HIROSHIMA MATH. J.

30 (2000), 191-203

Stationary solutions to boundary problem for the heat equations

A. Ya. DOROGOVTSEV
(Received August 26, 1997)
(Revised April 7, 1999)

ABSTRACT. The necessary and sufficient conditions for the existence of a stationary
solutions to the boundary value problem for an abstract heat equation with a stationary
disturbances and to the stochastic boundary value problem for such equation in the strip
are given. The existence of a bounded solutions to the deterministic boundary value
problem is also considered.

1. Introduction

In this paper we deal with an abstract stochastic heat equations, for which
one of the independent variables represents time. It is supposed that random
disturbances on the right-hand side are stationary with respect to the time
variable. We are interested in solutions which are stationary with respect to
the time variable of a boundary value problem in the strip. Periodic solutions
for the deterministic partial differential equations are intensively studied, see, for
example, well known book [15]. The problem of the existence of stationary
solutions to a stochastic ordinary differential equation is also well understood,
see books [8], [5] and a survey [6] for more references. During the past years it
has become apparent that it is natural and more adequate in many applications
to consider an input source as a random source or random disturbances. Thus
investigations of stochastic partial differential equations are important. We
consider the stationary solutions to some boundary value problem for a heat
equation and will present some approach to obtain the existence theorem of
stationary solutions. This approach is based on the results from [3] and [4].
We will demonstrate it in a simple situation relative to the random disturbances.
Let (B, || ||) be a complex separable Banach space, 0 the zero element in B,
and S£{B) the Banach space of bounded linear operators on B with the operator
norm, denoted also by the symbol || ||. For a ^-valued function the continuity
and differentiability means correspondingly the continuity and differentiability
in the i?-norm. For an operator A the sets σ(A) and p(A) are its spectrum and
resolvent sets, respectively. Let / be the identity operator.

2000 Mathematics Subject Classification. 60G10, 60H15, 35K20.

Key words and phrases. Stationary solutions, heat equation, stochastic boundary value problem.
 192 A. Ya. DOROGOVTSEV

In what follows, we shall consider all random elements on the same

complete probability space (Ω,&,P). The uniqueness of the random process,
satisfying some equation, means its uniqueness up to stochastic equivalence.
We consider only ^-valued random functions which are continuous with prob-
ability one. All equalities with random elements in this article are always the
equalities which hold with probability one.

2. Stationary solutions to the boundary value problem with stationary

disturbances

Denote by S\ the set of all 2?-valued stationary processes ξ = {ξ(t)\t e R}

with continuous sample functions satisfying the following condition

SUP \\ξ(t)\\\ < +OD

for some δ > 0.

Define

CQ : = { # : [0,π] -> C | g(0) = g(π) = 0} Π Cι([0,π\); Q:=Rx [0,π].

Given g e CQ, A e S£{S) and ξ e S\ let us consider the problem

f «;(*, x) - u'^it, x) = Au(t, x) + ξ{t)g{x), (ί, x) e β;

DEFINITION 1. A B-valued random function u defined on Q is a solution of

the problem (1) if u, u't and u"x are continuous with probability one on Q and
equality (1) holds with probability one.

DEFINITION 2. A B-valued random function u defined on Q is stationary

(with respect to t) if

W GR V«eN Vίfr,*!),..., (*„,*„)} c Q V{A, , A,} <= »{S) :

P< [){ω : u(ω tk + t,xk) e Dk}\ \ = p\\ f] {ω : u(ω; tk,xk) e Dk} I,

where 0f(B) is Borel σ-algebra of B.

We come now to the main result of this section.

THEOREM 1. Let A e S£(JS). Then the following two statements are

equivalent:
 Stationary solutions to boundary problem 193

(i) For any stationary process ξ e S\ and function g e CQ the boundary

value problem (1) has a unique stationary solution u such that
sup £||w(0,x)|| <+oo.
0<x<π

(ii) {k2 + /a I k e N, α e R} c />(Λ).

PROOF, (ii) implies (i). Let £ e S\ and g e Cj be given. If the condition
(ii) holds then we have, for some ko e N

where
σi c {zeC|Rez< 1},
σ2 cz {ze C| 1 <Rez<4};

Now let P* (or P^) be the Riesz projectors corresponding to the part of the
spectrum σ\ U UG> (or σ^+i U U ^ ) , k = l,...,ko — I. Note that if, for
example, σ\ = σ(A), then P} = / and P+ is the zero operator. It is known that
pk + pA: = jr

for 1 < k < ko — 1 and

| | ^ - * o ' ) ' | | < Le~a\ t >0

with some numbers L > 0, α > 0, which are fixed below.

The sequence {sinfcx, x e [0,π] : k > 1} is complete in CQ1, which means
that for any g e CQ,

fifW = y^^sinfcx, xe[0,π]; gk=-\ g(x)sinkxdx, k>Λ,

π J
kΞ\ o
where the series on the right-hand side is uniformly convergent.
Consider the random function
p+oo \
u(t,x) := J2(\ °kit- s)P-ξ{*)ds - s)P*ξ(s)ds
Gkk(t- s)P*ξ(s)ds gksinkx

5Z Gk{t - s)ξ(s)dsgk sinkx, (t, x) e Q,

J-co
k=ko
 194 A. Ya. DOROGOVTSEV

where
Gk(ή := teR

All integrals for u are convergent as a Bochner integral with respect to Lebesgue
measure (or improper Riemann integrals for continuous 2?-valued function) with
probability one. To see this, observe that for 0 < k < ko — 1

' E\\Gk{t-s)P*ξ(s)\\ds< Γ
— 00 J —
—00

P+00 Γ +00

J E\\Gk(t - s)P*ξ(s)\\ds < J

and

(2) [ E\\Gk{t-s)ξ{s)\\ds< -(a + k2-k20)(t-s))E\\ξ(0)\\ds

J

The series for w converges uniformly on [M,V] X [0,π] for any u < v, v — u <δ
with probability one. Indeed, we have

E\ sup Gk(t - s)ξ(s)ds =E\SUV \+CO Gk(p)ξ(t - p)dp\)

\u<t<v Jo |J

\ sup ί+
lu<ί<υJθ

+
<ί sup
Jo
E sup k>k0.
l_0</<<5
Hence the random function u is continuous with probability one. The random
function u is stationary with respect to t, see [5].
By a similar reasoning, it is verified that

k2l)
- (ϊ
*d-l
(3) Σ
k=\
- s)P*ξ(s)ds

M-oo

- J Gk{t - s)P*< gksinkx

 Stationary solutions to boundary problem 195

ί sinto

£
k=\
*
- P+ζ(ή)gk sinkx + ^
k=k0
ξ(t)gk ύnkx

= Au(t,x) + ξ{ήg(x)

f+0° \
- Gk(t- s)P*ξ{s)ds gkάnkx
h J
Gk{t - s)ζ(s)dsgk sinkx, (ί, x) e Q;
J — 00
k=k0
and

(4) ι&(/,*) =
-oo
p+oo
p+

- II
00 i ί
G
+ Σ *(* " s)ζ(s)dsgk(-k2) sinkx, (t,x) e Q
co
k=k0J-

with probability one. The series appearing on the right hand sides of (3) and
(4) converge unifofmly on [w,υ] x [0,π] for any u < v, v — u <δ with probability
one. Indeed, by conditions ξ e S\ and g e Cj we have
oo / II f' I\
V £ sup μ - k2l) Gk(t- s)ξ{s)dsgk sin to
k=ko \u<ί<v,0<x<π || J-oo ly

II f + 0 ° II
sup Gk(p)ξ(t-p)dp\\-\gk
IU II

/ II H I

sup \\k2\ Gk(t-s)ζ(s)dsgk sinkxl

u<t<v,0<x<π || J-oo l
 196 A. Ya. DOROGOVTSEV

The convergence of the series

a
k=k0 + k ~ ^o

00
can be justified by showing Σ \θk\ < oo for any g e Cj. Applying the formula
k=\
for integration by parts, we have

By the Cauchy-Schwarz inequality and the ParsevaΓs identity the last series
converges. Now (1) follows from (3) and (4).
Let us establish the uniqueness of stationary solution for the problem (1)
by contradiction. Given a solution u of (1), let

Vk(ή:=-\ u(t,x)sinkxdx, ίeR, k>\

be its Fourier coefficients. Then they satisfy the following equation

2
' + ξ{t)1 teR.
By the uniqueness of the Fourier expansion, the non-uniqueness of (1) implies
the non-uniqueness of the solution of (5) for some k > 1. But under condition
(ii) every equation (5) has a unique stationary solution by Theorem 1 in [3].
(i) implies (ii). Let k eN and α e R be given. Let u be a unique sta-
tionary solution of (1) for

g(x) = sinfcjc, jc e[O,π];

Define

Γ ίeR,
ΰk{t) := u(t,x) sinkxdx,
Jo
the last integral is a Riemann integral for ^-valued continuous function with
probability one. It can be easily checked, that v/c e S\, see [5] for more
details. From (1) we have

(6) v'k(ή = ( A - ^

with probability one. The stationary process v^ is a unique solution of (6),

because if the equation (6) had two different stationary solutions, then one could
construct, following the method described in part (ii) => (i) of the proof, two
 Stationary solutions to boundary problem 197

different stationary solutions of the problem (1). By Theorem 1 in [3] we have

2
σ(A - k l) Π ίR = 0 .
This completes the proof of Theorem 1.

3. Deterministic problem

In this section we shall formulate conditions for the existence of a bounded

solution u to the deterministic boundary problem of the form

{ u't(t, x) — u"x(t, x) = Au(t, x) + v(t, x),

u{t,x)=u(t,π) = O, teR,
(ί, x) e Q,

where v is a bounded function on Q.

Set

{v,v'x} <=C(β, 5); sup ||i;|| < + 0 0 , "

ρ
C M := < 1; : β - 5
t;(/,0) = !>(*, π) = 0, ίeR

{w,w;,w;;} c C(β,iϊ);sup||M|| < +00,

Q

u(t,O) = u(t,π) = 0 , ίeR

THEOREM 2. Let A e Se{B). Then

Ίυ e Cb\ 3!w e Cbn, u is a solution of (7)

The proof of Theorem 2 is similar to that of Theorem 1 except for one

difference. Instead of Theorem 1 from [3] we must use the M. G. Krein
Theorem [11], p. 54:
For De <£{B) the equation

x'(ή=Dx{t)+f(t), teR

has a unique bounded solution x in B for every bounded function / if and

only if
σ(D) Π iR = 0 .

4. Stationary solutions to the stochastic boundary value problem

We now consider the boundary problem (1), where ξ is a process of the

"white noise" type. More precisely, such a problem can be formulated in the
198 A. Ya. DOROGOVTSEV

following way. Suppose that B = H is a complex separable Hubert space and

consider an //-valued Wiener process {w(ή : t e R}, see, for example, [1] or [5].
In particular, we have

Ew(t) = 0, E\\w(t) - w(s)\\2 = \t-s\ tr W, {t,s} c= R,

where W is a nuclear operator. Define J*v : = σ(w(s) : s < ή, t e R.

We would remind the reader that a j£?(ϋΓ)-valued random function h on R
is a nonanticipatίng function if for every t ER the random element h{i) is J v
measurable. For such a function h the stochastic integral

\'h(r)dw(r)
Js

is defined in the usual way, see, for example, [1], [7], [5]. The definition of a
nonanticipating //-valued function is given in a similar manner.

DEFINITION 3. An H-valued random function u is a solution to the boundary

value problem

u't{t,x) - w" x (ί,x) = Au{t,x)+g(x)w'(t), (f,*) e Q,

(8)
u(t,0) = u(t,π) = 0 , teR
with g e CQ if the function u is nonanticipating, the functions u, u'^x are continuous
with probability one and, for every s < t, x e [0, π] we have

u(t,x)-u(s,x)-\ u"x(r,x)dr=\ Au(r,x)dr + g(x)(w(t) - w(s)),

(9) Js Js

w (f,0) = u(t,π) =0

with probability one.

Let us define now the class

Co3 := {g : [0,π] -> C | ^ ( 0 ) = g^{π) = 0, k = 0,1,2} Π C 3 ([0,π]).

Now we can formulate the main result of this section.

THEOREM 3. Let A e S£{H). Then the following two statements are
equivalent:
(i) For any Wiener process w and a function g e CQ the boundary problem
(8) has a unique stationary, nonanticipating solution u such that

sup £ | | W ( 0 , J C ) | | 2 < +oo.

0<Λ:<π

(ii) σ(A-I)^{zeC : Rez < 0}.

 Stationary solutions to boundary problem 199

PROOF, (ii) implies (i). First note that

for some L > 0 and a > 0.

Let a Wiener process w and a function g e CQ be ginen. Then
00

g(x) = V^ gk sin /DC, x e [0, π]

k=\
with {#£ : k > 1} c C. The last series converges uniformly on [0,π],
Now define the random function
oo /»/

(10) u(t,x) := ] Γ Gk(t-s)dw(s)gk sin kx, (t,x) e Q.

The integrals on the right-hand side are to be understood in the sence of

stochastic integrals with respect to the Wiener process w. These integrals exist,
since, for any t eR and k e N, we have

Moreover, these integrals as functions of t are continuous on R in //-norm with

probability one, the proof of this assertion being derived analogously to [1], see
also [5] and [2], [8], [10], [14] for general results. It can be verified, that these
integrals are stationary connected //-valued processes, defined as follows [5].
The processes £1,^2? >£™ a r e called the stationary connected processes if a
vector process (ξ\, ξ2,..., ζm) is a stationary process. Each of these integrals is
nonanticipating with respect to w. Let —oo<b<c<+oo be given. To
establish the uniform convergence of the series (10) on Q{b,c) : = [b,c] x [0,π]
with probability one, we prove that the series

(11) J2E( sup f Gk(t-s)dw{s)gksinkx

f^i \teQ(b,c) II J-oo
is convergent. Using the properties of stochastic integral, we find

E[( sup Gk(t-s)dw(s)gk sin kx 1

\teQ(b,c)\\J-oD
e 11/

<E^ sup^\Gk(t-b)\\ • 11* Gk(b-s)dw(s) |))|0t

-h^ί sup Gk(t-s)dw(s)\\)\gk

\b<t<c b
200 A. Ya. DOROGOVTSEV

<L\gk\E I Gk(b-s)dw(s

+ \gk\E( sup \\w(t) - Gk(t - b)w{b) + {'(A - k2I)Gk(t - s)w(s)ds

\6<ί<cll h

This relation implies

(12) E\ sup Gk(t - s)dw(s)gksinkx

V
s 1/2
\\Gk{b-S)\\2tϊWds\

\ b<t<c
f \\A- sup |
b<t<c

Consider the random process {w(ί) : / G [έ, c]} as a random element in

C([b,c],H). This element is Gaussian. As a consequence, we have

El sup ||w(0|| <+oo,

\b<t<c J
see, for example, [13] or [12].
The uniform convergence of the series (10) on Q(b,c) now follows from
the estimate (12) and the convergence of the series (11). Hence, the random
function {u(t,x) : (t,x) e Q} is continuous with probability one.
The continuity of the random functions

{!&(*,*) : {t,x) e Q}

and the equalities

u'x{t, x) = \ - s)dw(s)kgk coskx,

k=\ ^ -

r
«^(ί, x) = - - s)dw(s)k2gk ύnkx,

can be established by use of a similar argument, except for the following. For
a proof of convergence of a series for u"x it is enough to check that
 Stationary solutions to boundary problem 201

|0*| <+oo.
k=\

We use the assumption geC^. Let a/c(f), £&(/), k>\ are the Fourier
coefficients of / e CQ. Then

and the series

k=\

converges by the ParsevaΓs identity. By the Cauchy-Schwarz inequality the

series

>I + IW")I
δ .
is convergent.
The random functions u and uxx are stationary random functions.
With the aid of the above expressions for u and uxx, we obtain for the
right-hand side integral of (9) the following representation

ί s
Au(r,, x)dr

=
Σ (
£_1 Js \J-o

(A-k2I)Gk(r-p)dW(p)]drgkύnkx- \iTJr,
—oo / J

Σ['[ ((( d G(k { dr ~ p ) dr

= u(t, x) - u(s, x) - g(x)(w{t) - w(s)) - u"x{r, x)dr
Js

with probability 1. Therefore u is a solution of the problem (8).

202 A. Ya. DOROGOVTSEV

To show the uniqueness, assume that ύ is also a solution of the problem

(8). Then p : = u — ΐi satisfies the following conditions

p't(t,x) - p"xx = Ap(t,x), (t,x) e β;

p(t,0)=p(t,π)=0, teR

with probability 1 and for

Γ
pk(t) := p(t,x)sinkxdx, /eR,
Jo
we have the equation
= {A-k2I)pk{t)
which has under condition (ii) only one and hence trivial stationary solution by
Theorem 4.1 from [4], k>\. The rest of the proof of the uniqueness is the
same as in the proof Theorem 1.
(i) implies (ii). Let a Wiener process w and a function g e CQ be given.
Suppose that u is a unique stationary nonanticipating solution of (6) corre-
sponding to w, g and define, for fceN,

Γ
:= u(t,x)sinkxdx,
Jo
The function Vk is a stationary nonanticipating /f-valued process such that
2
< +00. From (9) we have

(13) f v (r)dr = f
vk(ή - vk(s) + k2 f vkk(r)dr = f ^^(r)dr + gk(w(ή - w(s))
Js Js

for s < t with probability one. Note that the equation (13) has unique non-
anticipating stationary solution, the proof is similar to the one of Theorem 1.
Now by Theorem 4.1 in [4] we have
2
σ(A - k l) c { z e C : R e z < 0 } .

Since this inclusion holds for all fceN, we obtain

σ(A-I) <=:{zeC: Rez < 0}.

The proof Theorem 3 is complete.

REMARK. We note, that second statements of Theorem 1 and 3 are

different. Theorem 3 suggests that the existence of the nonanticipating sta-
tionary solution is a stronger condition than the existence of a stationary
solution.
 Stationary solutions to boundary problem 203

5. Acknowledgment

The author would like to express his gratitude to the referee for his helpful
improvements on the text and suggestions.

References
[ 1 ] R. F. Curtain and P. L. Falb, Stochastic Differential Equation in Hubert Space, J.
Differential Equations 10 (1971), 412-430.
[ 2 ] G. Da Prato, Regularity Properties of a Stochastic Convolution Integral, Atti. Acad. Naz.
Lincei Rend. Sc. fis. mat. e. natur. 72 (1982), 217-219.
[ 3 ] A. Ya. Dorogovtsev, Periodic Solutions of an Evolutional Differential Equations Perturbed
by Random Processes, Ukrainskii Matematicheskii Zhurnal 41 (12) (1989), 1642-1648 (in
Russian).
[ 4 ] A. Ya. Dorogovtsev, Stationary and Periodic Solutions of Stochastic Difference and Dif-
ferential Equations in Banach Space, New Trends in Probability and Statistics, Proc.
Bakuriani Colloq. in Honor of Yu. V. Prohorov edited V. V. Sasonov and T. Shervashidze,
MOKSLAS, Vilnius 1 (1991), 375-390.
[ 5 ] A. Ya. Dorogovtsev, Periodic and Stationary Regimes of Infinite-Dimensional Deterministic
and Stochastic Dynamically Systems, Vissha Shkola, Kiev, 1992 (in Russian).
[ 6 ] A. Ya. Dorogovtsev, Periodic Processes: a Survey of Results, Theory of Stochastic Pro-
cesses 2(18) (3-4) (1996), 36-53.
[ 7 ] I. I. Gikhman and A. V. Skorokhod, Stochastic Differential Equations and its Applications,
Naukova Dumka, Kiev, 1982 (in Russian).
[ 8 ] B. Goldys, On some Regularity Properties of Solutions to Stochastic Evolution Equations,
Colloquium Mathematicum 58 (2) (1990), 327-338.
[ 9 ] R. Z. Khasminskii, Stochastic Stability of Diifferential Equations, Sijthoff and Nordhoff,
Alpen aan den Rijn, 1980.
[10] P. Kotelenez, A Submartingale Type Inequality with Applications to Stochastic Evolution
Equations, Stochastics 8 (1982), 139-151.
[11] M. G. Krein, Lectures on the Stability Theory of Solutions of Differential Equations in
Banach Space, Inst. Techn. Inf., Academy Sci. Ukrainian SSR, Kiev, 1964 (in Russian).
[12] Hui-Hsiung Kuo, Gaussian Measures in Banach Spaces, Springer-Verlag, 1975.
[13] A. V. Skorokhod, Integration in Hubert Space, Nauka, Moscow, 1975 (in Russian).
[14] L. Tubaro, Regularity Results of the Process X{t) = Jo' U(t, s)g{s)dW{s), Rend. Sem. Mat.
Univ. Padova 39 (1982), 241-248.
[14] O. Vejvoda et al., Partial Differential Equations: Time-Periodic Solutions, Noordhoff,
Alpen aan den Rijn, 1981.

Kiev Institute of Business and Technology

Kiev, 01601, Ukraine