+ All Categories
Home > Documents > nieto2006.pdf

nieto2006.pdf

Date post: 12-Jan-2016
Category:
Upload: ho-nhat-nam
View: 214 times
Download: 1 times
Share this document with a friend
Popular Tags:
25
International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems Vol. 14, No. 6 (2006) 687-709 © World Scientific Publishing Company LINEAR FIRST-ORDER FUZZY DIFFERENTIAL EQUATIONS JUAN J. NIETO Departamento de Andlisis Matemdtico, Facultad de Matemdticas, Universidad de Santiago de Compostela, Santiago de Compostela, 15782, Spain [email protected] ROSANA RODRIGUEZ-LOPEZ Departamento de Andlisis Matemdtico, Facultad de Matemdticas, Universidad de Santiago de Compostela, Santiago de Compostela, 15782, Spain amrosana@usc. es DANIEL FRANCO Departamento de Matemdtica Aplicada I, Escuela Tecnica Superior de Ingenieros Industrials, Universidad Nacional de Educacion a Distancia, Apartado de Correos 60149, Madrid, 28080, Spain. dfranco @ind. uned. es Received 14 December 2004 Revised 30 October 2006 We give the expression for the solution to some particular initial value problems in the space E 1 of fuzzy subsets of K. We deduce some interesting properties of the diameter and the midpoint of the solution and compare the solutions with the corresponding ones in the crisp case. Keywords: Fuzzy sets, Fuzzy systems, Linear systems, Fuzzy differential equations, Fuzzy initial value problems. 1. Introduction Expression for the unique solution to the initial value problem for linear first-order ordinary differential equations is well known. For I a real interval, M G i a constant, a : I > R, and u : I > R, the two following equations are equivalent: v!(t) + Mu(t) = cr(t), u'(t) = -Mu(t)+a(t), ^ World Scientific www.worldscientific.com 687 Int. J. Unc. Fuzz. Knowl. Based Syst. 2006.14:687-709. Downloaded from www.worldscientific.com by UNIVERSITY OF HAIFA on 06/14/13. For personal use only.
Transcript

International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems Vol. 14, No. 6 (2006) 687-709 © World Scientific Publishing Company

L I N E A R F I R S T - O R D E R FUZZY D I F F E R E N T I A L E Q U A T I O N S

JUAN J. NIETO

Departamento de Andlisis Matemdtico, Facultad de Matemdticas,

Universidad de Santiago de Compostela, Santiago de Compostela, 15782, Spain

[email protected]

ROSANA RODRIGUEZ-LOPEZ

Departamento de Andlisis Matemdtico, Facultad de Matemdticas,

Universidad de Santiago de Compostela, Santiago de Compostela, 15782, Spain

amrosana@usc. es

DANIEL FRANCO

Departamento de Matemdtica Aplicada I, Escuela Tecnica Superior de Ingenieros Industrials,

Universidad Nacional de Educacion a Distancia, Apartado de Correos 60149, Madrid, 28080, Spain.

dfranco @ind. uned. es

Received 14 December 2004 Revised 30 October 2006

We give the expression for the solution to some particular initial value problems in the space E1 of fuzzy subsets of K. We deduce some interesting properties of the diameter and the midpoint of the solution and compare the solutions with the corresponding ones in the crisp case.

Keywords: Fuzzy sets, Fuzzy systems, Linear systems, Fuzzy differential equations, Fuzzy initial value problems.

1. In t roduct ion

Expression for the unique solution to the initial value problem for linear first-order ordinary differential equations is well known. For I a real interval, M G i a constant, a : I —> R, and u : I —> R, the two following equations are equivalent:

v!(t) + Mu(t) = cr(t),

u'(t) = -Mu(t)+a(t),

^ World Scientific www.worldscientific.com

687

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

688 J. J. Nieto, R. Rodriguez-Lopez & D. Franco

independently of the sign of the constant M. The solution to those equations for an initial condition u(0) = no, UQ £ R, (0 G I) , can be obtained as:

u(t) = uoe~Mt + / <r(s) eM(*"*) ds,tel,t> 0. Jo

For the fuzzy case, the corresponding equations are not equivalent, even in the particular case where a(t) = X{o}? f° r t £ I, and X{o} the characteristic function of {0}. Recall that, if x, y £ E1 are such that x-\-y = X{o}> then x, y are real numbers and y = — x, but x + (—x) = X{o} is n ° t necessarily true for x £ E1 (take, for instance, x = X[o,i])- Equation

u'(t)+Mu(t) = X{o}

implies that the solution is crisp and

u'{t) = -Mu(t),

but the converse is not true in general. Besides, the multiplication of x G E1 by a negative constant implies the inversion of the endpoints of the level sets, that is,

-l[x]a = -l[XahXar\ = \-Xar,-Xal], Va G [0 ,1] ,

and this fact produces that the expressions of the solutions for

ur(t)+u(t) = X{0},

and

u'{t) -u(t) = X{0},

are completely different. In this paper, and for M G M, M > 0, we analyze the existence of solution for

the initial value problems associated to the fuzzy equations

u'(t)+Mu(t) =cr(t), te I,

Uf(t) = -Mu(t) + cr(t), t G / ,

uf(t) =Mu{t) + a{t), t G J,

u\t) -Mu(t) =cr(t), te I,

and compare their solutions by calculating the midpoint and the diameter of their respective level sets. We point out that the study of midpoints for fuzzy sets has many applications in Artificial Intelligence (Ref. 1). In Refs. 2-5, the expression of the solution to analogous problems for a(t) = X{o>5 t £ I, and a fixed M is studied. Other references dealing with fuzzy equations or their integral equivalent formulation are Refs. 6-9, and, recently, Refs. 10-15, while the basic theory about fuzzy metric spaces is included in Ref. 4.

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

Linear First-Order Fuzzy Differential Equations 689

For M > 0, we calculate the expression of the solution to the initial value problems corresponding to equations

Uf(t) = -Mu(t) + cr(t), t G I,

u'{t) =Mu(t)+a(t), t e J,

whose existence and uniqueness can be deduced in the case of I bounded from the application of the fuzzy Picard-Lipschitz Theorem (Theorem 3.2.1 in Ref. 5, Theorem 13.2.1 in Ref. 4 and Ref. 8), since functions f,g : I x E1 —> E1 given, respectively, by

f(t,x) = -Mx + a(t),

and

g(t,x) = Mx + a(t),

are Lipschitzian in x and continuous in (£, x) for a continuous. Indeed, let x,y G E1

and t G J,

doo(/(*, *) , /(*» V)) = dU-Mx + d(t), -My + a{t))

= doo(-Mx,-My) =MdOQ(x1y)1

and analogously for g. As we show, existence of the solution for initial value problems relative to

u'(t)+Mu(t) = cr(t), tel,

and

u\t) -Mu(t) = cr(t), tel,

with M > 0, is subject to the verification of some conditions of compatibility involving the constant M, the function a and the initial data UQ. In fact, if we try to determine the existence and uniqueness of solution for

u'(t)+Mu(t) = cr(t), tel,

by using the fuzzy Picard-Lipschitz Theorem, we can write the equation in its equivalent expression

u'(t) = a(t) -H Mu(t), t G I,

whose right-hand side function /*(£, x) = a(t)—nMx would represent a Lipschitzian function in case it made sense, but the Hukuhara differences a(t) —H Mu(t) in the previous equation are not well-defined unless we can guarantee that

diam[a(t)]a > Mdiam[u(t)]a, Vt,Va,

and additional hypotheses are required in order to obtain that the level set Hukuhara differences define a fuzzy real number. This shows that it is not a trivial question and that further analysis should be made.

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

690 J. J. NietOf R. Rodriguez-Lopez & D. Franco

2. Expression of Solutions

2.1. uf{t) + Mu(t) = o-(t), t E I, M > 0.

Let M > 0, J a real interval i" = [0,T] with T > 0 or J = [0,+oo), cr G C^.E1), uo <E E1 and consider problem

V(£)+Mw(£) = cr(t), £ G J,

(1) w(0) = 1̂ 0-

Theorem 1. Problem (1) has a unique solution in I, given by

U{t) = U0X{e-Mt} + / C r(^)X{eM(s-*)} ds> t e ^ ( 2 ) JO

if, for each t £ I, there exists f3 > 0 such that the Hukuhara differences

u(t + h) —H u(t) and u(t) —H u(t — h)

exist, for all 0 < h < (3.

Proof. Taking

[u(t)}a = [u(t)ahu(t)ar}J

and

Ual(t) = u(t)aU Uar(t) = u(t)ar,

(1) is written levelwise as

u'cd(t) + Mual(t)=<ral(t), tel,

< r ( t ) + Muar(t) = <rar(t), t e I,

Ual(0) = (U0)ah ^a r (O) = (u0)ar.

Using an integrating factor, we get

uai(t) = (u0)aie-Mt + f aal(s)eM^-^ ds, Jo

and, analogously for uar(t), producing (2). Now we study the differentiability of u in the sense of Hukuhara. Let t G I, and h > 0, then, for every a G [0,1],

u(t + h)-Hu(t)\ -Mtfe~Mh-l ~~ [uojaie h

rt / „-Mh _ i \ p-Mh rt+h + J a(s)ale

M^ ds (° h 'J + ^ 1 <s)aleM^ ds,

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

Linear First-Order Fuzzy Differential Equations 691

and

u(t + ft) —H u(t) ft

(uo)are~ -Mt -Mh

ft / 0-Mh _ i \ p-Mh rt+ti

+ J a(s)areM^ ds {* h

Lj + t-j- J a(s)areM^ ds.

The limits of these functions as ft —• 0 + , uniformly in a are, respectively, z(t)ai = <r(t)ai - M(u0)aie~ -Mt •M [ a{s)ale

M^s-tUs1 Jo

and

z(t)ar = (J(t)ar ~ M{u0)are-Mt - M f a ( s ) a r e M ^ ds, Jo

since (txo)a/> (uo)ary a r e bounded uniformly in a G [0,1], cr^ai, cr(s)arj are bounded on [0,£] uniformly in a (a is bounded in the compact [0,£] by continuity),

»-Mh _ 1 + Mh

lim ft

0,

and i pt+h

lim - / a(s)aleM^s-tUs = a(t)ah

h^o+ ft Jt

lim - / a(s)areM^s-tUs = a(t)ar,

h^o+ ft Jt

uniformly in a by continuity of a at t. The same behavior can be checked for the left-sided Hukuhara quotients

u(t) —H u(t — ft) ft

and

for ft > 0. This proves that

dH u(t + ft) —H u(t)

ft

U>(t) ~H u(t — ft) ft

[z(t)ahz(t)Q

as ft —>• 0+ , uniformly in a, so that

where, for £ G J, z(£) given levelwise by

[*(*)]°H*(*)aZ,*(*)ar],

is a fuzzy number since E1 is complete. We have proved that u'(t) = z(t) G .E1, for all t G J. It is easy to check that

u'(t) + Mu(t) = z(t) + Mu(t) = cr(t), t G i",

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

692 J. J. Nieto, R. Rodriguez-Lopez & D. Franco

and u(0) = uo, so that we obtain the solution to (1). •

Remark 1. If for t G / ,

diam([a(t)}a) > Me~Mt fdiam([u0}a) + / diam([a(s)]a)eMs dsj ,Va, (3)

and there exists (3 > 0 such that for 0 < /i < /?,

(«o)aze-M ' (e-M / l - 1) + /„' a(s)aZeM^-*) ds(e"Mfc - 1)

+ Jt a(s)aieM(8~(t+h" ds nondecreasing in a,

(u0)are-Mt(e-Mh - 1) + /„' a(s)a reM( s-*) d s ( e - M h - 1)

+ Jt a(s)areM(s~(t+h" ds nonincreasing in a,

(u0)ale-Mt(l - eMh) + J*~h a(s)aleM^ ds(l - eMh) ( }

+ ft-h a(s)aieM ^ ds nondecreasing in a,

(u0)are-Mt(l - eMh) + / J " a(s)areM^ ds(l - eMh)

+ Jt-/i c r(s)areM^s_ t^ ds nonincreasing in a,

then, if u is given by (2) and t G / , the Hukuhara differences

?/(£ + /i) — H u(t), u(t) —H u(t — h)

exist for 0 < h < /?. Indeed, for the Hukuhara differences of (2) to exist, it is necessary that

diam([u(t)]a) = diam([u0]a)e-Mt + / diam([a(5)] a)eM ( s- t ) ds,

Jo is a nondecreasing function in t. But this is a real differentiate function and, by

(3),

jdiam{[u{t))a) = diam([a(t)]a) - Me~Mt

diam([u0}a)+ f diam([a(s))a)eMs ds) > 0.

Thus, the level set Hukuhara differences exist. The intervals

[u(t + h)al- U(t)al, U(t ^h)^- U(t)ar],

[u(t)al ~ U>(t ~ h)ai, U(t)ar -U(t- h)ar],

define the level sets of a fuzzy set, for each t and 0 < h < j3:

• Using (3), we obtain that

U(t + h)al ~ U(t)al < U(t + h)ar ~ u(t)ar, Va,

u(t)ai - u(t - h)ai < u(t)ar - u(t - ft)or, Va,

• By (4), u(t + h)ai — u(t)ai, u(t)ai — u(t — h)ai, are nondecreasing functions in a and u(t + h)ar — u(t)ar, u{t)ar — u(t — h)ar are nonincreasing in a.

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

Linear First-Order Fuzzy Differential Equations 693

• Since u(t), u(t + /i), u(t — h) (h < t) are elements in E1, then

U(t + h)al ~ U(t)al, U(t + h)ar - U(t)ar,

U(t)al ~ U(t ~ h)au U{t)ar - u(t ~ ft)or,

are left-continuous in a.

R e m a r k 2. Condition (4) can be written as: for t G I, there exists f3 > 0 such that, for 0 < h < /?, and 0 < a < b < 1,

^ i frhMsfo - °(s)ai)eMs ds > (uo)bi - (uo)ai + J0 (o-(s)bi - cr(s)ai)eMs ds

e M h _ ! / t ( ^ ( s ) o r - C r ( s )6 r ) e M s d s

> (^o)ar - (^o)6r + JQ (a(s)ar ~ Cf(s)hr)eMs ds

!_e-Mh Jl_h{(j(s)u - (j(s)ai)eMs ds

> (uo)bl - (uo)al + Jo (a(S)bl ~ &(s)al)eMs ds

!_e-Mfe ft-h(a(s)ar ~ a(s)br)eMs ds

> Mar ~ Mbr + JQ (a(s)ar ~ Cf(s)br)eMs ds

R e m a r k 3. For x G E1, and a G [0,1], mp([x]a) denotes the midpoint of [x]a, that is, \{xai + xar). Note that n, the solution to (1), verifies that

diam([u(t)]a) = diam([u0]a)e-Mt + / ^am([cr(5)]a)eM ( s- t ) ds, t G I,

Jo and

(5)

mp(Kt)] a ) = mp( [^ 0 ] a )e - M t + / m p ( [ ( j ( s ) ] a ) e M ^ ^ s Jo

J((t*o)ai + Mar)e~Mt + \ j (a(s)al + a ( s ) a r ) e M ^ ) ds, £ G J. 2

R e m a r k 4. If <r = X{c}, then diam([cr(t)]a) = 0, for every t, a, and (3) is reduced to

Me~Mtdiam([u0]a) < 0, for all a, and all t,

that is,

^o = X{c /}-

In such a case, (4) is trivially true and the solution is crisp

u(t) = X{cfe-Mt + ^{l-e-Mt)}-

In the ordinary case, u$ G R, cr G C(I, R), conditions (3), (4) are valid, the Hukuhara differences always exist as ordinary differences and the unique solution is given by

U(t) = U0 e~Mt+ [ta(s)eM^s-tUs1teL Jo

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

694 J. J. Nieto, R. Rodriguez-Lopez & D. Franco

2.2. u'(t) = -Mu(t) + cr(t), t E I, M > 0.

Let M > 0, I = [0,T] with T > 0, or J = [0,+oo), a G ^ ( I , ^ 1 ) , u0 G £ \ and consider problem

u'(t) = -Mu(t) + cr(t), t G J, (6)

w(0) = u0.

Theorem 2. Problem (6) has a unique solution in I, given by the following expres­sion for all t G Ij a G [0,1],

eMt e-Mt

<*)ai = 2"tfi(*,a) + —2-U*(t>a)> (?)

eMt e-Mt <t)ar = —C/ i ( t , a ) + - ^ - E / 2 ( t , a ) , (8)

where

U1(t1a) = diam([u0]a) + [ diam([a(s)}a)e-

Ms ds, Jo

2 ( t , a ) = (w0)oZ + Mar + / (cr(5)aZ + S"0)ar ) Jo

[72(t, a) = (uo)nj + (Mn)ftr + / (o-(s)nj + cr(s)ftr) e s ds.

Proof. Take

[w(t)]° = [w(t)oZ,w(t)ar].

We obtain (Refs. 2, 3, 5) that the function w given, for t G I, a G [0,1], by

w(t)ai = - ^ i a m ( [ t / 0 ] a ) e M t + ± (K) a z + Mar) e~M\ (9)

w(*)ar = \diam{[u0}a)eMt + ^ ((n0)o/ + Mar) e~M\ (10)

is the solution to problem

wf(t) = -Mw(t), te I,

w(0) = u0.

If we calculate the solution to

v'(t) = -Mv(t) + cr(t), t G I,

^(0) = X{o},

then

(w + v)(0) = w(0) + v(0) = u0 + X{o} = ^o,

(11)

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

Linear First-Order Fuzzy Differential Equations 695

and, for t G I,

(w + v)'(t) = w'(t) + v'(t) = -Mw(t) + (-Mv(i)) + cr(t)

= -M(w(t) + v(t)) + cr(t) = -M(w + v)(t) + cr(t),

thus K; +1; is the solution to (6). We seek a solution to (11) of the type

At)ar)~\-eMte-^)\car(t))' (U)

for t £ I, a £ [0,1], such that

[v(0)}a = [v(0)ahv(0)ar} = {0}, Va,

that is, for a G [0,1],

/ 0 \ = / V ( 0 ) a A = / 1 l \ fcal(0)\ \0J \v{°)arj \~1 1 J \car(0) J '

which is a nonsingular homogeneous linear system, hence the unique solution is

cai(0)=0, c a r ( 0 ) = 0 , V a e [ 0 , l ] .

Now, for the expression

fv(t)al \ ( cal(t)eMt + car(t)e-Mt \

\v{t)ar) \-cal(t)eMt+car(t)e-Mt)> [L6)

to define an element in E1, it is necessary that cai(t) is a nonpositive function. Besides, given v by

[v(t)]a = [v(t)ahv(t)ar}1

where v(t)ai, v(t)ar are expressed in (13), for the existence of the Hukuhara differ­ences v(t + h) —H v(t) and v(t) —H v(t — /i), for h > 0 small enough, it is necessary that

diam([v(t)]a) = -cal(t)eMt + car(t)e-Mt - cal(t)e

Mt - c a r( t )e~M t = -2cal(t)eMt

is nondecreasing in t, (for instance, if cai(t) is a nonincreasing function in t). Note that, for t G I, a G [0,1],

Ht)]a = [cal(t)eMt + car(t) e~M\ -cal(t)e

Mt + car(t) e-M% (14)

Now, we calculate cai(t) and car(t) in order to obtain the solution of (11). Passing to the level sets, we obtain, for all a G [0,1], t G / ,

v,al(t) = -Mvar(t)+<jal(t),

v,ar(t) = -Mval(t) + aar(t),

and, using (14),

c'al(t)eMt + cal(t)MeMt + c'ar(t)e~Mt - Mcar{t)e~Mt

= Mcal(t)eMt - Mcar(t)e-Mt + aal(t),

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

696 J. J. Nieto, R. Rodriguez-Lopez & D. Franco

cfal(t)e

Mt - cal(t)MeMt + dar(t)e~Mt - Mcar(t)e-Mt

= -Mcal(t)eMt - Mcar(t)e-Mt + a a r ( t ) ,

that is.

which provides that

hence, integrating,

and

c'al(t)eMt+c'ar(t)e-Mt = aal(t),

-c'al(t)eMt+c'ar(t)e-Mt = aar(t),

c'al(t) = \e-Mt(aal(t)-aar(t)),

c'aAt) = \eMt {ciaiit) + aar(t))

1 /"' cal(t) = 2 / e M S (CT«*(S) ~ Var(s)) ds,

Cor(t) = \feMS Kl(s) + aar(S)) ds.

These calculations confirm the conditions

cai(t) < 0, for all t, a, nonincreasing function in t.

Taking into account (14), we get for t € I, and a G [0,1],

v(t)ai = - \ ( e-Msdiam([a(s)]a)ds eMt

* Jo + \ J eMS {aal{s) + aar{s)) ds e~Mt> (15) Mt l rt

v(t)ar = 7: e-Msdiam([a(s)}a)dse 2 Jo

+ l- J eMs (aal(s) + aar(s)) ds e~Mt. (16)

For checking that v(t) defines a fuzzy number, note that a is continuous, diam([(T(s)}a) and a(s)ar decrease in a and a(s)ai increases in a.

Now, we prove that v is differentiate in the sense of Hukuhara and that the derivative of v in the sense of Hukuhara at t is —Mv(t) + a(t). Note that it is trivially true that the levelset Hukuhara differences exist since

diam([v(t)]a) = -2cal(t)eMt = [ e-

Msdiam([a(s)]a) eMtds Jo

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

Linear First-Order Fuzzy Differential Equations 697

is nondecreasing in t. Moreover, the Hukuhara differences

v(t + h) -H v(t), v(t) -H v(t - h)

exist for h small enough. To prove this fact, let t £ I. Using (15) and (16), we get, for t £ I and a £ [0,1],

(v(t + h) -H v(t))al = \ f e-Msdiam([a(s)}a) ds eMt(l - eMh) ^ Jo

+ \ f eMs M*) + °ar(s)) ds e-M\e-Mh - 1) 1 JO

1 ft + h - / e-Msdiam([a(s)}a) ds eM^h^

i rt+h l-j eM°(aal(s)+aar(s))dse-M(t+h\

(v(t + h) -H v{t))ar = \ [ e-Msdiam([a(s)]a) ds eMt(eMh - 1) 2 Jo

+ \ j t eM° (aal(s) + aar(s)) ds e~M\e~Mh - 1)

1 ft+h =- I e-Msdiam([a(s)}a)dseM(t+V

1

^ It

>t+h

e-J(aal(s)+aar(s))dse-M^t+h\ Ms

which define the endpoints of the level sets for a fuzzy number. Indeed, condition

diam([v(t + h)}a) > diam([v(t)}a)

implies that

(v(t + K) -H V(t))ai = V(t + h)ai ~ v(t)ai

< v(t + h)ar - v(t)ar = (v(t + h) -H v(t))ar.

Besides, for a& —• a - ,

(V(t + h) -H V(t))aki —• (v(t + h) ~H v{t))aU

J art (v(t + h) -H v{t))akr —• (v(t + h) -H v(t))c

since the continuity of a implies that J0 <r(s)x{eRs} ds, Jt o-(s)x{eRs} ds are el­

ements in E1, for R = ± M , t £ I, h > 0. Finally, we have to prove that (v(t + h) —H v(t))ai is nondecreasing in a and (v(t + h) —H v(t))ar is nonincreasing in a. Take a, b £ [0,1], a < 6, then

(v(t + h) -H v(t))al < (v(t + h) -H v(t))u

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

698 J. J. Nieto, R. Rodriguez-Lopez & D. Franco

since

e-MseMt(l - eMh)(diam([a(s)]a) - diam([a(s)}b))

< eMse-Mt(e-Mh - l)(a(s)bl + a(s)br - a(s)al - a(s)ar),

for s G [0, t], and

- e-MseM^t+h\diam([a(s)}a) - diam([a(s)]b))

< eMse-M^t+h\a{s)bl + a(s)br - a(s)al - a{s)ar),

for s G [t, t + h]. Indeed, the first assertion is valid for s <t,

e2M(t- s) ( 1 _ e ^) (d i a m ( [ < T ( s ) ]a ) _ diam([a(s)]b))

< (1 - eMh)(diam([a(s)]a) - diam([a(s)]b))

< (e~Mh - l)(a(s)u + a{s)br - a{s)ai - a(s)ar),

due to

(2 _ eMk - e-Mh)(a(s)bl - a(s)al) < 0 < (e~Mh - eMh)(a(s)br - a(s)ar).

The second assertion is equivalent to

-e2M^t+h-s\diam([a(s)}a) - diam([a(s)]b)) < a(s)bl + a(s)br - a(s)al - a(s)ar,

which is trivially true for s G [t,t + h]. With a similar procedure, we achieve

(v(t + h)-Hv(t)) ar

is nonincreasing in a, and the same reasoning applies to the case of the Hukuhara differences v(t) —H v(t — h) with h > 0 small enough.

Now, for t e I,

71/r rt a\ J„ Mt

M f (-Mv(t) + a(t))ai = -— / e-Msdiam([a(s)]a)dse

* Jo M rf

y J eMs (aal(s) + aar(s)) ds e~Mt + aal(t),

M r (-Mv(t)+a(t))ar = — / e-Msdiam([a(s)}a)dseMt

2 Jo M rf

- Y J eMS Ms) + ^r(s)) ds e~Mt + aar{t).

Then,

lim dx (<* + h)-B<t) +

h^o+ V h

lim sup max{|(/9(t,/i, a)|, ^ ( t , / i , a) |}, ^ 0 + a G [ 0 , l ]

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

Linear First-Order Fuzzy Differential Equations 699

where

eMt _eMh + 1 + M h pMt _pMh , i i TUTfr ft

<p(t,h,a) = ?- ^ - — - / e-Msdiam{[a{s))a)ds 2 h Jo

[ eMs(aal(s)+aar(s))ds Jo

lo e-Mte-Mh_l + M h rt

h 0Mt 0Mh pt+h

e-Msdiam([a(s)]a)ds

pt+h / eMs (aal(s) + aar(s)) ds - <ral(t),

ft -Mt „-Mh pt+h

2 h Jt

which tends to zero as h —> 0+ , uniformly in a G [0,1], since a is bounded in [0,t] (&ah °ar are bounded on [0,t] uniformly in a),

,. - e M / l + l + M/i ,. e~Mh-l + Mh ^ hm = hm = 0,

/i-̂ o+ h h^o+ h

and, using the continuity of a at t,

- / e-Msdiam([a(s)]a) ds - • e-Mtdiam([a(t)}a)1

h Jt

i pt+h

j - J eMs (aal(s) + aar(s)) ds -+ eMt (aal(t) + aar(t)),

uniformly in a G [0,1]. Analogously, for

pMt pMh _ i _ jLfi. ft tl>(t,h,a) = — / e-Msdiam([a(s)]a)ds

e-Mte-Mh_1 + Mh p ^

h eMt eMh rt+h

f eMs (aal(s) + aar(s)) ds Jo

2 h Jt

e-Msdiam([a(s)]a)ds

e-Mt e~Mh pt+h + —^ ^— / eMs (aal(s) + aar(s)) ds - aar(t).

The same procedure is valid for the case of the left-sided Hukuhara derivative of v at t. This completes the proof, since v is differentiate in the sense of Hukuhara at every t and v'(t) = —Mv(t) +<r(£). Note that adding w to v, we obtain (7) and (8), which provide the solution to (6). •

Remark 5. Expressions (7) and (8) are valid for / bounded or / = [0, +oo), with aeCil.E1).

Remark 6. In particular, if a = X{o}, then (6) is

u'(t) = -Mu(t), u(0) = u0,

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

700 J. J. NietOf R. Rodriguez-Lopez & D. Franco

and the solution is given by (9)-(10). If UQ = X{o}> then the solution is v in the previous Theorem.

R e m a r k 7. If we denote by mp([x\a) the midpoint of [x]a, that is, ^(xai + xa

and by r([x]c

expressed as and by r([x]a) the radius of [x]a, that is, r([x]°) = hdiam{[x\a), then (7) can be

u{t)al = - (r([u0]a) + j r([a(s)]a)e-Ms ds ) eMt

mp([u0]a)+ f mp([a(s)}a)eMs ds

\ Jo

and (8) as

<t)ar = (r([u0]a) + f r([a(s)]a)e-Ms ds\ eMt

„Ms An \ -Mt + lmp({u0}a) + / mp([a(s)}a)eMsds) e

This solution is calculated taking into account the midpoint and the diameter of the level sets (of the initial data as well as function a), which characterize the level sets.

R e m a r k 8. If uo and a are crisp, then (3) and (4) are valid, and the solutions given by (2) and (7)-(8) are the same crisp function:

u(t) = u0e~Mt + / a(s)eMis~t) ds, t e I . Jo

Note that, in this case,

r([uo]a) = 0, mp([uo]a) = UQ, for all a,

and

r(W(s)]a) = °> mp([°(s)}a) = <Ks), for all s, and all a.

In fact, in the crisp case, (1) and (6) are equivalent problems.

R e m a r k 9. In the general case, if u is the solution to (6), then

diam([u(t)]a) = diam([u0]a)eMt + f diam{[a{s)]a)eM{t-s) ds, t e I,

Jo and

mp([u(t)]a) = mp([u0]a)e-Mt + / mp([a(s)]°)eM ( s- t ) ds

Jo

= \((uo)ai + {uo)ar)e-Mt + \ J (a(s)al + v(s)ar)eM(*-» ds, t e I.

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

Linear First-Order Fuzzy Differential Equations 701

(18)

2.3. u'(t) = Mu(t) + cr(t), t G I , M > 0

Let M > 0, I a real interval, J = [0,T] with T > 0 or I = [0,+oo), a G C{I,EX), uo £ E1 and consider problem

u'(t) = Mu(t)+a(t)J t el, (17)

u(0) = w0.

Theorem 3. Problem (17) has a unique solution in I, given by

u(t) = u0X{eMt} + / o-(s)x{eM(t-sn ds, t G I. Jo

Proof. Take [t^(t)]a = [u(t)ai,u(t)ar], and (17) is written levelwise as

u'al(t) = Mual(t)+aal(t), tel,

u'ar{t) = Muar(t) + <Tar(£), t G I.

Using an integrating factor, we get

uai(t) = (u0)aieMt + f aal(s)eM^ ds, Jo

and analogously for uar(t), then

[u(t)]a = [u0]aeMt + / [a(s)]aeM^-s) ds,

Jo

for every a G [0,1] and t G / , obtaining (18). It is obvious that u(t) defined by (18) defines a fuzzy number. Now, we check

that u is differentiable in the sense of Hukuhara at every point t with Hukuhara derivative at t equal to Mu(t) + cr(t), which is a fuzzy number for all t,

(Mu(t) + a(t))al = M(u0)aieMt + M [ a(s)al eM^ds + a{t)ah Jo

(Mu(t) + a(t))ar = M(u0)areMt + M [ a(s)ar eM^~sUs + a(t)ar.

Jo The Hukuhara differences u(t + h) —H u(t), u(t) —H u(t — h) for u given by (18) exist (at least for h small), since

diam([u(t)]a) = eMt (diam([u0}a) + [ diam([a(s)]a)e-Ms ds

is a nondecreasing function in t, so that

(U(t + h) -H U{t))ai < (U(t + h) ~H U(t))0

(u(t) ~H U(t ~ h))ai < (u(t) ~H U(t ~ h))0

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

702 J. J. NietOf R. Rodriguez-Lopez & D. Franco

and these expressions define left-continuous functions in a. Moreover, it is obvious by the definition of u that

(u(t + h) - H U(t))al, (u(t) ~H U(t - h))al

are nondecreasing in a and

(u(t + h) -H u(t))ari (u(t) -H u(t - h))ar

are nonincreasing in a. Let t E I be fixed, h > 0, and calculate the Hukuhara difference quotients

u(t + h)-Hu(t)\ _ ^ , ^Mt(eMh-l

h al

pt / Mh i \ r>Mh rt+h J <?(s)aieM^ ds ( ^ - j r - ^ J + ~ J *(s)ale

M^ ds

and

u{t + h)-Hu{t)\ , . MtfeMh~l ~ (Uo)are h Jar " "'"' V h

+ J a(s)areM^ ds (j—^-1) + V / t <K*)areM(t-8) ds.

Using the continuity of a, it is easy to prove that

'u(t + h) -H u(t) \ h^o+ h

and

'u(t + h) -H u{t) \ h^o+

h

uniformly in a, hence,

(Mu(t)+a(t))ah

an

and, similarly,

lim <t + h \ g " W = Mu(t)+a(t) in (E1 , <*«,),

lim " ( t ) g " ( * k) = Mu(t)+a(t) in (E1 , <*«,),

which proves that u is Hukuhara differentiable at t with derivative

M^(t)+cr(t) G ^ 1 ,

and the equation is satisfied. •

For a different interpretation of linear fuzzy differential equations, see Ref. 15.

Remark 10. Note that if we replace M by — M in (18), we obtain (2).

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

Linear First-Order Fuzzy Differential Equations 703

Remark 11. If tx is the solution to (17), then

diam([u(t)]a) = diam([u0]a)eMt + / diam{[a{s)]a)eM{t-s) dsy t E / ,

Jo and

mp([u(t)}a) = mp([u0]a)eMt + [ mp{[a{s)}a)eM{t-s) ds

Jo

)((uo)al + (uo)ar)eMt + \ J (a(s)al + a ( s ) a r ) e M ^ ds, t E I. Q W^u /uc i v ^ u / a / / ^ i Q

2.4. u'(£) - Mw(t) = cr(t), t € I , M > 0

Let M > 0, I = [0, T] or / = [0, +oo), cr G C(I, E^.UQEE1, and consider problem

Ur{t)-Mu{t) =cr(t), t E / , (19)

u(0) = UQ.

Theorem 4. Define

W1(tJa)=diam([u0]a)+ [ diam([a(s)]a)eMs ds,

Jo

W2(t, a) = (U0)al + (uo)ar + / (o-(s)ol + Cr(s)ar) e ~ M s d s .

Jo

Expressions e-Mt eMt

<t)ai = —Wi(t ,a) + — W2(t,a), (20)

e - M t eMt <t)ar = —^-W^a) + — W2(t,a), (21)

/or t E I, a E [0,1], represent the unique solution to problem (19) in I, if they define a fuzzy number, that is, if

u(t)ai nondecreasing and u(t)ar nonincreasing in a, (22)

and for every t E I, there exists (3 > 0 such that the Hukuhara differences

u(t + h) —H v>(t), u(t) —H u(t — h)

exist for 0 < h < (3.

Proof. Take

[u(t)]a = [u(t)ahu(t)ar]J

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

704 J. J. Nieto, R. Rodriguez-Lopez & D. Franco

and search for a solution of the type

'u(t)al\_f e~M* eM*\(cal(t)\

for t E I, a E [0,1], with

[u(0)]a = {(U0)ah(u0)ar}, Va,

that is, for all a E [0,1],

((uo)al \ = (u(0)ai \ = ( 1 lUcal(0)

Solving this system, we obtain the unique solution

Coz(0) = --diam([u0]a)J

car(0) = -((U0)al + Mar),

for a E [0,1]. Then, for all t and a,

[U(i)]a = [e-Mtcal(t) + eMtcar(t), -e~Mtcal{t) + eMtcar(t)}, (24)

and the following equations deduced from (19) must be satisfied

u'al(t) - Muar(t) = aal(t), t £ I, a £ [0,1], u'ar(t) - Mual{t) = aar(t), t £ I, a £ [0,1].

These conditions yield, for t £ J, a £ [0,1],

c'al(t)e-Mt + eMtc'ar(t)=Vai(t), -c'al(t)e-Mt + eMtc'ar(t) = aar(t),

and

c'al{t) = -\diam([a(t)Y)eM\ dar{t) = \{aal{t) + aar{t))e-M\

which taking into account the initial values cai(0) and car(0) give, for t E / , a E [0,1],

cai(t)

Car\t)

-\diam{[u0}a) - \ [ diam([a(s)}a)eMs ds,

1 1 ff

This joint to (24), produces (20) and (21). Hypothesis (22) and continuity of a guarantee that those level sets define a fuzzy number since it is obvious that

U(t)al < U(t) ar

and these functions are left-continuous in a.

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

Linear First-Order Fuzzy Differential Equations 705

For t G I, and h > 0 small enough, the Hukuhara difference quotients are given

by

'u{t + h) —H u(t)~

-Mh

— J (diam([u0]a) + f diam([a(s)]a)eMs ds

h

\„-Mt l~e

"2 1 / pMh _ 1 \ / /•*

" 2 ^ ( JT~ ) ( {Uo)al + {Uo)ar + J {a{s)al + , J ( s ) - ) e " M S ds

rt+h je-M(t+h)

2h ' — — ~- -

1 /»t+Al

- / diam([a(s)]a)eMsdsi

^ y ( a ( S ) a i + a ( S ) a r ) e - M ^ S e M ^ ) and

i/(£ + h) —H u(t) h

i / p - M h _ i \ / />t

: - e ~ M t ( J ( diam([u0}a) + / diam([a(s)]a)eMs ds

- 2 ^ ( — f t " ) ( M a / + M a r + jf (*(*)a* + ^ a r K ^ dfi

t+/i M{t+h) + — / diam([a(s)]a)eMs ds

+ 2hJ ^s)ai+^)ar)e~Msds

which converge as h —• 0 + uniformly in a, respectively, to g(£)a/ and g(t)a r , where

and

«(*)aJ = *(t)al + ^~MtWi(t,a) + ^-eMtW2(t,a),

q(t)ar = °{t)ar ~ ^e~MtWx(t,a) + ^eMtW2(t,a),

which define a fuzzy number q(t) by [q(t)]a = [q(t)ahq(t)ar], a £ [0,1]. The same element in E1 is the limit in d^ as h —> 0 + of the Hukuhara difference quotients

u(t) —H u{t — h) h *

We have proved that ur(t) = q(t), t £ I and, by the expression of q(t), it is easy to check that, for all a G [0,1], and t G I,

[?/(£) - Mu(t)]a = [q(t) - Mu(t)]a = [a(t)aha(t)ar] = [a(t)]a.

The initial condition u(0) = u$ is trivially verified, in consequence, u is the solution to (19). •

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

706 J. J. Nieto, R. Rodriguez-Lopez & D. Franco

R e m a r k 12. It is not always true that (20) and (21) define a fuzzy number. For instance, taking a(t) = x^o} crisp, then (20) and (21) are reduced to

e-Mt eMt

u(t)ai = —diam([u0}a) + — ((u0)ai + (^o)or),

e-Mt eMt u(t)ar = -^—diam([u0}

a) + — ((u0)ai + (u0)ar),

for t G I, a G [0,1], which do not define a fuzzy number for an arbitrary initial condition UQ. In fact, taking the following triangular fuzzy number u$ as the initial data

u°(t)= { - ^ ( t - loo), t e [0,100], 0, otherwise,

the level sets are given by

[«o]° =

hence, if we set

a — 1 100 :

-Mt

100(1 - a)

(100(1 -a

aG[0 , l ] ,

a-1

v(t,a)

2 V

QMt , a _ 1

100

100

+ 100(1 - a) ) ,

then

u(t)ai = - / i ( t ,a) + z/(£,a),

u(t)ar = /i(t,a) ^^(t.a),

for t G / , a G [0,1], and it is not valid that

[u(t)u,u(t)br] £ [u(t)al,u(t)ar], &! b > CL.

For this condition to hold, u(t)ai should be a nondecreasing function in a, that is,

—/i(t,a) + u(t,a) < —/i(t, b) + u(t, 6),

for a < 6, but this is equivalent, for 6 > a, to

-100e-M t + 100eMt < ^e~Mt + Mt

100

or

9999eMt < lOOOle" Mt

which is valid for 0 < t small, but it fails if t large, for instance, at t M> since

9999e2 > 9999-2 > 10001.

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

Linear First-Order Fuzzy Differential Equations 707

Then, if the interval / is large, (20)-(21) do not necessarily define a fuzzy number.

Remark 13. Taking into account that UQ G E1, the continuity of a, and the expression of

(u(t + h) -H U(t))ah (u(t + fy ~H U(t))ar,

(u(t) ~H U(t - h))ah (u(t) ~H U(t - h))arj

for h > 0, we obtain the following sufficient conditions for the existence of the Hukuhara differences u(t + h) — H u(h), u(t) — H u(t — h), for every t and h > 0 small

• Condition (3). • For t £ I, there exists j3 > 0 such that, for 0 < h < /?,

u(t + h)ai — u(t)ai nondecreasing in a u(t + h)ar — u(t)ar nonincreasing in a, . , u(t)ai — u{t — h)ai nondecreasing in a

u(t)ar — u(t — h)ar nonincreasing in a.

Note that the existence of the levelset Hukuhara differences of u comes from (3) and continuity of cr, since

diam[u(t)}a = diam([u0]a)e-Mt + / diam([a(s)}a)eMs ds e~Mt

Jo is nondecreasing in t. Its derivative is

diam([a(t)]a) - Me~Mt (diam([u0}a) + f diam([a(s)]a)eMs ds) > 0.

Remark 14. If u is the solution to (19), then

diam([u(t)]a) = diam([u0]a)e-Mt + / diam([(j(s)]a)eM(s- t ) ds, t e / ,

Jo and

mp([u(t)}a) = mp([u0}a)eMt + / mp([a(s)]a)eMit-s) ds

Jo

\{{u0)al + (u0)ar)eMt + \ j (a(s)al + a(*)ar)eM<*-> ds, t e I.

3. Relation Among the Solutions to Different Problems

In conclusion, for M > 0, initial value problems for equations

uf(t) = -Mu{t) + cr(t), t e I, (Eq. (6)),

u\t) = Mu(t) + <T(£), t E / , (Eq. (17)),

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

708 J. J. NietOf R. Rodriguez-Lopez & D. Franco

have always a unique solution. However, solutions relative to equations

u'{t) + Mu(t) = cr(t), t e I, (Eq. (1)),

u'(t) - Mu(t) = cr(t), t E I, (Eq. (19)),

exist under restrictive conditions, in fact, for a a crisp function, the initial data and the solution are necessarily crisp.

There are two elements which characterize the solution u(t) of a fuzzy problem, the diameter and the midpoint of each level set. If we compare the solution to (1) with the solution to (6), although their expressions seem to be very different, we can appreciate that the midpoint of each level set [u(t)]a is exactly the same. However, the diameter changes, replacing M by — M.

The same happens if we compare the solutions of (17) and (19). That is, the operation of passing the term depending on the constant to the other side of the identity produces a change in the diameter of the level sets of the solutions, but preserves the midpoint invariant.

Now, if we compare (1) and (19), which differ only in the sign of the constant (passing from a positive one to a negative one or vice versa), we find that the diameter of the level sets of the solutions is the same, but in this case, the midpoint of the level sets is different (we can pass from one expression to another, by writing —M instead of M). Similar considerations can be made for solutions of (6) and (17).

Note that the solutions of (1) and (17) have a similar expression, we can obtain one from the other replacing M by — M. The same change applies to the diameter and the midpoint of the level sets of the solution. Analogously for solutions to (6) and (19).

As we have shown, the expression of the solution changes if we change the sign of the constant M or if we change the term Mu(t) to the other side of the equation. Nevertheless, there is a close relation among the different solutions of these problems, that can be appreciated by comparing the diameter and the midpoint of their level sets.

Finally, note that for M = 0 conditions (3), (4), (22), and (25) are trivially fulfilled and the expressions for the solutions to the different problems (1), (6), (17), and (19) are equal to

u(t) = UQ + / a(s) ds, t e J, Jo

the unique solution to problem

( u'(t) = ait), t e / ,

u(0) = u0.

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

Linear First-Order Fuzzy Differential Equations 709

A c k n o w l e d g e m e n t s

Research partially supported by Ministerio de Ciencia y Tecnologia / FEDER,

project BFM2001-3884-C02-01; Ministerio de Educacion y Ciencia / F E D E R

project MTM2004-06652-C03-01; and by Xunta de Galicia / FEDER, projects

PGIDIT02PXIC20703PN and PGIDIT05PXIC20702PN.

References

1. J. J. Nieto and A. Torres, "Midpoints for fuzzy sets and their application in medicine", Artificial Intelligence in Medicine 27 (2003) 81-101.

2. T. G. Bhaskar, V. Lakshmikantham and V. Devi, "Revisiting fuzzy differential equa­tions", Nonlinear Anal. 58 (2004) 351-358.

3. P. Diamond, "Brief note on the variation of constants formula for fuzzy differential equations", Fuzzy Sets and Systems 129 (2002) 65-71.

4. P. Diamond and P. E. Kloeden, Metric spaces of fuzzy sets : theory and applications (World Scientific, Singapore, 1994).

5. V. Lakshmikantham and R. N. Mohapatra, Theory of fuzzy differential equations and inclusions (Taylor & Francis, London, 2003).

6. P. Diamond, "Time-dependent differential inclusions, cocycle attractors and fuzzy dif­ferential equations", IEEE Trans. Fuzzy Syst. 7 (1999) 734-740.

7. P. Diamond, "Stability and periodicity in fuzzy differential equations", IEEE Trans. Fuzzy Syst. 8 (2000) 583-590.

8. O. Kaleva, "Fuzzy differential equations", Fuzzy Sets and Systems 24 (1987) 301-317. 9. J. J. Buckley and T. Feuring, "Fuzzy initial value problem for iV-th order linear differ­

ential equations", Fuzzy Sets and Systems 121 (2001) 247-255. 10. D. N. Georgiou, J J. Nieto and R. Rodriguez-Lopez, "Initial value problems for

higher-order fuzzy differential equations", Nonlinear Anal. 63 (2005) 587-600. 11. B. Bede and S. G. Gal, "Generalizations of the differentiability of fuzzy-number-valued

functions with applications to fuzzy differential equations", Fuzzy Sets and Systems 151 (2005) 581-599.

12. J J. Nieto and R. Rodriguez-Lopez, "Bounded solutions for fuzzy differential and integral equations", Chaos Solitons Fractals 27 (2006) 1376-1386.

13. S. Abbasbandy and T. Allahviranloo, "The Adomian decomposition method applied to the Fuzzy system of Fredholm integral equations of the second kind", Internat. J. Uncertain. Fuzziness Knowledge-Based Systems 14 (2006) 101-110.

14. T. Allahviranloo, N. Ahmadya and E. Ahmady, "Numerical solution of fuzzy differential equations by predictor-corrector method", Inform. Sci. (in press) doi:10.1016/j.ins.2006.09.015.

15. B. Bede, I. J. Rudas and A. L. Bencsik, "First order linear fuzzy differential equations under generalized differentiability", Inform. Sci. (2006), to appear.

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

This article has been cited by:

1. Rosana Rodríguez-López. 2013. On the existence of solutions to periodic boundaryvalue problems for fuzzy linear differential equations. Fuzzy Sets and Systems 219, 1-26.[CrossRef]

2. Sneh Lata, Amit Kumar. 2013. A new method for solving differential equations with vagueparameters. Applied Mathematical Modelling 37:5, 3282-3295. [CrossRef]

3. Minghao Chen, Chengshun Han. 2013. Periodic behavior of semi-linear uncertaindynamical systems. Fuzzy Sets and Systems . [CrossRef]

4. A. Khastan, J.J. Nieto, Rosana Rodríguez-López. 2013. Periodic boundary valueproblems for first-order linear differential equations with uncertainty under generalizeddifferentiability. Information Sciences 222, 544-558. [CrossRef]

5. S. Salahshour, T. Allahviranloo. 2013. Applications of fuzzy Laplace transforms. SoftComputing 17:1, 145-158. [CrossRef]

6. Xiaobin Guo, Dequan Shang. 2013. Approximate Solution of th-Order Fuzzy LinearDifferential Equations. Mathematical Problems in Engineering 2013, 1-12. [CrossRef]

7. Minghao Chen, Chengshun Han. 2012. Some topological properties of solutions to fuzzydifferential systems. Information Sciences 197, 207-214. [CrossRef]

8. Marek T. Malinowski. 2012. Random fuzzy differential equations under generalizedLipschitz condition. Nonlinear Analysis: Real World Applications 13:2, 860-881. [CrossRef]

9. S. Salahshour, T. Allahviranloo, S. Abbasbandy. 2012. Solving fuzzy fractional differentialequations by fuzzy Laplace transforms. Communications in Nonlinear Science and NumericalSimulation 17:3, 1372-1381. [CrossRef]

10. Juan J. Nieto, Rosana Rodríguez-López, Manuel Villanueva-Pesqueira. 2011. Exactsolution to the periodic boundary value problem for a first-order linear fuzzy differentialequation with impulses. Fuzzy Optimization and Decision Making 10:4, 323-339.[CrossRef]

11. A. Khastan, J.J. Nieto, Rosana Rodríguez-López. 2011. Variation of constant formula forfirst order fuzzy differential equations. Fuzzy Sets and Systems 177:1, 20-33. [CrossRef]

12. T. Allahviranloo, S. Abbasbandy, S. Salahshour, A. Hakimzadeh. 2011. A new method forsolving fuzzy linear differential equations. Computing 92:2, 181-197. [CrossRef]

13. Minghao Chen, Daohua Li, Xiaoping Xue. 2011. Periodic problems of first order uncertaindynamical systems. Fuzzy Sets and Systems 162:1, 67-78. [CrossRef]

14. Marek T. Malinowski. 2010. Existence theorems for solutions to random fuzzy differentialequations. Nonlinear Analysis: Theory, Methods & Applications 73:6, 1515-1532. [CrossRef]

15. Jiuping Xu, Zhigao Liao, Juan J. Nieto. 2010. A class of linear differential dynamicalsystems with fuzzy matrices. Journal of Mathematical Analysis and Applications 368:1, 54-68.[CrossRef]

16. Juan J. Nieto, Rosana Rodríguez-López. 2010. Upper and lower solutions method for fuzzydifferential equations. SeMA Journal 51:1, 125-132. [CrossRef]

17. Ravi P. Agarwal, V. Lakshmikantham, Juan J. Nieto. 2010. On the concept of solution forfractional differential equations with uncertainty. Nonlinear Analysis: Theory, Methods &Applications 72:6, 2859-2862. [CrossRef]

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.

18. J.J. Nieto, A. Khastan, K. Ivaz. 2009. Numerical solution of fuzzy differential equationsunder generalized differentiability. Nonlinear Analysis: Hybrid Systems 3:4, 700-707.[CrossRef]

19. Luciano Stefanini, Barnabás Bede. 2009. Generalized Hukuhara differentiability ofinterval-valued functions and interval differential equations. Nonlinear Analysis: Theory,Methods & Applications 71:3-4, 1311-1328. [CrossRef]

20. Y. Chalco-Cano, H. Román-Flores. 2009. Comparation between some approaches to solvefuzzy differential equations. Fuzzy Sets and Systems 160:11, 1517-1527. [CrossRef]

21. T. Allahviranloo, S. Abbasbandy, N. Ahmady, E. Ahmady. 2009. Improved predictor–corrector method for solving fuzzy initial value problems. Information Sciences 179:7,945-955. [CrossRef]

22. Rosana Rodríguez-López. 2008. Monotone method for fuzzy differential equations. FuzzySets and Systems 159:16, 2047-2076. [CrossRef]

23. Minghao Chen, Yongqiang Fu, Xiaoping Xue, Congxin Wu. 2008. Two-point boundaryvalue problems of undamped uncertain dynamical systems. Fuzzy Sets and Systems 159:16,2077-2089. [CrossRef]

24. Rosana Rodríguez-López. 2008. Periodic boundary value problems for impulsive fuzzydifferential equations. Fuzzy Sets and Systems 159:11, 1384-1409. [CrossRef]

25. Rosana Rodríguez-López. 2008. Comparison results for fuzzy differential equations.Information Sciences 178:6, 1756-1779. [CrossRef]

26. Alexandru Mihai Bica. 2008. Error estimation in the approximation of the solutionof nonlinear fuzzy Fredholm integral equations. Information Sciences 178:5, 1279-1292.[CrossRef]

27. Jiuping Xu, Zhigao Liao, Zhineng Hu. 2007. A class of linear differential dynamical systemswith fuzzy initial condition. Fuzzy Sets and Systems 158:21, 2339-2358. [CrossRef]

28. Juan J. Nieto, Rosana Rodríguez-López. 2007. Euler polygonal method for metricdynamical systems. Information Sciences 177:20, 4256-4270. [CrossRef]

Int.

J. U

nc. F

uzz.

Kno

wl.

Bas

ed S

yst.

2006

.14:

687-

709.

Dow

nloa

ded

from

ww

w.w

orld

scie

ntif

ic.c

omby

UN

IVE

RSI

TY

OF

HA

IFA

on

06/1

4/13

. For

per

sona

l use

onl

y.


Recommended