equation (1.3) is stable with the Ulam constant

K=1|k=1p(|rk|1)|.K=\frac{1}{\left|\prod\limits_{k=1}^{p}\left(\left|r_{k}\right|-1\right)\right|}.

Later J. Brzdȩk, D. Popa and B. Xu proved that if the characteristic equation (1.6) has at least a root on the unit circle, then the equation (1.3) is not stable in Ulam sense ([4], [7], [8]).

A first approach of Ulam stability for a first order linear difference equation in topological vector spaces was considered by S. Moslehian and D. Popa in [11]. The results of the present paper extend and complement the results obtained in [11]. For more details on Ulam stability for linear difference equations we refer the reader to [1], [2], [5], [12], [15].

2. Ulam stability of the first order linear difference equation

First we give a result on Ulam stability for a first order linear difference equation of the form

(2.1) xn+1=axn+bn,n0,x_{n+1}=ax_{n}+b_{n},\ n\geq 0,

where a,a0a\in\mathbb{C},\ a\neq 0, and(bn)n0\ (b_{n})_{n\geq 0} is a sequence in X,x0XX,\ x_{0}\in X.

Lemma 2.1.

If (xn)n0(x_{n})_{n\geq 0} satisfies (2.1) then

xn=anx0+k=1nankbk1,n1.x_{n}=a^{n}x_{0}+\sum\limits_{k=1}^{n}a^{n-k}b_{k-1},\ n\geq 1.
Proof.

Induction on nn. ∎

A result on classical Ulam stability for the linear difference equation xn+1=anxn+bn,n0x_{n+1}=a_{n}x_{n}+b_{n},\ n\geq 0 was given in [11] in the case when (an)n0(a_{n})_{n\geq 0} is a sequence in K\{0}K\backslash\{0\}, and the field KK is \mathbb{R} or \mathbb{C}. In the next theorem we present a result on generalized Ulam stability for the case where (an)n0\left(a_{n}\right)_{n\geq 0} is a constant sequence in KK.

Theorem 2.2.

Let aa\in\mathbb{C} and (εn)n0(\varepsilon_{n})_{n\geq 0} be a sequence of positive numbers such that the series n=0εn|a|n\sum\nolimits_{n=0}^{\infty}\dfrac{\varepsilon_{n}}{\left|a\right|^{n}} is convergent. Then for every sequence (xn)n0(x_{n})_{n\geq 0} in XX satisfying

(2.2) xn+1axnbnεnV,n0,x_{n+1}-ax_{n}-b_{n}\in\varepsilon_{n}V,\ n\geq 0,

there exists a unique sequence (yn)n0,(y_{n})_{n\geq 0}, defined by yn+1=ayn+bn,n0,y0Xy_{n+1}=ay_{n}+b_{n},\ n\geq 0,\ y_{0}\in X, such that

(2.3) xnyn(m=1εm+n1|a|m)V,n0.x_{n}-y_{n}\in\left(\sum\limits_{m=1}^{\infty}\dfrac{\varepsilon_{m+n-1}}{\left|a\right|^{m}}\right)V,\ n\geq 0.
Proof.

Existence. Let (xn)n0(x_{n})_{n\geq 0} be a sequence satisfying (2.2). Then there exists cnV,n0c_{n}\in V,n\geq 0, such that xn+1axnbn=εncn\ x_{n+1}-ax_{n}-b_{n}=\varepsilon_{n}c_{n}. We get, according to Lemma 2.1,

xn=an(x0+k=1nbk1+εk1ck1ak),n1.x_{n}=a^{n}\left(x_{0}+\sum\limits_{k=1}^{n}\dfrac{b_{k-1}+\varepsilon_{k-1}c_{k-1}}{a^{k}}\right),\,n\geq 1.

First we prove that the series n=1εn1cn1an\sum\limits_{n=1}^{\infty}\frac{\varepsilon_{n-1}c_{n-1}}{a^{n}} is convergent. Let

sn=k=1nεk1ck1ak,n1.s_{n}=\sum\limits_{k=1}^{n}\frac{\varepsilon_{k-1}c_{k-1}}{a^{k}},\ n\geq 1.

We prove that (sn)n1(s_{n})_{n\geq 1}is a Cauchy sequence. Let WW be an arbitrary neighborhood of 0X0_{X}. Then there exists a ballanced and convex neighborhood UU of 0X0_{X} such that UWU\subseteq W. Since VV is bounded it follows that there exists α>0\alpha>0 such that VαUV\subseteq\alpha U.

We have:

sn+psn\displaystyle s_{n+p}-s_{n} =εnan+1cn++εn+p1an+pcn+p1\displaystyle=\frac{\varepsilon_{n}}{a^{n+1}}c_{n}+\ldots+\frac{\varepsilon_{n+p-1}}{a^{n+p}}c_{n+p-1}
εnan+1V++εn+p1an+pV\displaystyle\in\frac{\varepsilon_{n}}{a^{n+1}}V+\ldots+\frac{\varepsilon_{n+p-1}}{a^{n+p}}V
εnan+1αU++εn+p1an+pαU=\displaystyle\subset\frac{\varepsilon_{n}}{a^{n+1}}\alpha U+\ldots+\frac{\varepsilon_{n+p-1}}{a^{n+p}}\alpha U=
=α(εnan+1U++εn+p1an+pU)=\displaystyle=\alpha\left(\frac{\varepsilon_{n}}{a^{n+1}}U+\ldots+\frac{\varepsilon_{n+p-1}}{a^{n+p}}U\right)=
=α(εn|a|n+1U++εn+p1|a|n+pU)\displaystyle=\alpha\left(\frac{\varepsilon_{n}}{\left|a\right|^{n+1}}U+\ldots+\frac{\varepsilon_{n+p-1}}{\left|a\right|^{n+p}}U\right)
=α(εn|a|n+1++εn+p1|a|n+p)U\displaystyle=\alpha\left(\frac{\varepsilon_{n}}{\left|a\right|^{n+1}}+\ldots+\frac{\varepsilon_{n+p-1}}{\left|a\right|^{n+p}}\right)U

Since the series n=1εn1|a|n\sum\nolimits_{n=1}^{\infty}\dfrac{\varepsilon_{n-1}}{\left|a\right|^{n}} is convergent its remainder of order nn tends to zero as nn\rightarrow\infty, therefore there exists n0n_{0}\in\mathbb{N} such that

εn|a|n+1++εn+p1|a|n+p<1α,nn0,p.\dfrac{\varepsilon_{n}}{\left|a\right|^{n+1}}+\ldots+\dfrac{\varepsilon_{n+p-1}}{\left|a\right|^{n+p}}<\dfrac{1}{\alpha},\,\,n\geq n_{0},\ p\in\mathbb{N}.

Then

α(εn|a|n+1++εn+p1|a|n+p)UUW,nn0,p,\alpha\left(\dfrac{\varepsilon_{n}}{\left|a\right|^{n+1}}+\ldots+\dfrac{\varepsilon_{n+p-1}}{\left|a\right|^{n+p}}\right)U\subseteq U\subseteq W,\,\,n\geq n_{0},\ p\in\mathbb{N},

hence sn+psnW,s_{n+p}-s_{n}\in W, for nn0n\geq n_{0} and p.p\in\mathbb{N}. We conclude that (sn)n0(s_{n})_{n\geq 0} is a Cauchy sequence, thus it is convergent. Now let

k=1εk1ck1ak=s,sX.\sum\limits_{k=1}^{\infty}\dfrac{\varepsilon_{k-1}c_{k-1}}{a^{k}}=s,\ s\in X.

Define the sequence (yn)n0(y_{n})_{n\geq 0} by the relation

yn+1=ayn+bn,n0,y0=x0+s.y_{n+1}=ay_{n}+b_{n},\ n\geq 0,\ \,y_{0}=x_{0}+s.

Then, according to Lemma 2.1

yn=an(y0+k=1nbk1ak),n1y_{n}=a^{n}\left(y_{0}+\sum\limits_{k=1}^{n}\dfrac{b_{k-1}}{{a^{k}}}\right),\ n\geq 1

and

xnyn\displaystyle x_{n}-y_{n} =an(s+k=1nεk1ck1ak)=ank=n+1εk1ck1ak\displaystyle=a^{n}\left(-s+\sum\limits_{k=1}^{n}\dfrac{\varepsilon_{k-1}c_{k-1}}{a^{k}}\right)=-a^{n}\sum\limits_{k=n+1}^{\infty}\dfrac{\varepsilon_{k-1}c_{k-1}}{a^{k}}
=k=n+1εk1ck1akn=m=1εm+n1cm+n1am\displaystyle=-\sum\limits_{k=n+1}^{\infty}\dfrac{\varepsilon_{k-1}c_{k-1}}{a^{k-n}}=-\sum\limits_{m=1}^{\infty}\dfrac{\varepsilon_{m+n-1}c_{m+n-1}}{a^{m}}
m=1εm+n1amV=m=1εm+n1|a|mV\displaystyle\in-\sum\limits_{m=1}^{\infty}\dfrac{\varepsilon_{m+n-1}}{a^{m}}V=\sum\limits_{m=1}^{\infty}\dfrac{\varepsilon_{m+n-1}}{\left|a\right|^{m}}V
(m=1εm+n1|a|m)V.\displaystyle\subseteq\left(\sum\limits_{m=1}^{\infty}\dfrac{\varepsilon_{m+n-1}}{\left|a\right|^{m}}\right)V.

The existence is proved.

Uniqueness. Suppose that for a sequence (xn)n0(x_{n})_{n\geq 0} satisfying (2.2) there exist two sequences (yn)n0(y_{n})_{n\geq 0} and (zn)n0(z_{n})_{n\geq 0} in XX satisfying recurrences yn+1=ayn+bn,y_{n+1}=ay_{n}+b_{n}, zn+1=azn+bn,n0,z_{n+1}=az_{n}+b_{n},\ n\geq 0, such that

xnyn(m=1εm+n1|a|m)V,n0,x_{n}-y_{n}\in\left(\sum\limits_{m=1}^{\infty}\dfrac{\varepsilon_{m+n-1}}{\left|a\right|^{m}}\right)V,\ n\geq 0,
xnzn(m=1εm+n1|a|m)V,n0.x_{n}-z_{n}\in\left(\sum\limits_{m=1}^{\infty}\dfrac{\varepsilon_{m+n-1}}{\left|a\right|^{m}}\right)V,\ n\geq 0.

Let

Mn:=(m=1εm+n1|a|m)V.M_{n}:=\left(\sum\limits_{m=1}^{\infty}\dfrac{\varepsilon_{m+n-1}}{\left|a\right|^{m}}\right)V.

Then

znynMnMn,n0.z_{n}-y_{n}\in M_{n}-M_{n},\,\,n\geq 0.

Taking account of Lemma 2.1 it follows

an(z0y0)MnMn,n0,a^{n}(z_{0}-y_{0})\in M_{n}-M_{n},\ n\geq 0,

or

(2.4) z0y01anMn1anMn,n0.z_{0}-y_{0}\in\dfrac{1}{a^{n}}M_{n}-\dfrac{1}{a^{n}}M_{n},\ n\geq 0.

On the other hand

1anMn=m=1(εm+n1an|a|m)V,n0.\dfrac{1}{a^{n}}M_{n}=\sum\limits_{m=1}^{\infty}\left(\dfrac{\varepsilon_{m+n-1}}{a^{n}\left|a\right|^{m}}\right)V,\ n\geq 0.

The relation

|m=1εm+n1an|a|m|m=1εm+n1|a|m+n,n0,\left|\sum\limits_{m=1}^{\infty}\dfrac{\varepsilon_{m+n-1}}{a^{n}\left|a\right|^{m}}\right|\leq\sum\limits_{m=1}^{\infty}\dfrac{\varepsilon_{m+n-1}}{\left|a\right|^{m+n}},\ n\geq 0,

leads to

limnm=1εm+n1an|a|m=0\lim_{n\rightarrow\infty}\sum\limits_{m=1}^{\infty}\dfrac{\varepsilon_{m+n-1}}{a^{n}\left|a\right|^{m}}=0

since Rn=m=1εm+n1|a|m+n,n0,R_{n}=\sum\limits_{m=1}^{\infty}\dfrac{\varepsilon_{m+n-1}}{\left|a\right|^{m+n}},\ n\geq 0, is the remainder of order nn of the convergent series n=1εn1|a|n.\sum\limits_{n=1}^{\infty}\dfrac{\varepsilon_{n-1}}{\left|a\right|^{n}}. Now letting nn\rightarrow\infty in (2.4) and taking account of the boundedness of VV it follows y0=z0,y_{0}=z_{0}, hence yn=zn,n0.y_{n}=z_{n},\forall n\geq 0.

As a consequence it follows a result on classical Ulam stability for the equation (2.1)(see [11]).

Corollary 2.3.

Let a,|a|1,ε>0.a\in\mathbb{C},\ \left|a\right|\neq 1,\ \varepsilon>0. Then for every (xn)n0(x_{n})_{n\geq 0} in XX such that

xn+1axnbnεV,n0,x_{n+1}-ax_{n}-b_{n}\in\varepsilon V,\ n\geq 0,

there exists (yn)n0(y_{n})_{n\geq 0} in X,X, yn+1=ayn+bn,n0y_{n+1}=ay_{n}+b_{n},\ n\geq 0

xnynε||a|1|V,n0.x_{n}-y_{n}\in\dfrac{\varepsilon}{\left|\left|a\right|-1\right|}V,\ n\geq 0.

Moreover if |a|>1,\left|a\right|>1, then (yn)n0(y_{n})_{n\geq 0} is unique.

Proof.

If |a|>1,\left|a\right|>1, the result is a consequence of Theorem 2.2 for εn=ε,n0.\varepsilon_{n}=\varepsilon,\ n\geq 0. Let now |a|<1\left|a\right|<1 and (xn)n1(x_{n})_{n\geq 1} satisfying xn+1axnbnεV,n0x_{n+1}-ax_{n}-b_{n}\in\varepsilon V,\ n\geq 0. Then there exists cnV,n0c_{n}\in V,\ n\geq 0 such that

xn+1=axn+bn+εcn,n0,x_{n+1}=ax_{n}+b_{n}+\varepsilon c_{n},\ n\geq 0,

and, according to Lemma 2.1 we get

xn=anx0+k=1nank(bk1+εck1),n1.x_{n}=a^{n}x_{0}+\sum\limits_{k=1}^{n}a^{n-k}\left(b_{k-1}+\varepsilon c_{k-1}\right),\ n\geq 1.

Take (yn)n1,yn+1=ayn+bn,n0,y0=x0(y_{n})_{n\geq 1},\ y_{n+1}=ay_{n}+b_{n},\ n\geq 0,\ y_{0}=x_{0}. Then

yn=anx0+k=1nankbk1,n1.y_{n}=a^{n}x_{0}+\sum\limits_{k=1}^{n}a^{n-k}b_{k-1},\ n\geq 1.

We get

xnyn\displaystyle x_{n}-y_{n} =εk=1nankck1εk=1nankV\displaystyle=\varepsilon\sum\limits_{k=1}^{n}a^{n-k}c_{k-1}\in\varepsilon\sum\limits_{k=1}^{n}a^{n-k}V
=εk=1n|a|nkV=ε(k=1n|a|nk)V\displaystyle=\varepsilon\sum\limits_{k=1}^{n}\left|a\right|^{n-k}V=\varepsilon\left(\sum\limits_{k=1}^{n}\left|a\right|^{n-k}\right)V
=ε1|a|n1|a|Vε||a|1|V,n0.\displaystyle=\varepsilon\dfrac{1-\left|a\right|^{n}}{1-\left|a\right|}V\subseteq\dfrac{\varepsilon}{\left|\left|a\right|-1\right|}V,\ n\geq 0.

In the next theorem we obtain the best Ulam constant for the equation (2.1).

Theorem 2.4.

If VV is closed and a,|a|1a\in\mathbb{C},\ \left|a\right|\neq 1, then the best Ulam constant of the equation (2.1) is

KR=1||a|1|.K_{R}=\dfrac{1}{\left|\left|a\right|-1\right|}.
Proof.

Suppose that equation (2.1) admits an Ulam constant K<KR.K<K_{R}.

Let ε>0\varepsilon>0, uVKKRV.\ u\in V\setminus\frac{K}{K_{R}}V. Let (xn)n0(x_{n})_{n\geq 0} be given by

xn+1axnbn=εan+1|a|n+1u,n0.x_{n+1}-ax_{n}-b_{n}=\varepsilon\dfrac{a^{n+1}}{\left|a\right|^{n+1}}u,\ n\geq 0.

Since VV is a balanced set, it follows that εan+1|a|n+1uεV\varepsilon\dfrac{a^{n+1}}{\left|a\right|^{n+1}}u\in\varepsilon V. Then xn+1axnbnεV,n0,x_{n+1}-ax_{n}-b_{n}\in\varepsilon V,\ n\geq 0, so there exists (yn)n0(y_{n})_{n\geq 0} satisfying yn+1=ayn+bn,n0,y_{n+1}=ay_{n}+b_{n},\ n\geq 0, with the property

(2.5) xnynKεV,n0,x_{n}-y_{n}\in K\varepsilon V,\ n\geq 0,

in view of the supposition of stability with the Ulam constant K.K. From Lemma 2.1 we obtain

xn\displaystyle x_{n} =an(x0+k=1nbk1+εuak|a|kak),n1\displaystyle=a^{n}\left(x_{0}+\sum\limits_{k=1}^{n}\frac{b_{k-1}+\varepsilon u\frac{a^{k}}{\left|a\right|^{k}}}{a^{k}}\right),n\geq 1
yn\displaystyle y_{n} =an(y0+k=1nbk1ak),n1.\displaystyle=a^{n}\left(y_{0}+\sum\limits_{k=1}^{n}\frac{b_{k-1}}{a^{k}}\right),n\geq 1.

Then

xnyn\displaystyle x_{n}-y_{n} =an(x0y0+εk=1n1|a|ku)\displaystyle=a^{n}\left(x_{0}-y_{0}+\varepsilon\sum\limits_{k=1}^{n}\frac{1}{\left|a\right|^{k}}u\right)
=an(x0y0+εu|a|n1|a|n(|a|1)),n1.\displaystyle=a^{n}\left(x_{0}-y_{0}+\varepsilon u\frac{\left|a\right|^{n}-1}{\left|a\right|^{n}(\left|a\right|-1)}\right),n\geq 1.

10.1^{0}.\ Let |a|>1\left|a\right|>1. We have

limn(x0y0+εu|a|n1|a|n(|a|1))=x0y0+εu|a|1:=lX.\underset{n\rightarrow\infty}{\lim}\left(x_{0}-y_{0}+\varepsilon u\frac{\left|a\right|^{n}-1}{\left|a\right|^{n}(\left|a\right|-1)}\right)=x_{0}-y_{0}+\varepsilon\frac{u}{\left|a\right|-1}:=l\in X.

If l0l\neq 0 it follows that (xnyn)n0\left(x_{n}-y_{n}\right)_{n\geq 0} is unbounded, so xnynKεVx_{n}-y_{n}\notin K\varepsilon V.

If l=0,l=0, then we get

x0y0=εu|a|1,x_{0}-y_{0}=-\dfrac{\varepsilon u}{\left|a\right|-1},

therefore

xnyn\displaystyle x_{n}-y_{n} =ε1|a|1(an|a|nu)\displaystyle=\varepsilon\dfrac{1}{\left|a\right|-1}\left(-\frac{a^{n}}{\left|a\right|^{n}}u\right)
=εKR(an|a|n)u,n0.\displaystyle=\varepsilon K_{R}\left(-\frac{a^{n}}{\left|a\right|^{n}}\right)u,\ n\geq 0.

Then, according to (2.5), we get

uKKR(|a|nan)V=KKRV,u\in\frac{K}{K_{R}}\left(-\frac{\left|a\right|^{n}}{a^{n}}\right)V=\frac{K}{K_{R}}V,

contradiction with uVKKRV.u\in V\setminus\frac{K}{K_{R}}V.

20.2^{0}.\ Let |a|<1\left|a\right|<1. Then

xnyn=an(x0y0)+εu1|a|n1|a|an|a|n.x_{n}-y_{n}=a^{n}(x_{0}-y_{0})+\varepsilon u\frac{1-\left|a\right|^{n}}{1-\left|a\right|}\frac{a^{n}}{\left|a\right|^{n}}.

From (2.5) we obtain

ε1|a|n1|a|an|a|nuεKVan(x0y0),n0,\varepsilon\frac{1-\left|a\right|^{n}}{1-\left|a\right|}\frac{a^{n}}{\left|a\right|^{n}}u\in\varepsilon KV-a^{n}(x_{0}-y_{0}),\ n\geq 0,

so

uKKR11|a|n|a|nanV1εKR|a|n1|a|n(x0y0).u\in\frac{K}{K_{R}}\frac{1}{1-\left|a\right|^{n}}\frac{\left|a\right|^{n}}{a^{n}}V-\frac{1}{\varepsilon K_{R}}\frac{\left|a\right|^{n}}{1-\left|a\right|^{n}}(x_{0}-y_{0}).

Since |a|nanV=V\frac{\left|a\right|^{n}}{a^{n}}V=V we have

uKKR11|a|nV1εKR|a|n1|a|n(x0y0).u\in\frac{K}{K_{R}}\frac{1}{1-\left|a\right|^{n}}V-\frac{1}{\varepsilon K_{R}}\frac{\left|a\right|^{n}}{1-\left|a\right|^{n}}(x_{0}-y_{0}).

Letting nn\rightarrow\infty it follows

uKKRV¯=KKRV,u\in\frac{K}{K_{R}}\overline{V}=\frac{K}{K_{R}}V,

contradiction with uVKKRVu\in V\setminus\frac{K}{K_{R}}V. The theorem is proved. ∎

3. Ulam stability of a pp order linear difference equation

In this section we give stability and nonstability results for the linear difference equation (1.3).

Theorem 3.1.

Suppose that the characteristic equation (1.6) admits the roots r1,,rpr_{1},\ldots,r_{p}, |rk|1, 1kp\left|r_{k}\right|\neq 1,\ 1\leq k\leq p. Then for every ε>0\varepsilon>0 and every sequence (xn)n0(x_{n})_{n\geq 0} in XX satisfying

xn+pa1xn+p1apxnεV,n0,x_{n+p}-a_{1}x_{n+p-1}-\ldots-a_{p}x_{n}\in\varepsilon V,\ n\geq 0,

there exists a sequence (yn)n0(y_{n})_{n\geq 0} in XX with the properties

yn+p\displaystyle y_{n+p} =a1yn+p1++apyn,n0,\displaystyle=a_{1}y_{n+p-1}+\ldots+a_{p}y_{n},\ n\geq 0,
xnyn\displaystyle x_{n}-y_{n} ε|k=1p(|rk|1)|V.\displaystyle\in\frac{\varepsilon}{\left|\prod\limits_{k=1}^{p}\left(\left|r_{k}\right|-1\right)\right|}V.
Proof.

We prove the theorem by induction on pp. For p=1p=1, the conclusion of Theorem is true in virtue of Theorem 2.4. Suppose now that the Theorem holds for a fixed p1p\geq 1. We have to prove the following:

Suppose that the characteristic equation

(3.1) rp+1=a1rp++apr+ap+1r^{p+1}=a_{1}r^{p}+\ldots+a_{p}r+a_{p+1}

admits the roots r1,,rp+1r_{1},\ldots,r_{p+1} and |rk|1, 1kp+1\left|r_{k}\right|\neq 1,\ 1\leq k\leq p+1. Then for every sequence (xn)n0(x_{n})_{n\geq 0} in XX satisfying

(3.2) xn+p+1a1xn+papxn+1ap+1xnεV,n0,x_{n+p+1}-a_{1}x_{n+p}-\ldots-a_{p}x_{n+1}-a_{p+1}x_{n}\in\varepsilon V,\ n\geq 0,

there exists a sequence (yn)n0(y_{n})_{n\geq 0} in XX with the properties

yn+p+1\displaystyle y_{n+p+1} =a1yn+p++apyn+1+ap+1yn,n0,\displaystyle=a_{1}y_{n+p}+\ldots+a_{p}y_{n+1}+a_{p+1}y_{n},\ n\geq 0,
xnyn\displaystyle x_{n}-y_{n} ε|(|r1|1)(|rp+1|1)|V.\displaystyle\in\frac{\varepsilon}{\left|\left(\left|r_{1}\right|-1\right)\cdots\left(\left|r_{p+1}\right|-1\right)\right|}V.

We rewrite the relation (3.2) in the following form

(3.3) xn+p+1(r1++rp+1)xn+p+(1)p+1r1rp+1xnεV,n0.x_{n+p+1}-(r_{1}+\ldots+r_{p+1})x_{n+p}-\ldots+(-1)^{p+1}r_{1}\cdots r_{p+1}x_{n}\in\varepsilon V,\ n\geq 0.

Denoting zn:=z_{n}:= xn+1rp+1xn,n0x_{n+1}-r_{p+1}x_{n},\ n\geq 0 we obtain from (3.3)

(3.4) zn+p(r1++rp)zn+p1++(1)pr1rpznεV,n0.z_{n+p}-(r_{1}+\ldots+r_{p})z_{n+p-1}+\ldots+(-1)^{p}r_{1}\cdots r_{p}z_{n}\in\varepsilon V,\ n\geq 0.

By using the induction hypothesis, it follows that there exists a sequence (wn)n0(w_{n})_{n\geq 0} in XX with the properties

(3.5) wn+p=a1wn+p1++apwn,n0w_{n+p}=a_{1}w_{n+p-1}+\ldots+a_{p}w_{n},\ n\geq 0
znwnε|(|r1|1)(|rp|1)|V,n0.z_{n}-w_{n}\in\frac{\varepsilon}{\left|\left(\left|r_{1}\right|-1\right)\cdots\left(\left|r_{p}\right|-1\right)\right|}V,\ n\geq 0.

Hence

(3.6) xn+1rp+1xnwnε|(|r1|1)(|rp|1)|V,n0,x_{n+1}-r_{p+1}x_{n}-w_{n}\in\frac{\varepsilon}{\left|\left(\left|r_{1}\right|-1\right)\cdots\left(\left|r_{p}\right|-1\right)\right|}V,\ n\geq 0,

and in view of Theorem 2.4, it follows from (3.6) that there exists a sequence (yn)n0(y_{n})_{n\geq 0} in XX, given by the recurrence

(3.7) yn+1=rp+1yn+wn,n0y_{n+1}=r_{p+1}y_{n}+w_{n},\ n\geq 0

that satisfies the relation

xnynε|(|r1|1)(|rp+1|1)|V.x_{n}-y_{n}\in\frac{\varepsilon}{\left|\left(\left|r_{1}\right|-1\right)\cdots\left(\left|r_{p+1}\right|-1\right)\right|}V.

By (3.7) we get

wn=yn+1rp+1yn,n0,w_{n}=y_{n+1}-r_{p+1}y_{n},\ n\geq 0,

and replacing in (3.5) we get

yn+p+1=a1yn+p++ap+1yn,n0.y_{n+p+1}=a_{1}y_{n+p}+\ldots+a_{p+1}y_{n},\ n\geq 0.

The theorem is proved. ∎

To prove the nonstability result for the equation (1.3) we need the following Lemma, concerning the nonstability of the linear difference equation of order one.

Lemma 3.2.

Suppose that aa\in\mathbb{C} and|a|=1\ \left|a\right|=1 and let ε>0.\varepsilon>0. Then for each x0Xx_{0}\in X there exists a sequence (xn)n0(x_{n})_{n\geq 0} satisfying the relation

xn+1axnbnεV,n0,x_{n+1}-ax_{n}-b_{n}\in\varepsilon V,\ n\geq 0,

such that for every sequence (yn)n0 (y_{n})_{n\geq 0\text{ }} in X,X, given by yn+1aynbn=0y_{n+1}-ay_{n}-b_{n}=0, the sequence (xnyn)n0\left(x_{n}-y_{n}\right)_{n\geq 0} is unbounded.

Proof.

We fix vV,v\in V, v0Xv\neq 0_{X} and take (xn)n0\left(x_{n}\right)_{n\geq 0} such that xn+1axnbn=εvan+1,n0x_{n+1}-ax_{n}-b_{n}=\varepsilon va^{n+1},\ n\geq 0. Then according to Lemma 2.1,

xn=anx0+k=1nank(bk1+εvak).x_{n}=a^{n}x_{0}+\sum\limits_{k=1}^{n}a^{n-k}(b_{k-1}+\varepsilon va^{k}).

Let (yn)n0 (y_{n})_{n\geq 0\text{ }} be an arbitrary sequence satisfying yn+1aynbn=0,n0y_{n+1}-ay_{n}-b_{n}=0,\ n\geq 0. We have

yn=any0+k=1nankbk1,n0.y_{n}=a^{n}y_{0}+\sum\limits_{k=1}^{n}a^{n-k}b_{k-1},\ n\geq 0.

Then

xnyn\displaystyle x_{n}-y_{n} =an(x0y0)+εk=1nank(vak)=\displaystyle=a^{n}(x_{0}-y_{0})+\varepsilon\sum\limits_{k=1}^{n}a^{n-k}\left(va^{k}\right)=
=an(x0y0)+εk=1nvan\displaystyle=a^{n}(x_{0}-y_{0})+\varepsilon\sum\limits_{k=1}^{n}va^{n}
=an(x0y0)+εnvan\displaystyle=a^{n}(x_{0}-y_{0})+\varepsilon nva^{n}
=an(x0y0+εnv).\displaystyle=a^{n}(x_{0}-y_{0}+\varepsilon nv).

So it is easy to see that (xnyn)n0\left(x_{n}-y_{n}\right)_{n\geq 0} is unbounded.

We are able now to give the nonstability result for the equation (1.3).

Theorem 3.3.

Suppose that the characteristic equation (1.6) admits the roots r1,,rp, 1kpr_{1},\ldots,r_{p}\in\mathbb{C},\ 1\leq k\leq p and at least one of them is of modulus 11. Then there exists a sequence (xn)n0(x_{n})_{n\geq 0} satisfying

xn+pa1xn+p1apxnεV,n0,x_{n+p}-a_{1}x_{n+p-1}-\ldots-a_{p}x_{n}\in\varepsilon V,\ n\geq 0,

for some ε>0\varepsilon>0, such that for every sequence (yn)n0(y_{n})_{n\geq 0} fulfilling the linear difference equation (1.3), it follows that (xnyn)n0\left(x_{n}-y_{n}\right)_{n\geq 0} is unbounded.

Proof.

For p=1p=1 the conclusion of Theorem 3.3 is true in virtue of Lemma 3.2.

For p2,p\geq 2, we assume that |r1|=1.\left|r_{1}\right|=1. From Lemma 3.2 it follows that there exists a sequence (x¯n)n0(\overline{x}_{n})_{n\geq 0} in XX satisfying the relation

(3.8) x¯n+1r1x¯nεV,\overline{x}_{n+1}-r_{1}\overline{x}_{n}\in\varepsilon V,

such that for every sequence (y¯n)n0(\overline{y}_{n})_{n\geq 0} with

(3.9) y¯n+1=r1y¯n,n0,\overline{y}_{n+1}=r_{1}\overline{y}_{n},\ n\geq 0,

the sequence (x¯ny¯n)n0\left(\overline{x}_{n}-\overline{y}_{n}\right)_{n\geq 0}\ is unbounded.

Further there exists a sequence (xn)n0(x_{n})_{n\geq 0} in XX such that

(3.10) xn+p1(r2++rp)xn+p2++(1)p1r2rpxn=x¯n,n0x_{n+p-1}-(r_{2}+\ldots+r_{p})x_{n+p-2}+\ldots+(-1)^{p-1}r_{2}\cdots r_{p}x_{n}=\overline{x}_{n},\ n\geq 0

(it suffices to take x0==xp2=0,xp1=x¯0x_{0}=\cdots=x_{p-2}=0,x_{p-1}=\overline{x}_{0}, and (xn)n0(x_{n})_{n\geq 0} can be determined step by step).

Then (3.8) implies that the sequence (xn)n0(x_{n})_{n\geq 0} satisfies the relation

(3.11) xn+p+(1)(r1++rp)xn+p1++(1)pr1rpxnεV,n0,x_{n+p}+(-1)(r_{1}+\ldots+r_{p})x_{n+p-1}+\ldots+(-1)^{p}r_{1}\cdots r_{p}x_{n}\in\varepsilon V,\ n\geq 0,

which is equivalent to (1.4). Let now (yn)n0(y_{n})_{n\geq 0} be an arbitrary sequence defined by (1.3) and (y¯n)n0(\overline{y}_{n})_{n\geq 0} be the sequence given by

(3.12) y¯n=yn+p1+(1)(r2++rp)yn+p2++(1)p1r2rpyn,n0.\overline{y}_{n}=y_{n+p-1}+(-1)(r_{2}+\ldots+r_{p})y_{n+p-2}+\ldots+(-1)^{p-1}r_{2}\cdots r_{p}y_{n},\ n\geq 0.

Then (3.9) hold and (x¯ny¯n)n0(\overline{x}_{n}-\overline{y}_{n})_{n\geq 0} is unbounded.

In order to prove that (xnyn)n0\left(x_{n}-y_{n}\right)_{n\geq 0} is unbounded we suppose the contrary, i.e., (xnyn)n0\left(x_{n}-y_{n}\right)_{n\geq 0} is bounded.

From (3.10) and (3.11) it follows that

x¯ny¯n\displaystyle\overline{x}_{n}-\overline{y}_{n} =(xn+p1yn+p1)(r2++rp)(xn+p2yn+p2)++\displaystyle=\left(x_{n+p-1}-y_{n+p-1}\right)-(r_{2}+\ldots+r_{p})\left(x_{n+p-2}-y_{n+p-2}\right)+\ldots+
+r2rp(xnyn).\displaystyle+r_{2}\cdots r_{p}(x_{n}-y_{n}).

Then (x¯ny¯n)n0\left(\overline{x}_{n}-\overline{y}_{n}\right)_{n\geq 0} is bounded as a finite sum of bounded sequences. This is a contradiction with the unboundedness of (x¯ny¯n)n0\left(\overline{x}_{n}-\overline{y}_{n}\right)_{n\geq 0}\ . ∎

References

  • [1] D.R. Anderson, M. Onitsuka, Hyers-Ulam stability for a discrete time scale with two step sizes, Appl. Math. Comput., 344-345 (2019), 128-140.
  • [2] A.R. Baias, F. Blaga, D. Popa, Best Ulam constant for a linear difference equation, Carpatian J. Math., 35 (2019), No. 1, 13-21.
  • [3] N. Brillouët-Belluot, J. Brzdȩk, K. Ciepliński, On Some Recent Developments in Ulam’s Type Stability, Abstract and Applied Analysis Volume 2012, Article ID 716936, 41 pages.
  • [4] J. Brzdȩk, D. Popa, I. Raşa, B. Xu, Ulam stability of operators, Academic Press, 2018.
  • [5] J. Brzdȩk, P. Wojcek, On approximate solutions of some difference equation, Bull. Aust. Math. Soc., 95(3) 2017, 476-481.
  • [6] J. Brzdȩk, D. Popa, B. Xu, The Hyers-Ulam stability of the nonlinear recurrence, J. Math. Anal. Appl., 335 (2007), 443-449.
  • [7] J. Brzdȩk, D. Popa, B. Xu, Note on the nonstability of the nonlinear recurrence, Abh. Math. Sem. Univ. Hamburg 76 (2006), 183-189.
  • [8] J. Brzdȩk, D. Popa, B. Xu, On nonstability of the linear recurrence of order one, J. Math. Anal. Appl., 367 (2010), 146-153.
  • [9] D.H. Hyers, On the stability of the linear functional equations, Proc. Nat, Acad. Sci., USA, 27 (1941), 222-224.
  • [10] G.J.O. Jameson, Convex series, Proc. Camb. Phil. Soc. (1972), 37-47.
  • [11] M.S. Moslehian, D. Popa, On the stability of the first-order linear recurrence in topological vector spaces, Nonlinear Anal., 73 (2010), 2792-2799.
  • [12] M. Onitsuka, Influence of the stepsize on Hyers-Ulam stability of first order homogeneous linear difference equations, Int. J. Differ. Equ., 12(2) (2017), 281-302.
  • [13] G. Polya, G. Szegö, Aufgaben und Lehrsatze aus der Analysis I, Julius Springer, Berlin, 1925.
  • [14] D. Popa, Hyers-Ulam stability of the linear recurrence with constant coefficient, Adv. Differ. Equ.-NY (2005), 101-107.
  • [15] D. Popa, Hyers-Ulam-Rassias stability of the linear recurrence, J. Math. Anal. Appl., 309 (2005), 591-597.
  • [16] D. Popa, I. Raşa, Best constant in stability of some positive linear operators, Aequationes Math., 90 (2016), 719-726.
  • [17] W. Rudin, Functional Analysis, 2nd ed., in: International Series in Pure and Applied Mathematics, McGraw-Hill Inc., New York, 1991.
  • [18] S.M. Ulam, Problems in Modern Mathematics, Wiley, New York, 1964.
2021

Related Posts