Automatic Control Via Thermostats Of A Hyperbolic Stefan Problem With Memory

Automatic control via thermostats of a
hyperbolic Stefan problem with memory
Pierluigi Colli1 Maurizio Grasselli2 J
urgen Sprekels3
November 11, 1996
1991 Mathematics Subject Classi cation. 35R35, 35R70, 45K05, 93C20.
Keywords. Feedback control, Stefan problems, memory kernels, hyperbolic heat conduc-
tion.
1 Dipartimento di Matematica, Universit
a di Torino, Via Carlo Alberto 10, I{10123 Torino, Italy
2 Dipartimento di Matematica, Politecnico di Milano, Via Bonardi 9, I{20133 Milano, Italy
3 Weierstrass Institute for Applied Analysis and Stochastics, Mohrenstrasse 39, D{10117 Berlin, Germany
0

Abstract. A hyperbolic Stefan problem based on the linearized Gurtin{Pipkin heat conduction law
is considered. Temperature and free boundary are controlled by a thermostat acting on the boundary.
This feedback control is based on temperature measurements performed by real thermal sensors located
into the domain containing the two{phase system and/or at its boundary. Three di erent types of
thermostats are analyzed: ideal switch, relay switch, and Preisach hysteresis operator. The resulting
models lead to formulate integrodi erential hyperbolic Stefan problems with nonlinear and nonlocal
boundary conditions. In all the cases, existence results are proved. Uniqueness is also shown, unless in
the situation corresponding to the ideal switch.
1. Introduction
Consider a two{phase system which occupies a bounded domain R
N (N 1) at
any time t 2 [0;T] (T 0): Letting QT := (0;T); we denote by # : QT ! R the
relative temperature (rescaled in order # = 0 be the critical temperature at which the
two phases can coexist) and by : QT ! [0;1] the concentration of the more energetic
phase (e.g., water in a water{ice system). Within the framework of the study of memory
e ects in heat conduction phenomena, the following integrodi erential model has been
proposed (cf. [4])
@t('0# + ' # + ) ;k # = f in QT; (1:1)
2 H(#) in QT; (1:2)
where '0 is a positive constant, '; ; k : [0;+1) ! R are smooth relaxation (or
memory) kernels, and indicates the time convolution product on (0;t): Besides,
f : QT ! R is a known function which depends both on the heat supply and on the
past histories of # and up to t = 0 (supposed to be given), while H stands for the
Heaviside graph, that is,
H(s) =
8

:
f0g if s 0;
[0;1] if s = 0;
f1g if s 0:
System (1.1{2) endowed with suitable initial and boundary conditions produces a
hyperbolic Stefan problem provided that k(0) 0 (cf. [4, Section 2]). In the case of
homogeneous Dirichlet boundary conditions, this problem has been already investigated
in some detail, proving both strong and weak well{posedness among other things (see
[3{6]). Here we are interested in analyzing the heat exchange at the boundary ; := @
under the in uence of a thermostat, and to this aim we have to deal with a boundary
condition of the third type. Then, by taking the initial condition
('0#)(;0) = e0 in (1:3)
(the datum e0 expressing an enthalpy density), we are going to play with the following
relation
;k #n = (#; ;#e) on T := ; (0;T); (1:4)
1

where #n is the outward normal derivative of # on ;; denotes a positive constant,
#; is the trace of # on the boundary, and #e represents the external temperature
coupled with another term depending on the prescribed past history of # up to t = 0:
As we shall see, #e plays a crucial role in the control problems described below.
Suppose that we are able to measure the temperature # by a real system of thermal
sensors placed in some xed positions inside the body or on its surface. This fact can
be made schematic by assuming that the quantity
M(#)(t) :=
Z
0
#(x;t)!I(x)dx +
Z
;0
#;(y;t)!S(y)d; (1:5)
is known at any time t 2 [0;T]: Here 0 and ;0 ; are the involved sets, with
positive Lebesgue domain and surface measures, respectively. Moreover, the notation
R is used to indicate the mean value, and !I : 0 ! [0;+1); !S : ;0 ! [0;+1) are
weight functions determined by the characteristics of the sensors.
To control the evolution of the free boundary, a thermostat device acts modifying #e
on account of M(#): According to [11] (see also [8, 12]), the behavior of the thermostat
is described by
#e(y;t) = u(t)#A(y;t)+ #B(y;t); (y;t) 2 T; (1:6)
u : [0;T] ! R being a heating (or cooling) device whose dynamics obeys
u0
+ u = W(M(#)) + #C in [0;T]; (1:7)
u(0) = u0: (1:8)
The functions #A; #B : T ! R and #C : [0;T] ! R are given (with a sign property
for #A); is a positive parameter, W models the action of the thermostat, and
u0 2 R: We have to specify the operator W yet. Here, referring to [7{8, 11{14], we
are going to consider three di erent cases W1; W2; W3:
(A) Simple switch
A critical time{dependent value %(t) provides a jump discontinuity so that, for r 2
C0([0;T]) and t 2 [0;T];
W1(r)(t) := H(r(t) ;%(t)) =
8

:
+1 if r(t) %(t);
w(t) if r(t) = %(t);
;1 if r(t) %(t);
(1:9)
where w 2 L1
(0;T) ful lls ;1 w(t) 1 for a.a. t 2 (0;T): Obviously, % 2
C0([0;T]) is xed inside the thermostat. As we will postulate later, the selection of w
is purely random and ;H is nothing but the opposite of the maximal graph resulting
from the sign function.
2

(B) Relay switch
In this case there are two thresholds %L; %U 2 C0
([0;T]) with %L %U and such that,
for any r 2 C0
([0;T]); W2(r) changes its value at time tc from ;1 to +1 or vice
versa according to the rule
W2(r)(tc) :=
+1 if r(tc) = %L(tc) and W2(r)(t) = ;1 just before,
;1 if r(tc) = %U (tc) and W2(r)(t) = +1 just before.
(1:10)
The meaning of just before is made precise later (see Section 4), as well as the de nition
of W2(r) in the case when r coincides with one of the two quantities %L or %U on
an open subset of [0;T]:
(C) Hysteresis operator of Preisach type
To introduce W3; we partly follow [13, Chapter IV]. For r 2 C([0;T]) and for any pair
(%1;%2) 2 R2
satisfying %1 %2; we set
H(%1;%2)(r;)(0) :=
8

:
+1 if r(0) %1;
(%1;%2) if %1 r(0) %2;
;1 if r(0) %2
with : (%1;%2) 7! (%1;%2) 2 f;1;+1g being a given Borel measurable function. In
addition, if t 2 (0;T] we let
Tt := f 2 (0;t] : r() = %1 or r() = %2g
and
H(%1;%2)(r;)(t) :=
8

:
H(%1;%2)(r;)(0) if Tt = ;;
+1 if Tt 6= ; and r(max Tt) = %1;
;1 if Tt 6= ; and r(max Tt) = %2:
One can easily check that the functions z = H(%1;%2)(r;) are continuous on the right
in [0;T) and have nite total variation on [0;T]; i.e., z 2 C0
r ([0;T)) BV (0;T): We
are thus led to consider the mapping H(%1;%2)(;) : C0
([0;T]) ! C0
r ([0;T))BV (0;T)
which is called delayed relay operator (compare with (B)). Now, if is a nonnegative
Borel measure on the plane P := f(%1;%2) 2 R2
: %1 %2g; the associated Preisach
operator is speci ed by
W3(r)(t) :=
Z
P
H(%1;%2)(r;)(t)d(%1;%2): (1:11)
The main properties of such transformation will be recalled in Section 5.
Next, let us come to our control problems. They can be roughly formulated saying
that we are looking for a triplet (#;;u) ful lling (1.1{4), (1.6{8) with M prescribed
3

by (1.5) and W Wj; j = 1; 2; 3: Problems of this kind have been studied by several
authors (see, e.g., [7, 8, 11, 12] and the references therein). Nevertheless, the present
paper reports the rst attempt to investigate the thermostat control of a hyperbolic
Stefan problem with memory e ects.
It is convenient to put the feedback control problems we have just described in a
more general form (see [8]). Indeed, looking at (1.7{8), it is easy to observe that u is
given by a Volterra operator, namely
u(t) =
Z t
0
e;(t;)= (Wj(M(#))()+ #C())d + u0e;t= (1:12)
for any t 2 [0;T] and for j = 1; 2; 3: On account of (1.12), equation (1.6) can be
written as
#e = F[Wj(M(#))] on T; (1:13)
where, for any r 2 L2
(0;T);
F[r](y;t) :=
Z t
0
E(y;t;)r()d + E0(y;t); (y;t) 2 T; (1:14)
with (in the special case of (1.12))
E(;t;) = e;(t;)= #A(;t);
E0(;t) =
Z t
0
e;(t;)= #C()d + u0e;t=

#A(;t) + #B(;t)
almost everywhere on ;; t and both varying in [0;T]: Consequently, the feedback
control problems reduce to hyperbolic integrodi erential Stefan problems with a non-
linear and nonlocal boundary condition. More precisely, for j = 1; 2; 3 we shall deal
with
Problem (Pj). Find a pair (#;) satisfying (1.1{3) and
;k #n = (#; ;F[Wj(M(#))]) on T: (1:15)
Taking advantage of the xed{point techniques used in [11] (see also [8, 12]), we
prove the existence of a solution to (P1). Uniqueness is not expected in this framework,
due to the random behavior of the model. Regarding (P2), the inductive argument
developed in [11] allows to show existence and uniqueness. In the case of (P3), we
can apply the Schauder xed{point theorem to derive existence of solutions. Besides,
under further restrictions on the measure and for !S 0 in (1.5) (no boundary
measurements) we deduce uniqueness via suitable contracting estimates.
4

In order to prove these results, we rst need a careful analysis of the well{posedness
of the Stefan problem (1.1{4). Also, we have to state Problems (Pj ), j = 1; 2; 3; in a
more rigorous way. Let us present the plan of the paper. In Section 2, we give a precise
formulation of (1.1{4) and of the related results, which are shown and fully detailed in
Sections 6{8. In addition, we establish the crucial properties of the operators M and F :
The subsequent Sections 3, 4, 5 are devoted to Problems (P1), (P2), (P3), respectively.
In each of these sections, the feedback control problem is settled and discussed up to
the proof of the related main result.
2. Preliminaries
Here and in the sequel RN
(N 1) is an open, bounded, and connected set
with a smooth boundary ; (for instance, ; of class C
2 ). We put H := L2( ) and
V := H1( ); recalling that V ,! H ,! V0 with dense and compact injections provided
H is identi ed with its dual space H0: The duality pairing between V0 and V and
the scalar product in H are both denoted by h ; i; while ( ; ) stands for the scalar
product in HN
: The norm either in H or in HN
is simply indicated by k k : Also, let
k k
; be the norm in L2(;):
We remind that v; stands for the trace of a function v 2 V on ;; the normal
derivative on ; being represented by vn whenever v is regular enough. Moreover,
prime speci es the derivatives of functions depending only on time, whereas the position
(1 z)( ;t) :=
R t
0 z( ;s)ds holds for any t 2 [0;T] and any z 2 L1(0;T;V0):
Now, we introduce the assumptions on the data (cf. [4, Section 2]). Assume that
' 2 W1;1(0;T); (2:1)
2 W2;1(0;T); (0) 0; (2:2)
k 2 W2;1(0;T); k(0) 0; (2:3)
f 2 W1;1(0;T;H); (2:4)
e0 2 V; (2:5)
#e 2 H1(0;T;L2(;)); (2:6)
#e( ;0) = ';1
0 e0; a.e. on ;: (2:7)
Consequently, accounting for the identity k # = k(0)(1 #) + k0 (1 #); we can
rigorously formulate the hyperbolic Stefan problem (1.1{4).
Problem (SP). Find the pair (#;) such that # 2 L1(0;T;V) W1;1(0;T;H) and
2 L1(0;T;H) satisfy
1 # 2 L1(0;T;H); (2:8)
5

#n 2 L2(T ); (2:9)
#; 2 H1(0;T;L2(;)); (2:10)
@t('0# + ' # + ) ;k # = f a.e. in QT ; (2:11)
2 H(#) a.e. in QT ; (2:12)
('0#)(;0) = e0 a.e. in ; (2:13)
;k #n = (#; ;#e) a.e. on T ; (2:14)
where Qt := (0;t) and t := ; (0;t) for t 2 [0;T]:
Remark 2.1. Note that 2 L1(QT ) thanks to (2.12). Owing to (2.8), the normal
trace #n has a a meaning in H;1(0;T;H;1=2(;)) and (2.9) asserts that is even a
square integrable function. By (2.10) it turns out that (2.14) holds almost everywhere
on T :
Existence and uniqueness for (SP) are given by
Theorem 2.1. Let (2.1{7) hold. Then Problem (SP) has a unique solution (#;):
Moreover, there exists a positive constant 1 such that
k#kL1(0;T ;V)W 1;1(0;T ;H) + k1 #kL1(0;T ;H) + k#;kH1(0;T ;L2(;))
+ k#nkL2(T ) 1
n
1 + kfkW 1;1(0;T ;H) + ke0kV + k#ekH1(0;T ;L2(;))
o
; (2:15)
this constant depending only on ; ;; T; '0; k(0); ; and on the norms k'kW 1;1(0;T ) ;
k kW 2;1(0;T ) ; kkkW 2;1(0;T ) :
A further useful result concerns the Lipschitz continuity of the map associating #e
with the solution component #; namely
Theorem 2.2. Let (2.1{5) hold. Consider #i
e; i = 1; 2; ful lling (2.6{7) and denote
by (#i;i) the corresponding solution to Problem (SP). Then there exists a positive
constant 2 such that for any t 2 [0;T] one has
k#1 ;#2kC0([0;t];H) + k1 (#1 ;#2)kC0([0;t];V)
+ k(#1 ;#2);kL2(t) 2 #1
e ;#2
e L2(t) : (2:16)
Moreover, 2 depends only on T; '0; (0); k(0); ; k'kW 1;1(0;T ) ; k kW 2;1(0;T ) ; and
kkkW 2;1(0;T ) :
The last part of this section is devoted to establish some basic properties of the
already introduced operators M; de ned by (1.5), and F; taken as in (1.14) for suitable
functions E(y;t;); E0(y;t) (cf. (2.18) below) not necessarily coinciding with those of
the Introduction. We require that
!I 2 L2( 0;[0;+1)); !S 2 L2(;0;[0;+1)) (2:17)
6

and that
E; Et; E 2 L2(; (0; T )2); E0 2 H1(0; T ;L2(;)); (2:18)
noticing that Et; E represent the partial derivatives of E with respect to the time
variables. We also set the compatibility condition (cf. (1.13) and (2.7))
'0E0(; 0) = e0; a.e. on ;: (2:19)
Here are two statements concerning M and F; respectively.
Proposition 2.1. Under the assumption (2.17), M : V ! R is linear and continuous
and can be naturally extended to spaces of functions from (0; T ) to V: In particular,
for all v1; v2 2 V there holds
jM(v1) ;M(v2)j k!IkL2( 0) kv1 ;v2k+ k!SkL2(;0) k(v1 ;v2);k; : (2:20)
Moreover, whenever v 2 L1(0; T ;V)W 1;1(0; T ;H) and v; 2 H1(0; T ;L2(;)); then
M(v) belongs to C0;1=2([0; T ]) and satis es
jM(v)(t);M(v)()j 3
n
kvtkL2(QT ) + k@tv;kL2(T )
o
jt ;j1=2 (2:21)
for any pair (t; ) 2 [0; T ]2; where, for instance, 3 = k!IkL2( 0) + k!SkL2(;0) :
Proof. As (2.20) is a trivial consequence of the de nition of M; it suces to check
(2.21). Letting t ; from (2.20) it turns out that
jM(v)(t) ;M(v)()j = jM(v(; t));M(v(; ))j
3
Z
t
(kvt(; s)k+ k@tv;(; s)k;)ds
and the H
older inequality allows us to infer (2.21).
Proposition 2.2. Under the assumptions (2.18{19), F is a continuous operator from
L1(0; T ) to H1(0; T ;L2(;)) such that
F[r](; 0) = ';1
0 e0; a.e. on ;; 8 r 2 L1(0; T ):
Moreover, one can determine a constant 4; depending only on T; kEkL2(;(0;T)2) ;
kEtkL2(;(0;T)2) ; kE0kH1(0;T;L2(;)) ; on the L2(T ) norm of the (diagonal) trace func-
tion (y; t) 7! E(y; t; t); and on the quantity max0tT kE(; t; )kL2(T ) ; such that
kF[r]kH1(0;T;L2(;)) 4
n
1 + krkL1(0;T)
o
; (2:22)
k(F[r1] ;F[r2])(t)k; 4 kr1 ;r2kL2(0;t) (2:23)
7

for any t 2 [0; T ] and for all r; r1; r2 2 L1(0; T ):
Proof. In view of (1.14) and (2.18), it is clear that
@t(F[r])(y; t) = E(y; t; t)r(t)+
Z t
0
Et(y; ; t)r()d + @tE0(y; t)
for a.a. (y; t) 2 T : Hence, observing that
Z T
0
Z t
0
Et(; t; )r()d
2
;
dt
Z T
0
krk2
L2(0;t)
Z t
0
kEt(; t; )k2
; d dt
T krk2
L1(0;T ) kEtk2
L2(;(0;T )2) ;
it is a standard matter to recover (2.22). Regarding (2.23), by the H
older inequality we
have that
k(F[r1] ;F[r2])(t)k;
Z t
0
kE(; t; )k; j(r1 ;r2)()jd
kE(; t; )kL2(t) kr1 ;r2kL2(0;t)
and easily achieve the proof.
3. Problem (P1): simple switch
As we have already mentioned in the Introduction (see (A)), we prescribe that at the
times when r reaches the threshold %; then W1(r) takes some value between ;1 and
+1 in a purely random way. We assume that
% 2 C0([0; T ]) (3:1)
and consider the set{valued convexi cation of the ranges in (1.9) putting
H(s) :=
8

:
f+1g if s 0;
[;1; 1] if s = 0;
f;1g if s 0;
(3:2)
so that ;H : R ! 2R is a maximal monotone graph. Now, let W1 be the multivalued
operator from C0([0; T ]) to L1(0; T ) de ned by
w 2 W1(r) if w(t) 2 H(r(t) ;%(t)) for a.a. t 2 (0; T ): (3:3)
It is straightforward to verify that W1(r) is nonempty for any r 2 C0([0; T ]); since the
existence of a measurable selection w (obviously bounded) is ensured by the continuity
of r and %:
8

Hence, Problem (P1) can be precisely formulated as
Problem (P1). Find # 2 L1(0;T;V) W1;1(0;T;H); 2 L1(0;T;H); and z 2
L1(0;T) satisfying (2.8{13) and
;k #n = (#; ;F[z]) a.e. on T ; (3:4)
z 2 W1(M(#)): (3:5)
The result we are going to prove is
Theorem 3.1. Let (2.1{5), (2.17{19), and (3.1{2) hold. Then there exists a solution
to Problem (P1).
Proof. We introduce the set
YT :=

2 H1(0;T;L2(;)) : (;0) = ';1
0 e0; a.e. on ; (3:6)
and the mapping
S1 : YT ! 2YT ; S1() := fF[w]; w 2 W1(M(#()))g; (3:7)
where
(#();()) represents the solution of (SP) with #e = : (3:8)
These positions are plainly justi ed by Theorem 2.1, Propositions 2.1{2, and (3.3).
Moreover, from (2.22) and (3.2) it follows that
kkH1(0;T ;L2(;)) 24 8 2 S1(); 8 2 YT : (3:9)
Thus, setting
UT :=
n
2 YT : kkH1(0;T ;L2(;)) 24
o
; (3:10)
we have that S1 is a multivalued operator from UT to UT : Thanks to (3.4{5), (3.8),
and (2.14), we realize that any xed point of S1 makes the pair in (3.8) solve (P1).
Therefore, to show Theorem 3.1 we just need to nd 2 UT such that 2 S1():
Reasoning as in [8], we use the Glicksberg xed{point theorem (see [9]). This
tool works for set{valued mappings under suitable convexity, compactness, and closure
hypotheses. Let us omit the statement here and check carefully the assumptions in our
frame.
Denote by H1
w
(0;T;L2(;)) the space H1(0;T;L2(;)) endowed with the weak
topology. It is a locally convex topological vector space. Due to the boundedness
property in (3.10), UT is a nonempty, convex, and compact subset of H1
w
(0;T;L2(;)):
In addition, UT is also sequentially compact. For an arbitrary 2 UT we claim that
9

S1() is nonempty (because of (3.7) and (3.3)) and convex. Indeed, if 1; 2 2 S1();
then there are w1; w2 2 L1(0; T) such that i = F[wi] and wi 2 W1(M(#()))
for i = 1; 2: On the other hand, by construction W1(M(#())) is convex so that
w1+(1;)w2 2 W1(M(#())) whenever 0 1: As F is an ane transformation,
we conclude that F[w1 + (1 ;)w2] = 1 + (1 ;)2 2 S1() for any 2 [0; 1]:
It remains to deduce that the graph of S1;
G(S1) := f(; ) : 2 UT ; 2 S1()g; (3:11)
is closed in the product space (H1
w
(0; T;L2
(;)))2
: To this aim, take a Moore{Smith
sequence f(a; a)ga2A; A being a directed set, ful lling (a; a) 2 G(S1) for any
a 2 A and
lim
a2Aa = 1; lim
a2Aa = 1
for some pair (1; 1) 2 (H1
(0; T;L2
(;)))2
: Owing to (3.9{10), f(a; a)ga2A is
bounded, hence weakly sequentially compact, in (H1
(0; T;L2
(;)))2
: Therefore, we can
extract a subsequence f(j ; j )gj2N such that
j ! 1; j ! 1 weakly in H1
(0; T;L2
(;)) (3:12)
as j % 1: Remark that 1 and 1 are both in UT : Since j 2 S1(j ); we can
x zj 2 L1(0; T) satisfying j = F[zj ] and zj 2 W1(M(#(j ))) for any j 2 N: By
(3.2{3) we have that kzj kL1(0;T )
1; and consequently a subsequence of fzj g admits
a weak star limit z1; i.e.,
zj ! z1 weakly star in L1(0; T): (3:13)
On account of (3.12{13) and Proposition 2.2, the (strong and weak) continuity of F
implies that 1 = F[z1]: At this point, to obtain (1; 1) 2 G(S1) it suces to
prove that (cf. (3.7{8))
z1 2 W1(M(#(1))): (3:14)
By virtue of (3.10), (2.15) and (2.12) there exist #1 and 1 such that, possibly taking
subsequences,
#(j ) ! #1 weakly star in L1(0; T;V)W1;1(0; T;H); (3:15)
1 #(j ) ! 1 #1 weakly star in L1(0; T;H); (3:16)
#;(j ) ! #1; weakly in H1
(0; T;L2
(;)); (3:17)
#n(j ) ! #1n weakly in L2
(T ); (3:18)
(j ) ! 1 weakly star in L1(QT ) (3:19)
10

as j % 1: The convergence (3.15) and the generalized Ascoli theorem entail
#(j) ! #1 strongly in C0([0; T ];H): (3:20)
With the help of (3.12), (3.15{20), and (2.1{3) it is not dicult to infer that (#1; 1)
solves Problem (SP) for #e = 1; so that (by Theorem 2.1 and (3.8)) #1 = #(1):
In fact, from (3.19{20) it results that
ZZ
QT
(j)#(j) !
ZZ
QT
1#1; (3:21)
and then (3.21) and the maximal monotonicity of H (as induced operator in L2(QT ))
ensure that (see, e.g., [2, Lemma 1.3, p. 42]) 1 and #1 ful ll the nonlinear con-
dition (2.12). Now, (3.15), (3.17), and Proposition 2.1 enable us to conclude that
fM(#(j))gj2N is a bounded and equicontinuous (cf. (2.21)) subset of C0([0; T ]); and
thus M(#(j)) converges to M(#(1)) not only weakly star in L1(0; T ) but strongly
in C0([0; T ]): Finally, referring to (3.13), (3.2{3) and arguing as above, one can exploit
the maximal monotonicity of ;H to get (3.14).
Hence, S1 has a closed graph in (H1
w(0; T ;L2(;)))2: The Glicksberg theorem now
applies and yields the existence of at least one solution to (P1) .
4. Problem (P2): relay switch
The response of the thermostat in this case ((B) in the Introduction) needs to be better
speci ed following [11, Section 5]. First of all, consider the two critical functions
%L; %U 2 C0([0; T ]); (4:1)
which are supposed to satisfy
%U (t) ;%L(t) 0 8 t 2 [0; T ]; (4:2)
for some xed bound : Then, in view of (1.10), (1.15) and (1.3{5), we can assume, for
instance, that
M(e0) =
Z
0
e0!I +
Z
;0
e0;!S '0%L(0); (4:3)
and the relay is initially switched on, namely
W2(M(#))(0) = +1: (4:4)
Here the notion of solution for (P2) is made precise .
11

Problem (P2). Find # 2 L1(0;T;V) W1;1(0;T;H); 2 L1(0;T;H); z 2
L1(0;T) and a nite sequence fthgm
h=0 of switching times with
0 =: t0 t1 ::: tm = T
satisfying (2.8{13), (4.4),
;k #n = (#; ;F[z]) a.e. on T; (4:5)
z(t) = (;1)h if t 2 [th;th+1[; (4:6)
th+1 is exactly the in mum of the set fTg[Kh+1; (4:7)
where
Kh+1 :=

t 2 (th;T] : M(#)(t) =
%U(t) if h is even,
%L(t) if h is odd

for h = 0;:::;m ;1:
Remark 4.1. Note that Kh+1 may be empty for some h and in this case th+1 = T
and m = h+1: On the other hand, as the di erence M(#);%L is uniformly continuous
in [0;T] (see Proposition 2.1 and (4.1)), the switching times are at most nitely many
owing to (4.2).
In analogy with [11, Theorem 5.1], we have
Theorem 4.1. Let (2.1{5), (2.17{19), (4.1{3) hold. Then there exists a unique solu-
tion to Problem (P2).
Before proving Theorem 4.1, we establish a preliminary result about the number of
switching times. The next assertion improves that of Remark 4.1, taking advantage of
the fact that there is a modulus of continuity for M(#) uniform with respect to any
admissible #:
Lemma 4.1. Under the same hypotheses as in Theorem 4.1, for
w 2 B :=
n
r 2 L1(0;T) : krkL1(0;T) 1
o
let (#(w);(w)) be the unique solution to (SP) with #e = F[w]: Then there exists some
number 0 such that, for any w 2 B and for all times tL; tU 2 [0;T] ful lling
M(#(w))(tL) = %L(tL); M(#(w))(tU) = %U(tU); (4:8)
there holds
jtL ;tUj ;
and consequently the function M(#(w)) commutes at most [T= ] times between the
threshold functions %L and %U ([T= ] denoting the integer part of T= ):
12

Proof. Since w lies in B; thanks to (2.22), (2.15), and (2.21) there is a constant 5
such that
jM(#(w))(tL) ;M(#(w))(tU )j 5jtL ;tU j1=2
; (4:9)
where 5 depends only on 1; 3; 4; kfkW 1;1(0;T ;H) ; and ke0kV : Recalling (4.1),
let 1 0 be such that
j%L(t) ;%L()j
2 whenever t; 2 [0; T]; jt ;j 1:
Therefore, from (4.8{9) it may happen that either jtL ;tU j 1 or
5jtL ;tU j1=2
j%L(tL) ;%U (tU )j %U (tU ) ;%L(tU ) ;j%L(tL) ;%L(tU )j
2
because of (4.2). Hence we can choose := min

1; 2
=(25)2
: Concerning the
last part of the statement we also remind (2.13) and (4.3), which imply that, provided
t1 T; there is some 0 2 [t0; t1) with M(#(w))(0) = %L(0):
Proof of Theorem 4.1. We basically reproduce the inductive argument devised in [11,
Proof of Theorem 5.1]. On account of (4.4), we begin picking
w0(t) = +1 8 t 2 [0; T]
and taking the triplet (#(w0); (w0); w0); where (#(w0); (w0)) solves Problem (SP)
when #e = F[w0] (cf. Theorem 2.1). Consider the set
D1 := ft 2 (0; T] : M(#(w0))(t) = %U (t)g:
If D1 ;; then (#(w0); (w0); w0) provides the unique solution to (P2). Otherwise,
t1 := inf D1 is a minimum due to the continuity of M(#(w0)) and Lemma 4.1 yields
t1 : Further, we de ne
w1(t) :=

w0(t) if t 2 [0; t1);
;1 if t 2 [t1; T];
the solution (#(w1); (w1)) of (SP) with #e = F[w1]; and the set
D2 := ft 2 (tL; T] : M(#(w1))(t) = %L(t)g:
Reasoning as above, we conclude that (#(w1); (w1); w1) gives the unique solution to
(P2) unless D2 6= ;: In this alternative situation we introduce t2 := inf D2 and
w2(t) :=

w1(t) if t 2 [0; t2);
+1 if t 2 [t2; T]:
13

Similarly, we realize that t2 2 D2 and t2 2 : Then we can start again by considering
(#(w2);(w2);w2): Proceeding by induction, it is clear that there exist m 2 N; a triplet
(#(wm);(wm);wm); and a sequence of switching times fthgm
h=0 such that m T= ;
tm = T; and the triplet represented by # = #(wm); = (wm); z = wm uniquely
satis es (2.8{13) and (4.4{6).
Remark 4.2. As one can easily observe, our analysis covers the degenerate cases when
M(#) just touches one of the thresholds %L; %U or coincides with one of them on a
time interval.
5. Problem (P3): hysteresis operator of Preisach type
The case (C) of the Introduction is characterized by position (1.11) for suitable choices
of the measure and of the two{valued function : The latter has to ful ll
: P ! f;1;1g is Borel measurable. (5:1)
Regarding the former, in order to prove an existence result for Problem (P3) (which is
stated precisely below) we require that
is a nonnegative Borel measure with bounded density; (5:2)
(f%1gR) = (Rf%2g) = 0 8 (%1;%2) 2 R2
: (5:3)
Within this framework, the operator W3 enjoys two important properties.
Proposition 5.1. Under the assumptions (5.1{3), there hold
kW3(r)kL1(0;T) (P) +1 8 r 2 C0
([0;T]); (5:4)
W3 is strongly continuous from C0
([0;T]) to C0
([0;T]): (5:5)
Moreover, denoting by ` the bidimensional Lebesgue measure, if
(A) `(A) for all Lebesgue measurable sets A P; (5:6)
for some constant ; then there exists a constant 6; depending only on (P) and
; such that, for all r1; r2 2 C0
([0;T]);
j(W3(r1) ;W3(r2))(t)j 6 kr1 ;r2kC0([0;t]) 8 t 2 [0;T]: (5:7)
Proof. As H(%1;%2)(r;) 1 a.e. in (0;T) (see the Introduction, above (1.11)),
(5.4) is a straightforward consequence of (5.2). Concerning (5.5), we just notice that it
follows from (5.3), referring to [14, Section IV.3, Theorem 3.2] for details. The stronger
14

Lipschitz continuity (5.7) is ensured by assumption (5.6) (which implies (5.3), of course)
thanks to, e.g., [13, formula (1.14) and Proposition 8]. More general conditions yielding
(5.7) are examined in [14, Section IV.3].
After these preliminaries, let us come to the formulation of (P3).
Problem (P3). Find # 2 L1(0;T;V)W1;1(0;T;H) and 2 L1(0;T;H) satisfying
(2.9{13) and
;k #n = (#; ;F[W3(M(#))]) a.e. on T : (5:8)
By exploiting just (5.5) (and not (5.7)), we are able to show existence for (P3).
More precisely, we have
Theorem 5.1. Let (2.1{5), (2.17{19), (5.1{3) hold. Then there exists a solution to
Problem (P3).
Proof. Let us rst put Problem (P3) in a convenient xed{point setting by introducing
the operator
S3 : C0
([0;T]) ! C0
([0;T]); S3(r) := M(#(r)); (5:9)
where
(#(r);(r)) is the unique solution to (SP) with #e = F[W3(r)]: (5:10)
On account of Propositions 2.1{2 and Theorem 2.1, one can readily check that S3 is
well de ned. Besides, observe that solving (P3) is equivalent to nding a xed point for
S3: By means of (5.4), (2.22), (5.10), (2.15), and (2.21) we deduce that S3(C0
([0;T]))
C0;1=2
([0;T]) and we are able to nd a constant 7; depending on (P) and on the
same quantities as 5 does (cf. (4.9)), such that
kS3(r)kC0;1=2([0;T ]) 7 8 r 2 C0
([0;T]): (5:11)
Hence, due to the Ascoli theorem, S3 maps C0
([0;T]) (which is nonempty, closed,
and convex) into a relatively compact subset of C0
([0;T]): Therefore, if we prove the
continuity of S3; then we achieve the proof by the Schauder theorem. To this aim,
let frj gj2N be a sequence converging to some r in C0
([0;T]): Then (5.5), (2.23),
(5.10), (2.16), (2.20), and (5.9) allow us to infer, step by step, that W3(rj ) ! W3(r) in
C0
([0;T]); F[W3(rj )] ! F[W3(r)] in C0
([0;T];L2
(;)); #(rj ) ! #(r) in C0
([0;T];H)
and #;(rj ) ! #;(r) in L2
(0;T;L2
(;)); and nally
S3(rj ) ! S3(r) in L2
(0;T) (5:12)
as j % 1; all convergences being strong. On the other hand, in view of (5.11), by
compactness there are a subsequence fjhgh2N and an element w such that
S3 (rjh
) ! w in C0
([0;T]) (5:13)
15

as h % 1: Combining (5.12) and (5.13), the uniqueness of the limit entails w = S3(r)
and
S3(rj ) ! S3(r) in C0([0;T])
for the whole sequence, that is the desired conclusion.
Remark 5.1. In the study of Problem (P1) we could not argue this way since the
operator W1 was set{valued and, further, without strong continuity properties.
Under additional assumptions on and on the location of the thermostat sensors,
we can also prove uniqueness.
Theorem 5.2. Let (2.1{5), (2.17{19), (5.1{3), (5.6), and
!S 0 (5:14)
hold. Then Problem (P3) admits a unique solution.
Proof. By contradiction assume that there are two pairs (#i;i); i = 1;2; solving (P3).
By comparing (5.8) and (2.14), we put #i
e = F[W3(M(#i))]); i = 1;2; and use (2.23),
(5.7), and (2.20) to obtain
(#1
e ;#2
e)(t) 2
;
Z t
0
j4(W3(M(#1)) ;W3(M(#2)))(s)j2
ds
(46)2
Z t
0
kM(#1) ;M(#2)k2
C0([0;s]) ds

46 k!I kL2( 0)
2Z t
0
k#1 ;#2k2
C0([0;s];H) ds
for any t 2 [0;T]: Now, we can invoke (2.16) and get
(#1
e ;#2
e)(t) 2
; 8
Z t
0
(#1
e ;#2
e)(s) 2
; ds 8 t 2 [0;T];
with, for instance, 8 =
;
246 k!I kL2( 0)
2
T: Then the Gronwall lemma enables
us to establish that #1
e = #2
e a.e. in T ; whence the thesis is an outcome of Theo-
rem 2.1.
Remark 5.2. The uniqueness theorem works in a restricted framework essentially
because of (5.14). We do not know whether or not this restriction can be removed,
having at our disposal only an inequality like (5.7) for the mapping W3: In fact, the
reader may check that if (5.7) was replaced by
j(W3(r1) ;W3(r2))(t)j 6 j(r1 ;r2)(t)j; (5:15)
16

or more generally by
k(W3(r1) ;W3(r2))kL2(0;t) 6 j(r1 ;r2)jL2(0;t) ; (5:16)
then the above argument could be suitably modi ed in order to apply to the stan-
dard situation where !S 2 L2
(;0;[0;+1)): On the other hand, a Preisach hysteresis
operator is not expected to enjoy properties like (5.15) or (5.16).
6. Auxiliary parabolic problems
The last part of the paper is concerned with the proof of Theorems 2.1 and 2.2. To
sum up, we have to show the well{posedness of Problem (SP) and, in particular, to
derive estimates (2.15) and (2.16). Existence of a solution is proved passing through
a parabolic regularization of (SP). The aim of this section is that of preparing some
technical results for two related problems, one linear and the other nonlinear.
Therefore, we x a parameter 0 (subject to tend to 0 elsewhere) and, letting
the data '; F; #0; #e ful ll
' 2 H1
(0;T); (6:1)
F 2 H1
(0;T;H); F(;0) 2 V; (6:2)
#0 2 V; #0 2 V; (6:3)
#e 2 W2;1
(0;T;L2
(;)); (6:4)
;#0n = (#0; ;#e(;0)) a.e. on ;; (6:5)
we consider
Problem (LP). Find # 2 C1
([0;T];V)H2
(0;T;H) satisfying
# 2 H1
(0;T;H); (6:6)
#n 2 C1
([0;T];L2
(;)); (6:7)
@t('0# + ' #) ;# ;k # = F a.e. in QT ; (6:8)
#(;0) = #0 a.e. in ; (6:9)
;#n ;k #n = (#; ;#e) a.e. on T : (6:10)
Regarding this linear problem, we can state
Theorem 6.1. Under the assumptions (2.3) and (6.1{5), Problem (LP) has a unique
solution.
17

Proof. Taking (6.10) into account, let us rewrite equation (6.8) as
'0h#t;vi+ (r#;rv) +
Z
;
#;v; + hR(#);vi
= hF;vi+
Z
;
#ev; 8 v 2 V; a.e. in(0;T); (6:11)
where the mapping R is speci ed by
hR();vi := h@t(' );vi+ (k r;rv)
for any 2 L2(0;T;V): Recalling the well{known formulas
a b = a(0)(1 b) + at 1 b; @t(a b) = a(0)b + at b; (6:12)
which hold whenever they make sense, by (6.1) and (2.3) one easily veri es that R is
linear and continuous from C0([0;T];V)H1(0;T;H) to L2(0;T;H)+W1;1(0;T;V0):
Then, an application of [1, Teorema 6.1] allows us to deduce that there exists one and
only one function # 2 C0([0;T];V) H1(0;T;H) solving the Cauchy problem (6.11),
(6.9). From (6.11) we plainly recover (6.8) in the sense of distributions, and so the
condition
# + k # 2 L2(0;T;H) (6:13)
follows. Thanks to the classical theory of Volterra integral equations, it turns out that
# 2 L2(0;T;H): Thus, (6.13) ensures the validity of (6.8) and of the equality for the
traces, that is,
;#n ;k #n = (#; ;#e) in L2(0;T;H;1=2(;)): (6:14)
Moreover, owing to (6.4) and to the regularity of #; (6.14) implies
#n 2 C0([0;T];L2(;)) (6:15)
and consequently (6.10) is ful lled.
It only remains to infer the further smoothness of #: Observe that u := @t#
formally satis es (see (6.12))
'0hut;vi+ (ru;rv) +
Z
;
u;v; + hR(u);vi = hFt ;'0#0;vi
;k(r#0;rv) +
Z
;
(@t#e)v; 8 v 2 V; a.e. in (0;T) (6:16)
in addition to
u(;0) = ';1
0 (F(;0) ;'(0)#0 + #0) a.e. in : (6:17)
18

We note that (6.17) has been obtained by reading (6.8) at the initial time and using
(6.9). Due to (6.1{4), the abstract result in [1, Teorema 6.1] applies to (6.16{17) as well.
Therefore, this initial value problem has one and only one solution, say e
u; belonging to
C0([0;T];V)H1(0;T;H): Besides, arguing as before (cf. (6.13{15)) we achieve that
e
u 2 L2(0;T;H) and e
un 2 C0([0;T];L2(;)) (6:18)
(here (6.5) plays a role). At this point, it is not dicult to check that # #0 + 1 e
u
whence # 2 C1([0;T];V)H2(0;T;H) and e
u u @t#: Since (6.18) and (6.3) entail
(6.6{7), the proof is completed.
Next, we show a similar result for a nonlinear version of (LP). More precisely, let
G be a (possibly nonlinear and nonlocal) operator such that
G : H1(0;t;H) ! H1(0;t;H) 8 t 2 (0;T]; (6:19)
G[](;0) 2 V whenever 2 H1(0;T;H) and (;0) 2 V: (6:20)
A constant 9 is supposed to exist in order that
kG[]kL2(0;T;H) 9; (6:21)
kG[1] ;G[2]kL2(0;t;H) 9 k1 ;2kL2(0;t;H) (6:22)
for any t 2 (0;T] and for all ; 1; 2 2 H1(0;T;H): In this setting we introduce
Problem (NP). Find # 2 C1([0;T];V) H2(0;T;H) satisfying (6.6{7), (6.9{10),
and
@t('0# + ' #) ;# ;k # = G[#] a.e. in QT : (6:23)
We are still able to prove
Theorem 6.2. Let (2.3), (6.1), and (6.3{5) hold. Moreover, let G be as in (6.19{22).
Then there exists a unique solution to Problem (NP).
Proof. Once more we exploit a xed{point technique. Taking 2 H1(0;T;H) with
(;0) 2 V; due to (6.19{20) and Theorem 6.1 we can consider the function
# = #() 2 C1([0;T];V)H2(0;T;H) solving (LP) for F = G[]: (6:24)
Multiply the corresponding equation (6.8) by # and integrate by parts in space and
time over Qt; t 2 (0;T]: With the help of (6.9{10) and (6.12) one easily obtains
'0
2 k#(;t)k2 +
ZZ
Qt
jr#j2 + k(0)
2 k(1 r#)(;t)k2
+
ZZ
t
j#;j2 = '0
2 k#0k2 +
4
P
j=1
Ij(t); (6:25)
19

where
I1(t) := ;
ZZ
Qt
('(0)# + '0
#)#; I2(t) := ;
Z t
0
(r(k0
1 #)(;s);r#(;s))ds
I3(t) :=
ZZ
t
#e#;; I4(t) :=
ZZ
Qt
G[]#:
To estimate I1(t) we have recourse to the Young inequality for the convolution product,
namely
ka bkLr (0;T ;X) kakLp(0;T ) kbkLq(0;T ;X) 8 a 2 Lp
(0;T); b 2 Lq
(0;T;X); (6:26)
letting X denote a real Banach space and 1 p; q; r 1 ful ll 1=r = (1=p)+(1=q);1:
It is straightforward to get
jI1(t)j

j'(0)j+ k'0
kL1(0;T )
Z t
0
k#(;s)k2ds: (6:27)
By (6.12) it is a standard matter to verify that
I2(t) = ;((k0
1 r#)(;t);(1r#)(;t))
+
Z t
0
;
k0
(0)k(1 r#)(;s)k2 + ((k00
1 r#)(;s);(1r#)(;s))

ds;
from which we derive
jI2(t)j k(0)
4 k(1 r#)(;t)k2 + 10
Z t
0
k(1 r#)(;s)k2ds; (6:28)
where 10 is a positive constant only depending on k(0); T; and kkkW 2;1(0;T ) : Note
that here (6.26) has been used twice, the rst time for r = 1 and p = q = 2:
Concerning I3(t) and I4(t); we have that
jI3(t)j+ jI4(t)j 2
ZZ
t
;
j#;j2 + j#ej2
+ 1
2
Z t
0
;
kG[](;s)k2 + k#(;s)k2
ds: (6:29)
Combining (6.27{29) with (6.25), then applying the Gronwall lemma, one nds a con-
stant 11 (whose dependences are clear) such that
k#k2
C0([0;t];H) + kr#k2
L2(0;t;HN ) + k1 r#k2
C0([0;t];HN ) + k#;k2
L2(t)
11
n
k#0k2 + k#ek2
L2(t) + kG[]k2
L2(0;t;H)
o
; (6:30)
this inequality obviously holding for any t 2 [0;T]:
20

Besides estimate (6.30), we need a higher order estimate which is obtained by
multiplying equation (6.23) by ;# (this is admissible owing to (6.6)) and integrating
over (0;t); with t 2 (0;T]: In this case, (6.9{10) and the Green formula help us
to infer that
'0
2 kr#(;t)k2 +
Z t
0
k#(;s)k2ds + k(0)
2 k(1 #)(;t)k2
= '0
2 kr#0k2 +
8
P
j=5
Ij (t); (6:31)
where
I5(t) :=
ZZ
t
('0@t#; + '(0)#; + '0
#;)#n;
I6(t) := ;
Z t
0
(r('(0)# + '0
#)(;s);r#(;s))ds;
I7(t) := ;
Z t
0
h(k0
1 #)(;s)#(;s)ids; I8(t) := ;
ZZ
Qt
G[]#:
As @t#; = ;@t#n ;k(0)#n ;k0
#n + @t#e a.e. on T because of (6.10) and (6.7),
playing on I5(t) with (6.26) and the elementary Young inequality, it is not dicult to
get
I5(t) ;'0
2 k#n(;t)k2
; + '0
2 k#0nk2
; ; k(0)'0
2 k#nk2
L2(t)
+12
Z t
0
k#nk2
L2(s) ds + k#ek2
L2(t) + k#;k2
L2(t)

(6:32)
for some constant 12 which depends exactly on k(0); '0; ; kk0
kL2(0;T ) ; j'(0)j; and
k'0
kL1(0;T ) : On the other hand, arguing as in the deduction of (6.27{28), we are led to
jI6(t)j (j'(0)j+ k'0
kL1(0;T ))
Z t
0
kr#(;s)k2ds; (6:33)
jI7(t)j k(0)
4 k(1 #)(;t)k2 + 10
Z t
0
k(1 #)(;s)k2ds: (6:34)
It remains to point out that
jI8(t)j
2
Z t
0
k#(;s)k2
ds + 1
2
Z t
0
kG[](;s)k2
ds: (6:35)
Now, we estimate the right hand side of (6.31) with the aid of (6.32{35) and also of
(6.30). Moving the negative terms on the left hand side and invoking the Gronwall
21

lemma, we realize that
kr#k2
C0([0;t];HN ) + k#k2
L2(0;t;H) + k1 #k2
C0([0;t];H)
+ k#nk2
C0([0;t];L2(;)) + k#nk2
L2(t)
13
n
k#0k2
V + k#0nk2
; + k#ek2
H1(0;t;L2(;)) + (1 + 1=)kG[]k2
L2(0;t;H)
o
(6:36)
for any t 2 [0;T]; the constant 13 being independent of : Moreover, on account of
(6.30), (6.36), (2.3), and (6.24), a comparison in (6.8) yields an analogous bound for
k#tk2
L2(0;t;H):
Therefore, thanks to (6.21) there is a constant 14; depending only on 9; ; ;; T;
'0; k(0); ; k'kW1;1(0;T) ; kkkW2;1(0;T) ; k#0k2
V ; k#0k2
H ; k#ek2
H1(0;t;L2(;)) ; and (we
remind that is xed in this section) such that (cf. (6.24))
k#()kH1(0;T;H) 14 for all 2 H1(0;T;H) satisfying (;0) 2 V: (6:37)
Then, if we endow the set
XT :=
n
2 H1(0;T;H) : (;0) = #0 a.e. in ; kkH1(0;T;H) 14
o
with the distance function
dXT
(1;2) := k1 ;2kC0([0;T];H) ; 1; 2 2 XT;
it turns out that the operator N : z 7! #(z) acts from XT into itself (by virtue of
(6.9) and (6.37)) and, due to the weak lower semicontinuity of the norm kkH1(0;T;H) ;
XT is a complete metric space. Thus, to prove the theorem it is sucient to show that
either N or a suitable power of it is a contracting mapping. Letting 1; 2 2 XT and
reasoning as for (6.30), we can easily conclude that
kN(1) ;N(2)k2
C0([0;t];H) 11 kG[1] ;G[2]k2
L2(0;t;H) 8 t 2 [0;T]:
Hence, from (6.22) it follows that
jdXT
(N(1);N(2))j2
15
Z t
0
jdXs
(1;2)j2
ds 15 jdXt
(1;2)j2
t (6:38)
for any t 2 [0;T] and for 15 = 11(9)2; the de nition of dXt
being obvious. As it
is known, inequalities like (6.38) allow you to determine some m 2 N such that Nm is
a contraction from XT into itself. Hence an application of the generalized Contracting
Mapping Principle ends the matter.
22

7. Proof of Theorem 2.1
Since the main aim here is to recover the existence of one solution to Problem (SP), we
go directly to implement our approximation procedure in terms of the parameter 0:
Then we will derive estimates independent of and pass to the limit as 0:
Henceforth, we let (2.1{7) hold. For 0 1 consider some regularizing sequences
f'g; ffg; f#
0g; f#
eg ful lling
' 2 H1(0;T) 8 2 (0;1]; (7:1)
' ! ' strongly in W1;1(0;T) as 0; (7:2)
f 2 H1(0;T;H); f(;0) 2 V 8 2 (0;1]; (7:3)
f ! f strongly in W1;1(0;T;H) as 0; (7:4)
e
0 ! e0 weakly in V as 0; (7:5)
#
e 2 W2;1(0;T;L2(;)); #
e(;0) = ';1
0 e
0; 8 2 (0;1]; (7:6)
#
e ! #e weakly in H1(0;T;L2(;)) as 0: (7:7)
Moreover, for any 2 (0;1] we introduce the solution #
0 2 V of the elliptic variational
equality
h#
0 ;';1
0 e
0;vi+ (r#
0;rv) +
Z
;
(#
0; ;#
e(;0))v = 0 8 v 2 V: (7:8)
One readily sees that #
0 solves the boundary value problem
#
0 ;#
0 = ';1
0 e
0 a.e. in ; (7:9)
;#
0n = (#
0; ;#
e(;0)) a.e. on ;: (7:10)
In addition, choosing the test function v =
;
#
0 ;';1
0 e
0

= in (7.8), thanks to (7.5{6)
it is straightforward to infer that
2

#
0 ;';1
0 e
0
2
+ kr#
0k2
+ 2

k#
0; ;#
e(;0)k2
;
r
;
';1
0 e
0
2
16 f1 + ke0kVg
for any 2 (0;1] and for some constant 16 depending only on '0: Consequently, the
convergences
#
0 ! ';1
0 e0 weakly in V and strongly in H as 0 (7:11)
and the boundedness
k#
0nk2
; ( =2)16 f1 + ke0kVg 8 2 (0;1] (7:12)
23

are entailed by (7.5) and (7.10) (we remind that V is compactly embedded into H):
Next, let us approximate the Heaviside graph H by (cf. [5, Appendix and especially
Remark 9.1])
H(s) :=
8

:
0 if 0;
;
3s2 ;2s3
=3 if 0 s ;
1 if :
(7:13)
Note that, for any 2 (0;1];
H : R ! [0;1] is a maximal monotone graph in R2; (7:14)
H 2 C1(R) W1;1
(R): (7:15)
Then our regularized version of (SP) reads
Problem (SP). Find # 2 C1([0;T];V)H2(0;T;H) and 2 L1
(QT ) satisfying
# 2 H1(0;T;H); (7:16)
#n 2 C1([0;T];L2(;)); (7:17)
2 C0([0;T];V)W1;1
(0;T;H); (7:18)
@t('0# + ' # + ) ;# ;k # = f a.e. in QT ; (7:19)
= H(#) a.e. in QT ; (7:20)
#(;0) = #
0 a.e. in ; (7:21)
;#n ;k #n = (#; ;#
e) a.e. on T : (7:22)
Referring to the previous section, now we let
G[] := f ;@t( H()) = f ; (0)H() ; 0
H() (7:23)
for 2 H1(0;T;H); and observe that this operator obeys (6.19{22) by virtue of (7.3),
(7.15), and (2.2) (obviously, the constant in (6.22) blows up as 1= does). Thus, on
account of (7.20), Problem (SP) is nothing but a particular case of Problem (NP)
with G given by (7.23). In view of (7.1), (7.5), and (7.9{10), one checks that Theorem
6.2 applies and therefore Problem (SP) admits a unique solution (#;):
The second step of the proof consists in proving some a priori estimates on #
which allow us to pass to the limit in (SP) as 0 and to get a solution to the limit
problem (SP). Let us start by noting that (6.30) and (7.23) directly yield
k#k2
C0([0;T];H) + kr#k2
L2(0;T;HN) + k1 r#k2
C0([0;T];HN) + k#;k2
L2(T)
17
n
1 + kfk2
L2(0;T;H) + k#
0k2 + k#
ek2
L2(T)
o
; (7:24)
24

where the constant 17 just depends on 11; T; and k kW1;1(0;T) : Then, multiplying
equation (7.19) by ;# and integrating over Qt (with t 2 (0;T]); we obtain the
identity corresponding to (6.31). However, here we deal with I8(t) in a di erent way.
Indeed, (7.23) and standard integrations by parts enable us to deduce
I8(t) = ;h(f ; 0
H(#))(;t);(1#)(;t)i
+
Z t
0
h(@tf ; 0
(0)H(#) ; 00
H(#))(;s);(1#)(;s)ids
;
ZZ
Qt
(0)H0
(#)jr#j2
+
ZZ
t
(0)(H(#));#n: (7:25)
As H0
0 in R; recalling (7.14), (2.2), (6.26), and the elementary Young inequality,
from (7.25) it is not dicult to nd a constant 18; independent of ; such that (see
(6.31{34) for our choice of coecients)
I8(t) k(0)
8 k(1 #)(;t)k2
+ k(0)'0
4 k#nk2
L2(t)
+ 18

1 + kf(;t)k2 +
Z t
0
(1 + k@tf(;s)k)k(1#)(;s)kds

: (7:26)
Using (7.26) in place of (6.35), we argue as for the derivation of (6.36) even though here
we need a generalized version of the Gronwall lemma (like, e.g., the one stated in [1]).
This procedure leads to
kr#k2
L1(0;T;HN ) + k#k2
L2(0;T;H) + k1 #k2
L1(0;T;H)
+ k#nk2
L1(0;T;L2(;)) + k#nk2
L2(T )
19
n
1 + kfk2
W1;1(0;T;H) + k#
0k2
V + k#
0nk2
; + k#
ek2
H1(0;T;L2(;))
o
; (7:27)
with the constant 19 having the same dependences as 1 does. At this point, the
estimates (7.24) and (7.27) entail an analogous bound for k@t#k2
L2(0;T;H) via a com-
parison in (7.19). Then, owing to (7.4), (7.11{12), and (7.7) there exists a constant 20
such that
k#kL1(0;T;V)H1(0;T;H) +
p
k#kL1(0;T;H) + k1 #kL1(0;T;H)
+
p
k#nkL1(0;T;L2(;)) + k#nkL2(T )
20
n
1 + kfkW1;1(0;T;H) + ke0kV + k#ekH1(0;T;L2(;))
o
; (7:28)
where 20 relies on the same quantities as 1 does.
25

Since (7.28) and (7.20) hold for any 2 (0;1]; we can pass to the limit along a
subsequence and thus infer the existence of # and such that
# ! # weakly star in L1
(0;T;V) and weakly in H1
(0;T;H); (7:29)
# ! 0 strongly in L2
(QT ); (7:30)
1 # ! 1 # weakly star in L1
(0;T;H); (7:31)
#n ! 0 strongly in L1
(0;T;L2
(;)); (7:32)
#n ! #n weakly in L2
(T ); (7:33)
! weakly star in L1
(QT ) (7:34)
as 0: In particular, (7.29) implies that (cf. (3.20))
# ! # strongly in C0
([0;T];H) as 0: (7:35)
Convergences (7.29{34) combined with (7.2), (7.4), (7.7), and (7.11) allow us to take
the limit in (7.19{22) and recover (2.11{14). The details are either trivial or developed
in [4, Appendix]. Therefore, the pair (#;) solves Problem (SP). Moreover, it satis es
(7.28) because of the weak star lower semicontinuity of norms, whence (2.15) follows by
additionally comparing the terms of (2.11) and (2.14) (refer to (6.12) too).
Finally, in order to complete the proof of Theorem 2.1, we just remark that unique-
ness is a consequence of (2.16) coupled with a proper reasoning on equation (2.11) to
conclude also for the other variable :
8. Proof of Theorem 2.2
We remind that (#1;1) and (#2;2) denote the solutions of (SP) corresponding to
the respective data #1
e and #2
e; which both ful ll (2.6{7). Setting := #1 ; #2 and
X := 1 ;2; from (2.11) one can easily derive
(0)X + 0
X = L() a.e. in QT ; (8:1)
where
L() := ;@t('0 + ' ) + k : (8:2)
It is known that (cf., e.g., [10, Chapter 2, Section 3]) (8.1) can be equivalently rewritten
as (0)X = L() ; L(); that is,
@t('0 + ' ) ;k + (0)X = ; L() a.e. in QT ; (8:3)
the function 2 W1;1(0;T) being the the unique solution to the integral equation
(0) + 0
= 0
in [0;T]:
26

We point out that
k kW 1;1(0;T ) 21 (8:4)
for some constant 21 which only depends on (0); T; and k 0
kW 1;1(0;T ) :
Next, multiply equation (8.3) by and integrate in space and time over Qt; for
t 2 (0;T]: On account of (2.13{14) (observe in particular that (;0) = 0 a.e. in );
by the Green formula and (6.12) it is straightforward to check that
'0
2 k(;t)k2 + k(0)
2 k(1 r)(;t)k2 + k;k2
L2(t)
+ (0)
ZZ
Qt
X =
11
P
j=9
Ij (t); (8:5)
where
I9(t) :=
ZZ
Qt
@t( ('0 + ' ) ;' );
I10(t) :=
Z t
0
(r(( k ;k)0
1 )(;s);r(;s))ds;
I11(t) :=
ZZ
t
(e + (; ;e));;
and e := #1
e ;#2
e: In order to estimate I9(t); for
@t( ('0 + ' ) ;' ) = ('0 (0) ;'(0)) + ('0 0
+ 0
' ;'0
) ;
one can exploit (8.4), (2.1), and (6.26) to determine a constant 22; whose dependences
are obvious, such that
jI9(t)j 22
Z t
0
k(;s)k2ds: (8:6)
Observing that k 2 W2;1(0;T) and comparing I10(t) with I2(t); we have that
(cf. (6.28))
jI10(t)j k(0)
4 k(1 r)(;t)k2 + 23
Z t
0
k(1 r)(;s)k2ds; (8:7)
the constant 23 being dependent on k(0); T; k kW 1;1(0;T ) ; and kk0
kW 1;1(0;T ) : On
the other hand, (6.26) and standard inequalities lead to
jI11(t)j 2 k;k2
L2(t) + kek2
L2(t)
+ 2 k k2
L2(0;T )
Z t
0

k;k2
L2(s) + kek2
L2(s)

ds: (8:8)
27

Now, we collect (8.5{8) and note that
(0)X 0 a.e. in QT
due to (2.2), (2.12), and to the monotonicity of H: Thus, there is a constant 24;
having the same dependences as 2 does, such that
'0
2 k( ;t)k
2 + k(0)
4 k(1 r)( ;t)k
2 + 2 k;k
2
L2(t)
24

kek
2
L2(t) +
Z t
0

k( ;s)k
2 + k(1 r)( ;s)k
2 + k;k
2
L2(s)

ds

for any t 2 [0;T]: Finally, we just notice that it is not dicult to recover (2.6) taking
advantage of the Gronwall lemma.
References
[1] C. Baiocchi, Sulle equazioni di erenziali astratte lineari del primo e del secondo ordine negli
spazi di Hilbert, Ann. Mat. Pura Appl. (4) 76 (1967) 233{304.
[2] V. Barbu, Nonlinear semigroups and di erential equations in Banach spaces (Noordho ,
Leyden, 1976).
[3] P. Colli, G. Gilardi, and M. Grasselli, Weak solution to hyperbolic Stefan problems with
memory, NoDEA Nonlinear Di erential Equations Appl. (to appear).
[4] P. Colli and M. Grasselli, Hyperbolic phase change problems in heat conduction with mem-
ory, Proc. Roy. Soc. Edinburgh Sect. A 123 (1993) 571{592.
[5] P. Colli and M. Grasselli, Justi cation of a hyperbolic approach to phase changes in mate-
rials with memory, Asymptotic Anal. 10 (1995) 303{334.
[6] P. Colli and M. Grasselli, Convergence of parabolic to hyperbolic phase change models with
memory, Adv. Math. Sci. Appl. 6 (1996), 147-176.
[7] A. Friedman and K.{H. Ho mann, Control of free boundary problems with hysteresis,
SIAM J. Control Optim. 26 (1988) 42{55.
[8] K. Glasho and J. Sprekels, An application of Glicksberg's theorem to set{valued integral
equations arising in the theory of thermostats, SIAM J. Math. Anal. 12 (1981) 477{486.
[9] I.L. Glicksberg, A further generalization of the Kakutani xed point theorem, with appli-
cation to Nash equilibrium points, Proc. Amer. Math. Soc. 3 (1952) 170{174.
[10] G. Gripenberg, S.{O. Londen, and O. Sta ans, Volterra integral and functional equations,
Encyclopedia Math. Appl. 34 (Cambridge Univ. Press, Cambridge, 1990).
[11] K.{H. Ho mann, M. Niezg
odka, and J. Sprekels, Feedback control via thermostats of mul-
tidimensional two{phase Stefan problems, Nonlinear Anal. 15 (1990) 955{976.
[12] J. Sprekels, Automatic control of one{dimensional thermomechanical phase transitions, in:
J. F. Rodrigues, ed., Mathematical Models for Phase Change Problems, Internat. Ser.
Numer. Math. 88 (Birkh
auser, Basel, 1989) 89-98.
[13] A. Visintin, On the Preisach model for hysteresis, Nonlinear Anal. 8 (1984) 977{996.
[14] A. Visintin, Di erential models of hysteresis, Appl. Math. Sci. 111 (Springer, Berlin,
1994).
28

Automatic Control Via Thermostats Of A Hyperbolic Stefan Problem With Memory

More Related Content

Similar to Automatic Control Via Thermostats Of A Hyperbolic Stefan Problem With Memory (20)

More from Crystal Sanchez (20)

Recently uploaded (20)

Automatic Control Via Thermostats Of A Hyperbolic Stefan Problem With Memory