Internat. J. Math. & Math. Sci. VOL. 20 NO. (1997) 187-195
187
BOUNDARY
VALUE PROBLEMS
FORTHE DIFFUSION
EQUATIONWITH
PIECEWISE CONTINUOUS TIME DELAY
JOSEPH WIENER
Department
ofMathematics Uraversity ofTexas PanAmerican Edinburg,Texas78539, U.S.ALOKENATH DEBNATH
Department
of Mathematics Umversity of Central Florida Orlando,Florida32816,USA(ReceivedNovember 30, 1995 and inrevisedformJanuary20, 1996)
ABSTRACT. A study is made of partial differential equations with piecewise constant arguments.
’Boundary
valueproblemsfor three types of equationsarediscussed delayed; alternately ofadvanced and retarded type; and most importantly, an equation of neutral type (that is, including the derivative at differentvaluesoftimet).KEY WORDS AND PHRASES: Partial differential equations, piecewise constant arguments, oscillations,andstability.
1991AMSSUBJECTCLASSIFICATION CODES: 35A05, 35B25, 35L10,34K25.
1. INTRODUCTION
Functional differential equations (FDE)with delay provide a mathematical model fora physical or biological system in which therateofchange ofthe systemdepends uponits pasthistory. Thetheory of FDE with continuous arguments is well developed and has numerous applications in natural and engineering sciences. This paper continues our earlier work in an attempt to extend this theory to differential equations with discontinuous argument deviations. In
articles
[1-5], ordinary differential equations with argumentshavingintervals ofconstancy have been studied. Such equations representa hybrid ofcontinuous and discrete dynamical systems and combine properties of both differential and difference equations. They include as particular cases loaded and impulsive equations, hence their importanceincontroltheoryand in certain biomedicalmodels Continuity ofasolutionat apoint joining any two consecutive intervals implies recursion relations for the values of the solution at such points. Therefore, differential equationswith piecewisecontinuousargument(EPCA)
areintrinsically closer to difference rather thantodifferentialequations. In [6]boundaryvalueproblemsforsome linearEPCAin2. ACOMPARISONOF TWO EPCA Theequation
ut(z,t)
a2u=(z,t)
bu(z,t)
(2.1)describes heatflow ina rod with both diffusion
a2uzz
alongthe rod andheat loss (orgain) across the lateral sides of the rod. Measuring the lateral heat change at discrete moments oftime leads to the equationwithpiecewisecontinuousdelayut(z,t)
a2u(x,t)
bu(z,
[t]),
(2 2)which wasinvestigatedin[6]. Here
[-
designatesthegreatest integerfunctionand(x, t)
E[0, 1]
x[0,
oo) Theproblem posedin[6]forEq. (2.2)consistsoftheboundaryconditions(0. t)
0.
,(. t)
0 (23)and the initial condition
and the solution issoughtintheform
(, o)
uo(z),
(2 4)u(z, t)
X(z)T(t).
(2,5)Thenseparation ofvariablesleadstothe BVP
X’
+A2X
O, withtheorthonormalsetof solutionsX(O)
X(1)
O, (2.6)X(z)
V
sin(zrjz) (2.7)on
[0, 1],
andtotheequationr(t)
a27r2j2T:(t)
bT:([t]).
(2.8)Let
T(t)
denote a solutionof(2.8)
onthe intealn<_
t<
n+
1, where n is anonnegative integer ThenT:(t)
a2r2j2T.j(t)
bT.j(n),
(2.9)and the general solution ofthisequationis
T(t)
C.e
--Weputheret nand get
thatis,
where
At n
+
1 we haveb
a27r2.7.
T,j(n). (2.10)T(t)
G(t- ,)T,(,).
(2.11)T,,(n
+
1)
G(1)T,3(n)
E(t)
_oe,
b(
_o,,).
BOUNDARYVALUEPROBLEMS 189
and since
T,a(n
+
1)
T,+,,a(n +
1), thenT+,a(n
+
1)
Ea(1)Ta(n)
and
Ts(n
E(1)Toa(O).
(2 13)Therefore,
and
To(t)
Ea(t
n)E;(1)To,(O)
u,,(x,t)
Z
E’(1)Ta(O)Ea
(t
n)sin(rrjx),3=1
(214)
(2 15) where
un(x,t)
designatesthe solution ofBVP (2.2), (2.3), (2.4) in[0,1]
x[n,n
+
1].
Putting 0, n 0givesuo(x)
T0a(0)V
sin(Trjx)=1
To,(O)
V/
uo(x)sinQrjx)dx. andAlongwithEq. (2.2),westudytheequation
([1
u,
(z, t)
au=
(z, t)
bu z, t+
-
(2 16)underconditions(2.3), (2 4)
DEFINITION. A function
u(x,t)
is said to be a solution ofthe above BVP if it satisfies the conditions (i)u(x, t)
is continuousinG[0,1]
[0,
oo);
(ii)utandu
exist and arecontinuous inG, withthe possible exception ofthe points(x,
n+
),
whereone-sided derivatives exist(n
0,1, 2,...);
(ii)
u(x,t)
satisfiesEq. (2.16)inG,
withthe possible exception of the points(x,
n+
1/2),
andconditions (2.3),(2.4).Again, the solution is soughtinform (2.5),and separation ofvariables generates the eigenfunctions (2.7)andleadstotheEPCA
T’(t)
a27r2j2T(t)
bT
(
[t+l
)
(217)\L .j/
Eq.
(2.17)
isofconsiderableimerest, sincethe argument deviation 1(t)
t-t+
changesthesignineach interval
(n
,
n+),
withintegern. Indeed,r(t) <
0 forn<
<
nandr(t)
>
0 for n<
t<
n+
,
which meansthat Eq. (2.17)is alternately ofadvanced and retarded type. Assume thatT(t)
is a solutionof(2.17) on theinterval[n- 1/2,n
+
].
ThenEq. (2.17) changestoand since
then
Tn+l,,(:)
Ea(;-
rz-1)Tn+l,2(r/,
+
1),(1)
()
T,.,
n+-
ETr+.a(n
+
1).Furthermore,continuityofthe solution
T(t)
impliesandtherefore
whence
From here,
and
Ea(l/2)
Tn/l’a(n+l)
Ea(-
1/2)T,a(n).
1
(218)
where
E(t)
is definedby(2.12).
TItEOREM2.1. For
(2.21) the solution(2.15)ofEq. (2.2)tendstozero as oo,uniformlywith respect to z
PROOF. From (2.15)itfollows thatthe assertion is true if
IE(1)I
<
1. Solvingtheinequalities 1<
e_,,,
b(
a,,)
a27r3,
2 1e-
<
1forbprovesthe proposition. Furthermore,
u(x, t)
approacheszero withanexponentialrateTIIEOREM:L2. For
b
>
-a27r
2(2.22) the solution(2.20)of
Eq. (2.16)
tendstozero ast oo,uniformlywithrespectto x.PROOF. From (2.20)we seethat theassertion is valid
iflEj(/2)E(
/2)
<
,
whereE(t)is
given by(2.12). Consideringthe case
Ea(
1/2)
>
0givesb
>
2#ae
’
/
(e
’
1),
(2.23)1
T,.,a(t)
Ea(
n)E’
(
)
E’
(
-
)
Toa(O).
(2.19)The solution
un(z,t)
of BVP (2.17), (2.3), (2.4) in theregion[0,
1]
x[n-
,n+1/2]
is givenby the formulau(x,t)
Z
v/E;
E;
To,(O)&(t-
n)sin(rjx), (2.20)BOUNDARYVALUE PROBLEMS 191 withthe notation
i/A2
a2
71_2j2/2.
(2 24)From
E2(1/2
<
E2(
1/2)
weobtain theinequalityb
>
2#2
(225)which is stronger than(223), and so(2 23)should beomitted Furthermore,
E2(1/2
>
E2(
1/2)
implies that
b
>
-21.z2(e
m+
e-m)/(e
m+
e-m-2), and since this inequality is weaker than (2.25) it may be disregarded E1/2) <
0 isequivalenttob
<
2#jeu’/(e
u’1),
On the other hand, the case
(2.26) andthe inequality
Ea
(1/2) >
E
1/2)
leadstob
<
2#.
The latterinequality maybeignoredbecause it isweakerthan(2.26) From
E(1/2) <
E(
1/2)
itf611owsthat
b
<
2#,(e
m+
e-u,)/(e
u,+
e-u,2).
The latterinequality shouldbe omitted since (2.26)is morestringent. Onthe otherhand, (2.26)cannot holdtruefor all values ofj, and wemust retainonly(2.25). Thisinequalityis validforall j if(2.22)takes place. Comparing inequalities(2.21) and(2.22) shows that the stabilityintervalforEq. (2.17) islarger than thestabilityimervalforEq. (2.8).
TEIEOREM2:.3. Each solutionofEq. (2.8)hasazerointheinterval
In,
n+
1]
ifThe same istrueforEq (2.17)if
and
PROOF. From
Eq.
(2.13)wehave(,
+
)
7/()T0(0)
T,,,(n)T,,,(n +
1)=
g2n+l
(1)T.(O).
Hence,
T,v(n)T,,,(n
+
1) <
0 ifE(1)
<
0.The latterinequalityisequivalentto
b
>a27r2j2/(ear’-1),
whichholdstrueforall3under condition(2 27).inequality
Onthe otherhand, (2.28)isderivedby examiningthe
Inequalities(2.27)and(2 28)represent sufficient conditionsofoscillationfor
Eqs
(2 8)and(2 17),forall values ofjsimultaneouslyTHEOREM2.4. Each solutionofEq. (2.8)isnonoscillatory, forallj, if b
<
0 Foranyb>
0 and sufficientlylargej, thefunctionsT(t)
areoscillatory.PROOF. Theinequality
E
(1) >
0 isequivalenttob
<
aTrgjg-/(e’_
l)
and holds true for all j only if b
<
0 Since its fight-hand side tends to zero as j,
the above inequality breaks down forb>
0and allsufficiently largej. Therefore,in this case the solutionsT(t)
of (2 8) oscillate, in sharp comrast tothe functionsT(t)
inthe Fourier expansion for the solution of the equationu
ag-u
bu without timedelay.TItEOREM2.5. Each solution ofEq. (2.17)isnonoscillatory, forall j, if
-aTr/(e’’/-l)
<b<0. (2.29) Foranyb>
0and sufficiently largej, thefunctionsT(t)
areoscillatory Furthermore,ifthen the functions
T
(t),
Tnm
(t)
oscillate.PROOF. Fromtheinequalities
E(
1/2)
>
0andE(1/2)
>
0itfollowsthat2#e/(e
’
1)
<
b<
2#/(e
,
1),
where# isgiven by(2.24). Thelettofthese inequalities holdstreeforall j if b
>
-a27r/(e’’/- 1),
and thefight one is satisfied forall j if b
<
0. The inequalitiesE(-
1/2) <
0 andE(1/2)
<
0 are contradictory. Onthe otherhand,therightside oftheinequalityb
< 2#/(e
’
1)
tendstozero as j
--
o% andthereforetheinequality breaks down for any givenb>
0 and allsufficiently largej. Moreover,theleftsideoftheinequality-2#je
/(e
’
1) <
bmonotonicallydecreasesto ooas j oo. Hence,if thisinequalityfailsforj m, thesame istreealso forj
<
m.TEIEOREM2.6. If,for some integerm
_>
1,b=
a2Tr2m2(e’’m’+ 1)/(e
’’m’1),
(2.30) then the solutionTin(t)
ofEq.
(2.8)is aperiodic functionwithperiod2. The same istrueforthe solution ofEq (2.17)ifb 2/,
(e
+
e -u/ (2
e"
e-’"),
(2.3 l) where#,isgivenin(2.24).PROOF. Sincetheinitialvalue problem for
Eq.
(2.8)has auniquesolutiononeach unit interval with imegral endpoints,wehave onlytoshow thatT(n
+
2)
T,(,),
,
0, ,2,BOUNDARY VALUE PROBLEMS 19 3
and
T(t)
E.(t- n)E(1)To.(O),
Try(t)
Em(t-
-
2)E,/
(1)To(0),
n<t<n+l
n+2<t<n+3.
Therefore, whence
At t n
+,
Eq (3.3)gives intervaln<
t<
n+
1is Then the seriesand conditions (2.3) and (2.4). This equation is of neutral type since it includes the derivative ut at
differentvalues oft.
Let
un(x,t)
be the solution of the given problem on the interval n<
t<
n+
1. Thenun(x,t)
satisfies theequation
Ou(z,t)
a
+b
(z,n
+at Oz (32)
withboundaryconditions(2.3)and the initial condition
,(:,,+)
,(:),
where
u(z)
is yetunknown. Thesolution of(3.2)
issought inform(2.5), and separation ofvariables generatesBVP(2.6)witheigenfunctions(2.7)and theODET’,
,=aT,(t)
+
T’,(,+),
un(x,t)
E
X3(x)Tn(t)
(3 4)3=1
represents a formal solution of problem(3.2),
(2.3).
Turning toEq. (3.3), itsgeneral solution ontheA 7rj. (3.3)
b +
m,(t)
c.,-’,
(’-")+
T,;,(,
).
2
+ a
,
T,(n),
T,.,,(n
)=b-1
b
Tna(n).
T(t)
CeV,
(’-)+
bZ
1Since hypothesis (230) implies
E,(1)
1, then T,(n
+
2)
T,(n),
which proves the theorem for Eq (2.8) The proof for Eq (2 17) is analogous Condition (2.31) is equivalent toEm(1/2)
E,(
1/2)
Itremainsto notethatthe value orb in(2.30)ispositive, and the valueorb in(2.31)isnegative.3. APARABOLICEPCAOF NEUTRAL TYPE
Consider theboundaryvalueproblem(BVP)consisting oftheequation
ut(x, t)
a2uzz(x,
t)
+
but(x,
[t])
(3 1)At nand n
+
2theseformulasrespectively yieldAt n,wehave
that is,
and
Denote
At n
+
1,wehave1
C,,
1 b
T,(n)
l
(b e-d’(t-’))T,.u(n)
T’v(t)
b- 1
F3(t)
b- 1 (3.5)T,.u(n
/1)=
F3(1)T,(n
and since, by hypothesis(i)oftheabove definition,
T,,3(n
+
1)
Tn+lj(n+
1),
then
and
T,.,+,a(n
+
1)
Fa(1)T,2(n)
Ta(n)
Ff’
(1)Toa(0).
(3 6)Consequently,
T,.,(t)
Fj(t-n)Fj(1)To.(O)
and
(3 7)
u,(x,t)
E
VF;(1)Ta(O)Fa
(t
-n)sin(rjx). (3.8) Thefollowing theoremsillustratethefarmorecomplicatedsolutiontructure
ofEq.
(3.1)comparing withthe diffusionequationwithout timedelay.TREOREM3.1. Forb
>
1, eachfunctionT,(t)
ismonotoneunboundedasPROOF. If b
>
1, then (3.5) implies thatF:(t)>
1, for all t>
0. Hence,F"(1)
grows exponentiallyast+
oo,and theprooffollows from(3.7).TI]EOREM3.2. For
(1
+
e-Lx)
<
b<
1, (3.9)2
eachfunction
T,,
(t)
isunboundedandoscillating.PROOF. Frominequalities(3.9)itfollowsthat
F3(1)
<
1, andthereforeFJ
t](1)
isunboundedand hasazeroineachunit interval withinteger endpoints."rl]EOREM3.3. Forb
<
0, each functionT,(t)
monotonically tendstozero as PROOF. Theinequalities0
< F(1) <
1 (3 10)BOUNDARYVALUEPROBLEMS 195
Clearlythe right-hand sideof(3 11) vanishes asj oo, and therefore theconditionb
<
0 isnecessary and sufficientforallTn3(t)
toapproachzero monotonically.TItEOREM3.4. If
1
a2A
>In2 and e-’2x <b<,
(3 12) then every solution(3 7)tendstozeroand oscillatesast+
ooPROOF. Fromtheinequalities
-1<F3(1
<0, (3 3)whichimply asymptotic stabilityandoscillatorybehaviorofsolution(3 7),itfollowsthat
e-’’-a,
<b<
l+e-These inequalities holdtrueforall jsimultaneouslywith(3.12)
TtlEOREM 3.5. Foranycoefficientbthat satisfies theinequalities 1
-<b<l, 2
there existinfinitely manysolutions
T,3
(t)
whichareunbounded andoscillating.PROOF. For b
<
1, the inequalityF(1) <
1 whichimplies unboundedness and oscillation ofT3(t
),
isequivalenttoClearlythefight-handsideapproaches
1/2
as j oo.THEOREM3.6. Ifb
<
e-ax,
thenthesolutionsT1
(t),
T,(t)
ofEq.
(3.3)monotonically tendto zeroas -++
oo.ACKNOWLEDGEMENT. Thisresearchwaspartially supportedby NASAGrantNAG9-553
[]] [2]
[3] [4]
REFERENCES
AFTABIZADEH,
A.R. andWIENER, J.,
Oscillatory properties of first order linear functional differential equations,ApphcableAnal.20(1985), 165-187.AFTABIZADEH, A.R.,
WIENEK, J. andXU, J.-M.,
Oscillatory and periodic solutions of delay differential equations withpiecewise constantargument, Proc. Amer.Math. Soc. 99 (1987), 673-679.COOKE, K.L. andWIENER, J., Retarded differentialequationswithpiecewiseconstantdelays,J. Math. Anal.
App.
99(1984),265-297.COOKE, K.L. and
WIENER,
J., Anequationalternatelyof retarded and advanced type,Proc.Amer. Math.Soc. 99(1987),726-732.[5]
WIENER,
J. andCOOKE, K.L.,
Oscillations in systems ofdifferential equations with piecewise constantargument,J.Math.Anal.Appl.137(1989),221-239.[6]
WIENER,
J., Boundary-value problems for partial differential equations with piecewise constant delay,lnternat.J.Math.andMath.Sci. 14(1991),301-321[7] WIENER, J. and DEBNATH, L., Partial differential equations with piecewise constant delay, Internat. J.Math. and Math.Sct. 14(1991),485-496.