This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
IR3 satisfies
A
IR, j = 0,1,2, denote the harmonic vector fields defined by h0[ip]
x i-> x x V(/j(x),
hilx]
x M. V X - (£*)x + \{\*\2 ~ 1)V [(£ + \yl£} x M- Vf "1i9.
h2m
x,
Remember that
IR3. Let an : B —> M 3 be the harmonic extension of a. Then there exist unique harmonic functions (p,x,$ '• B ^>- TR, with ip(0) = x(0) = &(Q) = 0 such that aH = h0[(p] + h^x] + hzltf] + ex, with c = j ^ JdB(ax)
da. Note that if d = 0 and c = 0, then a is a tangent vector field.
Proof The proof is divided into several steps. 1. Step 1. Split the vector field a into a tangential and a normal component as follows : a = at + an + en, with c = ^ JgB(a• n) da, o„ = {a-n-c)n and at = a — an- en. Note that at is tangential. 2. Step 2. Define $ as the solution of the Dirichlet problem Ai?(x) = 0 , x € B,
•d(x) =an
— c,
xedB.
3. Step 3. Note that for any x e dB, the vector field a - V£ _1 t?(x) — ex is tangential. 4. Step 4. The harmonic extension x i-> aH(x) — 'V£~1'd(x)—cx satisfies the conditions of lemma 2. This means it can be uniquely written as /io[?] + hi[x] f° r suitable harmonic
0 Vw
(1.2)
If H denotes the following hyperplane of R w H = {w£KN/wfj,
= 0}
(1.3)
(w • n denotes the scalar product of w and fi) we assume t h a t there exist w\,...,
wN N elements of W(^>) such t h a t w{ eHVi
= l,...,N
66
(1.4)
67 0G
ri^CoK),^)
where Co(tui)^=1 is the convex hull of the uVs, riH(Co(«;i)-I1) tively to the topology of H.We denote by W„1,0°(fi) the set W^°°(Q) = {ve
(1.5)
denotes its interior rela-
W1-00^) I v(x) = 0 on T}.
Then we consider the following minimization problem inf
/
Ar = Span(uij - WI)J=2,...,JV.
(1-7)
where Span(a;)i denotes the vectorial space spanned by the vectors
N
t=l
i=l
By (1.5)
We adopt the following notations H+ = {weIlN N
/ w-n>0},
H. = {weR /w-li<0},
TJ+ = {w<EKN N
Tl- = {w£~R
W(
)* = % ) \ K
i= N
J
W/J,>0}
I
w-n<0}
l,...,N}
The case where W(ip) spans a proper subspace of R was studied by M. Chipot and C. Collins (see for example [C], [C.C.]). In this paper, we are concerned by the case where W(
0 = f [ Jn
ip(X) dvx{\)dx
JSIJR"
It follows that
L
1. N (yf) such that | 5 - « U < K ( t ) - 0 , ( 8 ) 1 ( 1 + 1^(2)1*) and V'(5)-^W<|6r(*)-6r(s)|(l d\z\2H, »)nB) T>Ot>T ), 0 (u) = 4>(u) if u e E°(Q) and 4>(u) = limn_KX)(/>(url), where un G E°(Q.) is an arbitrary sequence converging to u in E\oc. This is a linear bounded functional on E(fl). Denote all such functionals E. It is a linear subspace in E*. Suppose that E ^ E*. We take a functional ip £ E*, which does not belong to E. Let ip0 be a restriction of ip on E°. Then V'o S {E0)*. As above we can define the functional ip0 € (-E)*. By assumption i/i T^ 4>0. Denote ip = ip — xji0. Then V- = 0, 3a and sup |?'(r)| < \ja. Given any p £ 21 we introduce the Hanzawa transformation ftM._J*(J'(*).A(s) U W _ 2 and hence that there exists a p > 2sothat |Vu|, |VD| € LP((T,OO)XQ.) forsomep > 2 a n d r > 0. If we differentiate (6) with respect to t and set du/dt = w we have dw/dt = V• di(a:) Vw — k(x)udv/dt — k(x)vdu/dt. Similarly for dv/dt = z we have <9z/d£ = V-d,2(x)Vz+k(x)vdu/dt+k(x)udv/dt. Working with the nonlinearity we observe that k(x)udv/dt + k(x)vdu/dt e £OO((T, OO), L 2 (^))- We now use regularity and embedding results for analytic semigroups, [13], to argue that there is a p > 2 so that |Vw| = \Vdu/dt\, |Vz| = \Vdv/dt\ e L«,((r, oo);L p (n)). This together with previously obtained estimates and the analogous results for |Vv\ will insure that there exists a p > 2 so that lim|||Vu(. ) t)||| P i n = # ]im|||V«(.,t)||| P i n = 0. £—>oo E and G periodic then ipe —* M.((p), as e —• 0, in a weak-star L°°(fi) sense, i.e. for any ip G Ll(Q) Jnips(x)tlj(x)dx —» M.( 0 of the kinetics on the right hand sides of (13) towards the corresponding kinetics of (15) follows, weakly in L2(Q x (0,T)) and strongly in the dual space [if1 (£2 x (0,T))]'. The goal of homogenization techniques is to handle the behavior of such quantities as — V • dfVue when e —> 0. Then one can show that there exists a positive definite symmetric matrix D^1' depending solely on d^1', $1 and Cl such that upon extracting further subsequences dfVuE« —>• D^'Vu weakly in L2{Q. x (0,T)) as e" —• 0, see [2] and [15]. Identical arguments work for the equation for ve. At this point the convergence of a suitable subsequence of (uE, fe)o<e
0 be a given final time. Set T := <3fi, E := T x (0, T), Qt := fi x (0,t), for t e (0,T] and Q := QT. Set also H := L2(fi) and V := H\n) and endow the latter space with its usual scalar product. {{v,w)) := / vwdx+ Vu -Vwdx. (2) J n J n We identify H and its dual, in order that the compact inclusion H C V holds and (V, H, V) form a Hilbert triplet. Denote by (•, •) the scalar product of both H and Hd (fi), by |-1 the associated norms and by || • \\x the norm of the generic Banach space X. Finally, indicate by (.,.) the duality pairing between V and V and by ((., .))„ the associated scalar product on V. Given any £ € V, let us now set (n:=±(C,l); (.,0) c(«) n po,2, c0>2 also satisfying (23). Let also To G (0,T] and suppose (y>i,Ci,u)i,£i) and (v?2,C2, ^21^2) is a couple of solutions to (17) related to the different initial data and fulfilling dm, Bn, 8tcu BCl e LA (0,T0; H), Vw2 e L2 (0,T0;L3 (ft)), llyi|li4(0,ro;ff) + lla*Vlllt4(0,To;ff) + \\B
4
Cauchy problem for a semilinear equation
Consider the problem ut u(x,0)
= Au - ii\Vu\i + vP i e R " , ( > 0 , = u 0 (a;)>0 xeRN.
Herep, q > 1 and /u > 0. If fi = 0 and (N-2)p < N + 2 then all global solutions decay uniformly to zero (cf. [37]). (An earlier study of global solutions of this problem with fi = 0 was performed in [21] where sign-changing solutions with fast decay (as x —> oo) were considered). If /x > 0 then there exist global solutions such that (cf. [40]) : lim u(x,t) = oo
5
for all x e R w .
Stefan problem for a semilinear equation
Consider now the following reaction-diffusion problem with free boundary: ut u(x, 0) u(s(t),t) s'(t)
= = = =
uxx + v? xe (0, s(i)), t > 0, u0(x) > 0 x e (0, s 0 ), s(0) = s 0 > 0, ux{0,t) = 0 t > 0 , - U a ; (s(t),t) t > 0 ,
where p > 1. This problem can be viewed as a simple model of a chemically reactive and heat-diffusive liquid surrounded by ice. Here u > 0 represents the temperature of the liquid phase, and the ice is assumed to be at temperature 0. Global existence, stability and blow-up for this problem were studied in [18], for previous studies see references therein and in [15]. In [15] we showed that if u is a global solution then u < C ( s 0 , ||uo||ci([0,so]).
94
6
Cauchy problem for a semilinear equation in similarity variables
Consider a solution of the problem 11, = A« + « p u(x,0) = u0{x)>0
x e RN, xeRN,
0
which blows up at t = T. For a E R " one can rescale u using backward similarity variables as follows
s = -log(r-t),
j/ = - ^ L ,
w{y,s) =
(T-t)^u{x,t).
Then w satisfies y • Vw + wp
w. = Am
y
2 w(y,-logT)
= T^u0(a
-w y € R " , s > — logT, p-1
y
(3) yeRN.
+ VTy)
Under some assumptions on u0 and p, it has been shown in [27] that for 6 > 0 there exists C = C(p, N, 6) > 0 such that for any global solution of (3), it holds w
for s> -\ogT + S.
In the original variables this can be equivalently formulated as a universal bound (independent of «o) for the blow-up rate.
7 7.1
Nonlinear boundary conditions Heat equation
The following problem is studied in [4]: ut
= Au
—
= f(u)
u(x,0)
x G D, t > 0, x£dD,t>0,
= UQ(X) X&D.
Here I? is a bounded domain in R and v is the outer normal to dD. The result from [4] as applied to the particular case with f(u) = \u\p~1u says that global solutions are bounded if (N — 2)p < N. This means that in this case there are no global positive solutions since there is no positive steady state and the trivial steady state is unstable from above. The main result of [4] and several elements of its proof were generalized in [24].
95
7.2
Porous medium equation
In [25], Lieberman studied the problem A$(u) x e D, t > 0, dv u(x,0)
/(«)
xedD,
«o(^)
x £ D.
t > 0,
Here D is a bounded domain in R N and $ is a continuous increasing function. The results from [25] say that global solutions are uniformly bounded if the growth of / is super linear but subcritical (which may be faster than polynomial if N = 1,2). For the precise meaning of "superlinear" and "subcritical" see [25].
7.3
Porous medium equation with absorption
The large time behavior of solutions of the problem ut
= (um)xx - aup \x\ < 1, t > 0 , |x| = 1, t> 0,
u(x,0)
= UQ(X) > 0
W<1,
was discussed in [8]. Here a > 0 and p,q> m > 1. Among other things it was shown in [8] that if p < 2q — m or if p = 2q — m and ma < q then blow-up in infinite time does not occur for any initial data while blow-up in finite time occurs for some initial data. On the other hand, if p = 2q — m and ma = q then all solutions are global and unbounded. Corresponding results for m = 1 can be found in [3].
8 8.1
Dynamical boundary conditions Laplace equation
Consider next the problem du
Au du
u(x,0)
0
x e D, t> 0,
h(u)
x&dD,
u0(a;) > 0
t > 0,
xedD,
where A is the Laplacian in x, D is a bounded domain in R N , N > 2, with smooth boundary, u is the outer normal to dD, h e C1 and u0 belongs to some appropriate function space. We shall consider superlinear nonlinearities h for which blow-up in finite time occurs if initial data UQ are "large". As an example we can take h{u) = |w| p_1 u — au,
p > 1, a > 0.
It is shown in [13] that all global solutions are bounded provided the growth of h is subcritical, i.e. (N — 2)p < N if h is as above. But under this assumption only an
96 a priori bound for l i m s u p ^ ^ ||w(-,t)||C(n) i s obtained. To derive an a priori bound for su Poo llu('i*)llc(ft) a stronger growth restriction on h is needed in [14]. As an application of the latter result we proved existence of sign-changing solutions of the corresponding stationary problem.
8.2
Parabolic equation
The paper [14] is devoted to the study of the problem tH = Au + \\u\p-lu du
du
, q,
x£D,t>0,
i
„„
-sr + w- = nM u at dv u(x,0) = uo{x)>0
xedD,
t > 0,
xedD,
where Q is a bounded domain in R w , v is the outer normal on dQ, p, q > 1, A, fj, £ {—1,0,1}, max{A, fj,} = 1 and \p+/iq > 0. Under our assumptions on p, q, X and n, blowup in finite time occurs for some initial data. We proved in [14] that if (JV — 2)p < N + 2 and (N — 2)q < N then all global solutions are uniformly bounded. To obtain an a priori bound for global solutions in terms of suitable norms of M0 a stronger condition on p and q is needed. Namely, if N--^)p
+-
and
(N - - ) q < N + -,
then sup||u(-,£;u0)||z,~(n) < C x(|KIU~(0), IKII/fi(n)) > oo ' provided u is a global solution. As an application of this a priori bound we proved in [14] the existence of sign changing solutions of the problem Au - lu^u
= 0
^ dv
=
x £ D,
Wo-^
xedD,
if / 5\ 1 Kp
9
Weakly coupled system
Consider the system = vt = u = V = u(x,0) = v(x,0) = «t
Aw + ?)p Av + uq 0 0 u0(x) > 0 v0(x)>Q
x e D, t > 0, x e D, t > 0, xedD, t > 0. xedD, t>0,
x e ID, iefl,
97 where p, q > 1 and D is a bounded domain in RN. It was shown in [16] that if max(p,g) < 1 + 2/(N + 1) and r > 0 then there is a constant C(D,p,q,r) > 0 such that for all global solutions it holds u(x,t) +v(x,t)
< C(D,p,q,r),
x e D, t > T.
An a priori bound (which depends on initial data) for global solutions has been established in [35] under the assumption max(p,g) + l TV + 1 pq - 1 " 2 and under the weaker assumption max(p, q) + 1 pq-1
TV ~^
N
if D = R .
10
Methods
10.1
Energy methods
The basic idea of the proof of the boundedness of global solutions in [4]-[9], [11], [12], [22]-[24] is similar. We illustrate it here for the simplest problem considered in Subsection 2.1. We show by contradiction that the W/1,2(D)-norm of any global solution must be bounded uniformly in t, and then we use known results to conclude that the C(Z))-norm is also uniformly bounded. Suppose limsup ||u(-,t)||v^i,2(£)) = oo. t—*oo 2
A uniform L (D)-estimate (or the concavity method) enables us to rule out the possibility that hm||u(-,t) 11^1,2(0) = oo. In this part of the argument, the superlinear growth of the nonlinearity (as u —> oo) plays a crucial role. On the other hand, if limsup ||u(-,i)||vKi.2(D) < oo, then the o;-limit set of u contains for every number B large enough a stationary solution w with ||w||W^2(D) = B. (The assumption that the growth of the nonlinearity (as u —* oo) is subcritical is needed here.) This leads to a contradiction since there is a constant K (depending on UQ) such that ||u||iyi,2(£)) < K for every stationary solution v in the oi-limit set of u.
98 Both parts of the argument rely on the existence of a suitable Lyapunov functional. In general, this method does not yield any a priori estimate of ||u(-, i)||iv1.2(D) in terms of some norms of UQ. Other energy and interpolation arguments were used in [2]. The method from [2] was further developped in [31, 33] by bootstrapping and employing maximal regularity results.
10.2
Scaling arguments
The variational structure together with scaling invariance are the main tools in [19]. If we rescale a solution u of the equation ut = A H + up, as v(y, s) = \r-lu{x
+ Xy,t + As),
then v satisfies the same equation. This fact, combined with the existence of a Lyapunov functional, yield a contradiction argument to prove the a priori bound in [19]. A different contradiction argument, relying on scaling, energy and Hardy's inequality, gives the universal bound in [32]. In order to handle problem (3), this approach was extended in [27]. For problems without variational structure, a scaling argument was used in [35] in order to prove a priori estimates for global solutions.
10.3
Zero number techniques
The method of our proof of boundedness of global classical solutions in [12] is based on the intersection-comparison with a particular self-similar solution. It was inspired by the proof of Theorem 15.1 in [17] where a similar result was shown for a different problem. However, unlike the case in [17], the nonincrease of the number of zeros of the difference of two solutions, by itself, is not sufficient. We also employ the more subtle "zero number diminishing property" which says that the number of zeros drops when a multiple zero occurs. The nonincrease of the zero number also plays a crucial role in [41].
10.4
Smoothing effect in weighted Lebesgue spaces
To prove (2.2) in [16], we start from the classical argument of Kaplan which yields that for any r > 0 there is T\ £ (0, r/2) such that Ju"(T1,x)
(4)
where ipi > 0 is the first eigenfunction of —A in HQ(SV), normalized by / n ipi — 1. Since Ci<5 < ipi < C2<5, <5(a;) = dist(x,dD), the bound (4) suggests that it is rather natural to
99 consider the problem in L% = Lq(D, S(x) dx) spaces. With this motivation, we developped in [16] both linear and local nonlinear theories in these spaces. Our main result in the linear theory is an optimal Lj - Lrs estimate for the heat semigroup. In view of this estimate for the heat semigroup, the local existence and uniqueness theory in L\ is similar to the usual L9-theory in dimension JV+1. In particular, we proved local existence and uniqueness for solutions in L\ for q > qc = |(JV + l)(p — 1) and local nonexistence for q < qc. We also established a nonlinear version of the L\ — Lre estimate whose particular case q = p, r = oo, plays a crucial role in deriving the universal bound (2) from (4). A similar idea is used in [27] but weighted Lebesgue spaces are replaced by convolution Lebesgue spaces there.
11 11.1
Applications Blow-up
The first application concerns blow-up in finite time. It requires some knowledge of the corresponding stationary problem. For example, if one can show that there are no steady states then all solutions of the parabolic problem blow up in finite time (cf. [1, 5]). A slightly more sophisticated application of this type says that a solution blows up in finite time if it starts above a maximal steady state provided this steady state is unstable from above, cf. [3, 4, 6, 8], for instance.
11.2
Existence of stationary solutions
The second application goes in the opposite direction. Some knowledge of the parabolic problem is used to draw conclusions on the corresponding stationary problem. We describe roughly a simple illustration of this idea. Assume a stable steady state is known to exist. If Mo belongs to the boundary of its domain of attraction and the solution u starting from u0 is global then there must exist another (unstable) steady state which is in the w-limit set of u. For dynamical proofs of existence of solutions of elliptic problems we refer to [13, 14, 29, 30, 34].
11.3
Existence of periodic solutions
The a priori bound discussed in Subsection 2.6 was used in [34] to prove the existence of solutions which are periodic in t.
11.4
.^-connections
The result from [12] mentioned in Subsection 2.2 plays a fundamental role in the study of connections between equilibria by solutions which blow up but continue to exist in a weak sense (see [10, 11]).
100
References J. Bebernes, A.A. Lacey, Global existence and finite time blow-up for a class of nonlocal parabolic problems,Adv. Differ. Equations 2 (1997), 927-954. T. Cazenave, P.L. Lions, Solutions globales d'equations de la chaleur semilineaires, Comm. Partial Differential Equations 9 (1984), 955-978. M. Chipot, M. Fila, P. Quittner, Stationary solutions, blow up and convergence to stationary solutions for semilinear parabolic equations with nonlinear boundary conditions, Acta Math. Univ. Comenianae 60 (1991), 35-103. M. Fila, Boundedness of global solutions for the heat equation with nonlinear boundary conditions, Comm. Math. Univ. Carol. 30 (1989), 479-484. M. Fila, Boundedness of global solutions of nonlinear diffusion equations, J. Differ. Equations 98 (1992), 226-240. M. Fila, Remarks on blow up for a nonlinear parabolic equation with a gradient term, Proc. A.M.S. I l l (1991), 795-801. M. Fila, Boundedness of global solutions of nonlocal parabolic equations, Nonlin. Anal. TMA 30 (1997), 877-885. M. Fila, B. Kawohl, Large time behavior of solutions to a quasilinear parabolic equation with a nonlinear boundary condition, Adv. Math. Sci. Appl. 11 (2001), 113-126. M. Fila, H. A. Levine, On the boundedness of global solutions of abstract semi-linear parabolic equations, J. Math. Anal. Appl. 216 (1997), 654-666. M. Fila, H. Matano, Connecting equilibria by blow-up solutions, Discr. Cont. Dyn. Syst. 6 (2000), 155-164. M. Fila, H. Matano, P. Polacik, Existence of L1-connections between equilibria of a semilinear parabolic equation, preprint. M. Fila, P. Polacik, Global solutions of a semilinear parabolic equation, Adv. Diff. Equations 4 (1999), 163-196. M. Fila, P. Quittner, Global solutions of the Laplace equation with a nonlinear dynamical boundary condition, Math. Methods Appl. Sci. 20 (1997), 1325-1333. M. Fila, P. Quittner, Large time behavior of solutions of a semilinear parabolic equation with a nonlinear dynamical boundary condition, in: Topics in Nonlinear Analysis, The Herbert Amann Anniversary Volume, Birkhauser, Basel, 1998, 251272.
101 M. Fila, P h . Souplet, Existence of global solutions with slow decay and unbounded free boundary for a superlinear Stefan problem, Interfaces and Free Boundaries 3 (2001), 337-344. M. Fila, P h . Souplet, F. Weissler, Linear and nonlinear heat equations in L\ spaces and universal bounds for global solutions, Math. Ann. 3 2 0 (2001), 87-113. V. Galaktionov, J.L. Vazquez, Continuation of blow-up solutions of nonlinear heat equations in several space dimensions, Comm. P u r e Applied Math. 50 (1997), 1-67. H. Ghidouche, P h . Souplet, D. Tarzia, Decay of global solutions, stability and blowup for a reaction-diffusion problem with free boundary, Proc. A.M.S. 1 2 9 (2001), 781-792. Y. Giga, A bound for global solutions of semilinear heat equations,Comm. Math. Phys. 103 (1986), 415-421. S. Kaplan, On the growth of solutions of quasilinear parabolic equations, Comm. P u r e Appl. Math. 16 (1963), 327-330. O. Kavian, Remarks on the large time behaviour of a nonlinear diffusion equation, Ann. Inst. H. Poincare, Anal. Non Lineaire 4 (1987), 423-452. A. A. Lacey, D. Tzanetis, Global existence and convergence to a singular steady state for a semilinear heat equation,Proc. Roy. Soc. Edinburgh Sect. A 1 0 5 (1987), 289305. A.A. Lacey, D.E. Tzanetis, Global, unbounded solutions to a parabolic equation,J. Diff. Equations 101 (1993), 80-102. G.M. Lieberman, Study of global solutions of parabolic equations via a priori estimates. P a r t I: equations with principal elliptic part equal to the Laplacian,Math. Methods Appl. Sci. 16 (1993), 457-474. G.M. Lieberman, Study of global solutions of parabolic equations via a priori estimates II: porous medium equations,Comm. Appl. Nonlin. Anal. 1 (1994), 93-115. G.M. Lieberman, Study of global solutions of parabolic equations via a priori estimates III. Equations of p-Laplacian type, in Singularities and Differential Equations, Banach Center Publications, Vol. 33, pp. 199-221, P W N , Warszawa (1996). J. Matos and P h . Souplet, Universal blow-up estimates and decay rates for a semilinear heat equation,preprint. W.M. Ni, P.E. Sacks and J. Tavantzis, On the asymptotic behavior of solutions of certain quasilinear parabolic equations,J. Diff. Equations 5 4 (1984), 97-120. P. Quittner, Boundedness of trajectories of parabolic equations and stationary solutions via dynamical methods,Diff. Int. Equations 7 (1994), 1547-1556.
102 [30] P. Quittner, Signed solutions for a semilinear elliptic problem, Diff. Int. Equations 11 (1998), 551-559. [31] P. Quittner, A priori bounds for global solutions of a semilinear parabolic problem,Acta Math. Univ. Comen. 68 (1999), 195-203. [32] P. Quittner, Universal bound for global positive solutions of a superlinear parabolic problem,Math. Ann. 320 (2001), 299-305. [33] P. Quittner, Continuity of the blow-up time and a priori bounds for solutions in superlinear parabolic problems,Houston J. Math., to appear. [34] P. Quittner, Multiple equilibria, periodic solutions and a priori bounds for solutions in superlinear parabolic problems,NoDEA, to appear. [35] P. Quittner, Ph. Souplet, A priori estimates of global solutions of superlinear parabolic problems without variational structure, Discr. Cont. Dyn. Systems, to appear. [36] P. Rouchon, Boundedness of global solutions of nonlinear diffusion equation with localized reaction term, preprint. [37] Ph. Souplet, Sur l'asymptotique des solutions globales pour une equation de la chaleur semi-lineaire dans des domaines non bornes, C. R. Acad. Sci. Paris, Serie I 323 (1996), 877-882. [38] Ph. Souplet, Geometry of unbounded domains, Poincare inequalities and stability in semilinear parabolic equations, Comm. PDE 24 (1999), 951-973. [39] Ph. Souplet, A priori and universal estimates for global solutions of superlinear degenerate parabolic equations, Annali Mat. Pura Appl., to appear. [40] Ph. Souplet, F.B. Weissler, Poincare's inequality and global solutions of a nonlinear parabolic equation,Ann. Inst. H. Poincare, Anal. Non Lineaire 16 (1999), 335-371. [41] R. Suzuki, Boundedness of global solutions of one dimensional quasilinear parabolic equations,,!. Math. Soc. Japan 50 (1998), 119-138. [42] D. Tzanetis,Asymptotic behaviour and blow-up of some unbounded solutions for a semilinear heat equation,Proc. Edinb. Math. Soc. 39 (1996), 81-96.
Transmission problems for degenerate parabolic equations Takeshi Fukao Department of Mathematics Graduate School of Science and Technology Chiba University, 1-33 Yayoi-cho, Inage-ku, Chiba, 263-8522 Japan Email : [email protected] Abstract In this paper, we consider a transmission problem for nonlinear two degenerate parabolic PDEs, in such a setting that a fixed bounded domain in R 3 is separated by a moving interface into two subdomains. The degenerate parabolic PDEs are formulated on each of these subdomains and moreover the so-called transmission condition is imposed on the moving interface. The purpose of this paper is to prove the existence and the uniqueness of a solution.
1
Introduction
We consider the following initial-boundary value problem for degenerate parabolic equations in a cylindrical domain Q := (0, T ) x Q, where 0 < T < oo and fl is a bounded domain in R 3 with smooth boundary T. Moreover, we suppose t h a t 0 is separated by a time dependent smooth interface Tm(t) into two subdomains fii(£) and Ch{t), t h a t is : Q, : = n-i.(t) U n2(t) U Tm(t) for each t € ( 0 , T ) : «,-Aft(a)+v-Vu = / ft(«)=ft(«),
^
r
in Q; := U t e ( o r ) { i } x f2i(t),
+ ^ i ^ = °
i = l,2,
on E m := U ( 6 ( 0 , T ) W x r m ( t ) , (1)
^
^
+ noA(«) = p
o n S , := U e ( o , r ) W * W W
\ I W * ) } , » = 1,2,
u(0) = UQ on Q. Here ut := du/dt, A u := E L I ^ V ' ^ ? ) and V u := {du/dxudu/dx2,du/dx3). We denote by u+ := u+(t,x) t h e 3-dimensional unit vector normal to T m (i) at x e T m ( t ) , pointing to fi2(i), by v~ : = i/ _ (£,:r) t h e vector — v+ and by v := ^(a;) t h e 3-dimensional unit vector normal to V at x € T. no is a positive constant, / and p are given functions prescribed on Q and E, respectively. u0 is the initial d a t u m on Q. fa and fa are functions defined on R and which satisfy t h e following conditions:
103
104 • Pi • K —• K is a Lipschitz continuous and non-decreasing function for i = 1,2. • A(0) = 0 and \k(r)\ > c f t |rf - c^, where /3;(r) :=
JQr
Vr e R and i = 1,2,
f3i(s)ds, i = 1,2 and c ftl c^ are positive constants.
In the above problem, (1)2 describes a relation between Pi(u) and feW) on the moving interface T m (i). It is called a transmission condition. In our geometric setting, the following two cases are allowed, the first one is that <9fi;(*) = F u r m ( i ) and dQ,j{t) = Tm(t) for (i,j) = (I, 2) or (2,1), and the second one is that a part of <9f2;(i) is included in T for i = 1,2. For a typical example of the latter case, see [7]. In this paper, we assume that the following conditions are satisfied: ( A l ) For each t 6 [0,T], there exists a C2-diffeomorphism © := ©(£, •) on fi, such that 6 £ C2(Q) and Q(t,Qi(t))=Qi{0),
i = l,2,
O(i,r m (t)) = r m (0),
Vt€[0,T].
This assumption (Al) says that the motion of Tm(t) is prescribed and smooth in time and the C2-regularity of 6 is actually used in the existence and uniqueness proofs of solutions (Lemma 3.3, Lemma 4.2). (A2) We assume that the convection vector field v belongs to C 1 (Q) 3 and that the following properties are satisfied: div (v) = 0 in Q, +
v • v = vxm
(2)
on E m ,
(3)
v i / = 0 on£:=(0,T)xr,
(4)
where u Sm := v-Em(t,x) is the normal speed of Tm(t) at (t,x) € S m . We define the time-dependent function (3 : Q x R —* R P(t,x,r)
:= Pi(r)xQ1(t) + P2{r)xn2(t),
where x
dP(t x,u)
+
+ v - Vu
= f
onE;
=
«(0) =
in Q,
UQ
(5)
in O.
For the solvability of (5), it seems that the time-dependent subdifferential operator theory is useful (cf. [1, 4, 7, 8, 9, 10, 16]). In sections 3 and 4, we prove the existence and uniqueness of a solution.
105
2
Definition of t h e weak solution and t h e main result
We begin with the definition of a weak solution of (5) and then formulate our main result. To this end we introduce some function spaces. We denote by V the Sobolev space iJ 1 (Q) with norm j • |y given by 1/2
\u\v := {|Vu|£ 2(n) + n 0 | u | | 2 ( r ) |
Vu € V,
which is an equivalent norm to the standard one of H1 (f2). The dual space of V is denoted by V* and the duality pairing between V* and V is denoted by {-,-)v.y. The duality mapping F : V —> V* is defined by (Fu 1 ,u 2 )v«,v := / Vui • \7u2dx + n0 / uiu2aT Jn Jv
Vuj,u2 € V.
Moreover, we define an inner product (•, -)y in V* by (u\,u*2)v •= « , F~lv*2)v.y
\ful,u*2 6 V*.
Then V* is a Hilbert space with the following dense and compact imbeddings: V ^ L2(Q) ^
V*.
Now we give the definition of a weak solution of (5). Definition 2.1 u : [0, T] -* L2{0.) is called a weak solution of (5) ifue Wh2(0, T; V*) n L°°(0,T;L 2 (f2)), /3(u) := f3(t,x,u) G L2(0,T;V), and u satisfies (5) in the variational sense : - JQ ur]tdxdt + JQ Vp(u) • Vvdxdt 4 n0 J E f}{u)rjdTdt — JQ u(v • Vrfidxdt = JQ fndxdt + / s prjdTdt + Jn uori(0)dx V77 e W,
' ^
where W := {77 € Hl(Q) | rj(T) = 0 on fi}. Our main result is stated as follows. Theorem 2.2 Assume that (Al) and (A2) hold. Let f € L2{Q), p e Cl{Y,), u0 e L2{Q) and f3(u0) € if 1 (f7). Then there exists one and only one weak solution u of (5). The existence proof for (5) will be given in section 3, using the abstract theory of time-dependent subdifferential operators. The uniqueness proof will be given in section 4.
106
3
Existence proof
Let us prove the existence of a weak solution of (5), applying the existence result in [16] (or [8]) for a nonlinear evolution equations including time-dependent subdifferential operators and a perturbation of the form: f ut(t) + d(t>\u{i)) + g(t,u(t)) B f*(t) \ w(0) = u0 in V,
in V* for a.e. t £ [0,T],
(?)
where the functions (pf, g(t, •) and f*(t) for each t £ [0, T] are defined as follows: • cf/ : L2(Q) —> (—oo, oo] is defined by ftiz) '•=
/ / \
JJ fii(t)
01(z(x))dx+ f
iizeL2(Q),
j32(z(x))dx
JJ n f 2(t)
if z e V*\L 2 (fi).
Clearly 0* is a proper, lower semi-continuous and convex function on V. • g(t, •) : L2(tt) -> V* is defined by (ff(t,z),ri)vy •=-
I z(x)(v{t) • Vn)(x)dx Jn
L2(Q).
V77 &V,VZ£
• /* £ L2(0, T; V*) is defined by (f(t),ri}v*,v
:= { f(t,x)V(x)dx+ Jn
f p(t,x)n(x)oT Jr
Vr,eV.
It is easy to see that (5) is equivalent to (7). Therefore, it is enough to solve (7) with the above-defined functions $, g(t, •) and /*(*). The characterization of dcj/ is given by the following lemma. L e m m a 3.1 Let z £ D(90') and j3(z) := /3(t,x,z(x)). Then z* £ d<jt{z) in V* if and only if (F-1z*,w)^(fi) = {P{z),w)^(p) V»6l2(fi), (8) where (V)L 2 (S)) denotes the inner product in L2(Q). statement that j3(z) £ V and {z*,T))v.y
Moreover, this is equivalent to the
= / V/3{z) • VVdx + n0 J P(z)ndT Jn Jr
Vr? £ V.
(9)
The next lemma is concerned with the demi-closedness for g(t, •). Lemma 3.2 The operator g(t, •) : D(g(t, •)) = L2{Q.) -*V*,te 2
[0,T], satisfies :
\g(t,z)\ v. < C0<j>\z) + Ci Vt £ [0,T], Vz £ L 2 (Q), (10) where C$ and C\ are positive constants, and if {4>tn(zn)} is bounded and zn —> z in V* and tn —> t, when n —> oo for {zn} c L2(Q) and {tn} C [0, T], then g(tn,zn) —+ g(t,z) weakly in V*. Furthermore, the set Lr := (J0
107 The proofs of these lemmas are quite standard, so we omit them. In the following lemma, we check the time-dependence of ft and the conditions required in order to apply the abstract result in [16] (or [8]) to (7). Lemma 3.3 There exists a positive constant C2 with the following property : for every s, t € [0,T], s
+ \ft{z)\ll2),
t
\4> {zi)-ft{z)\
(11)
+ \ft{z)\).
(12)
Proof. For every z G D(ft) := L2(Q), we take zx(x):=z(Q-1(s,e{t,x)))
Viefi.
Then, by the geometric assumption (Al), zx e L2(fl) = D{ft) and the mapping &t,s '•= Q(s, -)~ L ° ©(*,-) i s a C2-diffeomorphism on fi for every s, t G [0,T] such that ©M (£k{t)) = &i{s), i = 1,2. Moreover there exist two positive constants C 3 and C4 such that | d e t ( j e t , ( z ) ) - 1 |
4
Uniqueness proof
In this section, we prove the uniqueness of a weak solution of (5) obtained in the previous section. Let u\ and u2 be two weak solutions. Then taking the difference of their weak variational forms (6) we have - JQ(ui - u2)rjtdxdt + JQ(V/9(wi) - V/?(u2)) • Vndxdt - jQ(ui - u 2 )(v • Vn)dxdt +nQ / E ( / ? K ) - P{u2))r,dYdt = 0 \fr,eW. Moreover, we define the function b : Q —> R as b{t,x):=i
( /?M*,s))-/?Mts)) Ul(t,x)-u2(t,x) [ 0
jf
^ (t } A ' l r ' if «i(t,a;) = « 2 (i,:r), (i
U
which is non-negative and bounded on Q. With this function b, Green's formula implies - f (u1-u2){nt Q
+ bAV+v-Vn}dxdt+
f {PiuJ-Piui)) E
l^+n0ri\ "
dTdt = 0
\fneW. (13)
108
Now we take smooth and positive approximations bE of b such that b
a.e. on Q, e
. .
where M0 is a positive constant. In order to prove the uniqueness, we consider the following backward problem for the auxiliary linear parabolic equation for any given ieD(Q): t]Et, + bEAr)£ + v • Vr?e = I in Q, ^+n0r,e = 0 on£, av 7jE(T,-) = 0 onO.
(15)
Using Ladyzenskaja, Solonnikov and Ural'ceva [11, Chapter 4, Theorem 5.3], problem (15) has a unique solution r)e € H2+a-1+a!2(Q) for a certain 0 < a < 1. In the next two lemmas, we give the L°°-estimate of {r)B(t)} in V and the boundedness of {bl/2AVe} in L2{Q). Lemma 4.1 There exists a positive constant Mi such that sup \r,e{s)\l+ 0<s
( Js
[ be\Ane\2dxdt < M1 f Ja
j v •V{rj2)dTdt
+M^^.y)
Js JT
for alls e [0,T] and e G (0,1]. Proof. Multiplying the first equation in (15) by Ane and integrating over Q with respect to x, with the help of Green's formula we have -|l»felv + 2 1
%\\AVe\2dx < (6|v| c l ( 5 ) 3 + l)|ife|», + \l\*v + 2n0 J
v • V(^)dT,
V i e [0,T].
(16) In fact, in order to get (16) we use the following equations and inequalities - / (v • Vr)s)Ar)£dx = V(v • Vr/£) • Vj)edx - f (v • Vr]e)^-dY J si J si \ r < 3|v| c l ( g ) , | n | V ^ d z + / n v • V ^ | V T ? E | 2 J dx+^J^v V(V2)dT and by virtue of (2) and (4) jf v • V (\\^vA
dx = Jdiv
Q | V % | 2 v ) dx = J i | V ^ | 2 ( v • v)dT = 0.
Finally we integrate (16) over [s, T] in time and apply Gronwall's inequality to get the conclusion. •
109 Lemma 4.2 There exists a positive constant M2 such that:
/r
v(<) ' V(77e2(i))dr < M2\nE(t)\2L2{T)
Vt e [0,71, Ve € (0,1].
Proof. Our geometric condition (Al) ensures that without loss of generality we may assume by taking an approximate local coordinate transformation and a partition of unity on r that: • Q c { x = ( i i , i 2 , x3); x3 < 0} and T contains a relatively open set T0 in K2 = {x | x3 = 0}. • supp (r/e) is compact infiU r 0 , Ve <E (0,1] and
In the above situation, we observe for any t £ [0, T] that /
v-V(7? 2 )dr
=
vV(V2)dXldx2
/ r
{vi,x! + v2,X2)ri2dx1dx2 + / 2v3ne^-dx1dx2 Jtf J R2 = - / (vi,xl+v2iX2+2n0v3)r]2dxidx2, J R2 where v := (^i, •02,^3)- Therefore, for any t € [0,T], =
-
/ vv -' Vv(?i ( r ;e2)ai )dr< ^ ( 2V^oi-^;|v|c.i n 0 + 2)|v| c l((Q)3|'/ Q ) 3| %E1|2L 2 (r) . Thus we have the conclusion of the lemma.
•
Now, on account of Lemmas 4.1 and 4.2 we have the uniform estimate 2 sup h(t)\ v+
f b£\Ane\2dxdt < M3, (17) Jo 0
Thus Uj = u2 a.e. on
•
Remark 4.3 TTie idea of our uniqueness proof is due to Ladyzenskaja, Solonnikov and Ural'ceva [11, Chapter 3, Section 3], and this was also extensively used in Niezgodka and Pawlow [12], Rodrigues and Yi [15] and Rodrigues [14] for the uniqueness proof of generalized Stefan problems and continuous casting problems. A similar technique was recently applied for the uniqueness proof of Stefan problems arising in the Czochralski process of crystal growth (cf. Fukao, Kenmochi and Pawlow [6, 7]).
110 Remark 4.4 In particular, if Pi and P2 are strongly monotone in M, namely (A(fi) - Pi{ri)){rx - r2) > m0\ri - r 2 | 2
Vr-j, r 2 e R and i = 1,2,
(18)
Wiere m 0 is a positive constant, then the uniqueness of a solution of (5) is a direct application of [16, Theorem 3.1]. The main point of the proof is to use the estimate / (uj. - u 2 )v • \7{F 1(ul - u2))dx < mi|v| c i ( Q )3 |ui -u 2 |i!i(n)|ui
~u2\v.,
2
(19)
yUl,u2eL (Q), where mi is a positive constant so that I V F - ^ l ^ ^ a < mi\F-1z\v(=
mi\z\v.)
Vz e V*.
In fact, for two solutions «! and u2, we observe that {u[(t) - u'2(t),r,)v.y J Dn
+ (F(f3(Ul(t)) -
(ui{t) - u2{t))(v{t) • Vri)dx = 0,
p(u2(t))),V)v.y VT? e V, a.e. t € [0,T].
Now, take i ? _ 1 (ui(i) — u2(t)) as n and use (18) and (19). Then ~ K ( t ) - «2(t)| 2 ^ + (mo ~ S)\ui(t) - u2(t)\2LHn) < - ^ M C H Q J S M * ) " fora.e.te [0,T],
u2(t)\2v,,
where 8 is an arbitrary positive number. Choosing m0/2 as S and applying the Gronwall's inequality, we infer that Uj = u2. •
5
Application t o Czochralski crystal growth process
This work is motivated by modeling the Czochralski process of crystal growth. The Czochralski pulling method is widely used for the production of a column of single silicon crystal from the melt. The idea for pulling method due to Czochralski is quite simple. A crucible, equipped with a heater, contains the melt substance and a pul-rod with seed crystal, which moves vertically and rotates flexibly, is positioned above the crucible. The rod is dipped into the melt, and then lifted slowly with an appropriate speed so that a meniscus surface is formed below the crystal seed and the melt attached to the crystal solidifies continuously. By controlling some thermal situations in the process, one obtains the growth of a single crystal column with a desired radius as well as a desired growth pattern of the solid-liquid interface and temperature pattern in the crystal in order to improve the crystal quality (see [3, 5, 6, 7, 13]). In such a model of crystal growth, if we assume that the motion of the material (solid-liquid) domain is prescribed and a 3-dimensional convective vector field, which is caused by the deformation of the material domain, is prescribed too, at least satisfying condition (A2), then our approach seems applicable to the Czochralski process. In this
111 case, we consider two equations in the crucible. One of them is a degenerate parabolic equation which describe the solid-liquid phase transition in the material region. The other is an usual heat equation in the gas region. These two equations are combined by the transmission condition on the common boundary. In this case the transmission condition implies that the temperature is possibly discontinuous on the solid-gas and liquid-gas interfaces. This is one of the important points which should be furthermore discussed from the mechanical point of view.
References [1] H. Attouch and A. Damlamian, Application des methodes de convexite et monotonie a l'etude de certaines equations quasi lineaires, Proc. Royal Soc. Edinburgh, 97A (1977), 107-129. [2] H. Brezis, Operateurs Maximaux Monotones et Semi-groupes de contractions dans les espaces de Hilbert. Math. Studies 5, North-Holland, Amsterdam, 1973. [3] A. B. Crowley, Mathematical modelling of heat flow in Czochralski crystal pulling. IMA J. Appl. Math., 30 (1983), 173-189. [4] A. Damlamian, Some results on the multi-phase Stefan problem, Comm. Part. Diff. Eq., 2 (1977), 1017-1044. [5] E. DiBenedetto and M. O'Leary, Three-dimensional conduction-convection problems with change of phase. Arch. Rat. Mech. Anal., 123 (1993), 99-116. [6] T. Fukao, N. Kenmochi and I. Pawlow, Two phase Stefan problems in non-cylindrical domains. In preparation. [7] T. Fukao, N. Kenmochi and I. Pawlow, Transmission problems arising in Czochralski process of crystal growth. In Mathematical Aspects of Modeling Structure Formation Phenomena, GAKUTO Intern. Ser. Math. Sci. Appl., Vol.17, pp.228-243 Gakkotosho, Tokyo, 2001. [8] A. Ito, N. Yamazaki and N. Kenmochi, Attractors of nonlinear evolution systems generated by time-dependent subdifferentials in Hilbert spaces. In Dynamical Systems and Differential Equations, Vol.1, ed. W. Chen and S. Hu, pp.327-349 Southwest Missouri State Univ., Springfield, 1998. [9] N. Kenmochi, Solvability of nonlinear evolution equations with time-dependent constraints and applications. Bull. Fac. Education Chiba Univ., 30 (1981), 1-87. [10] N. Kenmochi and I. Pawlow, A class of nonlinear elliptic-parabolic equations with time-dependent constraints. Nonlinear Analysis, 10 (1986), 1181-1202. [11] O. A. Ladyzenskaja, V. A. Solonnikov and N. N. Ural'ceva, Linear and quasilinear equations of parabolic type. Transl. Math. Monographs, Amer. Math. Soc, 23, 1968.
[12] M. Niezgodka and I. Pawlow, A generalized Stefan problem in several spaces variables. Appl. Math. Optim., 9 (1983), 193-224. [13] I. Pawlow, Three-phase boundary Czochralski model. In Mathematical Aspects of Modeling Structure Formation Phenomena, GAKUTO Intern. Ser. Math. Sci. Appl., 17, pp.203-227 Gakkotosho, Tokyo, 2001. [14] J. F. Rodrigues, Variational methods in the Stefan problem. In Phase Transitions and Hysteresis. Lecture Notes Math. Vol.1584, pp.147-212 Springer-Verlag, 1994. [15] J. F. Rodrigues and Yi-Fahuai, On a two-phase continuous casting Stefan problem with nonlinear flux. European J. Appl. Math., 1 (1990), 259-278. [16] K. Shirakawa, A. Ito, N. Yamazaki and N. Kenmochi, Asymptotic stability for evolution equations governed by subdifferentials. In Recent Developments in Domain Decomposition Methods and Flow Problems, GAKUTO Inter. Ser. Math. Sci. Appl., 11, pp.287-310 Gakkotosho, Tokyo, 1998.
Existence of solutions of a segregation model arising in population dynamics Gonzalo Galiano Maria L. Garzon Departamento de Matematicas, Universidad de Oviedo, 33007 Oviedo, Spain, Email: [email protected] .uniovi.es, maria@or ion. ciencias. uniovi.es Ansgar Jungel Fachbereich Mathematik und Statistik, Universitat Konstanz, 78457 Konstanz, Germany, Email: [email protected]
Abstract A strongly coupled cross-diffusion model for two competing species is analyzed. The unknowns are the population densities of the species. An existence proof for the evolution problem in one space dimension is sketched. The proof is based on a symmetrization of the problem via an exponential transformation of variables and the use of a new entropy functional. The solutions can be shown to be nonnegative, by means of the exponential transformation of variables. Moreover, necessary conditions for segregation phenomena for the steady-state problem are given and numerically illustrated.
1
Introduction
For the time evolution of two competing species with homogeneous population density, usually the Lotka-Volterra differential equations are used as an appropriate mathematical model. In the case of non-homogeneous densities, diffusion effects have to be taken into account leading to reaction-diffusion equations. Shigesada et al. proposed in their pioneering work [27] to introduce further so-called cross-diffusion terms modeling interspecific influence of the species. Denoting by rii the population density of the i-th species (i = 1,2) and by J; the corresponding population flows, the time-dependent equations
113
114 can be written as d t ni + div Jx 9jn2 + div J 2 Ji h
= = = =
gi{nun2), 52 ("1,^2), - V ((ci + anni + ai2n2)ni) + diriiVU, - V ((c2 + oci\nx + a22n2)n2) + d2n2VU,
(1) (2) (3) (4)
in the bounded domain Q, C Rd (d > 1) with time t > 0. Here, £/ = U(x) is the (given) environmental potential, modeling areas where the environmental conditions are more or less favorable [24, 27]. The diffusion coefficients c* and a*, are non-negative, and d j £ l (i,j = 1, 2). The source terms are in Lotka-Volterra form: 9i{ni,n2)
= (fli-7n«i-7i2«2)ni,
02 ("l, n 2 ) =
{R2 - 721711 - 722"2)«2,
(5) (6)
where i?, > 0 is the intrinsic growth rate of the i-th species (i = 1,2), 711 > 0 and 722 > 0 are the coefficients of intra-specific competition, and 712 > 0 and 721 > 0 are those of interspecific competition. The above system of equations is completed with mixed Dirichlet-Neumann boundary conditions and initial conditions: n-i = n-D,i on To x (0,00), Jj • v = 0 ni(-,0) = n0,j
o n f ^ x (0,00), in fl, i = 1,2,
(7) (8)
where v denotes the exterior unit normal to d£l. This means that the population density is fixed at a part of the domain boundary (due to emigration and immigration processes), whereas no flux boundary conditions are prescribed at the remaining boundary parts. Eqs. (l)-(4) contain various types of reaction-diffusion models. Indeed, in the case OLij = 0 for i, j = 1,2, they reduce to the drift-diffusion equations, which has been studied in various fields of application, e.g. electro-chemistry, [1, 3], biophysics [7] or semiconductor theory [23]. When c\ = c2 = 0 and «i2 = a2i = 0, Eqs. (l)-(4) are of degenerate type. These types of problems arise, for instance, in porous media flow [17], oil-recovery [8], plasma physics [16] or semiconductor theory [15]. In chemotaxis, related models appear [9, 25]. For Q12 > 0 and a 2 i > 0, the problem becomes strongly coupled with full diffusion matrix A(ni,n2)
= , 1
Ci + 20:1171! + Q i 2 n 2
012^1
a2\n2 c2 + 2a22n2 + a 2 ini Nonlinear problems of this kind are quite difficult to deal with since the usual idea to apply maximum principle arguments to get a priori estimates cannot be used here. Furthermore, the diffusion matrix is not symmetric. Up to now, only partial results are available in the literature concerning the wellposedness of the problem. We summarize some of the available results for the timedependent equations (see [29] for a review) and refer to [20, 21] for the stationary problem. We also refer to [6, 13, 18] for related results on various cross-diffusion models. Global
115 existence of solutions and their qualitive behavior for a n = a 22 = a 2 i = 0 have been proved in, e.g., [2, 22, 26]. In this case, Eq. (2) is only weakly coupled. For sufficiently small cross-diffusion parameters a 12 > 0 and a21 > 0 (or equivalently, "small" initial data) and vanishing self-diffusion coefficients a n = a 22 = 0, Deuring proved the global existence of solutions [5]. For the case Ci = c2, a global existence result in one space dimension has been obtained by Kim [19], Furthermore, under the condition 8«n > a 12 ,
8a 22 > a 21 ,
(9)
Yagi [30] has shown the global existence of solutions in two space dimensions assuming a 12 = a2\. A global existence result for weak solutions in any space dimension under condition (9) can be found in [10]. Condition (9) can be easily understood by observing that in this case, the diffusion matrix is positive definite: £TA(nltn2)£
> min{ci,c 2 }|£| 2
for all £ e R2,
hence yielding an elliptic operator. If the condition (9) does not hold, there are choices of ci: atj, 7ij > 0 for which the matrix A(ni, n2) is not positive definite, and it is therefore unclear if the problem (l)-(8) can be solved for these data. More recently, Ichikawa and Yamada [14] have improved the results of Yagi, replacing condition (9) by 6 4 a n a 2 2 > aua21
or
6 4 a u a 2 2 = a12ot2i > 0.
(10)
They use the same techniques as Yagi combined with suitable energy estimates. From the view-point of mathematical biology, conditions like (9) and (10) mean that self-diffusion or diffusion is dominant over cross-diffusion. The aim of this paper is to show how the existence of solutions of problem (l)-(8) can be obtained, without assuming conditions like (9) or (10). In fact, these conditions are just technical restrictions, needed in [30, 14], since the existence of solutions of the steady-state problem can be proved without this condition (see [20]). More precisely, we are able to show that for any Cj, a; > 0 and in one space dimension d = 1, there exists a weak solution U\,u2 to (l)-(4), (7)-(8) such that U\ and u2 are nonnegative. We stress the fact that the non-negativity property is obtained without the use of the maximum principle. The idea of the proof is as follows: The system (l)-(4) is first symmetrized via an exponential transformation of variables. A priori estimates are then derived by using a new entropy functional yielding H1 bounds which are independent of the solutions. The non-negativity property is obtained from the embedding i3"1(f2) <-)• L°°(fi), which holds only in one space dimension, and the transformation of variables. We sketch the proof in Section 2. For the detailed proof, we refer to [11]. Before we state the results and sketch the method of proof, we perform (for a smoother presentation) the following change of unknowns: Ui = a2ini,
u2 = ot\2n2,
and
q = —VU.
We assume that ai2 > 0 and a 2 i > 0 which is no restriction since if a.\2 = 0 or a2\ = 0, at least one of the equations (1), (2) is weakly coupled, and the results of [26] apply. Eqs.
116 (l)-(8) can be reformulated as dtUi - div(cjVu, + 2a;i4;Vuj + V(uiu 2 ) + diU{q) = fi(ui, u2), onrDx(0,T), Ui = uDti (ciVui + 2aiuiVui + V{ulu2)+diuiq)-v = Q onTN x (0,T), u(-,0) = uOi infi, i = 1,2,
(11) (12) (13) (14)
where T > 0, uD)i = a2\nD^, uD}2 = ai2nDt2, «? = a2i«o,i, "2 = ai2«o,2 and ax = 0:11/021, o2 = a 2 2 /ai 2 . The source terms are given by fi{u\,u2) = {Ri — Ai"i — Pi2U2)ui, with /?K = 7ii/«2i, /?2i = 72i/ai2, i = l,2.
2
Existence of solutions of problem (11)-(14)
In this section we assume: (Al) n C R is a bounded interval, dQ = TD U TN, and TD ^ III. (A2) u° € L°°(n) satisfies M? > 7 > 0 in 0, uD,i = const. > 0 on TD,i = 1,2. (A3) ahCi > 0, di € R (i = 1,2) and 9 e L2(f2 x (0,T)). (A4) /; : [0, oo)2 —> R (i = 1,2) is continuous and it holds for all s, a > 0, p,q > 0: fi(s,a)
P
+ Ms,
and h(s,o~) (---)+ \p s)
h(s,o)
(- - ±) < C3, \q a)
for some d , C2{p,q), C3{p,q) > 0. We observe that the Lotka-Volterra source terms satisfy condition (A4) if /3a > 0, i = 1, 2 and £12 = hi > 0. The main result is the following theorem. Theorem 1 Let T > 0. Under assumptions (Al)-(A^) there exists a weak solution (ui,« 2 ) 0/ (ll)-(U) satisfying uuu2 € L 2 (0,T; ffx(n)) n W ^ O . T ; tf_1(^)) and ui{x,t),
u2{x,t) > 0
for {x,t) 6 O x (0,T).
The proof consists of several steps. Step 1. We work with unknowns which symmetrize the elliptic operator. Introduce w = (wi,w 2 ) by defining ux = ew\u2 = eW2 and set b{w) = (&i(u>),&2H) = {eWl,eW2). With the diffusion coefficients au(w) = c{eWi + 2aie2w- +
m+w e
*,
i = 1,2,
al2(w) = a2i(w) =
Wl+W2 e
,
Eqs. (11)-(14) are formally equivalent to dtbi(w)-d\v
lj2aij(w)VwJ
+ d
Mw)q)
=Fi(w),
infix(0,T),
(15)
I ^2 Oij(w)Vw.,- + dibi(w)q J • v = 0
on TN x (0,T),
(16)
w = u>D w(0) = w°
onrDx(0,T), in fi,
(17) (18)
where F^w) = / i (e w l ,e W 2 ), wD>i = log(« Dii ), and w° = log(uj), i = 1,2. Step 2. In order to solve the above problem, we introduce a semi-discrete problem. Let TV e N and let r = T/N be the time step. Given wk~x, approximating w in the interval ((k — 2)T, (k — l)r], we are seeking solutions wk of the elliptic problem
k(wk) -uwk-i) _ div / £ ai.{wk)Vwk+dA{w^A ^
= Fi(w*} in n>
a^iu^Vur) + dibi{wk)qk \-v
(19)
= Q onTN,
(20)
wk = w D on TD,
(21)
for A; = 1 , . . . , N. Here we defined qk = \ / (ife 1 1)T (•, £)<&• Step 3. The idea of obtaining appropriate a priori bounds is to use an entropy functional. We introduce the discrete entropy (for k = 0 , . . . , N) Vk=Vi+
ar)k,
where a — 2min{ci,c 2 }, f]k is the discrete "physical" entropy 2
r
k
v = X) / ( 6 iK)K f c - ti»D,0 - h{wk) + bi{wD))dx > o, and Tjk is another discrete entropy-type functional:
4 = E [(^-^
- (wk - wD,i))dx > 0.
In the original variables, r]k can be written as f(ui(logUi — logup^) — ut + up^dx, which justifies the notion of "entropy". We can prove the following entropy-type estimate, which holds in any space dimension.
118 Lemma 1 Let wk € i7 x (Q;R 2 ) be a weak solution of (19)-(21). constant C > 0 such that for any k = 1 , . . . , N and any T > 0, ^ + T J2 f (iH^I
+
a|VeTO-/2|2 + ai\Vew" A
Then there exists a
dx < n^1 + CT.
(22)
The key of the proof of this lemma is to use (wf — v>D,i) + a(6j(—IBJ) — 6j(—«);)) € HQ(£1 U FJV) = {t)£ i71(Q) : D = 0 on To} as a test function in the weak formulation of (19)-(21) (see [11] for the details). Step 4. We use the Lax-Milgram's lemma together with the Leray-Schauder's fixed-point theorem to prove the existence of weak solutions of the semi-discrete problem (see [11]). Lemma 2 Let K;* -1 G L°°(fl;R 2 ), k > 1. H^Q-R2) of (19)-(21).
Then there exists a weak solution wk €
We notice that, since the solution satisfies wk e i7 1 (fi;R 2 ) «-> L°°(Q;R2) in one space dimension, the unknowns uk = exp(iwf) are well defined and elements of H1(Q). Hence (uk, uk) = (eWl, e"*2), k = 1 , . . . , N, is a solution of the discrete problem corresponding to (11)-(14). Step 5. We define the piecewise constant functions u/T) by w{T){x, t) = wk{x)
if {x, t) G Q x ((fc - 1)T, fcr]
T
and g( ' in a similar way. Observe that assumption (A3) implies q(T) - > ?
in L2(QT) as r -> 0.
(23)
An immediate consequence of Lemma 1 is the following result: Corollary 1 Let r > 0. Then we have ||?/T'||L<>°(o,:r;i,i(n)) < C and 2
J2 (a||e w - M/2 ||^(o,r;^(n)) + ailleffi,M|U2(o,T;ifHn))) < C, iti/iere C > 0 is independent of T, a = 2min{ci,C2}, and 2
»>(T)(t) = E
.
/ (^(^M)K!r)-«'fl)-''i(^T)) + &i(«'i>)+aft(«'lT))-^T)+^,0)W^-
We also need an estimate for the discrete time derivative. For this we define ftW(.)t)
=
f l _ I (p(wk) - biw"-1)) + b(wk),
t > 0,
and introduce aT, the shift operator: aTw(T){-,t) = wk~l Then we have (see [11])
if t e {{k-i)T,kr],
k=
l,...,N.
119 L e m m a 3 It holds
||6(u/M) - KoM^h^Ty)
< CT,
Hft6W||1.(o,Tiv) + I I ^ I I L W ; ^ ) ) < C,
where C does not depend on T and V* = (HQ(Q U IV))* is the dual ofV = Hg(fl U TN). S t e p 6. We already have all the necessary estimates to pass to the limit r —> 0. Since the embedding H1^!) <-> L°°(Q) is compact in one space dimension, we can apply Aubin's lemma [28] to ¥T\ in view of the uniform bounds of Lemma 3, to obtain, up to a subsequence which is not relabeled, dt¥T) ->• dtz 6
weakly in L 1 (0, T; V), weakly in L 2 (0,T; Hl{n)), strongly in L 2 (0,T; L°°(n)),
6(U/ T ) ) ->• u
weakly in L 2 (0, T; H^Q)).
(24) (25) (26) (27)
By Lemma 3 we have, as r —>• 0, ||6M - &(™ (T) )||LI(O,T ; V) < ||6(t«W) - 6(aTUJ(T))|UHo,r;v) -> 0,
and hence z = u. Finally, using the above estimates and convergences, we are able to prove that em-'T> = bi{w{T)) -» m strongly in L2{QT), i = 1,2. (28) Now we can let r -» 0 in the weak formulation of (19), i = 1,2, which reads for 0eL°°(O,T; (W^iil))*) : /" (dtb%(T),
-di f
(CiVera'T) + 2aiew'T)Ve,"-T).+ Ve",ST)+,"^)) • V^ctedt ew'r)q<-T'>-V
JQT
fi{ew^,ew^)cj)dxdt.
JQT
In view of (24)-(28), (23) and Assumption (A4) we obtain / {dtuh4>)dt+ / (ciVui + 2aiui'Vui + Jo JQT =
— di I JQT
u,q -V
V{u1U2))-V4>dxdt
Ji{u\,U2)
JQT
i.e. u = («i,U2) is a weak solution of (11)-(12). Moreover, the initial condition (14) is satisfied in the sense of V. As a consequence of the above proof we obtain a positivity-preserving numerical scheme. This scheme will be employed for numerical simulations of the evolution problem in a future work. For numerical stationary solutions, we refer to, e.g., [10].
120
3
A simple example
We consider in this section, as an example, the one-dimensional version of the stationary problem corresponding to problem (11)-(14), with }i{u\,u2) = 0, i = 1,2, and non-flux boundary conditions, as stated in [27]. Notice that this choice of boundary conditions and source terms implies the conservation of the number of individuals of each of the species. This may be regarded as a simple equilibrium model of a dynamics in which population pressure effects are much faster than growth and competition effects. For the study of the long-time behaviour of solutions in the case in which interaction of pressure and growth-competition terms are present (no environmental potential), the reader is refered to [24]. We shall see that this rather simple example already exhibits the phenomena of segregation of populations. We obtain necessary conditions relating transport and crossdiffusion coefficients. Sufficient conditions either involves the size of the solution or are just a consequence of a very steep difference between the transport coefficients of both equations. Integrating the equations that one obtains when dropping the time derivative and the source terms of equations (11) in (0,:r), with i e ( l , and using the boundary conditions, we obtain the following problem. Find {ui,u2) : O x Q —>• R^_ such that («i(ci + a1u1 + u2))' + dxqui = 0
.
[U2(c2 +
and
Ja
/ u2 ~ u2.
(30)
in
Here, U\ and u2 stand for the mass of u\ and u2, respectively. We have the following existence result, which is independent of the size of the non-negative parameters a*. Therefore, condition (9) is not necessary in this example, too. Proposition 1 Let c\ and c2 be positive and assume q e L°°(fi). There exists a solution (ui,u2) of Problem (29)-(30) such thatui,u2 > 0 in CI andu\,u2 6 W1,a°(Q). In addition, ifqe H""'°°(fl) then the solution (uuu2) satisfies uuu2 6 W™+1'°°(n). The proof is straightforward. We just rewrite equations (29)-(30) as gfei(Mi,M2)
g{ui,u2)
u,
'
ni,
_
_qh2{ui,u2)
g(uuu2)
U,
(31)
with g(ui, u2) := (ci + 2a1u1 + u2)(c2 + 2a2u2 + ui) - uiu2 > 0, hi(ui,u2) := di(c2 + 2a2u2 + ui) - d2u2,
(32)
h2{ux,u2) := d2(ci + 2ai«i + u2) — diU^. Then, we use an iterative scheme on the representation formulae of the solutions given by i \ m\ [x -qhi(u u ) Ui{x) = «i(0) exp{f / ——; lt 2 xr - } , JO »l(«l>«2)
[ fx u2{x) = w2(0) exp{ / Jo
-qh2{ui,u2), ^ r^}, MMl.Ua)
(33)
121 with ui(0),u2(0) determined by (30) (see [10] for details). We now discuss the notion of segregation in a simplified framework. We shall assume, following [27], that the environmental potential U(x) is a smooth function such that the corresponding enviromental flux satisfies q(x0) = 0 for a single point x0 € fi. Definition 1 We say that the stationary problem (29)-(30) has the property of segregation if there exists a point XQ £ Q such that u\ and u2 have a local maximum and minimum at XQ, respectively (or vice versa). Observe that, since q(xo) = 0, from (31) we find u\(x0) = u'2(x0) = 0, so we deduce the existence of a critical point, x0, both for ui and u2. Taking derivatives in (31) and evaluating at Xo we obtain g(u1(x0),u2{x0))u"(x0) = g(u1(x0),U2(x0))u2(x0) =
-qJ{x0)h1(ui(xo),U2{x0)), -q'{xo)h2(u1(xo),U2(xo)).
Therefore, segregation will take place if hi(ui(x0),u2(x0))h2(ui(x0),u2{xQ)) (32) we see that this is equivalent to one of the following conditions: u2(x0)(di - 2o2di) > di(c2 + ui(z 0 )),
< 0. ^From
ui(x0)(di - 2aid2) > rf2(ci + u2(x0)).
(34)
In the trivial case in which d\d2 < 0, segregation does always occur. Notice that in this case x0 is an attractive point (from the environmental point of view) for one of the populations and a repulsive point for the other. We may then analyze the case di > 0, i = 1,2 (the case di < 0 is similar). From (34) we obtain a necessary condition involving the diffusion coefficients, namely dt > 2atdj, i,j = 1,2, i ^ j . Observe that if none of these conditions are satisfied (no segregation) then the coefficients oi, a2 satisfy a\a2 > 1/4, which is a condition in the same spirit as (9). On the other hand, since the L°° norms of «; are bounded when di —> 0, i = 1,2, see [10], we conclude that a sufficient condition for segregation is that di/d2 is large or small enough. We discretized problem (11)-(14), with no-flux boundary conditions and zero source terms, by a semi-implicit finite difference method and let tend t —> oo in order to obtain steady-state solutions. The numerical experiments conducted show that, for a wide range of model parameters and initial conditions, the numerical scheme is stable, which allows us to show some different behaviours of the two interacting species included in the model. In particular, it is interesting to observe when segregation of the two species, due to habitat hetereogenity, does appear. We also used an iterative scheme based on the formulae (33) and could verify that the solutions of (11)-(14), obtained in the large-time limit, coincide with the solutions of the iterative scheme. We show here numerical experiments corresponding to the following cases: (a) Large and smallcross-diffusion terms. Set c; = dt = 1 and constant initial conditions Mio = 10, «2o = 20. Then small a* correspond to large cross-diffusion and vice versa. Figure 1 shows the numerical solution of model corresponding to values of Oi = 0,0.1,10.
122 (b) Segregation effects due to different transport coefficients, i.e. di 3> d2. Set ct = d2 = 1 and tiio = 10, u2o = 10. Figure 1 shows the numerical solutions corresponding to a i = 0,2 = 1 and d\ = 4,8,20,40. As case (a) shows, large cross-diffusion enhance segregation effects, therefore we repeat the last simulation with a; = 0.1. Figure 2 shows the same results as before also for this case. (c) Set Ci = 1, ai = 1, a2 = 0.01, di = 40, d2 = 1, Mio = W20 = 10 (Figure 2). Here, a neat segregation of the two species can be observed.
Acknowledgements The authors acknowledge partial support from the TMR Project "Asymptotic methods in kinetic theory", grant number ERB-FMBX-CT97-0157 and from the DAAD Project "Acciones Integradas". The first and second authors were supported by the Spanish Ministerio de Ciencia y Tecnologia, project number MCT-OO-BFM-1324. The last author was supported by the Gerhard-Hess Program of the Deutsche Forschungsgemeinschaft, grant JU 359/3, and by the AFF Project of the University of Konstanz, grant 4/00.
References [1] H. Amann and M. Renardy. Reaction-diffusion problems in electrolysis. Nonlin. Diff. Eqs. Appl., 1:91-117, 1994. [2] N. Boudiba and M. Pierre. Global existence for coupled reaction-diffusion equations systems. J. Math. Anal. Appl, 250:1-12, 2000. [3] Y. Choi and R. Lui. Multi-dimensional electrochemistry model. Arch. Rat. Mech. Anal, 130:315-342, 1995. [4] P. Degond, S. Genieys, and A. Jiingel. Symmetrization and entropy inequality for general diffusion equations. C. R. Acad. Sci. Paris, 325:963-968, 1997. [5] P. Deuring. An initial-boundary value problem for a certain density-dependent diffusion system. Math. Z., 194:375-396, 1987. [6] Z. Duan and L. Zhou. Global and blow-up solutions for nonlinear degenerate parabolic systems with crosswise diffusion. J. Math. Anal. Appl, 224:263-278, 2000. [7] R. Eisenberg and D. Gillespie. A singular perturbation of the Poisson-Nernst-Planck system: application to ion channels. Preprint, 1999. [8] G. Gagneux and M. Madaune-Tort. Analyse mathematique de modeles non lineaires de I'ingenierie petroliere. Springer, Paris, 1996. [9] H. Gajewski and K. Zacharias. Global behaviour of a reaction-diffusion system modelling chemotaxis. Math. Nachr., 195:77-114, 1998.
123
[10] G. Galiano, M. Garzon, and A. Jiingel. Analysis and numerical solution of a nonlinear cross-diffusion system arising in population dynamics. To appear in Revista de la Real Academia de Ciencias Exactas, Fis. y Nat, 2002. [11] G. Galiano, M. Garzon, and A. Jiingel. Semi-discretization in time and numerical convergence of solutions of a nonlinear cross-diffusion system arising in population dynamics. Submitted for publication, 2000. [12] M. T. Gyi and A. Jiingel. A quantum regularization of the one-dimensional hydrodynamic model for semiconductors. Adv. Diff. Eqs., 5:773-800, 2000. [13] H. Hoshino. Nonnegative global solutions to a class of strongly coupled reactiondiffusion systems. Advances Diff. Equ., 5:801-832, 2000. [14] T. Ichikawa and Y. Yamada. Some remarks on global solutions to some quasilinear parabolic system with cross-diffusion. Funkcialaj Ekvacioj, 43:285-301, 2000. [15] A. Jiingel. Quasi-hydrodynamic Semiconductor Equations. Progress in Nonlinear Differential Equations and Its Applications. Birkhauser, Basel, 2001. [16] A. Jiingel and Y.-J. Peng. A hierarchy of hydrodynamic models for plasmas. Zeroelectron-mass limits in the drift-diffusion equations. Annales H. Poincare, 17:83-118, 2000. [17] A. S. Kalashnikov. Some problems of the qualitative theory of non-linear degenerate second-order parabolic equations. Russ. Math. Surveys, 42:169-222, 1987. [18] J. Kamel and M. Kirane. Pointwise a priori bounds for a strongly coupled system of reaction-diffusion equations with a balance law. Math. Meth. Appl. Sci., 21:12271232, 1998. [19] J.U. Kim. Smooth solutions to a quasi-linear system of diffusion equations for a certain population model. Nonlin. Anal, 8:1121-1144, 1984. [20] Y. Lou and W.-M. Ni. Diffusion, self-diffusion and cross-diffusion. 131:79-131, 1996.
J. Diff. Eqs.,
[21] Y. Lou and W.-M. Ni. Diffusion, vs. cross-diffusion: an elliptic approach. J. Diff. Eqs., 154:157-190, 1999. [22] Y. Lou, W.-M. Ni, and Y. Wu. The global existence of solutions for a cross-diffusion system. Adv. Math., Beijing, 25:283-284, 1996. [23] P. A. Markowich, C. A. Ringhofer, and C. Schmeiser. Semiconductor Equations. Springer, 1990. [24] M. Mimura and K. Kawasaki. Spatial segregation in competitive interaction-diffusion equations. J. Math. Biol., 9:49-64, 1980.
124 [25] K. Post. A non-linear parabolic system modeling chemotaxis with sensitivity functions. Preprint, Humboldt-Universitat Berlin, Germany, 1999. [26] M. Pozio and A. Tesei. Global existence of solutions for a strongly coupled quasilinear parabolic system. Nonlin. Anal., 14:657-689, 1990. [27] N. Shigesada, K. Kawasaki, and E. Teramoto. Spatial segregation of interacting species. J. Theor. Biol., 79:83-99, 1979. [28] J. Simon. Compact sets in the space V(0,T; B). Ann. Math. Pura AppL, 146:65-96, 1987. [29] Y. Wu. Qualitative studies of solutions for some cross-diffusion systems. In T.-T. Li, M. Mimura, Y. Nishiura, and Q.-X. Ye, editors, China-Japan Symposium on Reaction-Diffusion Eqautions and Their Applications and Computational Aspects, pages 177-187, Singapore, 1997. World Scientific. [30] A. Yagi. Global solution to some quasilinear parabolic system in population dynamics. Nonlin. Anal., 21:603-630, 1993.
125
Figure 1: Stationary density distributions u\{x) (solid line) and u2{x) (broken line) for x e (0,3). Left figure: case (a); the numbers correspond to the values of a,. Right figure: case (b) with ai = 02 = 1; the numbers correspond to values of d\.
Figure 2: Stationary density distributions u^x) (solid line) and u2(x) (broken line) for x e (0,3). Left figure: case (b) with di = a2 = 0.1; the numbers correspond to values of d\. Right figure: case (c).
The nonstationary Stokes and Navier-Stokes equations in aperture domains Toshiaki Hishida Fachbereich Mathematik, Technische Universitat Darmstadt D-64289 Darmstadt, Germany Email: [email protected] Abstract We consider the nonstationary Stokes and Navier-Stokes flows in aperture domains fi c R",TI > 3. We develop the Lq-U estimates of the Stokes semigroup and apply them to the Navier-Stokes initial value problem. As a result, we obtain the global existence of a unique strong solution, which satisfies the vanishing flux condition through the aperture and some sharp decay properties as t —> oo, when the initial velocity is sufficiently small in the Ln space. Such a global existence theorem is up to now well known in the cases of the whole and half spaces, bounded and exterior domains. In this article we study the global existence and asymptotic behavior of a strong solution to the Navier-Stokes initial value problem in an aperture domain SI C K" with smooth boundary dSl: dtu + u-Vu V-u u|an u\t=o
= Au-Vp =0 =0 =a
(xeQ,t>0), (xen,t>0), (t > 0), (x 6 SI),
y
,. '
where u(x, t) = (ui(x, t), • • • , un(x, t)) and p(x, t) denote the unknown velocity and pressure of a viscous incompressible fluid occupying SI, respectively, while a(x) = (ai(x), • • • , an(x)) is a prescribed initial velocity. The aperture domain Q. is a compact perturbation of two separated half spaces H+ U i?_, where H± = {x = (Xi, • • • , xn) € R"; ±xn > 1}; to be precise, we call a connected open set Q C Rn an aperture domain (with thickness of the wall) if there is a ball B C Rn such that Q. \ B = (H+ U #_) \ J3. Thus the upper and lower half spaces H± are connected by an aperture (hole) M C fl D B, which is a smooth (n — l)-dimensional manifold so that fi consists of upper and lower disjoint subdomains ft± and M: SI = Q+ U M U 0_. The aperture domain is a particularly interesting class of domains with noncompact boundaries because of the following remarkable feature, which was in 1976 pointed out by Heywood [20]: the solution is not uniquely determined by usual boundary conditions
126
127 even for the stationary Stokes system in this domain and therefore, in order to single out a unique solution, we have to prescribe either the flux through the aperture M N • uda, JM
or the pressure drop at infinity (in a sense) between the upper and lower subdomains Q± [p] =
lim p(x) lim p(x), |x|-»oo,xen+ |x|—Kx),xen_
as an additional boundary condition. Here, N denotes the unit normal vector on M directed to fi_ and the flux
(2)
where v is the unit outer normal vector on dfi,. Here, the condition
128 \p(t)] is prescribed, Franzke introduced another kind of Stokes operator associated with the pressure drop condition, which generates a bounded analytic semigroup on the space {u e Lq(fi); V • u = 0, v • u\gn = 0} for n > 3 and n/(n — 1) < q < n (based on a resolvent estimate due to Farwig [12]). Because of this restriction on q, the Lq theory with q > n is not available under the pressure drop condition and thus one cannot avoid a regularity assumption to some extent on initial data. It is possible to discuss the L2 theory of global strong solutions for an arbitrary unbounded domain (with smooth boundary) in a unified way since the Stokes operator is a nonnegative selfadjoint one in L\\ see Heywood [22] (n = 3), Kozono and Ogawa [26] (n = 3) and Kozono and Sohr [27] (n = 4, 5). Especially, from the viewpoint of the class of initial data, optimal results were given by [26] and [27]. In fact, they constructed a global solution with various decay properties for small a € D{A% ). Here, we should recall the continuous embedding relation D(A" ) C I J . For the aperture domain fl their solutions should satisfy the hidden flux condition (f){u(t)) = 0 on account of u(t) e Ll(Q) together with (2). In his Doktorschrift [16] Franzke studied, among others, the global existence of strong solutions in a 3-dimensional aperture domain when either <j){u{t)) or \p{t)\ is prescribed. Indeed, the local strong solution in the L2 space constructed by himself [15] was extended globally in time under the condition that both a £ #o(0) (with compatibility conditions) and the other data are small in a sense, however, his method gave no information about the large time behavior of the solution. This article provides the global existence theorem for a unique strong solution of (1), which satisfies the flux condition <j>(u(t)) = 0 and some decay properties with definite rates as t —>• oo when the initial velocity a is small enough in L"(Q),n > 3. For the proof, as is well known, it is crucial to establish the Lq-Lr estimates of the Stokes semigroup \\e~tAf\\L^)
< Cr°||/||L,(n),
(3)
l|Ve-"/||L,(n) < Cr-^H/H^n),
(4)
for all t > 0 and / e Lqa(Q), where a = (n/q - n/r)/2 > 0. Recently for n > 3 Abels [1] has proved some partial results: (3) for 1 < q < r < oo and (4) for 1 < q < r < n. However, because of the lack of (4) for the most important case q = r = n, his results are not satisfactory for the construction of the global strong solution possessing various time-asymptotic behaviors as long as one follows the straightforward method of Kato [24]. Our main result on the Lq-U estimates is: Theorem 1 Let n > 3. 1. Let 1 < q < r < oo (q ^ oo). There is a constant C = C(Q,,n,q,r) > 0 such that (3) holds for allt>0 and f e £«(fi). 2. Let 1 < q < r < n or\ < q < n < r < oo. There is a constant C = C(Q, n, q, r) > 0 such that (4) holds for allt > 0 and f G Lqa(Q). 3. Letl
fast->0 j
M M ( X )
ifq
129
]|Ve-M/||r = oft-"-1'2) as t —> 0 as t —» oo
if q < r < oo, ifq
oo,
where a = (n/q — n/r)/2. Furthermore, for each precompact set K in Ll(Q) every convergence above is uniform with respect to f € K. By use of the Stokes operator A, one can formulate the problem (1) subject to the vanishing flux condition 4>{u{t)) =
N • u{t)da = 0,
t>0,
(5)
JM
as the Cauchy problem dtu + Au + P{u • Vu) = 0,
t> 0; u(0) = a,
(6)
in LJ(Q). Given a e L"(Q,) and 0 < T < oo, a measurable function u defined on SI x (0, T) is called a strong solution of (1) with (5) on (0, T) if u is of class u e C([0, T); /£(«)) n C(0, T; D(A n )) n C\Q, T; Lna(tt)) together with lim^o \\u(t) — a\\n = 0 and satisfies (6) for 0 < t < T in L"(fi). We make use of (3) and (4) to obtain: Theorem 2 Let n > 3. There is a constant 5 = 6(0,, n) > 0 with the following property: if a e £"(£2) satisfies \\a\\n < 5, then the problem (1) with (5) admits a unique strong solution u(t) on (0,oo), which enjoys \\u(t) ||r = o (t-l'2+nl2r)
forn
\\Vu(t)\\n =
o{r^),
||ft«(t)|| B + ||Au(t)|| B = o ( r 1 ) , as t —> oo. The strong solution is constructed as the solution of the integral equation u(t) = e~tAa - [ e-{i-T)AP(u Jo
• Vw)(r)dT,
t >.0,
by means of a standard contraction mapping principle along the lines of Kato [24]. The obtained solution is unique within the class u € C([0, oo); Z£(Q)),
Vu G C(0, oo; L"(Q)),
without assuming any behavior near t = 0 as pointed out by Brezis [7]. When one prescribes a nontrivial flux
130 a unique strong solution on (0, Tt) provided that a € Ln{Q) satisfies the compatibility conditions V • a = 0, v • a\gn — 0 and (j){a) = F(0). This improves a related result of Pranzke [17] and can be proved in the same manner as the proof of Theorem 2 with the aid of the auxiliary function of Heywood ([20], Lemma 11), which is used for the reduction of the problem to an equivalent one with the vanishing flux condition (5). Up to now we have the same global existence result as Theorem 2 for the whole space (Kato [24]), the half space (Ukai [32]), bounded domains (Giga and Miyakawa [18]) and exterior domains (Iwashita [23]) since the Lq-U estimates (3) and (4) are well established for these four types of domains. Especially for exterior domains with n > 3, based on (3) for q = r due to Borchers and Sohr [5], some partial results were given by Iwashita [23], Giga and Sohr [19] and Borchers and Miyakawa [4]; in particular, Iwashita proved (3) for 1 < q < r < oo and (4) for 1 < q < r < n, which made it possible to construct a global solution. Later on, due to the following authors, (3) for n > 2,1 < q < r < co (q ^ oo) and (4) for 1 < q < r < n = 2 were also derived: Chen [8] (n = 3), Shibata [31] (n = 3), Borchers and Varnhorn [6] (n = 2, (3) for q = r), Dan and Shibata [10], [11] (n = 2), Dan, Kobayashi and Shibata [9] (n = 2,3) and Maremonti and Solonnikov [28] (n > 2). In the proof of the Lq-U estimates, it seems to be heuristically reasonable to combine some local decay properties near the aperture with the Lq-Lr estimates of the Stokes semigroup for the half space (Ukai [32]) by means of a localization procedure since the aperture domain n is obtained from H+ U H_ by a perturbation within a compact region. Indeed, Abels [1] used this idea that was well developed by Iwashita [23] and, later, Kobayashi and Shibata [25] in the case of exterior domains. We should however note that the boundary dQ is noncompact; thus, a difficulty is to deduce the sharp local energy decay estimate
\\e'tAf\\w^(nR) < Ct-n'"\\f\\ma),
t > 1,
(7)
for / e Ll(Q), 1 < q < co, where Q,R = {x G CI; \x\ < R}, but this is the essential part of our proof. The estimate (7) improves the local energy decay given by Abels [1], in which a little slower rate t~n/2li+e was shown. In [1], similarly to Iwashita [23], a resolvent expansion around the origin A = 0 was derived in some weighted function spaces. To this end, Abels made use of the Ukai formula of the Stokes semigroup for the half space ([32]) and, in order to estimate the Riesz operator appearing in this formula, he had to introduce Muckenhoupt weights, which caused some restrictions although his analysis itself is of interest. On the other hand, Kobayashi and Shibata [25] refined the proof of Iwashita in some sense and obtained the Lq-U estimates of the Oseen semigroup for the 3-dimensional exterior domain. As a particular case, the result of [25] includes the estimates of the Stokes semigroup as well. To prove Theorem 1 we employ in principle the strategy developed by [25] (without using any weighted function space) and extend the method to general n > 3. The proof is divided into three steps and the outline is as follows. The first step is the investigation of the Stokes resolvent for the half space H = H+ or H-. We derive some regularity estimates near the origin A = 0 of (A + A f f ) _ 1 P f f / when / e Lq(H) has a bounded support, where AH = — P#A is the Stokes operator for the half space H. Although the obtained estimates do not seem to be optimal compared with those shown by [25] for the whole space, the results are sufficient for our aim and the
131 proof is rather elementary: in fact, we represent the resolvent (A + A#) _ 1 in terms of the semigroup e~tA" and, with the aid of local energy decay properties of this semigroup, we have only to perform several integrations by parts and to estimate the resulting formulae. One needs neither Fourier analysis nor resolvent expansions. In the next step, based on the results for the half space, we proceed to the analysis of the Stokes resolvent for the aperture domain fi. To do so, in an analogous way to [23], [25] and [1], we first construct the resolvent (A + A)~1Pf near the origin A = 0 for / € Lq(Q) with bounded support by use of the operator (A + AH)~1PH, the Stokes flow in a bounded domain and a cut-off function together with the result of Bogovskii [2] on the boundary value problem for the equation of continuity. And then, for the same / as above, we deduce essentially the same regularity estimates near the origin A = 0 of (A + A)~lPf as shown in the previous step. In the final step we prove (7) and thereby (4) for q = r G (l,n] as well as (3) for r = oo, from which the other cases follow. Some of the estimates obtained in the previous step enable us to justify a representation formula of the semigroup e~tAPf in Wl'q(QR) in terms of the Fourier inverse transform of 9™(is + A)-1 Pf when / G Lq(Q) has a bounded support, where n = 2m + 1 or n = 2m + 2. We then appeal to the lemma due to Shibata ([30], [25]), which tells us a relation between the regularity of a function at the origin and the decay property of its Fourier inverse image, so that we obtain another local energy decay estimate \\e~tAPf\\wl„{nR) q
< Cr"/2+l/||L,(n),
t > 1,
(8)
for / G L (Q,), 1 < q < oo, with bounded support, where e > 0 is arbitrary. Estimate (8) was shown in [1] only for solenoidal data / G Z/'(0) with bounded support, from which (7) with the rate replaced by t~n/2q+s follows through an interpolation argument. But it is crucial for the proof of (7) to use (8) even for data which are not solenoidal (so that the support of Pf is unbounded). In order to deduce (7) from (8), we develop the method in [23] and [25] based on a localization argument using a cut-off function. In fact, we regard the Stokes flow for the aperture domain Q as the sum of the Stokes flows for the half spaces H± and a certain perturbed flow. Since the Stokes flow for the half space enjoys the L'-L°° decay estimate with the rate i~"/2« (Borchers and Miyakawa [3]), our main task is to show (7) for the perturbation part. In contrast to the case of exterior domains, the support of the derivative of the cut-off function touches the boundary dil; indeed, this difficulty occurs in all stages of localization procedures in the course of the proof and thus we have to carry out such procedures carefully. Furthermore, the remainder term arising from the above-mentioned localization argument involves the pressure of the nonstationary Stokes system in the half space and, therefore, does not belong to any solenoidal function space. Hence, in order to treat this term, (8) is necessary for non-solenoidal data, while that is not the case for the exterior problem. Acknowledgment. The present work was partially supported by the Alexander von Humboldt research fellowship.
132
References [1] Abels, H., Lq-Lr estimates for the non-stationary Stokes equations in an aperture domain, Z. Anal. Anwendungen (to appear). [2] Bogovskii, M. E., Solution of the first boundary value problem for the equation of continuity of an incompressible medium, Soviet Math. Dokl. 20, 1094-1098 (1979). [3] Borchers, W., Miyakawa. T., L2 decay for the Navier-Stokes flow in halfspaces, Math. Ann. 282, 139-155 (1988). [4] Borchers, W., Miyakawa. T., Algebraic L2 decay for Navier-Stokes flows in exterior domains, Acta Math. 165, 189-227 (1990). [5] Borchers, W., Sohr, H., On the semigroup of the Stokes operator for exterior domains in L'-spaces, Math. Z. 196, 415-425 (1987). [6] Borchers, W., Varnhorn, W., On the boundedness of the Stokes semigroup in twodimensional exterior domains, Math. Z. 213, 275-299 (1993). [7] Brezis, H., Remarks on the preceding paper by M. Ben-Artzi "Global solutions of two-dimensional Navier-Stokes and Euler equations", Arch. Rational Mech. Anal. 128, 359-360 (1994). [8] Chen, Z. M., Solutions of the stationary and nonstationary Navier-Stokes equations in exterior domains, Pacific J. Math. 159, 227-240 (1993). [9] Dan, W., Kobayashi, T., Shibata, Y., On the local energy decay approach to some fluid flow in an exterior domain, Recent Topics on Mathematical Theory of Viscous Incompressible Fluid, 1-51, Lecture Notes Numer. Appl. Math. 16, Kinokuniya, Tokyo, 1998. [10] Dan, W., Shibata, Y., On the Lq-LT estimates of the Stokes semigroup in a 2dimensional exterior domain, J. Math. Soc. Japan 51, 181-207 (1999). [11] Dan, W., Shibata, Y., Remarks on the Lq-L^, estimate of the Stokes semigroup in a 2-dimensional exterior domain, Pacific J. Math. 189, 223-239 (1999). [12] Farwig, R., Note on the flux condition and pressure drop in the resolvent problem of the Stokes system, Manuscripta Math. 89, 139-158 (1996). [13] Farwig, R., Sohr, H., On the Stokes and Navier-Stokes system for domains with noncompact boundary in L'-spaces, Math. Nachr. 170, 53-77 (1994). [14] Farwig, R., Sohr, H., Helmholtz decomposition and Stokes resolvent system for aperture domains in L?-spaces, Analysis 16, 1-26 (1996). [15] Franzke, M., Strong solutions of the Navier-Stokes equations in aperture domains, Ann. Univ. Ferrara Sez. VII. Sc. Mat. 46, 161-173 (2000).
133 [16] Franzke, M., Die Navier-Stokes-Gleichungen in Offnungsgebieten, Technische Universitat Darmstadt, 2000.
Doktorschrift,
[17] Franzke, M., Strong L'-theory of the Navier-Stokes equations in aperture domains, Preprint Nr. 2139 TU Darmstadt (2001). [18] Giga, Y., Miyakawa, T., Solutions in Lr of the Navier-Stokes initial value problem, Arch. Rational Mech. Anal. 89, 267-281 (1985). [19] Giga, Y., Sohr, H., On the Stokes operator in exterior domains, J. Fac. Sci. Univ. Tokyo Sect. IA 36, 103-130 (1989). [20] Heywood, J. G., On uniqueness questions in the theory of viscous flow, Acta Math. 136, 61-102 (1976). [21] Heywood, J. G., Auxiliary flux and pressure conditions for Navier-Stokes problems, Approximation Methods for Navier-Stokes Problems, 223-234, Lecture Notes in Math. 771, Springer, Berlin, 1980. [22] Heywood, J. G., The Navier-Stokes equations: on the existence, regularity and decay of solutions, Indiana Univ. Math. J. 29, 639-681 (1980). [23] Iwashita, H., Lq-Lr estimates for solutions of the nonstationary Stokes equations in an exterior domain and the Navier-Stokes initial value problems in Lq spaces, Math. Ann. 285, 265-288 (1989). [24] Kato, T., Strong LP solutions of the Navier-Stokes equation in R m , with applications to weak solutions, Math. Z. 187, 471-480 (1984). [25] Kobayashi, T., Shibata, Y., On the Oseen equation in the three dimensional exterior domains, Math. Ann. 310, 1-45 (1998). [26] Kozono, H., Ogawa, T., Global strong solution and its decay properties for the NavierStokes equations in three dimensional domains with non-compact boundaries, Math. Z. 216, 1-30 (1994). [27] Kozono, H., Sohr, H., Global strong solution of the Navier-Stokes equations in 4 and 5 dimensional unbounded domains, Ann. Inst. H. Poincare Anal. Nonlineaire 16, 535-561 (1999). [28] Maremonti, P., Solonnikov, V. A., On nonstationary Stokes problem in exterior domains, Ann. Sc. Norm. Sup. Pisa 24, 395-449 (1997). [29] Miyakawa, T., The Helmholtz decomposition of vector fields in some unbounded domains, Math. J. Toyama Univ. 17, 115-149 (1994). [30] Shibata, Y., On the global existence of classical solutions of second order fully nonlinear hyperbolic equations with first order dissipation in the exterior domain, Tsukuba J. Math. 7, 1-68 (1983).
134 [31] Shibata, Y., On an exterior initial boundary value problem for Navier-Stokes equations, Quart. Appl. Math. 57, 117-155 (1999). [32] Ukai, S., A solution formula for the Stokes equation in R™, Commun. Pure Appl. Math. 40, 611-621 (1987).
Global attractors for multivalued flows associated with sub differentials Noboyuki KENMOCHI Department of Mathematics Faculty of Education, Chiba University, 1-33 Yayoi-cho, Inage-ku, Chiba, 263-8522 Japan Email: [email protected] Noriaki YAMAZAKI Department of Mathematical Science Common Subject Division Muroran Institute of Technology 1-27 Mizumoto-cho, Muroran, 050-8585 Japan Email: [email protected]
Abstract In a real Hilbert space we consider a class of non-autonomous evolution equations including time-dependent subdifferentials and their perturbations. We are interested in the case when the Cauchy problems have in general multiple solutions, namely they may lose the uniqueness of solutions. In such a situation the associated flows (or dynamical systems) are multivalued, and there have not obtained so many theoretical results especially in the non-autonomous case. In this paper we investigate the large time behaviour of the multivalued flows from the view-point of attractors. In fact, we construct a global attractor for our multivalued flows, and discuss the relationship to the one for the flows associated with the limiting autonomous system.
1
Introduction
In this paper let us consider a nonlinear evolution equation in a real Hilbert space H of the form: u'(i) + dip*{u{t)) + g(t, u{t)) 3 f(t) in if, t > s ( > 0), (1.1)
135
136 where d(p* is the subdifferential of time-dependent proper l.s.c. and convex function tp* on H, g(t, •) is a single-valued operator in H which is rather small relative to ip* and / is a forcing term given in Lfoc([0, +00); H). We assume that ip*, g(i, •) and f(t) respectively converge to a convex function ip°° on H, an operator g°°(-) in H and an element /°° in H in appropriate senses as t —>• +00. Then, we have the limiting autonomous system (as s —>• +00): v'{t) + d
intf,
t > 0.
(1.2)
In this paper, assuming that the Cauchy problems for (1.1) and (1.2) have in general multiple solutions, we shall discuss the large-time behaviour of solutions of (1.1). In such a situation the solution operator E(t,s) (0 < s < t < +00) for (1.1) is multivalued. Namely, E(t,s) (0 < s < t < +00) is the multivalued operator from D(ips) into D{ipl) which assigns to each u0 G D(
There is a solution u of (1.1) on [s, +00) 1 ^ ^ ^(t) ^ j . u{s) = ^ ^
., „. (1.3)
Of course, the solution operator 5(i) (i > 0) for the limiting autonomous system (1.2) is similarly defined as a multivalued operator in Z?(
2
:= sup inf \x — y\H-
Global boundedness of solutions
In what follows, system (1.1) is referred as (Es) and its Cauchy problem associated with initial value u0 as (E S ;K 0 ), namely (E • u ) I U ' W + ^ ' M * ) ) + 9(t, u(*)) 3 f(t) in H, t > s (> 0), s ' 0> \ u(s) = MoDefinition 2.1. Let J be any interval in R+ with initial time s. Then (i) A function u : J -> H is called a solution of (E,) on J, if u E C(J; H) n W^(J; <^()(u(-)) e L}0C(J), u{t) e Didip*) for a.e. t e J and f(t) - u'(t) - g(t,u(t)) e V ( w ( t ) ) ,
a.e. t G J.
H),
137 (ii) A function u : J —> H is called a solution of the Cauchy problem (Es; w0) on J with given initial value u0 G H, if it is a solution of (Es) on J satisfying u(s) = u0. Now, let {o r } := {ar; r G R+} and {&,; r G R+} be two families of absolutely continuous (real) functions ar, br on R+, with parameter r G R+, such that a'r G Ll{R+) n L2{R+),
b'r G L^R+y,
(2.1)
by (2.1), the limits a r (+oo) := lim aJt) and 6 r (+oo) := lim br(i) exist, so ar and br t—>+oo
£—y+oo
are considered as continuous functions on R+ := [0,+oo]. With these families {ar} and {br}, we specify a class $({a r }, {6r}) of families {?'} := {
+ |^W|).
For {<£*} G $({a r }, {&,•}), d{-, •) and /(•), we further require the following conditions: ( A l ) There exists a positive constant C\ > 0 such that
Vi G R~;, VZ G Z % ' ) .
(A2) For every finite r > 0 and t G i? + , the level set {z £ H ; tpt(z) < r} is compact in H. (A3) Diip1) C D(g(t, •)) C .ff for any t G R+, and (-,u(')) is strongly measurable on J for any interval J C -R+ and u G Lfoc(J; H) with <£>(''(«(•)) G L}0C(J). (A4) There are positive constants C 2 , C3 such that lff(*,2)l!r < < V ( * ) + Cs,
Vi G S^, Vz G D(v»t)-
(A5) (demi-closedness) If z„ G D((ptn), {tn} C f? + , {
b)E > C4(B)\z\j, - C5(B),
Vt eR~;,\/ze
Dtf),
V6 G B.
138 As to the existence and global boundedness of a solution for (E s ; u0) on [s, +00), we have the following theorem, which can be proved in the same way as in [8]. Theorems 2 . 1 . [8, Theorems 2.1, 2.2] Let f £ Lf0C(R+;H) and assume that {?'}€ $({a r }, {br}) and (A1)-(A6) hold. Then, for each s > 0 and u0 e D(
sup \u(t)\2H + sup /
ipT(u(r))dT < JVi(l + S) + |u 0 |lf),
where Ni is a positive constant independent of f, s > 0 and «o 6 D(ips). Moreover, for each 6 > 0 and each bounded set B C H, there is a constant Ni{5, B) > 0, depending only on 5 > 0 and B, such that sup \u'\Wt+1.H) t>s+s
+ sup f\u{t)) t>s+e
< N2{5,B),
as long as s > 0 and «o € D(tps) D B.
3
Global attractor of the non-autonomous system
By Theorem 2.1 we see that the solution operator E(t,s) (0 < s < t < +00) for (Es) is well defined as a multivalued operator from D(tps) into D(ipl) which assigns to each Uo 6 D(tps) the set E(t, s)uo given by (1.3). Also, it is easy to check the following evolution properties (El) and (E2): ( E l ) E(s, s) =1 (the identity) on D(
as t -* +00.
(3.1)
l
Moreover, note that {
(3.2)
Then, by assumptions (A5), (3.1) and (3.2), as s —>• +00 the limiting system of (E„) is of the form: (Eoo)
i/(t) + 0 p o o ( i ; ( t ) ) + f l o > ( t ) ) 9 / " i n # ,
t > 0,
where g°° := g(oo,-). By (Eoo^o) we denote the Cauchy problem for (Eoo) associated with initial value v0.
139 The limiting autonomous system (E m ) can be considered as a special case of (Es) with
s(t)v0 := L e Dip*)
Theie
!\i f t r " o f ( E r ; S o n R+
w v ^ ' such that u(0) = v0 and ti(i) = z. Clearly, the following properties are satisfied: (51) S(0) = 7 o n % ° ° ) . (52) S{t + s)z = S{t)S{s)z, \/z € D{
Moreover, we can easily show the closedness of S(-) and E (•){•) in the following sense. Lemma 3.1. (i) Assume that tn,t G R+ with tn —> £, ^on, v0 6 D{ip°°) with v0n —> D0 in i? and an element zn 6 S(£„)i;on converges to some element z in H as n —> +oo. Then, z G S(t)vo. (M) Assume £fta£ s n , s, £„, t G -R+ tuitfj s„ —> s and i„ —>• i, won £ D((ps"), u0 G D(ips) with ^on -^ u0 in H and an element zn G S(i n + s n , sn)«on converges to some element z in H as n -¥ +oo. 27»en, z G £(£ + s, s)w0- J" particular, if s = +oo, tten z G S(£)itoNext, to construct a global attractor for the non-autonomous system (E s ), we need the concept of w-limit sets under E(t, s) or S(t). Definition 3.1. Let B(H) be a family of bounded subsets of H. Then, for each B € B(H) the sets w s ( B ) : = n U E(t + s,s)(D(^jnB) T>Ot>T,!>0
and
__=^___ us(B):=n\JS(t){D(
are called respectively the w-limit sets of B under E(t, s) and S(t). Remark 3.1. (cf. [5, Lemma 1]) By the definition of the w-limit set U>E(B), it is easy to see that x € COE(B) if and only if there exist sequences {£„} C R+ with tn —> +oo, {s n } C R+, {zn} C B with zn G £)(3s«) and {z n } C # with i „ G U(t„ + s„, s„)z„ such that xn —>• a; in if as n —> +oo. So far as the construction of a global attractor for {S(t)} is concerned, we can do it by modifying the standard technique established in [2] and [9]. The key is the following lemma: Lemma 3.2. (cf. [3, Lemma 4.1]) There exists a compact and convex subset BQ O/-D(
140
and for each bounded set B e B(H) there is a finite time Tg > 0 satisfying S(t) (D(ip°°) n f l j c B o
for all
t>TB.
By the above lemma, we can construct a global attractor for {S(t)}. In fact, the global attractor A o of {£>(*)} is given by Ax. = Us{B0). For a detail proof, see [9]. Here, by a global attractor A o of {£(*)} we mean that (i) A o is a non-empty and compact subset of H; (ii) for each bounded set B e B(H), distH(S(t)z, Aoo) —• 0
uniformly i n z e D{ip°°) n B
as t —• oo; (iii) Aoo = S(t)Ao for all t > 0. Now, let us mention our main result in this paper. Theorem 3.1. Let {>'} e $({a r }, {br}) and assume that (A1)-(A6) and (3.1) hold. Then the set A:= U wE(B) (3.3) BeB(H)
satisfies that (i) At C Aoo and At is non-empty and compact in H; (ii) for each bounded set B G B(H), distn{E{t, s)z, At) —> 0
uniformly in s > 0 and z € D(ips) n B
as t ^r +co; (iii) A C S ( i ) A /or any t e R+. In our proof of Theorem 3.1 one of the important steps is to show the following lemma on the relationship between w-limit sets under E(t,s) and an absorbing set B0 for S(t). Lemma 3.3. Let Ba be the compact absorbing set for {S(t);t e R+} obtained in Lemma 3.2. Then, uE{B) C B0 for each B € B(H). In the proof of Lemma 3.3 the time dependence condition (*) on {?'} is essentially used, and the proof is a modification of that of [3, Lemma 4.1] to the multivalued case. Proof of Theorem 3.1: We give a sketch of the proof. By Lemma 3.3, we observe that At C Bo- From this relation together with assertion (iii) proved below we have (i). Also, (ii) of Theorem 3.1 follows from the definition (3.3) of A and Remark 3.1 in a way similar to that of the proof of [3, Theorem 5.1].
141
Next, let us show (iii) of Theorem 3.1. By Lemma 3.1, we easily see that for each XCB0, S(t)X C S(t)X,
Vt G R+.
(3.4)
Now, let x be any element of A*. By the definition of A,, there exist sequences {Bn} C B(H) and {xn} C H with xn G U>E{B„) such that x„ —>• a: in H
as n —> +oo.
(3.5)
It follows from Remark 3.1 that for each n, there exist sequences {tn,j} C R+ with tn,j -> +oo, { s n j } C R+, {znj} C B„ with znj G D(ips"J) and {«Jnj} C if with u„j G E(tnj + snj, snj)zn<j such that »„j —• a;n in H
as j —> +oo.
Let t be any time in R+. We note that
for j with tnJ- > t + 1. Hence there is a wnj G E(tnj + s„j — i, snj)znj
such that
Here by the global estimates obtained in Theorem 2.1, {w n j e ff ; j 6 iV with ^nj > £ + 1} is relatively compact in if, so that we may assume that the sequence wnj converges to some element wn G H as j —> +oo. Clearly, tun G LOE(Bn). Therefore from (ii) of Lemma 3.1 we observe that
xn e S{t)wn c S{t)uE(Bn), which implies that xn G (J S(t)u)B{Bn),
Vn > 1.
n>l
On account of (3.4) and (3.5), we see that x G [J S(t)cuE(Bn) = S(t) | J ujB(Bn) C S(t) (J uE{Bn) C S{t)A,. n>l
n>l
n>l
Hence A, is semi-invariant under S(t), namely A, C S(t)A, for all £ G R+.
Q
Theorem 3.1 says that the attracting set At is semi-invariant under S(t) associated with the limiting autonomous system (Eoo), in general. We now give a sufficient condition in order that A, is invariant under S(t). Definition 3.2. Let z G D((f°°). Then, we say that S(t)z is regularly approximated by E(t + s, s) as s —> +00, if for each finite T > 0 there are sequences {sn} C R+ with sn —> +00 and {zn} C H with zn G D((pSn) and zn —» z in i i satisfying the following property: for any function u G W l,2 (0, T; ii) satisfying u(t) G 5(i)z for all t G [0, T] there is a sequence {«„} C W 1,2 (0, T; ii) such that un(t) e,E(t + sn, sn)zn for all t G [0, T] and u n —> u in C([0, T]; i/) a s n - > +00.
142 Now, let us mention a result on the invariance of A, under S(t). Theorem 3.2 Assume that for any z of A*, S(t)z is regularly approximated by E(t + s, s) as s —> +00. Then, A, = S(t)A* C Aoo
for all T 6 R+.
(3.6)
Moreover, if S(t)z is regularly approximated by E(t+s, s) as s —• +00 for all z e A^, then A, = Aoo. (3.7) Proof. By Theorem 3.1, the attracting set At is semi-invariant under S(t). So we have only to show that S(t)A* C A* for any t 6 R+. Now let y e S(t)A». Then, by (iii) of Theorem 3.1 we see that y e S{t)A. C S{t + T)A»
for all r > 0,
which shows that there are sequences {T„} with T„ —> +00 and {xn} c A, such that y€S(t
+ Tn)xn.
Now by our assumption we observe that for each n there are sequences {s n j} C R+, {xn,j) C H and {ynj} C H such that snj->+00,
xnj- e D(
xn in H
and )xn,j,
Vn,j —• 2/
in -ff
as j —> +00. Therefore, by the usual diagonal argument, we can find a subsequence {jn} of {j} such that xn := aJnj„> Vn '•— yn,j„ a n d s n := snjn satisfy \xn - XH\H < - , n
Vn € £ ( i + T„ + s„, s n )x„,
\yn - y\H < n
for every n = 1, 2,.... Since {£„} is bounded in # , say {£„} C i? for some B £ B(H), the above fact implies (cf. Remark 3.1) that y e uE{B) c -A». Thus 5(f).4„ C A*. Now here, we have At = S(t)A* for all t € i?+. Therefore it follows from the invariance of A* and ./"loo under S(t) that A, = S(t)A, C S(t)Aoo = Ax> /or aH t e JR+. Thus (3.6) holds. Assertion (3.7) is proved by a slight modification of that of (3.6).
We say in this paper that the set A*, having the properties in Theorem 3.1, is a global attractor for {E(t, s)}.
4
Examples
In order to illustrate our theorems we give two simple examples of non-autonomous flows in the one dimensional space H = R, which show that .4, ^ Aoo in general. For a given interval [a,b] we denote by I[a,b]{-) the indicator function on [a,b].
143 Example 1. We consider a family {ip1} of proper l.s.c. convex functions on R and a function g(t, u) on R defined as follows: f /[!,!](•) '(•) := | /[!_!](•) [ /[o,i](•)
for 0 < t < 2, for 2 < t < +oo for t = +oo
and g(t, u) := —\/u
for 0 < u < 1.
It is easily checked that they satisfy all the conditions in section 2. Now, consider the non-autonomous and autonomous evolution equations (Es) with / = 0 and (Eco) with f°° = 0, and the corresponding non-autonomous and autonomous systems E(t, s) and S(t), respectively. Also, let A* and A*, be the global attractors associated with them. Then, observing that E(t + s, s)u 0 = min{l, -At + c)2} with c = 2^j% e [-, 1] for s > 2, S(t)v0 = min{l, -(t + c)2} with c = 2 0 ^ G (0,1] and S(i)0 is the set [0,1] for all sufficient large t > 0, we have At = {1} and Aao = [0,1]. By the way, in this example, S(t)0 is not regularly approximated by E(t+s, s) as s —> +oo. Example 2. We next consider {?'} and g(t,u) as follows: /[_!,!)(•)
for
0 < t < 2,
! /[_!,!](•) for 2 < t < +oo, -y/u7[o,i](-) for 0 < fort t < +oo, = +co0 < u < 1, u 1 for 0 < t < 2, - - < u < 0, g(t,u) 2 - - ' 2 ~ u 1 for 2 < t < oo, < u < 0. Just as in Example 1, they satisfy all the conditions in section 2, and we can generate the non-autonomous and autonomous systems E(t, s) and S(t) associated with the evolution equations (Es) with / = 0 and (Soo) with /°° = 0, respectively. In this case it is easy to see that A, = A^, = [0,1] and for any z 6 [0,1], S(t)z is regularly approximated by E(t + s, s) as s —> +oo. It seems difficult to give any example showing that A* / S(t)A* in such simple situations as above.
and
References [1] H. Brezis, Operateurs Maximaux Monotones et Semi-Groupes de Contractions dans les Espaces de Hilbert, North-Holland, Amsterdam, 1973.
144
[2] J. K. Hale, Asymptotic Behavior of Dissipative Systems, Mathematical Surveys and Monographs 25, Amer. Math. Soc, Providence, R. I., 1988. [3] A. Ito, N. Yamazaki and N. Kenmochi, Attractors of nonlinear evolution systems generated by time-dependent subdifferentials in Hilbert spaces, pp. 327-349, Dynamical Systems and Differential Equations, Missouri 1996 Volume 1, ed. W. Chen and S. Hu, Southwest Missouri State University, 1998. [4] A. V. Kapustian and J. Valero, Attractors of multivalued semiflows generated by differential inclusions and their approximations, Abstract and Applied Anal., 5(2000), 33-46. [5] V. S. Mel'nik and J. Valero, On global attractors of multivalued semiprocesses and nonautonomous evolution inclusions, Set-Valued Anal., 8(2000), 375-403. [6] U. Mosco, Convergence of convex sets and of solutions variational inequalities, Advances Math. 3(1969), 510-585. [7] M. Otani, Nonmonotone perturbations for nonlinear parabolic equations associated with subdifferential operators, Cauchy problems, J. Differential Equations, 46(1982), 268-299. [8] K. Shirakawa, A. Ito, N. Yamazaki and N. Kenmochi, Asymptotic stability for evolution equations governed by subdifferentials, pp. 287-310, in Recent Developments in Domain Decomposition Methods and Flow Problems, GAKUTO Intern. Ser. Math. Appl., vol 11, Gakkotosho, Tokyo, 1998. [9] R. Temam, Infinite Dimensional Dynamical Systems in Mechanics and Physics, Springer-Verlag, Berlin, 1988. [10] N. Yamazaki, A. Ito and N. Kenmochi, Global attractor of time-dependent double obstacle problems, pp. 288-301, Functional Analysis and Global Analysis, ed. T. Sunada and P. W. Sy, Springer-Verlag, Singapore, 1997.
Quasiconvex extreme points of convex sets M a r t i n Kruzik C e n t e r of A d v a n c e d E u r o p e a n Studies a n d Research, Friedensplatz 16, 53 111 Bonn, G e r m a n y * E m a i l : [email protected] and I n s t i t u t e of I n f o r m a t i o n T h e o r y a n d A u t o m a t i o n , A c a d e m y of Sciences of t h e Czech Republic, P o d v o d a r e n s k o u vezi 4, 182 08 P r a h a 8, Czech Republic. E m a i l : [email protected]
Abstract Zhang [17] showed that if the quasiconvex hull of a compact set in K m x " is convex then also its rank-1 convex hull is convex. In this note we show that a reason for that is in a special structure of quasiconvex extreme points of compact convex sets. In particular, we show that compact convex sets are lamination convex hulls of their quasiconvex extreme points. We also give a clear geometric characterization of quasiconvex extreme points of compact convex sets. Finally, we show that quasiconvex exposed points of convex compact sets coincide with their lamination convex extreme.
1
Introduction
It was proved in [8, 16] that generalized Krein-Milman type theorems ([7]) hold for compact quasiconvex and rank-1 convex sets in TRmxn (the Euclidean space of real matrices m x n). In particular there exist minimal generators of those sets called quasiconvex and rank-1 convex extreme points, respectively. An example in [8] shows that quasiconvex and rank-1 convex extreme points differ in a general situation. In this short paper we give a simple geometric description of quasiconvex extreme points on compact convex sets. More precisely, we show that quasiconvex extreme points on compact convex sets are those points which cannot be written as a convex combination of two mutually different rank-1 connected matrices. In other words, for convex compact sets the notion of lamination convex extreme points coincides with that of rank-1 and quasiconvex extreme points. This gives a concrete description of this, generally a very abstract, notion. * address for correspondence
145
146 Further, we get that convex compact sets are generated by the sets of their quasiconvex extreme points via sequential lamination. An example in [8] shows that this does not hold for nonconvex sets, however. Moreover, our result implies that if the quasiconvex hull is convex then also the rank-1 convex hull must be convex, the fact which was proved in [17]. Finally, we show that recently defined quasiconvex exposed points ([18]) coincide with quasiconvex extreme points in case of compact convex sets. Proofs of our results heavily rely on separation theorems for convex sets (see e.g. [14]) and on Krein-Milman type theorems for quasiconvex and rank-1 convex sets. We say that a function / : IR mx " ->• IR is quasiconvex if for any ip G W1-°°(]Rn; IRm), (0, l)"-periodic, and any A G IR mx " f(A)<[
f(A + VV(x))dx
.
(1)
Quasiconvexity plays a crucial role in the calculus of variations. Namely, the sequential weak* lower semicontinuity of / : W1-00^; IR") -> IR, I(u) = / n /(Vu(i)) dx, Q C IR" a bounded domain is equivalent to the quasiconvexity of / : M m x n -» H, see [1, 10, 11]. Further we say that / : R m x " -> IR is rank-1 convex if for any A, B G H m x " , rank(A - B) < 1 and any 0 < A < 1 f(XA + (1 - X)B) < Xf(A) + (1 - X)f(B) .
(2)
Rank-1 convexity is a necessary condition for quasiconvexity but it is not a sufficient condition if m > 3 and n > 2 as shown in [15]. If min(m, n) = 1 then both the notions are equivalent to each other and the case m — 2, n > 2 still remains open. Let if be a compact set in ]R mxn . Besides the convex hull C(K) we define the quasiconvex hull and rank-1 convex hull Q{K) and R{K) of K, respectively, as follows (see e.g. [12]) Q{K) := {A G H m x n ; f{A) < s u p / V/ : R. mx " -> IR quasiconvex } ,
(3)
K
R(K) := {A G IR mxn ; f(A) < s u p / V/ : H m x " ->• E. rank-1 convex } .
(4)
K
Quasiconvex hulls play an important role in the theory of martensitic transformations. Namely, the quasiconvex hull of the zero set of a material stored energy density is the set of stress free configurations. The rank-1 convex hull is an inner approximation of this set. Finally, we define the so-called lamination convex hull L{K) of K as L(K) := [J L{(K), ielNu{0}
where L0(K) := K and Li{K) := {XA + (1 - X)B; 0 < A < 1 , rank(A -B)
= l,A,Be
L^K)}
, i e IN .
Clearly, we have L(K) C R(K) C Q(K) and contrary to quasiconvex and rank-1 convex hulls the lamination convex hull is relatively easy to construct. The notion of quasiconvexity is closely related to gradient Young measures. Let Q C IR" be a bounded domain and K C IR mx " compact. It is known ([3, 6]) that for any
147 sequence {uk}ke^ C W1'°°(ft;Irlm) such that, for almost all x e Q, Vuk(x) 6 K there exists its subsequence (here denoted by the same way) and a family of probability measures {vx}x&i, supported on K such that for any continuous function v : K —> ]R and any geLl(Q) lim / v(Vuk(x))g(x) k-HxJn
dx = / / v(A)i>JdA)g(x) dx .
(5)
JSIJK
The family of probability measures {vx}xm f° r which the above limit passage holds is called a gradient Young measure generated by {Vw^jtejN If {vx}xm is independent of x we call such a measure a homogeneous gradient Young measure. It follows from the analysis by Kinderlehrer and Pedregal [6] who found an explicit characterization of gradient Young measures that a probability measure v supported on a compact set K is a homogeneous gradient Young measure if and only if for any quasiconvex / : M m x " —> IR
/ (JK Au(dA)j < jK }{A)u{dA) .
(6)
A homogeneous gradient Young measures which satisfies (6) even for all rank-one convex functions will be called a homogeneous laminate ; cf. [13]. It is well known that the quasiconvex (rank-1 convex) hull of a compact set K c Mm>
2
Properties of convex compact sets
From now on we suppose that K has more than one point. It holds for a compact set K c IR mx " such that Q(K) = K that K = Q{Kqfi), cf. [16], and, in general, L{Kqfi) C K, strictly. Surprisingly, if K is convex we have the following result.
148 Proposition 2.1. Let K c R m x " be a compact convex set. Then K = L ^m
j^(Kqye).
Proof. We proceed by induction. If dim K = 1 and Aff K has rank-1 connections then the assertion clearly holds due to the Krein-Milman theorem because Ke C Kqfi; cf. [16]. If Aff K has no rank-1 connections then K = ifg,e by [4, Th. 4.1] and K = Li(K). Suppose that the assertion holds for sets of the dimension less than d and that dim K = d. Take p £ dK and let H be a supporting hyperplane to K through p. KnH is compact, convex and its dimension is less than d. By the induction hypothesis KnH = L ^ ^ j ^ n jj((Kf] if),, e ). But if A G {K n H)q>e c K n H then A G if,,e because K is in one of the half spaces determined by H. Thus p E K n H C i ( j j m ft n ^-(if,, e ) C L d _i(if,, e )- If Aff A" has rank-1 connections then Lrf(if) = Li(dK). Therefore K C Li(I/
•
The proposition above shows how convex sets are generated by means of their quasiconvex extreme points and it answers a question raised in [16] in this special case. Corollary 2.2. Let K C R m x " be compact, dim K = d and such that Q(K) is convex. Then Ld{Kq,e) = Q{K). Proof. We have Q(if) = Ld{Q(K)qfi) Q(if),, e c if,,e; see [16]. '
because <2(if) is convex On the other hand, •
The following result has been first obtained by Zhang [17] and a new proof has recently been found in [5]. We get it here as a simple consequence of Corollary 2.2. Corollary 2.3. Let K c IR mx " be compact, dim K = d and such that Q{K) is convex. Then Ld(K) = Q{K). Proof. We have L d (if,, e ) C L d (if) c Q(K). But due to Corollary 2.2 Z,d(if„,c) = Q(if) and therefore Lrf(if) = Q(if). ' • We can give a simple geometric description of quasiconvex extreme points of convex sets. Proposition 2.4. Let K c M m x n be a compact convex set. Then Kqfi = Kre = Kifi. Therefore, A £ if is a quasiconvex extreme point of K if and only if there is no interval in K with distinct rank-1 connected endpoints and with A as an inner point. Proof. As in general Kqe C ifr,e C if|,e we must only show that K^e C if,jC. We proceed by induction. If dim if = 1 and Aff if has rank-1 connections then the assertion clearly holds. If Aff if has no rank-1 connections then if = if,je by [4, Th. 4.1] and the assertion holds again. Suppose that the assertion holds for sets of the dimension less than d and that dim K = d. If Aff if has no rank-1 connections then the argument is the same as for the dimension one, so suppose that Aff if has rank-1 connections. Take A G ifj,e. Then A must be in dK and let H be a supporting hyperplane to if through A. KnH is
149 convex, compact and its dimension is less than d. Moreover, by the induction hypothesis and lamination extremality A € (K n H)i,e C (K n H)qfi. If z/ is a homogeneous gradient Young measure supported on K with the first moment A it must be supported on KCiH as K entirely stays in one of the half spaces defined by H and we have that v = 5A, i.e., A e Kqfi. This yields K\fi c if,ie.
• There is a geometric description of quasiconvex extreme points by means of quasiconvex exposed points given in [18]. Recall that argmax M / := {A e M; f(A) = maxBgM f(B)} for a compact set M and a continuous function / defined on M. Definition 2.5. (see [18]) Let K c Wnxn be quasiconvex. A e K is called a level k quasiconvex exposed point of K if there is a sequence of compact subsets {lfj}*_0 such that K0 := K and K^ := {A} and quasiconvex functions {/i}*=1 such that Ki+l := {Ae K; Ae argmaxK.fi+1}
,i = 0,...,k-l
.
The set of all quasiconvex exposed points of finite levels is denoted by Kq>ex. Zhang showed that for a compact set K the set of level 2 quasiconvex exposed points is dense in Kq<e. It is not clear whether in general Kqje = Kq<ex. If we restrict ourselves to functions which are quasiconvex only on K, i.e., which satisfy the Jensen inequality (6) for all gradient Young measures supported on K then it is really true that Kqfi = Kqfix. Namely, as Kqfi is the Silov boundary of K, cf. [2, 8], we easily see that each point from Kq^e is even a level 1 quasiconvex exposed point. In general, for convex sets we can prove the following. Proposition 2.6. Let K c M m x n be compact and convex. Then Kq<e = Kqex = Kifi. Proof. It suffices to prove the first equality. We again proceed by induction. If dim K = 1 then if K has rank-1 connections it has only two extreme points and the assertion follows from the theorem [18, Th. 1.1] by Zhang. If K has no rank-1 connections then we can translate it such that Aff K is a subspace without rank-1 matrices and the result again follows by Zhang [18, Th. 1.3]. Suppose that dim K = d and that the assertion holds for sets up to the dimension d — 1. If K has no rank-1 connections we are done again; cf. [18, Th. 1.3], as all point in K are quasiconvex exposed points. Otherwise if A 6 Kq<e it must be at the boundary of K. Take a supporting hyperplane H through A to K. Then A e (H fl K)q<e and by the induction hypothesis A € (H C\ K)q>ex. By the definition above it means that there is a sequence of sets M 0 := (H n K),... M^ := {A} and a sequence of quasiconvex functions / i , . . . . / * such that Mi+1 = {P G Kti /i+i(P) = maxfi+l{B),B e #;}. But H n K = argmax^/, where / is an affine function such that H = {x; f(x) = c} and K is in the region where / < c. Therefore, if we define the sequence of sets N0 := K, Ni = M;_i, i = 1 , . . . , k + 1 and a sequence of functions v{ := f, vt = / i + i , we get that A is quasiconvex exposed point because the definition above is satisfied for {iV,} and {u;}. Q
150 Corollary 2.7. Let K C H m x n be compact. Then {Kq>e n C(K)„,e) C K,,CT. In particular, Ke c i f , ^ as Ke C {Kqfi n C^if),,,,). Proof. By the previous proposition C(K)qfi = C(K)qfix and, clearly, .K^nC^-fiT),^ c Remark 2.8. The assertion that K,>e = KqfiX for convex sets is implicitly contained in [16, Th. 1.5]. The fact that Ke C Kqfix has been pointed out in [18]. Note that here we extended definition 2.5 to nonquasiconvex sets. Acknowledgment: Partial support by the grant A 107 5005 (Grant Agency of the Academy of Sciences of the Czech Republic) is gratefully acknowledged.
References [1] E. ACERBI, N. Fusco, Semicontinuity problems in the calculus of variations. Arch. Rat. Mech. Anal. 86 (1986), 125-145. [2] E.M. ALFSEN, Compact convex sets and boundary integrals. Springer, Berlin, 1971. [3] J.M. BALL, A version of the fundamental theorem for Young measures. In: PDEs and Continuum Models of Phase Transition. (M.Rascle, D.Serre, M.Slemrod, eds.) Lecture Notes in Physics 344, Springer, Berlin, (1989), pp. 207-215. [4] K. BHATTACHARYA, N . B . FIROOZYE, R.D. JAMES, R.V. KOHN, Restriction on
microstructure. Proc. Roy. Soc. Edinburgh 124A (1994), 843-878. [5] G. DOLZMANN, B . KIRCHHEIM, J. KRISTENSEN, Conditions for equality of hulls
in the calculus of variations. Arch. Rat. Mech. Anal. 154 (2000), 93-100. [6] D. KINDERLEHRER, P . PEDREGAL, Characterizations of Young measures generated by gradients. Arch. Rat. Mech. Anal. 115 (1991), 329-365. [7] M. KREIN, D. MILMAN, On extreme points of regularly convex sets. Studia Math. 9 (1940), 133-138. [8] M. KRUZI'K, Bauer's maximum principle and hulls of sets. Calc. Var. and PDEs 11 (2000), 321-332. [9] J. MATOUSEK, P . PLECHAC, On functional separately convex hulls. Discrete Cornput. Geom. 19 (1998), 105-130. [10] C.B. MORREY, J R . , Quasi-convexity and the lower semicontinuity of multiple integrals. Pacific J. Math. 2 (1952), 25-53. [11] C.B. MORREY, J R . , Multiple Integrals in the Calculus of Variations. Springer, Berlin, 1966. [12] S. MULLER, Variational models for microstructure and phase transitions. Lecture Notes of the Max-Planck-Institute No. 2, Leipzig, 1998. [13] P . PEDREGAL, Laminates and microstructure. Europ. J. Appl. Math. 4 (1993), 121149.
151 [14] R.T. ROCKAFELLAR, Convex Analysis. 2nd ed., Princeton, New Jersey, 1972. [15] V. SvERAK, Rank-one convexity does not imply quasiconvexity. Proc. Roy. Soc. Edinburgh 120 (1992), 185-189. [16] K. ZHANG, On the structure of quasiconvex hulls. Ann Inst. H. Poincare, Ann. nonlin. 15 (1998), 663-686. [17] K. ZHANG, On various semiconvex hulls in the calculus of variations. Calc. Var. and PDEs 6 (1998), 143-160. [18] K. ZHANG, On the quasiconvex exposed points. ESAIM: Control Optim., Calc. Var. 6 (2001), 1-19.
The Richards equation for the modeling of a nuclear waste repository Olivier Lafitte and Christophe Le Potier CEA/DM2S, Centre d'Etudes de Saclay 91191 Gif sur Yvette Cedex, France E-mail : [email protected] Abstract We study a nonlinear degenerate parabolic equation (called the generalized Richards equation) modeling the water pressure in a nuclear waste repository. We show that this degenerate parabolic-elliptic equation has a unique solution, when considering inhomogeneous Dirichlet boundary conditions induced by the surrounding geological material. We describe this solution (as in [2]) as the limit of a piecewise linear function. When studying a more general model taking into account the mechanical behavior of the material, we show that the new equation on the water pressure h is a new Richards equation, but no longer degenerate. We deduce that any model that will take into account the mechanical behavior will be more regular than the Richards equation.
1
Introduction
Modeling t h e saturation processes in a high level waste deep geological repository is a crucial problem in t h e nuclear industry. This situation corresponds to t h e evolution of t h e water flow in a saturated-instaturated medium. T h e equation used in this set-up is t h e generalized Richards equation : dt{9(h))
= C{h)dth
= div(K{h)Wh),
(1)
which comes from t h e generalized Darcy's model U dt9{h) + div{U)
= =
-K(h)Vh, 0,
( ) (
'
where h stands for t h e water pressure in t h e medium, 9(h) is t h e saturation and K(h) is t h e permeability coefficient, see for example [8], [6], [4] for t h e justification of this model. This model does not take into account t h e mechanical behavior of t h e material, which is t h e aim of t h e third section of this paper.
152
153 The region h > 0 corresponds to the saturated region, where the function 9(h) is constant, equal to 6S. Hence, since C(h) = &, C(h) is equal to 0 in the saturated region and the elliptic-parabolic equation is degenerated. The first section contains classical results. We show, applying an easy extension of the results of Benilan and Wittbold [2] and that of Alt and Luckhaus [1], that the equation (1) with the initial Cauchy condition h(x,0) = hi(x), and the boundary conditions h(0) = h0, h(l) = hi, has a unique solution h(x,t) for (x,t) e [0,1] x K + , and that, for this unique solution, the quantity Q(x,t) = 8(h(x,t)) can be computed as the limit of a sequence (0„ (x, t))n, where Q„ is the solution of an elhptic problem. We give the method which builds the sequence hn(x,t) from the elliptic equation and we deduce 0„(x,£) = 8(hn(x, t)). It is interesting to notice that the scheme used by the CEA to solve numerically this generalized Richards equation is the same as the one used for the construction of the sequence hn. 1.00 .80 -SO
-to
.00 -.20 -.ao
-.80 -1.00 .00
.50
1.00
Function h ( x , t > , t - j * 1 0 0 0 a , j - 0 . , 6 , b l | « ) - 0 . B m
1.50
2.00
2.50
3.00
| 0 . , 1 . 5 a [ , b l U l — 0.8 on ]1.5m,3n)
Figure 1: Evolution of h„ (x,nAt).
2
Uniqueness of t h e solution of t h e generalized Richards equation
We assume that h0{0) = M0) and h0(l) = hj(l). We have : T h e o r e m 2.1 Assume that 9(hi(.)) belongs to ^([0,1]) andht belongs to W2'1^, 1). The equation (1) with the above Dirichlet and Cauchy conditions has a unique solution h (t, x) in W1'™ {[0,oo};I?{0,l)).
154 The sketch of the proof is presented in [7]. We introduce : h0(x) = h0 + (hi - h0)x. Under the assumption K(h) > C > 0, let us introduce Kirchoff's transform F of K : rv+tm(x)
F(x,v)=
/
K(s)ds
The function v \-> F(x,v) is strictly increasing. We denote by R(x,
- h0).
Hence the equation (1) is equivalent to the following equation on v(x,t) = h(x,t) — h0(x) : dt(0(v(x,t)
+ h0(x))) = d2x2(F(x,v(x,t)))
+ (hi -
h0)-^(K{h0(x)))
and v satisfies the homogeneous Cauchy condition and the initial condition : v(x,0) = hi(x) - h0(x). We first show that the equation : -^2(F(x,v(x)))
~ (h - h0)^(K(h0(x)))
= /(*),
(3)
has a unique solution for / G i^QO, 1]) and that / H-> V is increasing and continuous from L 1 to W2'1. We thus introduce, for u e L^O, 1), the set A u =
[ ~^2{F{x,v)) X \
- (In - h0)±[K(h0(x))], v e W^ n Wo1'1, 1 u = 9(h0(x) + v(x)). J
For / in Au, we check that there exists v such that 9(ho(x) + v(x)) = u(x) and
f(x) = - ^ ( ^ ( * . « ) ) + dx(K(h0(x))). This partial differential equation on v has a unique solution satisfying the homogeneous Dirichlet boundary condition. Hence v constructed by the procedure is unique. This proves the invertibility of A. The accretivity of A comes from the monotonicity of the operator / — i > v. We thus prove the invertibility and accretivity of the multi-valued operator A (in the sense described by Crandall and Ligget [3]). For all / , there exists (through [3]) a unique (through Lemma 2.2) u in L x (0,1) such that / belongs to u + eAu, e>0. The main tool is the following Lemma. L e m m a 2.2 For v0 G W0' , v'0 G BV, the following assertions are equivalent :
155 • existence of v in L°°([0, oo[; W2-1 n W01,2) such that 9{h0(x) + v(x,t)) belongs to iy1'°°([0, oo[; L1) and h0(x) + v(x,t) is a weak solution of (1), with the initial condition : v{x, 0) = vQ (x) (that is 9(h0(x) + v(x, 0)) = 0(fc/(x)) = 0i{x)), • existence of a solution u in W1,oo([0,oo[;Z/1) such that 0 € u{t) + Au(t), a.e. and u{0)=8(hi{x)). The proof of this Lemma comes from the construction of a sequence approximating the solution. We thus construct the sequence unE{x) = (I + eA)-n{Bi{x))
; u°(x) = Oi(x).
We notice that this construction can be done as follows : • find v\(x,t) solution of 8{vl{x) + h0(x)) - ei{x) •
d2x2 (F (x,vl(x)))
+ (h, - h0)£
(K (ho(x)))
• compute u\ (x) = 9{v\{x) + h0{x)) ; • consider u\ instead of 8i(x) in the first step. For e = 1/iV, we build uN(x,i) uN(x,t)
as the continuous function defined for t e [^, ^ - ] by = u$/N{x) + {Nt -
k)v*%{x).
As A is accretive, (u^(x,t))N converges, when N goes to infinity, to the same limit as (7 + jjA)~ (9i). This limit is denoted by u(x,t). Denote by v(x,t), for almost every t, the unique (through Lemma 2.2) solution of : -d2x,(F(x,v(x,t)))
- (hi - ho)^{K{ho(x)))
= ft«(a:.*)-
As0(u(x,t)+ho(a;)) = u(x,t) (using the definition of the limit), we check that h(x,t) = v(x,t) + ho(x) is a solution of Richards' equation (1). As v(0,t) = v(l,t) = 0, by construction, and v(x,0) = hi(x) — h0(x), h is a strong solution of the equation (1) with h(0,i) = ho, h(l,t) = h\ and the initial condition 9(h(x,Q)) = 9(hi(x)). We thus constructed h as this limit, and it is enough to construct v as the unique solution of (3) to obtain the result of Lemma 2.2. • Note that another initial condition gj(x) such that 9{hi{x)) = 9(gi(x)) leads to the same solution h(x, t) of the problem on ]0, +oo[, through the uniqueness result. We may also notice that the solution of this problem can be constructed in the following way : introducing 0n+1(a;) = F(x,vn+i(x)), we know that >n+1(0) = <j>„+i(l) = 0 and that :
156 -e^+1(x)
+ 8(h0(x) + R(x, >n+1(x))) = 9n(x) + efa - h0) —
(K(h0(x))).
We have thus a fixed-end nonlinear elliptic equation. For each e, we build the sequence (
3
Coupling with t h e mechanical behavior
The Richards equation was well studied by physicists, but it appears that it is not the best equation to model a porous medium which is deformed under the pressure of a liquid. Coupling Daxcy's law and the permeability law with the mechanical behavior of the medium yields interesting results; in particular, the generalized Richards equation obtained with this process is no longer a partial differential equation but can be seen as a pseudo-differential equation, and it is no longer degenerate. Recall that the displacement tensor e is given in terms of the stress tensor a through a linear relation with constant coefficients, which is (i and j denote one of the coordinates) e
u = ^ ( <>ii ~ v zl aii J 2(1 + 1/)
It is then easy to deduce the stress tensor in terms of the displacement (matrix D)
(l + */)(l-2z/) 0ii =
' (l-^eu + ^^T/r, ) ,
(1+ !/)(!-2«/) (2 " " J 6 *
When we have air and water pressure in the medium (pa and pw) as well as other forces 6, the linear relation between the stress tensor, the water pressure pw and the deformation tensor e is
157 da = D(de - hsd(pa - pa)), 1 (l\ where hs = — I * • We use the fundamental law of dynamics : H
° \ 1 /
div (dtcr) = -dtb, dk , dk : dlj to obtain the system on e and h = pw —pa. Using now e« = -^— and e;,- = —— + -—?-, we OTj era., oxi obtain t h e following equation for t h e displacement (lx, ly, lz):
+ Lwdtpw = -dtb -
hsdtVpa,
where KT is a matrix differential elliptic operator of order two and Lw is a vector field acting on dtpw- We use the relation between the porosity and the mechanical displacement (lx, ly, lz), given by u) = dxlx + dyly + djj. The total system is thus ' KTdtl + Lwdth dt(8(h))
= =
dt~b, ft(wS(/i))
=
S(h)dtui +
=
div{K(h)Vh),
u—oth ^
8tu = ftTt(e) = MB{x,V)dtl = dt(djx + dyly + dzlz). Note in this system that the function S(h) is the function 9(h) of the first part, the notation considered depends on the physical problem studied. We have two ways of rewriting the problem: • The first approach uses the pseudo-differential calculus because the matricial linear operator leads to a matricial symbol, and the calculus of K^1, where KT is elliptic, is easier in this set-up. This system leads to the equation:
"IK ~ S^MB(<X'
v
= div(K(h)Vh)
-
) ^ F l L " ) dth
,5)
S(h)MB(x,V)K^dt~b.
In general, the operator R = UJ^ — S(h)MB(x, V)K^ 1LU1 is a pseudo-differential operator of order 0. When the Lame coefficients depend on the position (which may be the case of a complicated geological structure), the pseudo-differential calculus easily gives the leading order term of this zeroth-order operator. 'Note that it is not exactly the porosity as physically introduced, because in tid we have u = gTre and we omitted, for simplicity, the coefficient \ in the porosity given here.
158
The second more classical approach relies on the particular form of the operators and of the constitutive relation of the material. In fact, considering the three equations of the fundamental law of dynamics, and taking the divergence, we easily obtain
A(
J l ^ f c , -
{l + u)(l-2u)
E dth {l-2v)Hs
which implies that the function (1+J)7112l/)^ai hence the equality
_
+ dAPa
+ 9tdiv(6) = 0,
(i-L)H,^tn ^ o e s
no
^ depend on h,
dt{e{h)) = s{h)dtw + Lo~dth an becomes, when d — 3
ww-l-f + irr^)**-^ Note that this equality becomes dt{6{h)) = L^f + because we have
D
1 0 0 0
Ht(1_„)ff(fe)]
dth+F when d = 2,
/ 1 \ 1 {l +
E v){\-2v)
0 0 0
d=2
v°/ D
1 1 0 0
w
E \-2v
/ 1 \ 1 1 0 0
d = 3.
Even in the region where S'(h) = 0 (which was the region where the Richards equation was degenerate when the mechanical behavior was omitted) the coefficient of dh in the equation (5) is greater than a constant equal to -=— s~£7~> where Ssat is the constant E (1 — v) Hs value of the function S(h) in the saturated region, equal to the minimum of S(h). We note that the Richards equation without the mechanical behavior is u)—dth = div(i;sr(/i)V/i) + F, where a; is a constant. Hence the coefficient of dth in this equation vanishes in the saturated region, because S(h) is constant. Hence, when the mechanical behavior is considered, u> is non longer a constant as we saw and the elliptic-parabolic equation is ' dS 1 1+v . dth = div (K (h) V/i) + G,
159 where G is a source term. Introduce the function 0(h) such that :
"W-i+ ££>>• As ©' > Co, the function © is strictly monotone. Hence we can consider U = 0(h) and h = 0 _ 1 ((7), with (0 _ 1 )'(/t) < 1/C0 and the equation (5) writes dtU = d&(B(U)) + G, which is a non-linear non degenerate elliptic parabolic equation. A similar idea was used by Le Potier [6] in the numerical computation of the solution of the coupled problem, splitting the term S(h)dtu> into two terms, one being positive and
dth (
leading to a non-degenerate Richards equation. Precisely, he writes dte — —
l
\
1 I =
D ldt<J. Hence using to = Tr(e) and dtcr = -MB (z,V) K^dtib), he obtains a new system, where the modified Richards equation has a "mechanical" source term and a non vanishing coefficient for the term dth, which is w— + —S(h). It is still a coupled system ah H3 (because a depends on h and is still in the equation for h), but the coefficient of dth is positive, and a usual numerical method for the heat equation is convenient in most cases.
References [1] H.W.Alt, S. Luckhaus, Quasilinear elliptic-parabolic differential equations. Math. Z. 183 (1983), 311-341. [2] P. Benilan, P. Wittbold, Sur un modele parabolique-elliptique. Math. Mod. Num. Anal. M2AN 33(1) (1999), 121-127. [3] M.G. Crandall, T. Liggett, Generation of semigroups of nonlinear transformations on general Banach spaces. Amer. J. Math. 93 (1971), 265-298. [4] P.S. Huyakorn, G. F. Pinder, Computational methods in Subsurface Flow. Academic Press, New York, 1983. [5] J. Kacur, Solution of some free boundary problems by relaxation schemes. SIAM J. Numer. Anal. 36(1) (1999), 290-316. [6] C. Le Potier, F. Cany, Couplage hydromecanique et transferts de gaz dans les milieux poreux. Rapport DMT/SEMT/MTMS/RT/00-115+ du 5/12/2000. [7] O. Lafitte, C. Le Potier, Quelques remarques sur le couplage du transfert hydrique et du comportement mecanique en milieu poreux. Rapport DM2S/SFME/MTMS/01005Mars 2001. [8] G. de Marshy, Quantitative Hydrogeology. Academic Press, New York, 1986.
Trace identities and universal estimates for eigenvalues of linear pencils* Michael Levitin Department of Mathematics, Heriot-Watt University Riccarton Edinburgh EH14 4AS, U. K. Leonid Parnovski Department of Mathematics, University College London Gower Street, London WC1E 6BT, U. K. Email : [email protected] ; [email protected]
Abstract We describe the method of constructing the spectral trace identities and the estimates of eigenvalue gaps for the linear self-adjoint operator pencils A — XB.
1
Introduction
Since 1950's, mathematicians have devoted a lot of efforts to the construction of the so-called universal eigenvalue estimates for elliptic boundary value problems. As an illustration, we recall the original result of Payne, Polya and Weinberger [5], who have shown in 1956 that if {A.,} is the set of (positive) eigenvalues of the Dirichlet boundary value problem for the Laplacian in a domain fi C M*1, then A
m
Am+i — Am < — y \j, ran '—'
(i)
3=1
for each m = 1,2,.... The term universal estimate is used in this context since (1) does not involve any information on the domain Q apart from the dimension n. The Payne-Polya-Weinberger inequality has been significantly improved over the course of several decades; similar universal estimates have been also obtained in the spectral problems for operators other then the Euclidean Dirichlet Laplacian (or Schrodinger "The research of M.L. was supported by the EPSRC grant GR/M20990. Both authors were also partially supported by the EPSRC Spectral Theory Network
160
161 operator), e.g. higher order differential operators in R", operators on manifolds, systems like Lame system of elasticity, etc. We refer the reader to the important paper [3] and the recent survey [1] for more details. In the recent paper [4], we have shown that the majority of the universal eigenvalue estimates known so far can be easily obtained from a certain operator trace identity which is valid not only for boundary value problems for PDEs but, under minimal conditions, for a general self-adjoint operator acting in a Hilbert space. Namely, we have proved the following Theorem 1.1 Let H and G be self-adjoint operators such that G(DH) C DH. Let \j and 4>j be eigenvalues and eigenvectors of H. Then for each j :
E KIM&M£ k
Xk X
,
.l[WM,^{M 2
~ i
= 5>*-Ai)|(Gfc,&>|2.
(2)
k
This result implies Theorem 1.2 Under conditions of Theorem 1.1, m
m
-(A m+1 - Am) £ ( [ [ # , G ] , G ] ^ ) < 2 ^ | | [ t f , G ] ^ | | 2 . 3=1
(3)
3=1
The inequalities (3) may be viewed as a family of estimates of the spectral gap A m+ i — Am for the operator H, with different estimates being obtained for each choice of the operator G. The particular choice of G depends, of course, on the problem, and cannot be prescribed. In [4], we give numerous examples of concrete estimates for boundary value problems for elliptic PDEs (and systems of PDEs) obtained using Theorem 1.2. In particular, the classical estimate (1) follows from Theorem 1.2 by taking G to be an operator of multiplication by the coordinate X[, summing the resulting equalities over I, and using some elementary bounds. The aim of this paper is to extend Theorems 1.1 and 1.2 to the spectra of linear self-adjoint operator pencils of the type A — \B. The statement of the problem and the main results are collected in Section 2; the proofs are in Section 3.
2
Statement of t h e problem and main results
We consider a linear operator pencil V{\) = A-XB,
(4)
acting in a Hilbert space H, equipped with the scalar product (.,.) and the corresponding norm ||.||. Here, A and B are self-adjoint operators such that DA C DB- We identify the domain of the pencil D-p with DA- By the spectrum of V we understand the set of
162 complex values A for which V(X) is not invertible. Throughout this paper we assume, for simplicity, that the spectrum consists of isolated real eigenvalues Ai < A2 < . . . (counted with multiplicity) which may accumulate only to +00, and that the system of corresponding eigenfunctions Uj such that V(\J)UJ = 0, is complete in H (the case of a pencil with a continuous spectrum can be treated as well, cf. Remark 2.5 in [4]). We refer to [2] for a general discussion of the spectral theory of operator pencils. An important property, which we shall use later on, is the fact that, under a technical condition (BUJ,UJ) ^ 0 (which is satisfied automatically if, e.g., B > 0), the eigenfunctions can be assumed to be normalised by the relations (Buj,uk) = Sjk.
(5)
Indeed, multiplying the equality V{\j)Uj = 0 by uk in H, and using the fact that A and B are self-adjoint, we get {Auj,uk) = (uj,Auk) = \j (Buj,uk)
= Afc
(uhBuk),
which implies (5). By completeness, any element / e K can be expanded in a convergent series / =
^{Bf,uk)uk, k
and for any
f,g£H,
{Bf,g) =
Y,{Bf:Uk)-{uk,Bg). k
Thus, by the density argument,
(f,g) = J2(f>u*)-(u*>B9)-
(6)
k
In what follows, [H, G] denotes the standard commutator HG — GH. Our main result is the following spectral trace identities for 'P(A). Theorem 2.1 Let V(X) = A — \B be a self-adjoint linear operator pencil with discrete spectrum \j and eigenfunctions Uj. Let G be an auxiliary self-adjoint operator with domain DQ such that G(D-p) C D-p C DQ. Then for each j
pmXi^U^=-lm^GhG]u3,u]}.
(7)
and £ ( A * - \,)\ (BGUj,uk)
| 2 = - i <[[7>(A,-),G] ,]«,•,«,•> •
(8)
163 Remark 2.2 Instead of the condition G(D(A)) C D(A) we can impose weaker conditions GUJ e D(A), G2Uj e D(A), j = 1,.... Moreover, the latter condition can be dropped if the double commutator is understood in the weak sense, i.e., if the right-hand side of (7) and (8) is replaced by ([V(X),G]UJ,GUJ) (see (13) below). The trace identities (8) imply the following universal estimate, which generalises the Payne-Polya-Weinberger inequality for linear operator pencils. Theorem 2.3 Under conditions of Theorem 2.1 m
771
-(Am+i-Am)^([p(A3),G],GK,U,)<2^||B-1[P(A3),GK.||2. J=l
3
(9)
3=1
Proof of t h e main results
We start with establishing (8). Obviously, we have [A - XjB, G) UJ = (A - \jB)Guj.
(10)
Therefore, ([A - XjB, G] Uj,GUj)
= {(A - XjB) GUJ,GUJ) .
(11)
Since A, B and G are self-adjoint, we have, using (6) {{A - XjB) GUJ,GUJ)
=
] T ({A - XjB) GUj,uk)
(uk,BGUJ)
k
= Y,^Gui^A~XiB">Uk^u>"BGu^ =
(12)
2
^(Afc-A^IBGuj.u*! . k
Using the fact that [A - XjB,G] is skew-adjoint, the left-hand side of (11) can be rewritten as (G [A - XjB, G] Uj,Uj)
= - {{[A - XjB, G], G] Uj,Uj) + {[A - XjB, G] GUj,Uj) = - ({{A - XjB,G],G]UJ,UJ) + (UJ, [A - XjB,G]Uj),
so (G[A - XJB,G]UJ,UJ)
= - - {[[A~XjB,G],G]uj,Uj)
(13)
(notice that (G [A - XjB, G] Uj,Uj) is real, see (11) and (12)). This proves (8). Since (10) implies {[A - XjB,G]Uj,uk) this also proves (7).
= (GUJ, [A - XjB,G]uk)
= {Xk - Xj)
{BGuj,uk),
164 We now proceed to the proof of (9). Let us sum the equations (7) over j = 1, ...,m. Then we have
^ g w^rf...£r(1)Wi%i„,
(14)
Parceval's equality implies that the left-hand side of (14) is not greater than
"m+1
—
Am
=1
This proves (9).
References [1] M.S. Ashbaugh, Isoperimetric and universal inequalities for eigenvalues, in Spectral theory and geometry (Edinburgh, 1998), E.B. Davies and Yu Safarov, eds., London Math. Soc. Lecture Notes, vol. 273, Cambridge Univ. Press, Cambridge, 1999, pp. 95-139. [2] I.C. Gohberg, M.G. Krein, Introduction to the Theory of Linear Nons elf adjoint Operators. Transl. Math. Monographs, vol. 18, AMS, Providence, 1969. [3] E.M. Harrell II, J. Stubbe, On trace identities and universal eigenvalue estimates for some partial differential operators. Trans. Amer. Math. Soc. 349 (1997), 2037-2055. [4] M. Levitin, L. Parnovski, Commutators, spectral trace identities, and universal estimates for eigenvalues. J. Funct. Anal, (to appear), preprint at http://www.ma.hw.ac.uk/~levitin/research.html. [5] L.E. Payne, G. Polya, H.F. Weinberger, On the ratio of consecutive eigenvalues. J. Math. Phys. 35(1956), 289-298.
Universal estimates for the blow-up rate in a semilinear heat equation Julia M A T O S « , Philippe SOUPLET (2-3) W Departement de Mathematiques, Universite d'Evry Val d'Essonne Boulevard des Coquibus, 91025 Evry Cedex, France (2)
(3)
1
Departement de Mathematiques, INSSET, Universite de Picardie, 02109 St-Quentin, France Laboratoire de Mathematiques Appliquees, UMR CNRS 7641, Universite de Versailles, 45 avenue des Etats-Unis, 78302 Versailles, France Email : [email protected] ; [email protected]
Introduction
We consider the semilinear heat equation ut = Au + \u\P~\ u (x, 0) = u0 (x)
0<(
ieR"
, . >
(
where p > 1 and u0 e L°°(RN). Let u be the unique classical solution of (1) which is denned on a maximal time interval [0, T). This solution u is bounded on [0, T"] x R ^ , for all 0 < T" < T. When T < oo, we have ||u(t)||oo —> oo as i —> T, where || • H^ denotes the norm in L°°(HN), and we say that u blows up in finite time T. Denote ps = ^ | if N > 3 and ps = oo if N = 1,2. It is well-known that if p < ps and u is a nonnegative solution of (1) which blows up in finite time T then there exists a constant C > 0, depending on u, such that
||«(t)lloo
(2)
Giga and Kohn [8] obtained this upper bound on the blow-up rate of u(t) and proved that (2) still holds for sign-changing solutions provided (3N-A)p < 3N+8. Furthermore, with no restriction on p > 1, the same estimate holds (cf. Bricher [3] and Matos [14]) for any nonnegative solution of (1) which is nondecreasing in time and such that its blow-up set is compact. (Recall that a G RN is in the blow-up set if u(tn,xn) —• oo for some tn —» T and xn —> a.) However, it is known that one may have a faster blow-up rate in dimensions N > 11 for sufficiently large values of p > ps (see Herrero and Velazquez [12])-
165
166 On the other hand, for global positive solutions of (1) on a smoothly bounded domain fi C R w , with homogeneous Dirichlet boundary conditions, a universal a priori bound of the form suplluWHoc^C^.p.r), f o r a l l r > 0 , (3) t>T
was recently established by Fila, Souplet and Weissler [4] under the assumption (N— l)p < N + 1, and by Quittner [19] provided iV < 3 and p < p$. The first goal of this note, motivated by (2) and (3), is to establish a global a priori bound for the blow-up rate of nonnegative solutions of (1) which is universal, that is independent of the solution u.
2
Main results
Theorem 2.1 Let p > 1 and e € (0,1). Let w0 G L°°(RN), u0 > 0, and assume that one of the following conditions holds: (i)p
+ %,
(ii) UQ is radially symmetric and nonincreasing as a function of r = \x\, and (N-2)p
+2
ifN<3, ifN>3.
{
'
There exists a constant C = C(p, N, e) > 0, independent ofu, such that if the classical solution u of (1) exists on [0,T) x HN, then lh(t)||oo
eT
(5)
It is unknown whether the upper blow-up rate (2) can fail for all N > 3 and all P > Ps- However it is interesting to note that when N > 3 and p > ps, the universal global blow-up rate (5) can be true for no value of e £ (0,1), even if one restricts to radially symmetric nonincreasing initial data. (Indeed this would imply that Theorem 2.2 below holds for such p - see after that Theorem.) Let us note that Theorem 2.1 holds whenever u exists on [0,T) x R " and that we do not need to assume that u blows up at t = T. Actually, Theorem 2.1 has several interesting consequences, not only related to blowing-up solutions. The first one concerns the decay rate in large time of global nonnegative solutions of (1). Theorem 2.2 Letp > 1 satisfy (4)- Assume thatuo € L°°, u0 > 0, is radially symmetric and nonincreasing as a function ofr = \x\, and that the classical solution u of (1) is global (i.e., T = oo,). Then there exists a constant C = C(p,N) > 0, independent ofu, such that \\u(t) ||oo < C f ^ , t>0. (6)
167 When p < ps and u 0 , Vu 0 satisfy a fast decay condition (namely, / \f(x)\2e^2/4 dx < oo), the estimate (6) was obtained by Kavian [13], with a constant C depending on u. The result of [13] relied on the method of forward self-similar variables. Interestingly, although a result on global solutions, Theorem 2.2 relies on the use of backward self-similar variables (and on some new ideas - see section 3). Later in [21], the second author proved that when p < ps, any global nonnegative solution of (1) (with w0 6 L 2 nL°°) must satisfy lim t _ 00 ||w(t)|| 00 = 0. On the other hand, we see that (6) cannot be true unless p < p$. Indeed, when N > 3 and p > Ps, there exist positive, classical stationary solutions of (1), which are radially symmetric nonincreasing (see e.g. Haraux and Weissler [10]). Also, note that the self-similar solutions to (1) (constructed by [10] for p > 1 4- 2/JV) decay precisely like ||«(*)lloo = C r ^ . For other sufficient conditions ensuring global existence and decay of positive solutions of (1), we refer to the recent paper by Gui, Ni and Wang [9] and to the references therein. In particular it is observed in [9] (see p. 590) that all previous results concerning the decay in time provide rates no slower than t~p^ when p < ps- (Some solutions with slower decay rates are constructed in [9] for N > 11 and sufficiently large values of p > p§.) From the works [13, 10, 9] and the references in [9], it thus seems natural to make the following conjecture: Conjecture 2.3 When p < p$, all global nonnegative classical solutions of (1) decay at least like i" 1 /^" 1 ' as t -> oo. Theorem 2.2 proves this conjecture in dimensions N < 3 in the radial symmetric case. Moreover, the estimate is global on (0, oo), with a universal constant. As a consequence of Theorem 2.2, we obtain a new kind of parabolic Liouville type Theorem, concerning solutions of (1) that are defined globally on (—00,00). Corollary 2.4 Let p > 1 and let u be a global nonnegative classical solution of ut = Au + up,
— 00 < t < 00,
x € RN,
Assume that (4) holds and that for all t, u(t,.) is radially symmetric and nonincreasing as a function of r — \x\. Then u = 0. Furthermore, we partially improve a parabolic Liouville Theorem of Merle and Zaag, concerning solutions of (1) that are defined globally in the past (see Corollary 1.6 in [16]). Corollary 2.5 Assume p > 1, T < 00 and let u be a global nonnegative solution of ut = Au + up,
- 0 0
i6RN,
with u(t,.) £ L°° for all t < T. Moreover, assume that one of the following conditions holds: (i) p < 1 + | ,
168 (ii) for all t
K
Then u = 0 or there exists T0 > T such that u(t,x) = (p-l)-i/(p-D.
= K(T0 — t ) _ 1 ^ _ 1 \ where
In Corollary 1.6 of [16], the same conclusion was obtained under the additional assumption that there exists C > 0 such that u(t,x)
- o o
leR",
But instead of (i) or (ii), it was only assumed that (N — 2)p < N + 2. We prove Corollary 2.5 as a consequence of Theorem 2.1 and of [16, Corollary 1.6]. Finally, we obtain an estimate for the initial blow-up rates of local solutions to ut = Au + up. Corollary 2.6 Assume p > 1, 0 < T < oo and let u be a nonnegative classical solution
of
ut = Au + uF,
0
ie
RN,
with u(t,.) e L°° for all t < T. Moreover, assume that (i) or (ii) in Corollary 2.5 is satisfied. Then \Ht)\\oo < Ct~&, 0 0. Results on initial blow-up rates for equation (1) were first derived by Bidaut-Veron [2]. Namely, it was proved there that (7) holds under the condition p < N(N + 2)/(N — l) 2 , without radial symmetry restriction. Note that this condition is better than ours when N > 4, worse when N = 2,3 and identical when N = 1. The method of [2] is completely different from ours and quite involved (relying on Bernstein type gradient estimates, Aronson-Serrin Harnack inequalities and multiplier arguments). Also for the CauchyDirichlet problem on a bounded domain, initial blow-up rates up to the boundary are studied by Quittner and the second author in [20]. Remark 2.7
a. Theorem 2.1 fails for e = 0.
b. It is well-known that all blowing-up solutions of (1) satisfy the following universal global lower bound:
H^lloo^er-i)"^ 1 , o
(8)
where K = (p - 1 ) - V ( P - I ) .
c. Assuming p > 1 and (N — 2)p < N + 2, and w0 £ i? 1 (R i v ), Merle and Zaag [17] derived the following uniform estimates of order one for the blow-up rate of a
169 nonnegative solution u of (1): for all e > 0, there exists t0 = t0(e) € (0,T) such that K
^(T-t)7blWu(t)\^^K+{^
+ £)l]og{T_t)l
forallte[t0,T).
N_K
The constant ^2 appearing in the term of order one is optimal and t 0 (e) depends on the H2 norm of the initial data UQ- However this estimate only holds on a neighbourhood of the blow-up time T ofu(t), whereas estimate (5) is global in the existence time interval of the solution. f
d. It seems that when p < 1 + 2/iV, the estimate (5) can be obtained by the methods of [1], where, among other things, the authors prove upper blow-up rates for a reactiondiffusion system which generalizes equation (1). The arguments of [1] are rather involved and rely on a Moser's iteration scheme. Our approach here is completely different and simpler, especially in the case p < 1 + 2/N. Throughout this note, C denotes various positive constants which depend only on the indicated arguments. In the next sections, we present the main tools and ideas used in the proofs of Theorems 2.1 and 2.2. For the detailed proofs the reader is referred to [15].
3
Main tools: self-similar variables and convolution Lebesgue spaces
The study of the blow-up rate of u(t) is done by using the method of backward self-similar variables, introduced by Giga and Kohn [7] and Galaktionov and Posashkov [5]. For each a G R/^, we rescale u by similarity variables around {T,a) by setting S=
"l0g(T-t)'
wa{s,y) =
y= x
W=t
(9)
(T-t)^u(t,x).
This function w = wa is defined in (s 0 , +oo) x R w , with s 0 = — logT, and satisfies
w, = Aw-\yVw
w (so,y)
=
+ \w\!'~1w-f^,
s>s ,
0 w 2/eR ,
jeR",
,1Q.
where
170 Let us briefly sketch our basic ideas to establish (5). We start from the simple fact that, for each a £ R N , the rescaled nonnegative global solution wa of (10) has to satisfy /
!"«(«, y)p(y) dy < (p- \)~7^ = K for all s > s0,
with p(y) = (47r)-T e - V4 .
Since wa(s, y) = WQ{S, y + ae as follows
S//2
(11)
), this can be restated in terms of convolution product
H/o* w0(s)||oo = sup / w0(s,r])p(r] — aes'2)dn < K for all s > s0. o e R « JR."
(12)
This estimate leads us to consider the problem (10) in convolution Lebesgue spaces. Definition 3.1 For all 1 < q < oo and any function f e L]oc(RN), we set \\f\\lP = IIP* l/l'll^* = f sup j where || • ||, denotes the norm in Lq = Lq(RN). LqpA(RN) is then defined by
\}{y)\"P{a - y) dy)
*,
(13)
The convolution Lebesgue space L9pA =
Ll* = {feLlc(RN)--\\P*\f\qC<™}These are Banach spaces with the norm \\-\\*p- Also, for q = oo, we define L^
(14) =
L°°(RW).
The central part of our arguments is then the analysis of the smoothing properties of the linear problem zs — Az - \y • Vz in (0,+oo) x R w , . . z(0,y) =
171
4
Linear and nonlinear smoothing effect in convolution Lebesgue spaces
Let A be the operator defined in the Hilbert space L2 by V {A) = {/ e L2p : V / G [iff A(f)=Af-\yVf,
,
Af-^y-Vf£L2
in the distributional sense! ,
feV(A),
where L2p is the weighted Lebesgue space defined on the usual way corresponding to the Borel measure p(y)dy. The operator —A is self-adjoint with dense domain in L2 and it is the generator of a semigroup of contractions on that space which we denote by (5(s))s>oIt is easily proved that (5(S)^)(2/) = ( T ( l - e - ^ ) ( e - ^ ) i
0 e L
J,
yeR",
where T(t) = (Gt)* denotes the linear heat semigroup in RN and G t (x) = ( 4 7 r i ) - ? e - ] ^ ,
t > 0.
(16)
Note that d = p. We prove that the linear semigroup (S(s)) s > 0 denned in L2 admits a uniquely, densely defined extension to L* still denoted by (<S(s))s>o. In particular, if <> / £ L}pik and <j> > 0, then S(s)cj) > 0. Moreover, for each 1 < q < oo, (S(s))s>o restricts to a semigroup of contractions on LqPii,. Our main result in the linear theory is a smoothing effect of the semigroup («S(s))s>0 on convolution Lebesgue spaces L* . Theorem 4.1 . Let <j> e L*^, tuifft 1 < g < oo. Then S(s)<j> 6 L ^ /or all q < r < oo, and ||S( S M|;„ < (1 - e - s ) - ^ ( i - » ||0||; ip , /or a// s > 0. We should point out that this estimate does not hold if we replace the Lqpik norms by L? norms (see e.g. [11]). Next, we are concerned with the nonlinear problem (10) (where we assume that so = 0) with initial data <j> e L9 , 1 < q < oo. Note that, if w is a classical solution of (10) (that is, w E C1-2{{0,oo) x R>) and w e L°°{{0,T),L°°(RN)) for each T > 0), then it solves the integral equation w(s)=S(s)
J'S{s
- T) (\W{T)\P-^T)
- ^Yj
dr,
(17)
for all s > 0. An essential ingredient to the proof of our main result Theorem 2.1 and its consequences is the following smoothing property for the equation (10) in L9 spaces.
172 Theorem 4.2 . Let qc = N(p — l)/2 and let q > 1 satisfy qc < q < oo. Let M > 0. For any T > 0, any <j> € L°° with \\<j)\\qtP < M, and any solution w e L°°((Q,T),L':o) of (17), it holds
IMs)IU < C8-%M\\*qiP,
0 < 8 <min(T, Sl ),
where C > 0 depends only on N, p, q and Si > 0 depends only on N, p, q, M. The proof of Theorem 4.2 relies on the linear estimates of Theorem 4.1. Actually, from these estimates, we can also derive a local existence-uniqueness result for problem (10) in LP spaces (in the spirit of Theorem 4 in [4] and based on arguments of [22]).
5
Universal bounds a n d uniform decay rates
The result of Theorem 2.1, case (i) (p < 1 + 2/N) is a direct consequence of the estimate (12), which can be restated in terms of convolution Lebesgue norms as lko(s)Hi,p < « for all s > s 0l and of Theorem 4.2 applied with q = 1 > N(p — l)/2. In view of case (ii) of Theorem 2.1, we also need to derive a uniform a priori bound for the blow-up rate of nonnegative solutions of (1). More precisely, for any (N — 2)p < N + 2, we prove that the constant C in (2) depends only on the L°° norm of the initial data and on an upper bound of the blow-up time T. This result was already obtained in [18] for the vector-valued problem associated to (1) under the condition (3N — 4)p < 3iV + 8 (in the scalar problem no sign restriction on the solution was made). In terms of the rescaled problem (10), we prove the following theorem. Theorem 5.1 . Assume p > 1 and (N - 2)p < N + 2, and let C0 > 0. Let
(18)
The conclusion then follows by combining (18) with Theorem 4.2, applied for q = p > N(p - l)/2, and Theorem 5.1. The proof in the case (N — 2)p < N + 2, N < 3 is inspired from the method of [19] and is somehow more involved. Beside convolution Lebesgue spaces, it relies on rescaling and contradiction arguments and on energy estimates similar to those in [6] and [8].
173 Finally, Theorem 2.2 follows easily from Theorem 2.1. Since the solution u of (1) exists globally on [0, oo) x RN, by applying Theorem 2.1 with e = 1/2 for each T > 0, we deduce that K * ) l | » < C ( J V , p ) ( T - t ) - 1 / < ' ' - 1 \ T/2
For each fixed t0 > 0, applying (19) with T = 2t0 and t — t0 yields
and the result follows.
References D. Andreucci, M. A. Herrero, J. J. L. Velazquez, Liouville theorems and blow up behaviour in semilinear reaction diffusion systems, Ann. Inst. Henri Poincare 14, No. 1 (1997), 1-53. M.-F. Bidaut-Veron, Initial blow-up for the solutions of a semilinear parabolic equation with source term, in: Equations aux derivees partielles et applications, articles dedies a Jacques-Louis Lions. Gauthier-Villars, Paris. (1998) pp. 189-198. S. Bricher, Blow-up behaviour for nonlinearly perturbed semilinear parabolic problems, Proc. Royal Soc. Edinburgh 124A (1994), 947-969. M. Fila, P. Souplet, F. Weissler, Linear and nonlinear heat equations in Lp6 spaces and universal bounds for global solutions, Math. Ann. 320 (2001), 87-113. V. A. Galaktionov, S. A. Posashkov, The equation ut = uxx + u13. Localization, asymptotic behavior of unbounded solutions. Akad. Nauk SSSR Inst. Prikl. Mat. Preprint 1985, n°97, 30 pp (in Russian). Y. Giga, A bound for global solutions of semilinear heat equations, Comm. Math. Phys. 103 (1986), 415-421. Y. Giga, R. Kohn, Asymptotically self-similar blow-up of semilinear heat equations, Comm. Pure Appl. Math. 38 (1985), 297-319. Y. Giga, R. Kohn, Characterizing blowup using similarity variables, Indiana Univ. Math. J. 36 (1987), 1-40. C. Gui, W.-M. Ni, X. Wang, Further study on a nonlinear heat equation, J. Differ. Eq. 169 (2001), 568-613. A. Haraux, F. B. Weissler, Non-uniqueness for a semilinear initial value problem, Indiana Univ. Math. J. 31 (1982), 167-189. M. A. Herrero, J. J. L. Velazquez, Blow up behaviour of one dimensional semilinear parabolic equations, Ann. Inst. H. Poincare, Anal. Nonlin. 10 (1993), No. 2, 131-189.
174 [12] M. A. Herrero, J. J. L. Velazquez, Explosion de solutions des equations paraboliques semilineaires supercritiques, C. R. Acad. Sci. Paris, t. 319 (1994), 141-145. [13] O. Kavian, Remarks on the large time behaviour of a nonlinear diffusion equation, Ann. Inst. H. Poincare, Anal. Nonlin. 4 (1987), 423-452. [14] J. Matos, Convergence of blow up solutions of nonlinear heat equations in the supercritical case, Proc. Royal Soc. Edinburgh 129A (1999), 1197-1227. [15] J. Matos, Ph. Souplet, Universal blowup estimates and decay rates for a semilinear heat equation, Preprint. [16] F. Merle, H. Zaag, Optimal estimates for blowup rate and behavior for nonlinear heat equations, Comm. Pure Appl. Math. 51 (1998), 139-196. [17] F. Merle, H. Zaag, Refined uniform estimates at blow-up and applications for nonlinear heat equations, GAFA, Geom. Funct. Anal. 8 (1998), 1043-1085. [18] F. Merle, H. Zaag, A Liouville theorem for vector-valued nonlinear heat equations, Math. Ann. 316 (2000), 103-137. [19] P. Quittner, Universal bound for global positive solutions of a superlinear parabolic problem, Math. Ann. 320 (2001), 299-305. [20] P. Quittner, Ph. Souplet, Initial blow-up rates for a nonlinear heat equation, to appear. [21] Ph. Souplet, Sur I'asymptotique des solutions globales pour une equation de la chaleur semi-lineaire dans des domaines non bornes, C. R. Acad. Sci. Paris, Serie I 323 (1996), 877-882. [22] F. B. Weissler, Local existence and nonexistence for semilinear parabolic equations in V, Indiana Univ. Math. J. 29 (1980), 79-102.
Asymptotic behavior of curves evolving by forced curvature flows Hirokazu Ninomiya Department of Mathematics and Informatics, Ryukoku University, Seta, Otsu 520-2194 JAPAN. The motion of an interface is one of most important and attractive topics in applied mathematics. Especially, the mean curvature flows have been studied by many researchers. In the two dimensional case, it is called a curve shortening flow. Asymptotic behaviors of curve shortening flows are well-known. If a curve is a simply closed one, it becomes convex eventually and is shrinking to a point (see [7, 8]). If it possesses two asymptotes, Huisken has proved that it converges to a self-similar solution (see [5, 10]). In this paper, we consider how the behavior changes, if we add a constant driving force. First we will define an interface. A pair (T(t),v) is called an interface provided that there exists a family of connected open sets D(t) in RN such that F(i) is a smooth connected boundary of D(t) and v is a outer normal vector on T(t) pointing from D(t) to D(t)c. By the definition, an interface is not allowed to have any self-intersection points. We consider here an interface (T(t),u) which satisfies V =H +k
(1)
where V is a normal velocity in the direction of u, H is the curvature, and k is a given constant. This equation appears in the several fields of applied mathematics. For examples, this system represents the motion of transition layers of the Allen-Cahn equation [6], and the filamentray vortex of the Ginzburg-Landau equation confined in a plane [4], the BZ reaction [12]. In the curve shortening flow (i.e., k = 0 and N = 2), the property of non-existence of self-intersection points preserves in time (see [7, 8]). In the case k ^ 0, unfortunately, some interfaces may possess some self-intersection points eventually, even if the initial interface has none. Actually, in the case k > 0, take the initial interface T(t) and D(t) as in Figure 1, i.e., £>x(0) C D(0) C D2{Q), where
£>!(*) :=
s((°),-Ri(*))U5((_°0),fli(*)).
D2(t) := B((J),fl 2 (i))UB((~ 0 6 ),W) ,
175
176
§1'<%gm VAP*
§L
W
(®?\
y//y/z
IS
Figure 1: Example of an initial interface which will possess self-intersection points.
== -J^t)~k>
i^W
RM = R°,
where B(x, r) is a disk of radius r centered at x. It follows from the comparison principle (see [14, Lemma 4.1]) that Dx{t) C D(t) C D2{t) as long as D(t) possesses no selfintersection points. We can choose positive numbers a, Ri,R% and b as follows: a > R° > -, k kR\ - log \kR\ + 1| > k{a - E?) + log b2>(Rl
+
ka — 1
kRl-
a)2-a\
(e.g. a = Z/k, b = 7/k, Rf( = 2/fc, R\ = 4/fc). It follows from these assumptions that two circles in D?(t) are not extinct and the circles in Di(t) never touch those of D2(t) before two disks in D\(t) collide. Therefore the interface T(t) becomes self-intersection points in finite time. It is natural to think that D(t) possesses two holes if the solution can be extended after the singularity (see [2]). However, one can also consider the self-crossing interfaces (see [1]). It is also remarked that the self-intersection may occurs in even the mean curvature flow (k = 0) in R 3 (see [9]). If the interface is represented by the graph y = u(x,t), the equation (1) is reduced to the following Cauchy problem ut
l + ul
+ fc^/l + u\
i£R,i>0,
(2)
with initial condition u(x, 0) = u0(x)
iGR.
(3)
177 The existence and uniqueness of solutions of this equation is studied in [3, 4] (also see [11, 13]). Note that the maximum principle holds for this equation (cf. [15]). We will give the definition of the traveling front of (1). As an example, we take the line y = xta,n9 with v = t(—sm6,cos6) where 0 < 9 < ir/2. It is easily seen that this interface moves not only with speed k in the direction i/, but also with speed k/cos9 in the direction '(0,1). So, it is natural that the velocity of the traveling front should be specified. The interface (T(t),v) is called a traveling front with velocity v, if r(i) = 17(0) + vt. The above example is regarded as the traveling front with velocity kv or velocity '(0, fc/cos#„). Deckelnick et al in [4] proved the existence of the traveling curved front and studied the stability of the front under some restricted assumptions for «o- The author and Taniguchi relaxed the assumption for the initial data and classified all the traveling fronts in [13, 14]. They proved the following (see [13, Proposition 1.1 and Theorem 1.2]). Theorem 1 Any traveling front of (1) with velocity '(0, c) is one of the three, after appropriate translations, (i) lines y = xta,u9t, and y = —xtan6» (ii) a traveling curved front Tc(t) which possesses two asymptotes y = irrtanft,, (iii) stationary circles with radius l/\k\ only in the case c = 0, where 9„ = arctan(\/c 2 — k2/k). Moreover the explicit form of the traveling curved front Tc(t) with speed c(> k) is given in 1 +
x{9; c)
y(9;c)
cVc 1
k2
log
1/
c+ k 9 r tan -
\j c-
k
2
/c + fc 9 ! " 1 \j/ c-k r tan 2fccos9 — k
•cl0Eb^r
for9e (-0.A)The traveling curved front Tc(t) is "V-shaped", which connects two asymptotes. The existence of this traveling front is also reported in [1,4]. The proof of this theorem consists of two parts. First it was proved that the traveling front with non-zero speed should be an entire graph (see [13, Lemma 2.4]). Next all the traveling front of (2) are classified (see [13, Lemma 2.3] for the details). In the case (ii) where the initial interface possesses two asymptotes, we denote the exterior angle between two asymptotes by 9", which is equal to 7r — 29,, if their inclinations are ±tan#„. It is checked that the speed c of the traveling curved front Tc(t) is uniquely determined by the angle 9* (or 9„), i.e., c = k/ cos 9*. This traveling front can be observed in a liquid BZ reaction (see [12]). The asymptotic stability of the curved traveling front in (2) is discussed in [4, 14].
178 T h e o r e m 2 The traveling curved front is asymptotically stable, if the initial perturbation is restricted to BCl := {v e CX(R)
sup
(|n(o;)| + |vT(x)|) < oo, lim v(x) = 0}.
-0O<X<00
|l|->0O
Furthermore, it is shown in [14] that the interface which have two asymptotes with angle 9* converges to the corresponding traveling curved front as t tends to infinity, provided that 9* < IT. This means that Tc(t) is globally asymptotically stable. If we do not take BCg as the perturbation space, the situation changes. In the class BC1 :={ve
C^fR) |
sup
(\v{x)\ + \vx{x)\) < oo},
—oo<:r
the traveling curved front is not asymptotically stable. Namely, for any e > 0, there exists an interface T(i) of (1) such that
v(t)n (rc(t)- ( ° ) ) ^ 0
dist(r(o),rc(o))<4e, T{t)n (rc(t) + ( ° \\?9,
for all t > 0 (see [14, Theorem 4.1]). This interface oscillates at infinity around Tc(i). The very interesting result [16] by Yagisita should be remarked. He considered nearlyspherically expanding fronts of the Allen-Cahn equations. Though the front is known to converge to the sphere after the rescaling of the radius to unity, he proved that the difference between the front and the expanding sphere does not decay as t tends to infinity. Next we will explain the case where the angle between the two asymptotes is greater than n. Using the method of functional analysis, Deckelnick et al also proved in [4] that, for u0(x) = — |x|tan0*, u(x, t) - tQ ( -
+ 0,
(4)
as t tends to infinity, where \s\ < fcsinft,,
x/F Q(s) :=
\s\ tan#* -
COS0*
\s\ > ksin9t,
(5)
where 9, = (9* — 7r)/2. This result indicates that the top order of the solution is tQ{x/t). We can expect the difference between u and tQ(x/t) does not decay as in [16].
References [1] P . K. BRAZHNIK, Exact solutions for the kinematic model of autowaves in twodimensional excitable media, Physica D, 94, (1996) 205-220. [2] Y.-G. Chen, Y. Giga and S. Goto, Uniqueness and existence of viscosity solutions of generalized mean curvature flow equations, J. Diff. Geom. 33, (1991), 749-786.
179 [3] K.-S. Chou and Y.-C. Kwong, On quasilinear parabolic equations which admit global solutions for initial data with unrestricted growth, Calc. Var. Partial Differential Equations 12-3, (2001), 281-315. [4] K. Deckelnick, C. M. Elliott, and G. Richardson, Long time asymptotics for forced curvature flow with applications to the motion of a superconducting vortex, Nonlinearity 10 (1997), 655-678. [5] K. Ecker, and G. Huisken, Mean curvature evolution of entire graphs, Ann. of Math. 130-3 (1989), 453-471. [6] P. C. Fife, Dynamics of Internal Layers and Diffusive Interfaces (CBMS-NSF Reg. Conf. Ser. Appl. Math. 53), SIAM (1988). [7] M. Gage, and R. S. Hamilton, The heat equation shrinking convex plane curves, J. Diff. Geom. 23 (1986), 69-96. [8] M. A. Grayson, The heat equation shrinks embedded plane curves to round points, J. Diff. Geom. 26 (1987), 285-314. [9] M. A. Grayson, A short note on the evolution of a surface by its mean curvature, Duke Math. J. 58 (1989), 555-558. [10] N. Ishimura, Curvature evolution of plane curves with prescribed opening angle, Bull. Austral. Math. Soc. 52-2 (1995), 287-296. [11] O.A., Ladyzenskaja, V.A. Solonnikov, and N.N. Ural'ceva, Linear and Quasilinear Equations of Parabolic Type, Translations of Mathematical Monographs 24, Providence RI, (1968), American Mathematical Society. [12] V. Perez-Munuzuri, M. Gomez-Gesteira, A. P. Munuzuri, V. A. Davydov, and V. Perez-Villar, V-shaped stable nonspiral patterns, Physical Review E 51-2 (1995), 845-847. [13] H. Ninomiya and M. Taniguchi, Traveling curved fronts of a mean curvature flow with constant driving force, in " Free boundary problems: theory and applications, i" GAKUTO Internat. Ser. Math. Sci. Appl. 13 (2000), 206-221. [14] H. Ninomiya and M. Taniguchi, Stability of traveling curved fronts in a curvature flow with driving force, to appear in Methods and Application of Analysis. [15] M. H. Protter and H. F. Weinberger, Maximum Principles in Differential Equations, (1984) Springer-Verlag. [16] H. Yagisita, Nearly spherically symmetric expanding fronts in a bistable reactiondiffusion equation, J. Dynam. Differential Equations, 13-2 (2001), 323-353.
Existence of non-steady flows of an incompressible, viscous drop of fluid in a frame rotating with finite angular velocity M. Padula and V.A. Solonnikov Department of Mathematics, University of Ferrara, Via Machiavelli 35, 44100, Ferrara, Italy Emails : [email protected] ; [email protected] Abstract We give an existence theorem of global non-steady solutions of the Navier-Stokes equations governing flows of a incompressible viscous fluid totally bounded by free surface. Specifically, we prove that, in the presence of surface tension, if a rigid rotation admits a stable configuration in the sense of capillarity theory, then to any suitably small initial perturbation to the shape and to the velocity field there exists a unique global non-steady solution which eventually goes to the stable rigid rotation as time goes to infinity, with exponential decay rate. Our proof is based on a existence theorem of local regular solutions proved by [13], and on some a priori estimates on the solution, uniform in time for weak norms of the solution in the wake of [7]. The result does not assume smallness on angular velocity of the drop, only smallness on initial data!
1
Introduction
Plateau's work inaugurated in 1863 t h e study of shape and stability of a drop driven by rotation of surrounding liquid. Plateau's drop was pierced by a shaft and immersed in a tank of liquid having almost the same density of the drop. T h e shaft was mounted vertically and turning it, Plateau could bring the drop into rotation. T h e drop is driven to rotate at an imposed angular velocity UJ. W h e n increasing the rate of this rotation, the drop progresses through a sequence of shapes, axisymmetric at first, then ellipsoidal and two lobed, and then a most remarkable toroidal ring t h a t remains intact for a short time. Shapes and stability of isolated, self-gravitating masses rotating freely in space have been much analyzed, e.g. by Maclaurin 1742 (perfect spheroids), Jacobi (1834) (ellipsoids), and t h e equilibrium shapes are analogous to those found by Plateau.
1.1
Results on Equilibrium Shapes
In theory, t h e shape r t of the drop in rigid rotation would have been strictly set by the balance of the capillary force t h a t comes from surface tension of t h e curved drop surface,
180
181 and the gyrostatic pressure force which is exerted on the surface by the rotating liquid owing to ordinary centrifugal force, under the constraint to have constant given volume \ilt\. Mathematically, this balance is expressed, in spherical coordinates, by the functional
G(R,t) = a\rt\-^j
R5{y/ t)
^' ^n2edS1-Po\Qt\,
(1)
where a represents the surface tension, p0 is a Lagrange multiplier and Si is the unit surface. For steady flows : G(R, t) = G(R). The basic flow is a steady flow characterized by a shape f2(,, a gyrostatic pressure pt,, and a uniform rigid rotation with angular velocity cj = u;e3, which is considered as a parameter. The fluid is supposed to be in equilibrium in the frame rotating with the drop, therefore the shapes are also called equilibrium shapes. Several equilibrium shapes of rotating liquid drop are theoretically possible. The equilibrium shape physically realizable is that which renders minimum the potential G. In capillarity theory, the range of "stability" of the different shapes is investigated when LU is varying, and it is shown that rigid rotational stable shapes exist, in correspondence of a finite value of angular speed w, may be they are not unique ! Some interesting analysis can be found, e.g., in [3], [1], [2]. The aim of this paper is to prove that the above said configurations, which are stable in the class of small fluctuations in shape, are also asymptotically stable in the class of non-steady flows with respect to small initial fluctuations in shape and in velocity inside the drop. We quote the papers [4], [6], [7], and [8]- [13] concerning the existence problem for the full free boundary problem. In particular in [7], it is proved, for the first time, existence of an unsteady regular motion of rotating liquid drop, exponentially decaying to a rigid steady configuration corresponding to a given angular momentum. In that paper, it is assumed smallness on the size of angular momentum. Here we intend to remove such a restriction, at least for star-shaped domains. Precisely we prove the following. Theorem 1.1 Let T0 = 3fio be given by equation R = Ro (y) with R(y,0) = Ro(y), y e Slt Ro e C ^ S i ) , a e (0,1), and letv(x,0) = v0{x), with v 0 £ C 2 + a ( n 0 ) . Moreover, let Rf, (w) satisfy the stability hypothesis of capillarity theory, that is Ri, is an isolated local minimum for G (R) at fixed u>, and set v<, = UJ x x. Assume additionally that VQ satisfies the compatibility condition
/
x x v0dx = na
x x vbdx, J nb
and let |vo - v t l c a + a ^ ) + |.Ro - Rb\c*+°(Si) - e>
(2)
with e sufficiently small. Then the fluid drop has a unique solution (R,v,p) defined in an infinite time interval t > 0 and possessing the following properties : R (.,t) € C3+a (Si), v(.,t) € C 2 + a (ft t ), p(.,t) 6 C 1 + Q (fi ( ), Vt > 0, where Tt = dQt is given by equation
182 .R = R(y,t),
y € Si. The solution satisfies the inequality
sup | w t ( . , t ) | c , „ ( n ) + sup | w ( . , t ) | c 2 + „K( n ) + sup te{0,T) te(0,T) ' te(o,T) +
™P ^ (••*) "
ftlc+«(n,)
^
c
\p{.,t)-pb\cl+a(n) l "
(l w o| C 2 + « ( n o ) + |#o - ifc| c »+.(s 1 )) c ~"»
V T G
(0. ° ° l . (3)
where w = v — V&, iwrfi some constants 6 > 0 and c independant
ofT.
We give the proof for star-like configurations. As it will be clear from the proof, our result heavily relies on the assumption that the basic configuration Rb renders minimum the energy functional of capillarity theory. Therefore, in principle, our proof can be enlarged to more general equilibrium configurations, like the toroidal ones.
2
Position of t h e problem
Let a drop of incompressible viscous fluid occupy a star-shaped domain Q. = {y G R 3 : \y\ < R(y/\y\)}, r = 90. We wish to determine the configuration fit C R 3 , of prescribed volume 47r/3, admissible with a steady rigid rotation of angular velocity u;e3.
2.1
Equilibrium results
We remind that the energy functional, given by (1), must assume a minimum in Rb. Denote by Vgj the derivatives along the unit sphere S\, in spherical coordinates, and set g = R2 + \VSlR\ , OR 1 OR Vsj = e 9 — ++ ee,f ,
~de
' '^e'^^
n = (-VSlR,R)
l
y/R* + \VSlR\ 2
Also, it holds : frdS = Js Ry/gdS\. The condition that Rb is a minimum for G implies, in particular, that it must be a stationary point for G, that is, for R = Rb + tp, the first derivative of G(R) with respect to t at t = 0 must vanish, i.e. : 0 = 5G(Rb) = ^-\ =
pR*sin26dSi\
+ -p0
k
=°
Sl
VR-VpildStl h ) =° [ J Si
pR2dS1\
l4
(4)
. =°
Integrating by parts, and owing the arbitrariness of p, we deduce the classical eigenvalue problem in unknown Rb and po : H{Rb)+pb(Rb)+p0
= 0,
(5)
where the double mean curvature of the surface H(R) has the following expression :
H
1„
(VSlR\
2
™=a** (-%-)-v?
(6)
183
Also, weset: K{R) = H(R)+pb{R).
Thus the first variation of: SG(R+tp)\t=0
=
^
is given by : 6G(R) = JSipR2(K.(R) + p0)dSi, and the hypothesis that G assumes a minimum at Rb implies that SG(Rb) must vanish, i.e. :
SG{Rb) = J PRl{K{Rb) + Po)dSl = 0, Vp. Si
Hence we find, at the equilibrium position RflC(Rb) = R2b[H{Rb) +Pb(Rb)] =
-p0Rl
(7)
Since, in the sequel, we need a precise computation of the second variation of G, here we make some exact computations. Set p = R — Rb, we have
f
G(R) - G(Rb)
dG\ dt \Rb+tp
dt.
(8)
J o
Also, we set
d2G
SG(Rb) = ^§\ , 62G(Rb) = ds, 22 |\R . dt \R b
(9)
b
(8) implies 1 G(R)-G(Rb) = Jf fdG\ ( dt \R +tp _dG\ dt \R b
I I I. r (£2. J o
=
dt + 5G{Rb) b
ds dt + 6G{Rb) - p J 2 (d G\ d G\ ]dg 2
Rb+sp
ds2 \Rb
dt + -82G{Rb) + SG(Rb)
2 b) + nl, SG{Rb) + -82G{R
J o V ds
(10) where nl denote nonlinear terms. We now compute the second variation of G at Rb. Set F = R^/R? + \VSlR\2, V e = de, V
+2
/ F)2 F
/ ORdVeR
Sl
{avj?^2
pVeP + 2
,),
Vp,
f)2F
+
d dVeRpVvP
B2 R.
WJT2^2 +
^dp2
+
2
e^RV°pv*p
~ 2Lj2R3b Sln2 9p2 ~ 2poRbp)
dSl
that is a norm for p in W^iSi) if Rb is a minimum for G. In particular, from (10), at the minimum configuration Rb, we also have [6G(R) - SG(Rb)\ = 62G(Rb) + / f ^ ? | - &2G(Rb[ ds. J o _| as \ Rb+sp
184
This yields SG(R)-SG(Rb)
=
/ Q
=
2.2
2
P[R
K.(R) - R2bIC(Rb)]dS1 = S2G{Rb) fill v /
+ nonlinear terms IIPIIM'1(S ) + nonlinear terms.
Equation of motion
Let a drop of incompressible viscous fluid occupy a star-shaped domain Qt = {y € R 3 : \y\ < R(y/\y\)}- We wish to determine the domain Qt C R 3 , the velocity vector field (y,t) and the scalar pressure field p(y,t), satisfying the problem ' v t + (v • V)v - i/V2v + Vp = 0 V•v = 0 Tn(v,p)n =
0, y € fit, t > 0,
w2 aH{R)n + — R2 sin2 6 - pe,
nrRt(y/\y\,t) = vn, R(y/\y\,0) = R0(y/\y\), v|t=o = v 0 (y)
^
y G Q0,
where T(v,p) = —pI+2vS(v), v is the constant kinematic viscosity (the density has been choosen equals one), S is the symmetric part of the gradient of velocity. Furthermore, Yt is the boundary of Qt, n is the exterior normal to Tt, u: is the angular velocity of the tank, and pe is the constant external pressure. Moreover, since the pressure is defined up to a constant we can fix uniquely the pressure by setting pe = —po in (12)3.
2.3
Conservation laws
Let us observe that problem (12) admits the conservation laws of mass, momentum, angular momentum, and energy expressed by
ij^vMv
= 0,
(13)
AC
jtJ
[ y x v ( s , f s = 0,
d
-{T[y](t) + G(R)} + V(t) = 0, where
G(R) = a|rt| - £ j
s
«
TM(t) = f ^ d l V(t) = 2v I 8(v(y,t))dy. J fit
M
sin' 6dSl -
PM
185 Through a suitable choice of measure unity, we fix the origin 0 of the frame at the center of mass of tob, and assume that the center of mass of to0 is still in 0. Furthermore, we fix the initial data such that
lai = l^o| = y , l /
ydy = / vdy = 0, «« JJk v{y,t)dy = / v0(y)dy = 0,
[y x v{y,t)]dy
=
J fit
/
(14)
[yxv0(y)]dy=
J f!0
[yxvb{y)]dy. J f!b
All conservation laws are expressed in terms of the flow v, p, not in terms of the perturbation w = v — V(„ r = p — pb ! All conservation laws are independent of the choice of reference frame where problem is written ! In the frame rotating with angular velocity w, call it Rb, the basic flow reduces to the rest for the velocity field with a mere perturbation of the spherical surface for the shape of the domain !
2.4
Basic flow
The system (12) admits as steady solution the rigid rotationVb(y) = w x y, Pb(y) = \ W.'|2 where \y'\ denotes the distance of y with respect to the 3/3-axis, in the domain Qb = {y : \y\ < Rb}, with Rb solution of the Young-Laplace equation of interface configuration (5).
3
T h e equations of p e r t u r b a t i o n
We now look for the best reference frame where to write the perturbed problem.
3.1
Choice of t h e reference frame
Consider the noninertial frame 1Z rotating with angular velocity o;e3, and set v(j/,t) = w(x, t) + vb(x). In 7?., the fluid has velocity w, and problem (12) is written as follows:
w t + (w • V)w - i/V2w + Vp = V•w Tn(w,p)n nrRt{x/\x\,t) R{x/\x\,0) w| t = 0
= = = = =
V ( — ~ - j - 2toe3 x w, 0 {aH{R)+pe)n, w n Ro{x/\x\), w0(a;) = v 0 (x) - V|,(x)
x € tot, t > 0, (15) leT,, x € too,
where \x'\ denotes the distance of x from the a;3-axis, and the terms — 2ue3 x w, V ( " '% ) represent the Coriolis and the centrifugal force, respectively.
186
3.2
Initial D a t a
Notice t h a t in 71 t h e basic velocity is zero, therefore w is the perturbation t h a t will be assumed initially small. Concerning the shape Qt it will be the perturbation (a slight deformation) of the basic stable shape Q,b (ui), and it will be supposed initially close to fit, say R0(x/\x\) — Rb is small. Concerning t h e initial data, due to t h e conservation laws (13) they must statisfy some conditions. Precisely, we remind t h a t in "R, the center of mass, t h e volume, angular momentum are constants of the motion, see (14). Therefore, we choose initial d a t a as small perturbations of t h e basic flow, paying attention to choose f2p such t h a t it has the same center of mass O (origin of the reference frame) as flb, and same volume f2o = 47r/3. Due to t h e invariance of relations (14) 1 , we have the following relations 47T
L
xdx
/
x d x = 0, O.0
vt(x,t)dx
/
= =
fit
/ [xx w(x,t)]dx J «t
/
w0(x)dx
= 0,
fi§\
J «o — / [xxvb(x)}dx J sib
— / J nt
=
vb(x]r3drdSi.
/ [e r x J SiJ R
[xxvb(x)]dx
In particular from (16)4 if follows t h a t J"n [x x w(a;,t)]d3; < c \\R — RbW^ts y
4
T h e equation of energy
Notice t h a t equations (15) are not yet the perturbation equations, because b o t h p and R denote the pressure and t h e radius of the flow in the rotating frame. However, they are enough to write energy identity. Actually, multiplying (15)i by and integrating over fi( by transport theorem we deduce /
\w\2dx + v [
|S(w)|2dr (17)
= [
w
n{oH(R)
+ Pe)dS
+ f
w •V ( ^ L \
dx.
It is well known t h a t
/ wn(aH{R)+pe)dS=-4-
JTt 1
f dS.
dt JTt
Denoting by at, eit i = 1,2,3 two basis respectively of R and 7£, for any vector x we have Y^i=\ £ , a « : Yli=i ZtkVkZl'ei = S i = i Vk^k- Hence, by the orthogonality of Z we have Jn x x wdx = Ja y x vfdy.
187 Hence from the identity .,2
(17) yields
d_ [r ^j-dx iwi2.
dt Jat
d r ._ w 2 /• R5
+ 1 f dS-1^f ^-sin2edSl at JTt I JSl 5
*
+ v
f | S (w)| 2 dx = 0. Jnt
(18)
We obtain easily a norm for the perturbation p = R-Rb, to the surface Rb = Rb(x/\x\) if we assume that G(R) assumes a local isolated minimum in Rb. Indeed,, for regular function R(x/\x\,t), in a small neighborhood of Rb, it holds G(R)-G(Rb)
= \\R-Rb\\2wUSi).
(19)
Finally, since G(Rb) is constant in time we deduce the following energy equation for the perturbation |
{llwllL(ft) + G(R) - G(Rb)} + H|S(w)|| 2 2 ( n t ) = 0.
(20)
From Korn's inequality, we know that the gradient of w is increased by S(w) plus the square of momentum and angular momentum of w. The momentum is zero, cf. (16)3, while the angular momentum is increased by R — Rb, cf. (16)4. This allows us to state c
ll Vw lli2(nt) ^ l|S(w)|| 2 2(nt) + H-R- RbW^Si)-
(21) Till now we have obtained a classical energy estimate for the perturbation, with a dissipative term only for the deformation of the velocity field (not even for the velocity !). In the next section we shall provide a dissipative term for p = R — Rb and for w.
5
Auxiliary equation
In this section we wish to find a "dissipative" term for R — Rb, to this end we use the method introduced in [5] and already applied to free boundary problems in [6], [7]. We notice that problem (15) can be rewritten as follows w t + (w • V)w - i/V2w + Vr = -2we3 x w, V•w = 0 Tn(w,r)n = [a{H(R) - H{Rb)) +Pb{R) - Pb(Rb)]n, nrRt(x/\x\,t) = w-n
R(y,o) = Mv) w| t = 0
= w 0 (i) —^—
(22) x(zTt,
yeslt
x S fi0,
where r = p-[
x e fit, t > 0,
= V ~ Pb-
188 We solve an auxiliary problem. Find a solenoidal vector function w in W^fij) satisfying the problem V-v(x,t)
llvllwjtn,) + l|0 t v|| i2( no
= 0
x e fit,
< c(||p|| W; i (rj) + ||w • n|| L 2 ( r t ) ).
The compatibility condition is satisfied because the volume is constant. The construction of w can be done by reducing fit to a fixed domain, thus following the method outlined in [7]. Furthermore, we select v to satisfy the additional constraint
/ v-#ote=0,
V * = e* x x, i = 1,2,3. Jn This is possible. Indeed, we choose A = ]T}i=i c'(t)a^(x), with a^ G Cg°(Q.t), and since fit C -02, where Bi is a ball of radius 2, we can also assume ||ai||w2(nt) < 1, J n s^dx = 1. Then, if V satisfies (23), the vector field V = V + rotA, A ' S C£°(fit),'satisfies : J n V • tydx = 0, provided that — / V • * d x = / rotA • *cte = / rofr& • Adx. Jilt
JQt
Jilt
In particular, we have / rot(ei x x) • Adx = —2 / At = —2at = / e; • (x x V')da;, Jnt Jat Jnt which tells us that the coefficients a^t) are bounded by the L2-norm of V , and finally, V is bounded by \& and w in W^fi). Let us multiply (22)i by V, orthogonal to rigid motions, and integrate over fit. Using (23)2, (22)2 we get (w«, V) + ((w • V)w, V) + i/(S(w), VV)
L
V • n(a{H(R) - H(Rb))+Pb(R)
-Pb{Rb))dS
= -2(w x w, V),
(24)
rt where (.,.) denotes the scalar product in L2(fij). We notice that, by transport theorem, it yields (w t , V) + ((w • V ) w , V ) 4 ( w , V) - (V t ,w) - ((w • V ) V , w ) . at Thus, we rewrite (24) in the form ±(w,V)-J
(25)
V-n(a(H(R)-H(Rb))+Pb(R)-Pb(Rb))dS
= -i/(S(w), VV) + ((w • V)v, w) + (v4, w) - 2(w x w, V). Our aim is to increase the right hand side of (26) with terms of the kind either : ll s ( w )lli 2 («t) + H s ( w )lli2(r!oll-R _ ^HwjHsi)) o r nonlinear terms. Now we notice that, by
189
(23)3, embeddings theorems and Poincare's inequality, the first two terms at the right hand side are increased by ||S(w)||| 2 ( n 0 + ||S(w)|| i a ( n t )P - Rb\\w}(Sl), plus nonlinear terms as we wish. The term F(w) = (vj,w) - 2(w x w , V ) , presents some problems since it is increased only by the quadratic term ||Vw||i2(nt)||.R - Rb\\wi(Sly To overcome this difficulty, we remind that w is orthogonal to rigid motions and that the vector field w can be always decomposed into a sum of two vectors w' + h with w' such that ll Vw 'lli 2 (nt) — llS(w)ll!2(fit)> a n d h a r i 8 i d motion. The linearity of F(yv) implies that f ( w ' ) < ||S(w)||| 2 ( n t ) + ||S(w)|| I , 2(nt) ||fl-fl t || w , 2 i (Sl) as we wish. Therefore it remains the term F(h), with h = (f>xx. This will be handled employing the orthogonahty of w to rigid motions. Let R1 be a reference frame rotating with angular velocity —2u> in 1Z. Let us call e; a basis in TZ, and a^ a basis in R'. Poisson's formula implies : ^a^ = —2w x aj. Therefore, since the same vector w has different components V1, W', in the two frames, related by the identity V\t)ei = Wj{t)a.j(t) = V(t), we have
£v-™.-|c*<.M0>-^W-*x<*<.*,M.-| where §^V, |^V = |^V mean the time derivatives of the same vector w in the different frames 1Z. and R', and where ~V = |£V + 2w x V is the derivative in the frame rotating with angular speed 2u>. In this way, we write F(h) = ^ V + 2 ( w x V , h ) = ( ^ V , * x x Since we are in Eulerian coordinates, it holds dr ~
/ (it
f
-
dr
dT f
—V • x xdx at J nt dt dt J n t - [ (V • 4> x x ) w ndS = / V • [("V • <j> x x) w]dx.
ir,
Jn,
^
'
Therefore, F(h) is a nonlinear term. The latter term at left hand side in (24) requires some calculation because we don't want require smallness on the derivatives of Rb. Let us observe that
V • n{K{R) - K(Rb))dS
=
h - / (IC(R) V ' 03 y i s
= - / J Si
^~^(IC(R)
KiR^R^gdS, (27)
- ICiR^dS^
*
Our aim is to prove that (27) is equivalent to the norm of R — Rb in W^iSi). By hypothesis, Rb is a minimum configuration for G(R). Noting that R3 — Rb = (R — Rb){R2 + R2b + RRb), since we can choose p = R - Rb, by (11), (7), we rewrite (27) as follows
190 R3
/
_ P3 r =—*(/C(/Z) - IC{Rb))dSl = /
S l
2
22 T
1 ^ p (fl
fRtRRb
R* + Rl + RRb ftp + Rl3iP +
I. a
2
R2
_L
' , \ , ,-,21--/ r>\
-It-Hf,
+
p{R2K.{R) o2l-/
RltCiR^dS, nn j n
~ l) (#£(*) - RtlC(Rb))dS_ _
(28)
RRb
{Rl-tf)K{Rb)dS,
i ^ „ , f„™„,„.
= l|p||^ 2 i (5l ) + non linear terms +p0 //
2 n.\ p„2fn (.R +, .R 6)
R2
+
R2
+
RR
b ^—
t>1
R\dSi
= P\\pW^i,s •. + non linear terms, where the nonlinear term in p is at least of grade three and can be rendered small for small initial data as we shall see from estimates in the existence theorem. Gathering all informations we deduce - | ( w , V) +/?|M|^ 1 ( S l ) < c (||HI^(ft)IIS(w)|U 2(ni) + ||S(w)||i, ( n t ) ) , (29) where c is also function of p and S(w), because of nonlinear terms. Therefore, adding (29) multiplied by a suitable small constant e to (18) we obtain the inequality
|
{ll w IIL(n t) + G(R) - G(Rb) - e(w, V ) } + H|S(w)||| 2 ( n i ) + eP\\p\\w,(Sl)
(30)
< ce||p||W2i(Sl)l|S(w)|U2(f2t) +ce||S(w)||| 2 ( n t ) , and, for e small enough we get
Jt
£ + H|S(w)||! 2 ( n t ) + eP\\p\\wHSl) < 0,
(31)
with £ = 2 (l|w||i 2(ft[) + G(R) - G(Rb) - e(w, V)) = ||w||£ 2(nt) + | 11^(51)Applying Korn's and Poincare's inequahties to the term H|S(w)||w n > + ^IMIw'fSi)' we finally deduce ^-S + bS < 0, at and by Gronwall's lemma we obtain the desired decay estimate S(t) < e-bt£(0).
(32)
(33)
We remark that (33) has been obtained under regularity assumption on the basic flow, but not under smallness hypotheses on ui !
191
6
Reduction t o initial-boundary value problem in a fixed domain
We intend to prove that if u satisfies the hypothesis of stability required by capillarity theory, then for small perturbations of the rest state there exists a regular global solution to (12) that decays exponentially to zero as time goes at infinity. To this end, in our proof will play an essential role only smallness of perturbation, in this statement the gradient of the basic shape is not infinitesimal ! For this reason the perturbation equation will be linearized around basic flow carefully, say we write the equations in reference frame rotating with angular velocity u> = oie3. Of course, in Tt the velocity w is directly the perturbation to the rest, only we must consider a suitable linearization of Tt around i \ . We remind that the surface r ( is given by the equation |x| = R(x/\x\,t) and the origin O coincides with the center of mass of the liquid, i.e., Jn x^ds = 0, k = 1,2,3. Assume that Ot contains the ball BR^O) = {|x| < R{\ and is contained in R^R^O) = {\x\ < CQR\}, define a smooth function x(\x\) x such that x{\ \) = 0, if |x| < -Ri/2, xd^l) = 1) if W > -Ri. a n < i consider the coordinate transformation
—C 1 • * « © $ - . ) ) - • < • • « .
(34)
or, in spherical coordinates
This transformation is invertible, if R — Rb is not too large, and it maps Qb onto fi( and Tt onto Vb leaving BRl/2 invariant, and in some neighbourhood of I\ it holds : x = j^y. Let J = (I 32 -)
be the Jacobian matrix of the transformation y = y(x, t) inverse to
(34) and J = det J. It is clear that J is an inverse matrix to J - 1 = I -£?•) V
'
whose
V°W/i,j=l,2,3
elements are easily calculated by differentiation of (34). The vector field u = J w / J , with w(j/,t) = -w(x(y,t),t), is solenoidal2 and it holds W(x(y,t),t)
and
=
J(y,t)J-1(y,t)u(y,t)=J(y,t)u(y,t),
192 Hence :
wt = (ju)t - x ( M ) ^ J f f ( y ' JTvH)(J(y,t)u) and the equations (22) take the form J u t + J ( u - X{\y\) o/./L',V(y • J r V,)(J(2/,i)u)
Wlwl)
+ J ( u • V,,) J u - v{3TVy • J T Vj,)Ju + J T V,,r = - w e 3 x j u , V - u ( y , t ) = 0, where r = p — pb, p = R — Rb. The initial conditions are u| (= o = u0{y) = J 0 w 0 ,
p| t = 0 = po(y).
Further, we separate the tangential and the normal components in (22)3 which gives IlbIlS{u, p)n = 0,
-r + vn-S{u,p)n
= K,{R)-]C(Rb),
yeTb,
(35)
where nf = f — n(n • f) is a projection of the vector field f(x), x € F t , onto the tangent plane to r t , (analogously for Hb) and <S(u, p) = J T V ® J u + (J T V g> J u ) T , is a transformed matrix S(w) = V
= =
a((n-nb)-A{t)x + nb-(A(t)-Ab)y)+pb{R)-pb(Rb), anb-{A(t)-Ab)(erp(y/\y\,t)).
We note in particular that the expression rif,- (A:(£) — A 2 (t)) y
depends only on the
first derivatives of Rj and R2, because nb-dy/d^j = 0 (£1, £2 are arbitrary local coordinates on Td)- Therefore, Ti (p), i = 1,2 denote the sums of lower order and nonlinear terms with respect to p, respectively. Let us consider kinematic boundary conditions. The normal n (x(y, t)) to Tt and the normal n^ to ilb are related to each other by
»(*(*')) = | 5
(36)
and in the neighbourhood of Tb the elements of J are given by Rbf-i , td (Rb Jij = -^6} + y'-. 6 +y -R > dy-\~R. so, J = (Rb/R)
. Therefore condition (22)4 : (n • eT)Rt(y,t) n
„
Rl(y/\y\)
„
n
= w • n, is transformed into ,,,,*
193 Our free boundary problem reduces to the problem of finding u(y,t), r(y,t), y e flb, and p{y,t), y e Si, satisfying the relations
J U t + J4U -
. 3TVy)(3(y,t)u)
x{\y\)PMMA{y T
-v(J Vy
T
• 3 Vy)Ju
+ J(U • V , ) J u
T
+ J Vj,r = -2we 3 x Ju, V-u(y,t) = 0 yeflb,t>0, n 6 nS(u, p)n = 0, -r + vnS{u, p)n = K.{R) - K.{Rb), R2 (nb-eT)pt{y,i) = j±u-nb(y,t) y e Si,
u(y, 0) = 3(y, 0)v0(x(y, 0)) = u0(y) y € Qb, p{y, 0) = Ro(y) - Rb{y) = p0{y) y £ Fb, (38) or, which is the same u ( — i/V2u + Vr V u IL,S(u)n[, -r + vnb • S(u)n(, -cmb- Ab{p{y/\y\,t)er) nb • erpt u(2/,0) p(y,o)
= f (u, r, p) — 2o;e3 x u, = 0, = II 6 b(u,p), = = = =
d{u,p) + F{p), u • nb + h(u, p) u 0 (y) po(y)
(39)
2/ e r 6 , t > o, y e Qb, y e Si,
where f(u,g,p)
(I - J)ut + xM^/V'fhy
Mv/\y\)
• 3T^y)(3(y,t)u) - J t u - J(u • V„)Ju
= -2we 3 x (J - l)u + v ( ( J r V „ • J r V „ ) J u - V 2 )uJ + (I - J T )V. b(u, p) = rif,S(u)ni, - ILS(u, p)n, d(u,p) = v(nb-S(u)nb-n-S(u,p)n), T(p) = T1{p)+^2(p), (R? h(u,p) = u - n J - ^ - l (40)
7
Existence theorem
In this section we prove Theorem 1.1. For this we need to consider the linear problem
194 v t - ! / V 2 v + Vp V-v n»S(v)ni -p + vnb • S(v)n 6 -crnb- Ab{p{y/\y\,t)eT) nb • erpt v(y,0) p(y,o)
= f(y,t), = 0, = b(y,t), (41)
= d(y,t), = v • nb + h(y, t) y£Tb,t£ = v0(y) yeUb, = Po(y) yeSu
with some given i{y,t), h(y,t), d(y,t),
(0, T ) ,
h(y,t).
Theorem 7.1 Let f(y,t), d(y,t), h(y,t) be continuous in y, t, f(.,i) 6 Ca(flb), d e C1+a{rd), h e C2+a{Td), Vi e (0,T), T < oo, b e c,1+a-<1+0!>/2(rf> x (0,T)), v 0 e C 2 + a (0 6 ), po e C 3 + a (5i), and let the conditions V-v0 = 0,
n 5 S(vo)n 6 | S l =b(2/,0),
b • n b = 0,
(42)
be satisfied. Then problem (41) has a unique solution p(.,t) e C 3 + a (5i), v(.,£) € C2+a{ttb), p(-,i) £ C1+a(ttb) withvt(;t) e Ca{Qb) definedfort e (0,T), and this solution satisfies the inequality sup (\vt(.,t)\c°(sib) + |v(.,t)|cs+-(n,,) + b(-,*)|cJ+-(nt) + |p(-,<)|c^( S l ) < c(T) (|v0|c2+»(nt) + |po|c3+«(Si) + |b(-,*)lc1+».(1+»>/2(rbx(o,r))) +c(T)sup (|f(.,t)|c. ( n,) + \d(,t)y+°(rb) + |A(-,*)lc+-(rJ where c(T) is a non-decreasing function
(43)
ofT.
Proof. Due to (41)5, the boundary condition (41)4 can be written in the form -p + vnb • S(v)nb - anb • / A 6 e r Jo
v • nb + h(y,
T)
dr = anb • Aberp0(y) + d(y,t),
nb-eT
-p + vnb • S(w)nb - anb • / Abw(y,r)dr
+ / Zx(w)dr
J 0
J 0 A Ab-
= a m • A 6 e r p 0 (y) + d(y, t) + anb • 0
r n
'
dr.
6 ' er
where Ji(v) = nf, • Aber
n 6 • Abv, a.b - e r is the first order differential operator applied to v. The solvability of the initial-boundary value problem
195 v t - i / V 2 v + Vp = f(y,i) V - v = 0, n 6 S(v)n t = b{y,t), - p + vnb • S(v)n 6 -o-ni • / Abv{y,T)dr+ Jo
/ h(v (y,r))dT Jo
=
yeSlb,t€(0,T),
b(y,t)
+ [ B{y,T)dT Jo v(y,0) = v„(y)
yerb,te{0,T), ye Slb, (44)
and the estimate sup {\vrt(;t)\c°(n,,) + \w(;t)\c*+°(nb) + |p(-,t)|c+»(n6)) t
< c(T) (|w0|c2+»(n,,) + |polc3+«(S!) + |ty^*)lci+°.(i+°)/2(rtx(o,r))) (45) +c(T) sup (|f(-,0lo»(n,) + \b(;t)\c^(rb) + |S(-,t)|c-(r t ) ) , t
< c(T) (|w 0 | C 2+«(n 6 ) + |po|c3+=(Si) + |b(-i*)lc1+«.(1+«)/a(rbx(o,T))) +c(T) sup (|f(-,i)| c «(^) + \d{;t)\c^{Tb) + \h(;t)\C2+a(Tb)). Since p satisfies the elliptic equation (41)4 on Si, Schauder's estimate yields the same inequality for |p(-,t)|C3+«(Sl). The theorem is proved. • Now, we turn our attention to the nonlinear problem (39). We start with the estimation of the nonlinear functions (40) (major part of these estimates is obtained in ([7]) Let j(*' = (jjj\y,t)) be the Jacobian matrices of the transformations (34) with R = Rk(y,t),
k = 1,2. The differences J>}' — j\f satisfy the inequalities
W - 7? (1) C H . ( „ b ) I 7(2)
dt(
'3
*J ' c"=+»(nb)
< c\Pl - P2|o*«+-(tu)
k
<
A; = 0 , 1 .
c(\p2t - Pw|c*+>+«(ni) + IPI - P2|c*+i+«(n6))
= 0,1,2,
In particular : \Jij> — ^ij|c*+°(fib)
at^lc^tn,,)
c
^
-
c
lp|c*+»+»(ni,) c c+1+
(^'l "
'*(«i.)
A; = 0 , 1 , 2 , +
p c !+1+<
ll"
"(^))
k = 0 1
>-
The same inequahties hold for the elements of the matrices J(*', ( J ^ ) _ 1 , as well as for the functions j(*> = det J(*> and vector fields n
196 (but of course these vectors are evaluated on Tt). T h e coefficients of t h e difference of the Laplace-Beltrami operators A ^ ' ( t ) — A ( 2 ) (t) on the surfaces |x| = Rk(x/\x\,t) can be estimated in a similar way (see [12]). Omitting lengthy but elementary technical details we present the estimates we are going to use below. L e m m a 7.2 / / Y((u,r,p) ssup|us(-,s)|c«(n(,)+sup|u(-,s)|c.2+»(nl) s
k = 1,2,
then s u p | f ( u 2 , g 2 , P 2 ) - f ( u i , ? i > P i ) | c - ( n b ) + | b ( u 2 , p 2 ) -b(u 1 ,p 1 )| c .i + <,,(i+.)/2 ( r b X ( 0 i t ) ) + s u p | d ( u 2 , p 2 ) - d ( u i , p i ) | c i + « ( r b ) + s u p | / i ( u 2 , p 2 ) - ^(ui,Pi)lc 2 +»(r b ) +
SU
d
P ^ ( - ^ ( P z ) - -^i(Pi))lci+»(r t ) + sup \T2(p2) - ^2(pi)|c2+«(r b ) s
(46)
< criYt(u2 - U ! , r 2 - r1}p2 - pi) + c|p 2( - Pit|ci+°(r b )In particular, ifYt(\i,r,p)
c i+°(r b )
P K u > P)|d+»(r b )
+ ssup|^ 2 (p)|c2+«(r b ) <*
(47)
< cr?y t (u,r,p) + c|p f ci+°(r b )-
-sup | ( v^ ( P )
at
!<*<">
SU
c«(r„)
These estimates and Theorem 7.1 enable us to prove t h a t t h e problem (39) is uniquely solvable in a finite time interval. We restrict ourselves to the case of small initial data. T h e o r e m 7.3 Assume that the datauo g C2+a(Q.b), p € C 3 + a ( 5 i ) satisfy the compatibility conditions V - u 0 = 0, II(,S(uo)n(,|r 6 = n 6 b(u 0 ,po) and that |u 0 |c 2 +»(n b ) + |Po|c3+'"(s1) < E < 1. Then, there exists T = T(e) such that the problem (39) has a unique solution {u{y,t),r{y.t),p{y,t)) such thatue C2+a{nb), ut e C{Slb), r e C1+a{Qb), p 6 C 3 + a ( S i ) for arbitrary t 6 ( 0 , T ) , and YT(u,r,p)
=
sup(|u ( (-,t)|c«(n b ) + |u(-,i)| C 2+c ( « b )
<
C T
+ |P(->*)|C3+"(.91))
( ) (|uo|c 2 +»(n b ) + |po|c3+«(Si)) •
Proof. We write the problem in the form
"
K-,*)l ci+»(n„)
197 u f — vV2u
+ Vr V u rii,S(u)n(, -r + vnb • S ( u ) n 6 -anbAb{p{y/\y\,t)eT)
n;, • e r p f u(?/,0) p(y,o)
= = =
f (u, r, p) — 2cje 3 x u, 0, Ili,b(u,p),
=
d{u,p) + JF^po)
r d_ Txd,T J odt
= = =
(48)
+
Ti{p),
u • ni, + h(u, p) u 0 (y) po{y)
y£Tb,t>0,
and apply the method of successive approximations. We p u t u ' ° ' ( y , t ) = u 0 , r ' 0 ' = 0, p (0) (z/,<) = Po(y), and define ( u ( m + % , i ) y m + 1 > ( i / , i ) , p ( m + 1 % , t ) ) , m = 0 , 1 , . . . as a solution to the linear problem (m+l)
u,
• vV2u(m+1)
+ Vr( m + 1 > v.u(m+i) n 6 s(u( m+1 ')n b _ r (m+l) + ^ . S(u< m + 1 ')n 6 -anh-kb{p^+lKy/\y\,t)eT)
=
f(u( m >, r<m>, p(m>) - 2we 3 x u
=
0)
=
d(u<™\pM) + ^i(po)
= n 6 b(u( m ),p( m )), -/'J^i(p ( m ) )dT + ^2(p("
v.
Q ^(m+1)
n;, • e r p £
„(m+l)
„
— u v T ' • rij, „ ( - + ! % , 0) P(m+1)(y,0)
= =
h(u( m ',p( m > , > u0(y) po(y)
yeTb,t>0, y e fifc,
2/GSi, (49)
By virtue of (43), for arbitrary t G (0,T) : y ^ u W . r 1 ' , ^ 1 ' ) < ci (|u 0 | c »+«(n t ) + |Po|c3+«(Sl)) Let us estimate y ( ( u ( m + 1 ' , r m + 1 ) , P ( m + 1 ) ) assuming that yr(u(*V(A0,P(i;))<<S,
fc
(50)
= l,..,m.
For k = 1 this inequality holds, if Cj£ < 5. T h e differences ,(*+!)
=
„(fc+l)
satisfy the relations
,(*0
p(fc+i)
= r (fc+i)
_rW
£(*+!)
=
p(fc+l)
_ p< ,(*)
198 vf+1)
- i/VM fc+1 ) + Vp(*+1) = f (uO, r « , p « ) - 2ve3 x v ^ -f(u(fc-i)i7.(fc-i)iP(fc-i))i V-v(fc+1> = 0,
iuscv^ 1 ))^ = n6b(u(fc),PW)-ni,b(u(fc-1),p(*-1)),
_p(*+i) + j . n , . S(v(fc+1))n6 -anb
• Ab (V<=+D (lL,t\
er\
= d(u«,p«) - ^u^pC*"1))
n„ • e r d fc+1) - v
\y{k)(;r)\c^b)dr.
J 0
As a result, we arrive at y t (w<* +1 \p t+1 \f<* +1 >)
/"V T (wW,p fc) ^ (fc) )d 1
Hence m+l k=i
+c2 f*xiyt(ww,p*))ew)+ rx)VT(ww,p*),ew)dT
It follows that, if c25 < 1/2, then, by virtue of (49) and of Gronwall's inequality
y((u(™+Vm+1),P(m+1)) < Et^Yt(^k),^\ek)) <
(52)
C{T) (|u0|C2+«(S2t) + |Po|c'+"(Si)) •
Hence, condition (50) is satisfied also for k = m + 1, if c(T)e < 8\. We see that the parameters 5 and e should satisfy the condition : max(ci,c(T))e<(5< —-. 2c2
(53)
In this case we can repeat the above argument and show that (50) holds for all k = 1,2,3, the sequences (u' m ' l r'" l ',p' m ') are convergent in the norm Yr(-> •> •) to the solution of the problem (39) and the solution satisfies (52). The uniqueness of the solution also
199 follows from the inequality (43) applied to the difference of two solutions of the problem (39) (we omit the details). Next, we obtain uniform estimates for the solution of the problem (39) defined in an arbitrary time interval (0,T). To this end, we evaluate the norm Y
toAu,r>P)
=
SU
P
|u s (-,s)|c«(n 6 )+
to — T<S
+
sup
|u(-,s)|C2+„(fi6)
tQ—T<S
sup
|r(-,s)| c i+« ( n i ; ) +
(0— T<S<*0
sup
|p(-,s)| C 3+« (Sl) ,
to — T<S
with a fixed r € (0,1] and with i 0 > 2T. Theorem 7.4 If sup
|u s (-,s)| c =(n 1 ,)+
*0 — T<S
+
sup
|u(-,s)|c2+«(nb)
to~T<S
sup
|r(-,s)| c i+«(n t )
to—r<s
(^A.}
sup
|p(-,s)|C3+«(Sl) < <5,
to—T<s
with some sufficiently small 5 > 0, then the solution of the problem (39) satisfies the inequality y"to,r(u,r,p) < c ( sup ||u(-,s)|| L2( n b )+ sup \\p{-,s)\\wi(Sl) ) , \t1<s
t1<s
(55)
J
where t\ = t0 — 2T, T is a small positive number and c = C(T) is a constant independent
of t0. Proof. We set w =
q = r(x,
UCA,
where (,\(t) is an infinitely differentiable function such that CA(*) = 1 for£>£1 + 2A,
CA(*) = 0 for t < h + A,
\('x{t)\ < cA"\
and A € (0, r/2). The functions w and q satisfy the relations V w w ( - !/V2w + Vq noS(w)n 6 u| t = t 0
= = = =
0, f + u& y e Qb, * > h bCA, 0.
Moreover, integrating by parts in [n6 • T(u, r)nb - anb • Aberp\(x -
[nb • T(u,r)n b - an • Aberp]Cx{t')dt' J ti
= {d + F)Cx{t)- f (d + f)(x(t')dt', J t-l
(56)
200 we obtain nb-T(w,q)nb-anb-
Abwdt'+ J ti
= d(x+
j
h{Mv)dt' J ti
[nb • T ( u , r)nb - d]('x(t')dt'
+ f ^-F(p)(,(t')dt' + *nb- f
(57)
Ab^^df.
We consider w and r as a solution of the problem (55)-(56). Inequality (45) implies
sup
|ws(-,s)|c«(nt)+
tl<S
sup
|w(-,s)|C2+Q(nii)+
tl<S
< c
• • (
sup
|f|c-(n k ) +
sup
\ti+X<s
sup
\r(-,t)\ci+a(Qb)
ti<S
\h\C2+a(rb)
ti+A<s«o
sup ~^T + lblc'+-.<'+-)/*(r,,x(*1,t„)) + A-( 1+a >/ 2 sup |b(y,s)| t1+x<s
+
h+X<s
+A M
sup
ti+\<s
|u(-,s)| C i+a ( r ,,)+
\*i+A<s
sup
\r(;s)\c-(Th)
ti+A<s<4 0
j
(58) To estimate t h e norms of r and u in the right hand side, we use the boundary conditions and t h e interpolation inequalities. We have sup \b{y,s)\ rtx((i+A,fo) \r(-,s)\c°(rb)
< <
sup |Vu(2/,s)|, Six(ti+\*o) c(\u(;s)\ci+«(rb) + \d\c°(rb) +
< <
e|u(-,s)| C 2+»(n l ) + c e - 5 / 2 - a | | u ( - , s ) | | L 2 ( n 6 ) , £\p{-,s)\c3+°(s1) + ce~~5/2~a\\p(-,s)\\w}(s1)
and |u(-,s)| C i+a ( n i ) \p(-,s)y+-(Sl)
c
Finally, when we consider (39)4 as Schauder's estimates, we obtain
a n
(59) \p{;s)\e'+<'(s1))
elliptic equation for p on Si and make use of
\p{-, s)| c «+«(s 1 ) < c [\d + r - unb • S(w)n 6 )(-, s)| c i+a ( S l ) + \\p(-, s)\\wi(Sl)j Hence, t h e right hand side of (59)2 does not exceed ce(\d(-,s)\cl+a{rb)
+ \r{-,s)\ci+«{Tb)
+ |w(-,s)| C 2+«(n,,)) + \d\c°(rb) +c(e)(l|w|| i 2 (n 1 > ) + ||p(-,s
)ikc*>)Now, when we choose e = As' and taking account of (47), we obtain
.
201 sup
|u s (-,s)| C a (n ,,)+
ti+2A<s
+
sup
|u(-,s)| C 2+a (nt)
ti+2X<s
sup
|r(-,s)| C i+«(n k )+
sup
ti+2A<s<( 0
|p(-,s)| C 3 + « (Sl)
ti+2A<s
/ +Cl(e'+ 6) [
sup °
h+x<s
|u(-,s)| c 0 + « ( n i , ) +
I \
+
sup \r{-,s)y+^nb) ti+A<s<*o
SUP |p(-,S)|C3+a(Sl) tl+A<s
+c( £ ')A~ 7/2 - Q ( sup ||v(-, s )|U 2 ( s i t ) + sup Vi<S<«0
\
,finx 160)
I /
IIT/MH^SA
«l<S
/
We define the function of A /(A) = A 7 / 2 + a (
sup
|u 5 (-,s)| c «(n 6 )+
Vi+2A<s
+
sup
sup
|u(-,s)|C2+«(ni>)
ti+2X<s
|r(-,s)| c i+a (f!b )+
ti+2A<s
sup
|p(-,s)|C3+o.(Sl;
ti+2X<s
and we multiply (60) by \7l2+a. Since A < T/2, (60) implies f(X)
+ S)f(\/2)
+ K,
where K = c{e')[
sup ||u|| i2 (o 6 )+ sup \\p{-,s)\\wi{Sl) \tl<S
tl<S
i, /
is independent of A. We assume that c'27/2+a6 < 1/4, and, choosing e' such that c'e'2 7 / 2+Q < 1/4, we obtain
which furnishes (51), once we take A = T / 2 . The proposition is proved.
•
The following Corollary is a consequence of (33) and (55). Corollary 7.5 For any arbitrary solution (u,r,p) of the problem (39) with the data satisfying |u0|c2+<»(ftt) + |Po|c3+«(Si) < £ and the condition / x x wocte = / / x x vbdSi, Jfio JSiJ.R which is defined in time interval (0,T), T < oo and satisfies (53) there holds the uniform estimate swpYt{u,r,p) < ce~bt (|u0|c2+«(ni,) + |po|c3+=(Si)) . with the constants c and b independent of T.
202 Repeated application of the local existence theorem (Theorem 7.2) leads to our main result Theorem 1.1. • Acknowledgment The authors are grateful to professor Taddia for his valuable comments. The authors thank the GNFM-INDAM group for financial support. Padula thanks also the 60% contract MURST.
References [1] R.A. Brown and L.E. Scriven, Proc. R. Soc. London, A 371 (1980). [2] F. Brulois, Var. Meth. Free Surf. Int. P. Concus and R. Finn Eds. Springer-Verlag, Berlin. [3] S. Chandrasekhar, Proc. R. Soc. London, A 286 (1965). [4] I.Sh. Mogilevskii & V.A. Solonnikov, On the solvability of an evolution free boundary problem for the Navier-Stokes equations in Holder spaces of functions, Mathematical problems related to the Navier-Stokes equations, 11 (1992), 105-181. [5] M. Padula, On the exponential decay to the rest state for a viscous isothermal fluid, J. Fluid Mech. and Anal., 1 (1998). [6] M. Padula & V.A. Solonnikov, On the Rayleigh-Taylor Stability, Annali dell'Universita' di Ferrara, (sez.VIII, Sci. Mat.) 46 (2000), 307-336. [7] M. Padula & V.A. Solonnikov, On the global existence of nonsteady motions of a fluid drop and their exponential decay to a uniform rigid rotation Quaderni di Matematica, to appear. [8] V.A. Solonnikov, Solvability of the problem of evolution of an isolated amount of viscous incompressible capillary liquid, Zapiski Nauchn. Semin. LOMI, 140 (1984), 179-186. [9] V.A. Solonnikov, Unsteady motion of a finite mass of fluid bounded by a free surface, Zapiski Nauchnykh Sem. LOMI, 152 (1986), 137-157. [10] V.A. Solonnikov, On an evolution of a isolated volume of viscous incompressible capillary liquid for large values of time, Vestnik Leningrad Univ., ser.l, 3 (1987), 49-55. [11] V.A. Solonnikov, On non-steady motion of a finite isolated mass of self-gravitating fluid, Algebra and Analysis, 1 (1989), 207-249. [12] V.A. Solonnikov, On the justification of the quasistationary approximation in the problem of motion of a viscous capillary drop, Interfaces and free boundaries, 1 (1999), 125-173.
[13] A. V. Solormikov, Evolution free boundary value problems, Summer course in Punchal, 2000. To appear.
On a convection-diffusion equation with partial diffusivity * Andrea Pascucci Dipartimento di Matematica, Universita di Bologna ^
Abstract We consider the Cauchy problem for the nonlinear degenerate equation in RJV+1 div(AVu) + u(b • Vti) - dtu = f(-,u), where A > 0 is a constant symmetric matrix and ker(A) is generated by b. We prove the existence of a local viscosity solution u and we study the interior regularity of u in the framework of Hormander type operators.
1
Introduction
We consider the Cauchy problem for the nonlinear convection-diffusion equation div(AVu) + u(b • Vu)-dtu
= f{-,u),
in ST = R w x]0,T[,
(1.1)
with initial datum u{-,0)=g,
inR*.
(1.2)
We assume that / , g are globally Lipschitz continuous functions. Moreover we assume that the matrix A is constant, symmetric and positive semidefmite. The convection direction b is constant and generates ker(A). Equations of form (1.1) were studied by Escobedo, Vazquez and Zuazua [11] in order to describe the asymptotic behaviour as t —)• oo of solutions to a related parabolic equation with complete diffusion. We remark that, without loss of generality, by performing a suitable change of variables we may assume that A is diagonal so that b points along a coordinate axis, for instance b = eN. Then it is convenient to denote a point in M.N by (x, y) with x = (xi,..., x^-i) and y € R. Hence (1.1) becomes Lu = Axu + udyu-dtu
= f(-,u),
in ST,
where A x denotes the Laplace operator acting in the x variables. 'Keywords: nonlinear degenerate parabolic equation, interior regularity, Hormander operators. ^Piazza di Porta S. Donato 5, 40127 Bologna (Italy). E-mail: [email protected] 'Investigation supported by the University of Bologna. Funds for selected research topics.
204
(1.3)
205 Equation (1.3) also arises in mathematical finance, when studying agents' decisions under risk. The classical approach for this financial problem is based on the representation of agents' preferences in the framework of the utility theory and various models have been proposed, aiming to taking into account many aspects of the dynamics of the economy. Epstein and Zin in [10] propose a utility functional which is the solution of a backward stochastic differential equation. Recently Antonelli, Barucci and Mancino [1] propose a more sophisticated utility functional that considers some other aspects of decision making, such as the agents' habit formation, which is described as a smoothed average of past consumption and expected utility. In that model the couple of processes utility and habit is described by a system of backward-forward stochastic differential equations. In [1] is proved that there exists a unique solution u of such system, that satisfies some suitable initial and final conditions, and which is a viscosity solution, in the sense of the User's guide [9] of Cauchy problem (1.3)-(1.2). Moreover, in [1] it is proved that the solution u is defined in a suitably small interval of time [0, T[ and satisfies \u(x,y,t)-u{x',y',t)\
< c0(\x - x'\ + \y - y'\),
\u(x,y,t)-u{x,y,t')\
+
\(x,y)\)\t-t'\^
for every (x, y), (x1, y') e R", t, t' € [0, T], where Co is a positive constant that depends on the Lipschitz constants of / and g. Related problems also arise in stochastic control theory. For instance, the value function v of a suitable control problem is a semiconcave solution of the following Cauchy problem dxxv + ^(dyu)2-dtv
= ip,
v{-,0) = il>,
inR 2 x]0,T[, inR 2 ,
for some continuous functions
2
Main results
Our aim is to find a functional space where this problem is well posed. The main difficulty is the mixed parabolic-hyperbolic feature of equation (1.3) due to the lack of diffusion in the y-direction, so that it may include the Burgers' equation, when / = 0 and g = g{y). We explicitly note that the nonlinearity in (1.3) is not monotone, then a standard comparison principle does not hold and, as a consequence, the uniqueness of the solution
206 is not guaranteed. This fact also affects the existence of the solution. Indeed, when using the classical Bernstein's method, a maximum principle for the operator Lv + v2 (that occurs when we differentiate both sides of (1.3) w.r.t. y) is required. Yet also more sophisticated versions of that method (cf. Barles [3]) do not seem to work in our setting. On the other hand, in the space of functions characterized by conditions (1.4) the operator L in (1.3) does satisfy a comparison principle. Then we are able to prove the existence of strong solution of the Cauchy problemfor small times. More precisely, we have Theorem 2.1 Let f, g globally Lipschitz continuous. IfT>0is suitably small, then there exists a unique function u, verifying estimates (1.4) on ST = K" x [0,T] and assuming the initial datum g, such that
u e HiJSr),
AXU e Llc(ST),
and equation (1.3) is satisfied a.e. Let us remark that, in general, the linear growth of the initial datum g does not allow solutions which are defined globally in t. Indeed, let us consider the following simple example: for N = 2, take / = 0 and g(x, y) = x + y. A direct computation shows that u(x,y,t) = jz^ is the unique solution to (1.3)-(1.2) and it blows up as t —¥ 1. Our main results regard the interior regularity of the strong solution u of Theorem 2.1. Since L is a degenerate second order operator, the known results by Cabre e Caffarelli [5], Trudinger [24], Bian e Dong [4], Wang [25] do not apply. Therefore we set the problem in the framework of subelliptic operators on nilpotent Lie groups. We remark that L is an operator such that the matrix of the coefficients of the second order derivatives is only positive semi-definite. As one can expect, the solution u of the equation Lu = 0 is smooth in the directions in which the matrix is non-degenerate, but not in other directions. Consider for example the operator
Lx = dl + d2y
(2.1)
3
in the variables (x,y,t) e R . Every solution u of L\u = 0 is smooth with respect to the variables x and y, but is not regular in the variable t. However, as Hormander pointed out in the celebrated paper [16], there are other "regularity directions" for the solution u, and these directions are the ones of the commutators. For instance, let us consider the Kohn-Laplace operator in R3: L2 = (dx + 2ydtf + {dv - 2xdtf .
(2.2)
As before, there are only two directional derivatives, while the dimension of the space is three, but in this case every solution of L2u = 0 is smooth, not only in the directions of the derivatives X = dx + 2ydt and Y = dy — 2xdt, but also in the direction of their commutator [X,Y} = XY -YX = -4d ( . The operator considered above is a simple but meaningful example of the class studied by Hormander in [16]. Let Xo, ...,Xp be a set of linear first order operators (i.e. vector fields) defined as n
i=l
207 where a{j are smooth functions on some domain S l c K " and let also / G C°°(fi). Hormander proved in [16] that if u is a solution of the equation v ^XiXjU + X0u = f, in Q, (2.3) and the Lie algebra generated by the vector fields X0, ...,XP has rank n at every point of Q, then u € C°°(U). Hormander's result was the starting point of an extensive research aiming to investigate the regularity properties of the operators in (2.3) and their links with some suitable Lie group structures on R". The existence of a fundamental solution and of a control distance have been established in [20], [23], [17]. Using these properties, a general theory of the regularity both in Sobolev spaces and in spaces of Holder continuous functions has been settled down in [13], [14], [23] and [18]. Aiming to use the linear theory for the study of our problem we can try to consider the "linearized" operator Lu = Ax + udy - dt, where u is considered as a coefficient, but we immediately realize that the smoothness of the coefficients a^ is a fundamental assumption in the previous papers and, in our problem, we cannot assume that the coefficient u of the equation is C°°, since the smoothness of the solution u is exactly the goal of our study. Actually, in all the papers cited above it is crucial that the vector fields Xj are regular at least as it is sufficient to obtain, by commutation, n linearly independent vector fields at every point of R". For the first time, Franchi and Lanconelli [15] studied the properties of a control distance related to a family of non regular vector fields, aiming to adapt the classical Moser's iteration scheme to prove the Holder regularity of weak solutions of an equation of the form n
where the matrix (aij(x)) is positive semi-definite for every i 6 l " . Franchi e Lanconelli assume that the operator is "elliptic" w.r.t. a family X\, ...,Xp of Lipschitz continuous vector fields, in the sense that
3=1
i=i
for some positive constant A. This class of operators includes, for instance, in R2
d2x + \x\2adl, where a is a positive constant. We stress that a general theory for operators with nonsmooth coefficients is not available. Here we employ a technique introduced by Citti in [6] where the author considers the regularity of the solutions to the following equation of prescribed Levi curvature r
-
_L
, ul
+ u
l
Cu = uxx + uyy + Y^uu
, «uy - u*ut
+2 -
T T
^
u + u ut
(
{l + u} + ul + uy)m
-n 2x^ ^ 3v - u * = k
^
,
208 in the variables (x, y, ( ) e E 5 . In [6] it is pointed out that the principal part of the above operator can be written in the form (2.3) Cu = X2u + Y2u in terms of the nonlinear vector fields x
=
d
U_M-UxUi
y
=
U_x_^Uidt
d
1 + uf
1 + uf
Then, based on the notion of "intrinsic" Taylor expansion of the coefficients of the operator, a modification of the freezing method used by Rothschild and Stein in [23] is developed. Analogously, we remark that, by letting Xj = dXj,
j = 1,..., JV - 1 and X0 = udy - dt,
then L in (1.3) can be formally represented as L
= EX'+X°-
(2-4)
Since the commutator of Xj and X0 is [Xj,X0] = {dXju)dy, the Hormander condition is satisfied if dXju(x,y,t)^0,
(2.5)
for every (x,y,t), for some index j . Note that the regularity of the solution u in Theorem 2.1 is stated in condition (1.4), then dXju is defined almost everywhere and the above condition has to be considered only formally. Then we first state a regularity result for classical solutions to (1.3). Theorem 2.2 Let Q be an open set in M.N+1 and u a classical solution of (1.3) on Q with f e C°°. If (2.5) holds, then u e C°°(fi). If we do not require any assumption on the commutators (in particular, if we do not require anymore condition (2.5)) then the Lie algebra associated to the operator is completely unknown. However we consider L as a subelliptic operator with respect to some tentative Lie groups. This allows us to prove the existence of the derivatives dx.x.u and XQU, defined as the directional derivative with respect to the vector vz = (0, u(z), — 1) at the point z = (x, y, t) X0u(z) = —(z) = hm -^ Then we have
^
^-.
2.6
209 Theorem 2.3 The strong solution u to (1.3) in Theorem 2.1 is a classical solution in the sense that dxx.u, j = 1,..., N — 1, and XQU are continuous functions and the equation is satisfied at every point. This result is quite reasonable, since without assuming the Hormander condition we are able to prove the regularity of u only in the directions of the vector fields. Note that, although the derivative J^- can be obtained as a sum of more simple terms du dvz
udyU — dfU
it is not true in general that the terms udyu and dtu are continuous. Also note that, at this point, the Hormander condition (2.5) is meaningful since XjU are defined and continuous functions.
3
Existence
The proof of Theorem 2.1 is based on some estimates which can be obtained by adapting the classical Bernstein's method. We consider the regularized Cauchy-Dirichlet problem in a cylinder Llu = Axu + e2dyyu + vdyu-dtu
= f(;v), u = g,
in QT = Bx]0,T[, in dpQT-
(3.1) (3.2)
where e > 0, B is a ball in RN and 8PQT denotes the "parabolic" boundary of QT defined as (B x {0}) U (dB x [0,T]). By a standard density argument we may assume that / , j £ C°°nLip. We fix a positive constant Ci such that, for (x, y, t, v) e QT, it holds c\ > max{Lipschitz constants of /, g},
l(z,y)l
\f(x,y,t,v)\
+
\(x,y,t,v)\i.
Given a €]0,1[, we assume that the coefficient v in (3.1)-(3.2) belongs to Ci + Q (Q T ), the Holder space w.r.t. to the parabolic distance d((x,y,t), (x',y',t')) = \x — x'\ + \y — y'\ + \t — t'|5. Moreover we assume that v satisfies in QT the estimates \v(x,y,t)\<2Cl^l \dXiv\<2cu
+ \(x,y)\\ i = l,...,N-l,
|a„w| < 2ci.
(3.4) (3.5)
The following proposition is the key step in the proof of Theorem 2.1 which then follows by the Schauder's fixed point theorem and passing at limit as e goes to zero. Proposition 3.1 There exists T > 0 such that, under the above assumptions, every classical solution of (3.1)-(3.2) verifies the e-uniform estimates (3.4), (3.5). Proof. Let a be a classical solution of (3.1)-(3.2). We prove estimate (3.4) for u by applying the maximum principle to the functions H ±u where H is defined as follows H{x, y, t) = (Cl + fit) ^l + \{x,yW
210 and /j, is to be suitably fixed. Keeping in mind (3.3) and (3.4), is easily verified that L°vH(x,y,t)
< ( 1 + £ 2 ) ( ^ + ^ 2) yjl + \{x,y)\
+ ( ( c i + / j r ) c i
_ »)^l
\(x,y)\2<-\f(x,y,t,v(x,y,t))\,
+
if /i, i are suitably large. On the other hand, by (3.3), H\dpqT > \g\. Therefore, by the maximum principle, we infer that H ± u > 0 on QT, that is
\u\
if T <
<=i A*
Next we prove the estimate for the y-derivative of u. We differentiate equation (3.1) w.r.t. the variable y and then multiply it by e~2Xtdyu. Denoting u = (e~xtdyu) , we obtain LEvuj = e'2XtLl{dyu)2 = 2 (e-
2Af
+ 2\u
( I V ^ ^ I 2 + e\dvyuf
+ dyu {dyf + djdyv))
+ Xto)
2Xt
> 2 (e~ dyu {dyf + dvfdyv) + XCJ) .
(3.6)
2
Hence, by setting w = a; — c , we get from (3.6) L%w > 2V^ ( - \dyf\ - \dvvdj\
+ Av^)
(by (3.3), (3.5) and by the elementary inequality y^J > 2 ( Cl > V2uj (Cl (\ - 2V2ct - V2) +
+ s
Sn(u;)\/M))
\sgn(w)y/\w\)
(for A = A(cj) suitable large) > c^/uj\w\sgn(w), (3.7) for some positive constant c = c(ci). By contradiction, we want to prove that w < 0 in QT- It will follow that \dyu\ < cxext which implies (3.6) if T = T(ci) > 0 is sufficiently small. Let ZQ be the maximum of w on QT. If w(zo) > 0, then ZQ 6 QT \ OPQT, since by construction w < 0 on 8PQT- This leads to a contradiction since, by (3.7) 0 > Lcvw(z0) > cy/ui(z0)w(z0) > 0. By a similar technique, we prove estimate (3.5) of the ^-derivatives of u. We set ui = (e~xtdXku)
,
w = u> — c2.
Differentiating equation (3.1) w.r.t. xk and multiplying it by e~2XtdXku, we get L%w = e-2XtLl(dXku)2
+ 2Aw = 2 {e~2XtdXku {dxJ + dXkvdvf) + Aw)
(by (3.3),(3.5), and the estimate of dyu just proved) > V2u (ci fA - A/2 - 2A/2CI - 4\/2c 2 ) + Asgn(iu)y/\w\) (if A = A(ci) is suitable large) > c^/uj\w\sgn(w), for some positive constant c which depends only on cx. As before, we infer that w < 0 which yields (3.5). •
211
4
Regularity
Theorems 2.2 and 2.3 rely on some representation formulas for u and its derivatives in terms of the fundamental solution of a frozen operator. The freezing method is a well-known technique, classically used to study the regularity of solutions to linear parabolic equations. In this case, the associated frozen operator is simply obtained by evaluating the coefficients at a fixed point. This new operator is, up to a linear change of coordinates, the heat operator and its fundamental solution can be considered as a parametrix of the fundamental solution of the non-constant coefficients operator. A much more difficult argument was used to prove the existence of a fundamental solution for Hdrmander type operators (2.3). Indeed the frozen vector fields n
Xi>xo = ^2Oij(xo)dXj
i = 0, • • • ,p
i=i
commute, and the generated Lie algebra has dimension, in general, less than n. In this case the operator
E*L-*o,*„
(4-1)
is not hypoelliptic, and it has not a fundamental solution. Folland and Stein [14] first pointed out that the model operators in this case are operators of the form (2.3) such that the Lie algebra generated by X0,..., Xp is nilpotent and stratified. Later on Rothschild and Stein introduced in [23] an abstract and very general version of the freezing method. The choice of the frozen vector fields Xi>xo is made in such a way that their generated Lie algebra has, at low orders, the same structure as Lie(Xo, ...,XP). With this choice of vector fields, the operator in (4.1) is hypoelliptic and its fundamental solution r i 0 is a parametrix for (2.3). As said above, we employ here a modification of this technique, introduced by Citti in [6]. We define a frozen operator in terms of the notion of "intrinsic" Taylor expansion of the coefficients. We first consider L as a linearized operator N-l
Lu = 2^i Xj + Xo, j=i
where Xj = dx. and X0 = udy — dt. Then we let, for every zo = (x0, y0, t0) e RN+1, JV-l
.7 = 1
where Xo>20 = (u{z0) + (x — xo) • Vxu(zo))dv — dt. Under the assumptions of Theorem 2.2, this choice ensures that the frozen operator Lzo is a nilpotent Hormander type operator, it has a fundamental solution r z o and an associated control distance dzo. For simplicity we assume that / does not depend on u. We represent the solution u in terms of r z o :
«(«) = / rzo(2, C)Lzou(OdC = J rZ0(z, C)/(CR + J rzo(z, Q)KZO{Z, CR,
212 where Kzo(z,()
= (Lu - LZ0)u(Q = (u(C) - u(z0) - (£ - x0) •
Vxu(zo))dyu{Q.
Since Lu is a second order operator, we have to consider the term X0tZou as a second order derivative, whereas Xj is a first order derivative. As a consequence, the first order Taylor polynomial of u, with initial point at z0, is given by Pzou(z) = u(z0) + (x-
x0) • VZM(Z 0 ).
1,a
Now, by (1.4), dyu is bounded so, when u € C , we have KZ0{z,0
= O{d(z0,Cy+a)
as d(«b,C)-+0.
By choosing zo = z, this estimate allows us to differentiate up to 3 times the above representation formula under the integral sign: D3u(*o) = J Drzo(z0,0-D2/(CR
+ / D3rzo(z0, OKzo(z0, C R
and conclude that u 6 C3'a. A rather delicate argument, based on the use of some high order difference quotients allows us to iterate this argument and conclude the proof of Theorem 2.2. The proof of Theorem 2.3 follows the same lines, but there is another difficulty. Indeed, without the assumption of the Hormander condition (2.5), the frozen operator Lzo may not have a fundamental solution. In this case, it is convenient to approximate X0 by the vector field Xo,z0 = {u(zo) + {x- x 0 )i) dy - dt, where xx denotes the first component of the vector x. Then the operator N-l
Lzo = 2 ^ Xj + X0iZ0 does not depend on VXU(ZQ) and it is hypoelliptic. Note that, with respect to X0>zo, this vector field gives a less close approximation for Xo, since X0u{z) - X0iZ0u(z) = (u{z) - u(z0) - (x - x0)i)dyu{z)
=
O(d(z0,z)),
as d(z0, z) —> 0, however it is sufficiently accurate to prove Theorem 2.2.
References [1] ANTONELLI, F.; BARUCCI, E.; MANCINO, M.E., A Comparison result for FBSDE with Applications to Decisions Theory. Math. Methods Oper. Res., in press (downloadble from http://www.dm.unibo.it/~pascucci/web/Ricerca/Ricerca.html). [2] ANTONELLI, F.; PASCUCCI, A. On the viscosity solutions of a stochastic differential utility problem. J. Differential Equations, in press (downloadble from http://www.dm.unibo.it/~pascucci/web/Ricerca/Ricerca.html).
213 [3] BARLES, G., A weak Bernstein (1991), 241-262
method for fully non-linear elliptic equations, Diff. Int. Eq., 4-2,
[4] B I A N , B . ; D O N G , G., The regularity of viscosity solutions for a class of fully nonlinear Sci. China, Ser. A 34, No.12, 1448-1457 (1991).
equations,
[5] C A B R E , X.; CAFFARELLI, L. A., Fully nonlinear elliptic equations, Colloquium Publications. American Mathematical Society. 43. Providence, RI (1995). [6] ClTTI, G., C°° regularity of solutions of a quasilinear equation related to the Levi operator, Ann. Sc. Norm. Super. Pisa, CI. Sci., IV. Ser. 23, No.3, 483-529 (1996). [7] ClTTI, G.; PASCUCCI, A.; POLIDORO, S., On the regularity of solutions to a nonlinear parabolic equation arising in mathematical finance, Diff. Int. Eq. 14-6 701-738 (2001). [8] ClTTI, G.; PASCUCCI, A.; POLIDORO, S. Regularity properties of viscosity solutions Hormander degenerate equation, 3. Math. Pures Appl. 2 0 0 1 , 80 (9), 901-918 (2001).
ultra-
of a non-
[9] CRANDALL, M. G.; ISHII, H.; LIONS, P.-L.,User's guide to viscosity solutions of second order partial differential equations, Bull. Am. Math. S o c , New Ser. 27, No.l, 1-67 (1992). 10] L. EPSTEIN, S. Z I N , Substitution, risk aversion and the temporal behavior of consumption asset returns: a theoretical framework, Econometrica, 57 (1989), 937-969. [11] E S C O B E D O , M.; VAZQUEZ, J . L . ; ZUAZUA, E., Entropy solutions for diffusion-convection with partial diffusivity, Trans. Am. Math. Soc. 343, No.2, 829-842 (1994).
and
equations
[12] W . H. F L E M I N G , H. M. SONER, Controlled Markov processes and viscosity solutions, Applications of Mathematics, 2 5 , 1993, Springer-Verlag. [13] FOLLAND, G . B . , Subelliptic estimates and function spaces on nilpotent Lie groups, Ark. Mat. 13, 161-207 (1975). [14] FOLLAND, G . B . ; STEIN, E . M . , Estimates for the 5(, complex and analysis on the Heisenberg group, Commun. Pure Appl. Math. 27, 429-522 (1974). [15] B . F R A N C H I , E . LANCONELLI, Holder regularity theorem for a class of linear non-uniformly elliptic operators with measurable coefficients, Ann. Sc. Norm. Super. Pisa, CI. Sci., IV. Ser. 10 (1983), 523-541. [16] HORMANDER, L., Hypoelliptic second order differential equations, Acta Math. 119,147-171 (1967). [17] JERISON, D . S . ; SANCHEZ-CALLE, A., Estimates for the heat kernel for a sum of squares of vector fields, Indiana Univ. Math. J. 35, 835-854 (1986). [18] KRYLOV, N . V . , Holder continuity and Lp estimates for elliptic equations under general der's condition, Topol. Methods Nonlinear Anal. 9, No.2, 249-258 (1997).
Horman-
[19] MOSER, J., On Harnack's theorem for elliptic differential equations, Comm. Pure Appl. Math. 14, (1961). [20] N A G E L , A.; S T E I N , E . M . ; WAINGER, S., properties, Acta Math. 155, 103-147 (1985).
Balls and metrics defined by vector fields I: basic
[21] PASCUCCI, A., Hlder regularity for a Kolmogorov equation, preprint [22] PASCUCCI, A.; P O L I D O R O S., preprint
On the Cauchy problem for a nonlinear ultraparabolic
equation,
214 [23] ROTHSCHILD, L.P.; STEIN, E.M., Math. 137(1976), 247-320 (1977).
Hypoelliptic differential operators on nilpotent groups, Acta
[24] TRUDINGER, N.S., Holder gradient estimates for fully nonlinear elliptic equations, Proc. R. Soc. Edinb., Sect. A 108, No.1/2, 57-65 (1988). [25] WANG, L., On the regularity theory of fully nonlinear parabolic equations I and II, Commun. Pure Appl. Math. 45, No.l, 27-76 and No.2, 141-178 (1992).
Quasiconvexity and optimal design Pablo Pedregal ETSI Industrials Universidad de Castilla-La Mancha 13071 Ciudad Real, Spain
Abstract A typical optimal design problem in conductivity is examined where the cost functional is the square-mean deviation of the gradient of the electric potential from a given vector field. Through a convenient reformulation of the situation as a vector variational problem, we are able to provide a complete analysis of relaxation including a fully explicit formula for the quasiconvexification of the resulting integrand and associated optimal microstructures, both under no resource constraints and under a typical volume constraint. In particular, this analysis opens the gate to the numerical approximation of optimal structures via relaxation. To David Kinderlehrer
1. Introduction We would like to report on some recent advances on the analysis of some typical optimal design problems in conductivity where two different conducting materials with conductivities 0 < a < f3 are to fill out in given proportions a regular, simply-connected design domain Cl in R. so as to achieve a desired goal. One of the new features is the explicit dependence of the objective functional on derivatives of the underlying electric potential which is related to a given mixture through the equilibrium equation div ( ( a x ( x ) + 0 ( 1 - x ( i ) ) ) V t i ( i ) ) = f{x)
in
O,
together with boundary condition u = Uo on
dil.
Here the design field X is a characteristic function of a subset of Q such that / x(x) dx = A \Q\, Jn and A £ (0,1) is given. The integral functional which we want to minimize is I(X)=
f
Jn
IVu(x) - F(x)\2
215
dx,
216 where F is a given target field. Altogether, the optimal design problem we would like to treat is: Given Q C R 2 , simple-connected and regular, 0 < a < /3, A e (0,1), F e L2(Q), 1 f G H-^il), u0 e H ^), Minimize subject
I(x) = / \Vu(x)
- F(x)\2
dx,
to X,
a characteristic div ((aX{x)
/ x{x) dx = X \Q\ Jn + /9(1 - x ( x ) ) ) Vu(a:)) = f(x) in SI, u = UQ on dQ. function
of a subset ofQ,
Homogenization theory (see for instance [3] or [19]) has been the main tool in dealing with such optimization problems both analytical and numerically when cost functionals do not depend explicitly on derivatives of u. Such problems are rather well understood. Further extensions of the ideas of homogenization to tackle the dependence on derivatives of u can be found in [6], [10], [18]. Here we would like to describe an alternative approach based on a reformulation of the optimization problem as a purely vector variational problem and see how far we can reach in the analysis of such variational principle. This perspective has already been explored in [16]. As it turns out, the underlying variational problem is non-convex, which has been known since the pioneering work [11], and therefore its analysis proceeds by examining relaxation ([4]). Under the volume constraint, this relaxed problem involves a convex hull that must take into account such integral restriction and thus ought to depend on a new scalar variable in addition to the usual dependence on gradients. What is really remarkable is that the explicit form of this relaxed integrand can be computed explicitly and given in closed form, thus opening the gate to numerical approximation of optimal microstructures via relaxation. This contribution is organized as follows. Section 2 describes the reformulation of the problem as a constrained, vector variational problem with some interesting features. Sections 3 and 4 treat the same optimal design problem without the resource constraint
[ X(x)dx = \p\. Section 5 contains some ideas and statements on the relaxation of this non-convex variational problem under the above volume constraint as well as explicit formulae for the appropriate constrained convex-hull. Some final remarks are included in Section 6.
2. Reformulation An initial observation relevant to simplify the computations that will follow is that we can assume without loss of generality that both F and / are identically zero. The reason for taking F = 0 is
|V«(x) - F(x)\2 = \Vu(x)\2 - 2V«(x) • F(x) + \F{x)\2 ,
217 and only the first term (corresponding t o F s O ) is non-linear. The reason for taking / = 0 is that the general case involves a traslation by an auxiliary known function, and thus the computations are easily derived in the general case from those for / = 0. We thus take from now on F and / vanishing identically, so that the functional to be minimized is
I(X) = [ |Vu(.x) I dx, and the equilibrium equation is div {{aX(x)
+ P(l - x{x)))Vu(x))
= 0
in
O.
The reformulation of our problem starts with the realization that this equilbrium equation is equivalent (under the assumption of simple-connectedness of Q.) to the existence of an stream function v such that {aX{x)
+ /3{1 -
X{x)))Vu{x)
+ TVv{x)
= 0,
where T is the 7r/2, counter-clockwise rotation in the plane. Hence we can alternatively use as design variables the pair of vector fields («, v) satisfying the additional, important, pointwise restriction aVu(x) + TVv(x) = 0 or /3Vu{x) + TVv(x) = 0, (2.1) for a. e. x € Ci. Notice how we can go from an admissible x to s u c n a P 3 ^ ( u i ")> a n ( i conversely, from such a pair (u, v) verifying (2.1) to an admissible X- Om- aim now is to rewrite the optimal design problem in terms of these pairs (u, v) instead of XTo this end, collect both u and v in a single vector variable U = (L^ 1 ', LA 2 ') (u is identified with the first component [/t 1 ) while v is identified with £ / " ' ) . Define the two densities W, V : M 2 X 2 -> R* = R U {+00} , by putting W{A)
= \\AW\2' [. + o o , V(A)
XAeAaUAf,^ else,
= [l< (^ + o o ,
i f / else,
A 7 = J A € M 2 x 2 : 7 A W + TA^
6 A
«',
= o|,
7 = a or 0.
Then it is elementary to convince ourselves that the original optimal design problem is equivalent to the vector variational problem Minimize
UeH1^),
J{U) = f W{VU(x))
U^=u0
Jn
on 90,
dx
I V{VV{x))dx = X.
218 The main advantage of this formulation is two-fold. On the one hand, we somehow overcome the non-local effect associated with the equilibrium equation to the expense of treating a vector variational problem. But on the other, as such typical vector variational problem, we can utilize all the accumulated experience with non-convex variational problems. In particular, all ideas and techniques about relaxation, gradient Young measures, microstructure, etc. But there are also some new ingredients. We have an additional integral constraint coming from the initial volume constraint which is important and relevant. Moreover, both integrands W and V are not Caratheodory functions as they take on the value + 0 0 suddenly. In these notes, we will ignore this last difficulty as it is, we believe, essentially technical, and focus on how to deal with the first integral restriction.
3. Relaxation without volume constraint In this and the next sections we explore our original optimal design problem, but drop the volume constraint. The reason is that if this integral restriction is not taken into account our reformulation of the problem falls into the typical scenario of a vector variational problem. It is interesting to point out that when some very peculiar adjustment between F, f and a or /3, takes place, then our optimal design problem admits a unique optimal solution. Indeed, if / = a div F or / = f3 div F the variational problem Minimize
[/ IVufx) \Vu{x)- - F(x)\ F(x2
dx
subject to u — u0 G HQ(Q), admits a unique optimal solution which is a weak solution of the associated Euler-Lagrange equation div (Vu) = d i v F = 7 " 1 / ,
in &,
for 7 equal to a or fl. Therefore our optimal solution, when that close relationship between F and / occurs, is the solution of the Poisson's equation jAu
= f
in Q,
U —
U0€HQ(£Y).
Notice that this choice corresponds to taking x constant to 1 or 0, depending on whether 7 = a or 7 = /3, throughout fi. When such adjustment between F and / does not happen, then our analysis becomes significant. Our main result is the explicit computation of QW and the identification of optimal microstructures. Theorem 3.1 g(A) =a2p2
Put
\AW\4 I
I
-2a/3UW h(A)={a
+ |yl<2>|4 + ( a 2 + 6a/3 + /32) detA 2
I
I
Ll
2
(2)
+ 0)detA-ap\AW\
I
-2aj3(a
I
+ P)\A^\
- U(2)l ,
2
2
I
I
det A - 2(a +/3) \A^\
2
detA,
219 and consider the set of matrices r = {A G M 2 x 2 : h{A) > 0,g(A) > 0} Then the quasiconvexification ofW, QW, is given by Q W
1 2.P
^
a,flU«| - U ( 2 ) | +{a + /3)detA-y/^(A)
if A e T, and QW{A) = +oo otherwise. Moreover the unique optimal Young measure providing the value of this quasiconvexification, when A does not belong to A a UA^, is the first-order laminate
tSAa + (1 - t)8Ae where 1
t = 2x +
2(/3-a)detA
/ a U w f - U r o f + VffCA)
This result has been announced and published in [15]. We would like to sketch the proof of it. The strategy for the proof is the following. We know that in computing QW (A) we must care about (homogeneous) gradient Young measures supported in the set where W is finite and with first moment A ([12]). Thus we examine the minimization problem Minimize
( W » = /"
\xw\
dv{X)
where v runs through all gradient Young measures supported in A a U A^ with given first moment A. We now divide the proof in several steps summarized as follows: 1. By using the weak continuity of the determinant we enlarge the set of competing measures in the previous minimization problem and, at the same time, the cost functional is suitably modified by using Jensen's inequality, thus providing a lower bound for QW{A). 2. We show that the optimum of this new enlarged optimization problem is uniquely attained for a measure which is a laminate of first order, thus proving that QW(A) can actually be computed by examining the minimum of the above minimization problem for first order laminates, i.e. the previous lower bound turns out to be exact.
220 3. Explicit computations are carried out. Step 1. Let v be a gradient Young measure supported in A a U A^ with first moment A. Naturally we can decompose v into two parts v = sva + (1 - s)up,
s € [0,1],
where supp (va) C AQ,
supp (up) C Ap.
Fdua{F)eAa,
Ap=
In the same way, set Aa=
f
f
FdvpWeAp.
See Figure 3.1.
Figure 3.1 It is also easy to check the following det A
=-A&-TAW, i
1
deti4 = a A ( ) det J 4 = / 9 U ( 1 ) | det(sj4 a + (1 - s)Ap) — sdetAa
i2
if,4eAa, ifAeA^,
+ (1 - s)detAp - s(l - s)det(^4 Q - Ap).
This last formula for the determinant is only valid for 2 x 2 matrices. We pretend to make use of the weak continuity of det and see what conclusion we can reach. Indeed, we should have f
JM2"
detFdv(F) = det ( / VM"
2
Fdv(F))
/
.
By using all the decompositions and all the formulas written above, it is elementary to arrive at
221 as j
\FW\2 dva(F) +/3(1 - s) f
\FW\2
dv0(F)
2
as f
F^dua(F)
+P(l-s)
[
JAa
F^dvp(F)
- s(l - s) det(j4 a - Ap
JAff
By Jensen's inequality, we conclude det(Aa - Ap) < 0. On the other hand, the objective functional for our initial optimization problem can be rewritten, based on the preceding decomposition for our feasible probability measures, as tf \xW\2 dva(X) + {l-t) f \xW\2 dvp(X), and, once again by using Jensen's inequality, a lower bound for this expresion is t\A^\2
+
(l-t)\A^\\
Therefore the optimization problem i
Minimize
i2
t \A^
i
|2
+ (1 - t) U£°
where t, Aa and Ap are constrained by A = tAa + (l-
t)Ap, Aa e A a , Ap e Ap, t e [0,1] det(A a - Ap) < 0,
will provide a lower bound for QW(A). Step 2. Notice that the objective functional, under these constraints, can be written 1
t\A^\ + ( l - t ) | 4. ( D l
2
so that we will be concerned about ibout the optimization optimizatioi problem |12 2
Minimize
tlA^l
|
i, , , i 2
+ (1 - t) U i 1 '
subject to A = tAa + (l-
i)Ap, Aa£Aa,Ap£Ap,te det(A a - Ap) < 0.
[0,1]
222 By writing Aa=
Af3=
{aTz)'
\/3Tw)
for certain vectors z, w, it is elementary to find
*=^K,+™(a,)« l ic= — —(aAW+TA^\. \—ta—p\ I
For simplicity, let us put A1 = j - ^
(HAM + TA&)
,
A2 = - i ^ (aA*1) + TA<2)) .
Then we can rewrite the above optimization problem in the following terms •> 1
Minimize
•> 1
|Ai| - + | 4 2 |
subject to te(0,l),
2 a
2
a +
2 l A x' l ^ 4+- 1 I4 AI a l ^ — ^ - 4A i A - A a ^> ^ ^ O . ?+l^l ^3-Al-Aa^—1)
Notice that this last expression is precisely det(j4 a — Ag) and that the vectors Ai and Ai are constant. Put, for future reference
Since the objective function for this new formulation is convex in t, it tends to +oo as t —> 0 + and t —¥ 1~, and the function determining the constraint is continuous, the minimum value sought will correspond to equality in the restriction, provided the point of absolute minimum of the objective function on the whole interval (0,1), to, is such that
This is indeed an elementary calculus exercise. Since to —
\Ai\ \Ai\ + \A2\'
the previous expression simplifies to (a + P) {\AX\ + \A2\f (l - ^
. j i ? _ ) > o.
(3.1)
223 Note that A* cannot vanish unless A £ A„ U A^. We conclude that if v is a gradient Young measure supported in the set where Wo is finite and having first moment A, then we have (W,v) > m i n i m i 2 i + |A 2 | 2 ~
:
= min||^1|2i + |^ 2 | 2 T l 7 :^(i)=o} >QW{A). The last inequality is correct because when ip(t) = 0 we obtain a first-order laminate. By taking the infimum in v we get QW(A) = nun 11Ai| a \ + \A2\2 ~
:
Notice how first order laminates are the only possibility for which we can have equality.
Figure 3.1 Step 3. Computation of the previous minimum. The equation ip(t) = 0 is quadratic in t. Indeed, it can be rewritten as a ( l - i)2 |7li| 2 + fit2 \A2\2 - (a + )3)i(l - t)Ax • A2 = 0.
(3.2)
The value of this parabola for t = 0 and t = 1 is positive if A does not belong to either A a or A/3. In order for this quadratic equation to have real roots in the interval (0,1), we need to demand the discriminant to be non-negative, the leading coefficient to be (strictly) positive and the vertex to belong to (0,1). After a few computations we also have [a ] ^ ! | 2 + 0 \A2\2 + {a + p)A! • A2) t2 - (2a [A^2 + (a + P)AX • A 2 ) t + a | A 1 | 2 = 0;
224 or even further, bearing in mind the expressions for the vectors Ai and A2, detA t2 - ^ — (ap U ( 1 ) | 2 - U ( 2 ) | 2 + {P-a) + -—^—^ (aP2 \A^\2
+ a\A^\2
- 2aPdetA]
(p-ay \ 1 \ 1 1 Those three conditions mentioned above amount to 0(^4) > 0 ,
(a + P) det ,4 >
det/) t =0. /
*/3 U ^ - U<2>
|2
After some algebra, it is elementary to show that these two conditions together are equivalent to the ones defining the set T of the statement of Theorem 3.1. Notice that the equality h(A) = 0 must be allowed to incorporate matrices on A a and A/3. For a matrix not belonging to T, the quasiconvexification will be infinite. We need to clarify which one of the two roots of the above quadratic equation we are interested in. We know, to begin with, that the point of minimum to is either to the right or to the left of the interval determined by the those two roots. By elementary continuity arguments, it will always be to the same side provided that to is never the vertex of the quadratic equation. This possibility can only happen if, in addition, because of (3.1), A1-A2
= \A1\
\A2\.
Having this in mind, equating the vertex of the parabola with to leads to the equation lax1 + (a + P)xy _ x 1 2 lax + 2(a + P)xy + 2Py ~ x + y' where x = \Ax\,
y = \A2\.
It is elementary to check that that equation can never hold. In fact what is always true is 2ax2 + (a + P)xy x 2 2ax + 2(a + P)xy + 2Py2 < x + y and this in turn, implies that the largest root is the one we are interested in (see Figure 3.2). After some computations, it has the form given in the statement of the theorem. For the value of the minimum itself, we are in need of calculating 1
2 I V Tt + I^I ' ' i-t'
precisely for this value of t. Notice that if t is a root of at2 + bt + c = 0 so that
t =
—6 + v b2 — 4ac 2a '
225 then 1 _ - 6 + \/b2 - 4ac t~ 2c ' because 1/t is a root of the equation a + bs + cs2 = 0. On the other hand, (3.2) is invariant when changing t by 1 — t, a to P and Ax to A2, so that 1/(1 — i) has the same expression as 1/i but changing a to /3 and Ax to A^- After some careful arithmetic the expresion on the statement of the theorem is obtained.
4. Some more general situations If we review the key facts on the previous proof with the objective in mind of specifying more general situations where similar computations could be carried out, we immediately isolate the main ingredients needed. Indeed, even with the most general cost functional we can provide some definite results. For the characteristic function X °f a simply-connected domain in R 2 , we consider a cost functional I(x)=
/
i>(x,x(x)a+{l-x{x))P,u{x),Vu{x))dx
where div ((aX(x) + 0(1 - X(x))) Vu(z)) = / ( * ) , u = uo on 8Q.
in fi,
For simplicity we take / = 0, so that the auxiliary function V vanishes identically in f2. We are interested in minimizing I(x) among the class of all characteristic functions over Q,. Because of the nature of characteristic functions, we can rewrite the cost funcional as !{x) = / {xi.x)ipa{x,u{x),Vu{x))
+
(l-x{x))ip0{x,u{x),Vu(x))}dx
where ipa(x, u, Vw) = ip(x, a, u, Vu) and the same for tpp- It is straightforward to check that our new optimal design problem is equivalent to the variational problem Minimize
I{U) = [
W(x,U{x),VU{x))dx
where UeH1^),
Uw = uQ
ondfi,
226 and the density W is given by ' xpa(x, UW,AM), if aAW + TAW = o, rl)p{x,U^\AW)t XPAM+TAM = 0, W(x, U,A)= < min {ipa(x, U^\ 0), ip0{x, U<-1\ o)} , if A = 0, +oo, else. The same analysis leads to the following result. Again for the sake of simplicity, we drop the (x,u) dependence of tpa and ijjp, although this more general case is also covered by the following computations. The set F is the same as in the statement of Theorem 3.1. We also let Ti(A), i = 1,2, be the two roots of the quadratic equation (3.1) for AeT, i.e. n(A) = - + 2 r2(A)
2
2(0-a)detA
+
2(j3-a)detA
*P\AW\
- U(2)
*P\AW\
-U (2) +\/
-\/
Finally the matrices At, i = 1,2, are Al
= - i - (PAW + TAW)
Theorem 4.1 1. the functions
,
A2 = ^ i - ( a ^ 1 ) +
TAW)
Assume that
4(v«),
^(Vu)
are non-neg'atjVe and convex on Vu; 2. for all AeT, the functions 9A(t) = tipa (jAij
+ (1 - t) fa ( r z r ^
,
t e (0,1),
are such that min
te[ri(A),r2(A))a
<M(£)
=
min
94 ft).
(4.1)
t6{ri(A),r 2 (vl)}
Then
iA l+oo,
^» ( O T ^ 1 ) + (1 ~ n{m* (r^W 2 ) }'
itA G r
'
else.
In addition, if either ipa and ipp are strictly convex, or the minimum in (4.1) is only attained at one of the end-points of [ri (A), r2 (A)] then the only optimal microstructures are first order laminates.
227 Notice that the hypotheses on the functions gA(t) are valid when either QA(t) is concave or when it is monotone (either increasing or decreasing) over the interval [ri(A),r2(A)). It is interesting to look at some other particular examples where these hypotheses hold. We will always take ipa = ipp = tp and focus on the situation where ip only depends upon Vu and it is homogeneous of degree p, p > 0. Thus the functions (A2). If 0 < p < 1 these functions are concave so that, if ip is convex, we can apply Theorem 5.1, to obtain the appropriate quasiconvexincation. For example, for p = 0 and overlooking the difficulty appearing because of the singularity at the origin, we get f 4>(A2) + n ( A ) (V(A0 - 4>{A2)), if A e T and ^(Aj) > f(A2), QW(A) = \ j,(A3) + r2(A) (V»(Ai) - V(4»)), if A e T and ^(A1) < f(A2), { +00, if A <£ T. For p = 1, curiously enough we obtain QW(A) = i,{AJ + i,(A2) if A e T. For p > 1, the functions <7A(*) are now convex, and the further requirement V-(^i) 1/p V(^i)1/P + ^ 2 )
1 / P
£
(ri(A),r2(A))
must be enforced. When this is the case y>(^i) 1/p
n ^ - W i ) + (1 -nCA)) 1 "^^), if , , , , X P T / M W* ^ r i ^ ' QW{A) r 2 (A)^-P^(A0 + (1 - r2(A)y-^(A2),
if ^
^
^
i
^ > r2(A),
for A e r. Further details can be found in [2].
5. R e l a x a t i o n under volume constraint As pointed out, we would like to concentrate on how things change in the presence of an integral constraint as the one we have in our equivalent variational problem. Indeed, and again overlooking technical issues concerning the lack of continuity, the appropriate convex hull is
CQW(A, t) = inf | j^-J
W(A + V
228 This was studied in [14] and, independently, in [9]. If we envision a relaxation result, the relaxed problem will have integrand CQW(A,i) depending on gradient and local volume fraction. In order for this relaxed variational principle to be relevant we need to know that it admits optimal solutions, and this in turn, involves the appropriate convexity conditions. What are these for a functional depending on gradient A and some other parameter 4? The answer was given with D. Kinderlehrer in [5] motivated by a completely different problem. In general terms, a function ^(A,t) ought to satisfy the "joint convexity property"
*(A,t) < yjj- J *{A + Vv{x),i+0(x))dx,
9eL°°(D),
f 6(x) dx = 0, JD
for the corresponding variational principle to be lower semicontinuous with respect to weak convergence. What is interesting is that the functional CQW(A,t) coming from the relaxation of a variational problem under integral constraints does enjoy this joint convexity property ([14]). As a main consequence, the whole relaxation framework is valid for our initial design problem, and microstructures and microgeometries for the original optimal design problem can be understood via relaxation. In fact the following relaxation result is true under typical technical assumptions which we overlook here. Theorem 5.1 inf I f W(VU(x))
dx:Ue
Hx{n), C/(1) = u0 on dU, f V{VV(x))
min I f CQW(VU(x), 0
dx = \ \ =
t(x)) dx : U 6 # * ( « ) , Um = M0 on 80,, < 1, / t(x)dx = X>.
As is usual in the well-known non-convex variational scenario, the relaxed integrand CQW(A, t) locally encodes the information on optimal microstructures and microgeometries. What is rather remarkable is that CQW(A,t) can be explicitly computed for our equivalent variational problem. The following formulas are given for the purpose of completeness. They appear in the explicit form of the relaxed integrand CQW(A,t): g{A) = a 2 /3 2 U ( 1 ) I* + I A<2> I* + (a 2 + 6a/3 + /32) det A - 2a/3 \A^\2
n(A) V2(
]
~
U<2>| - 2a/3(a + /3) \A^\2
1 1 U(D 2 2{/3-a)detA 1 1 1/3AW 2 + 2(/3-a)detJ4
2
det A-
2(a + /3) U ^ f det A,
_ U<2)
-JKA)
- U(2)
+Vg(A)
229
_(l-ri(A))[t(l-rj(A))-a-t)rj(A)} t{l-rj(A))-(l-Ti{A))ri(A)
,
* + 3-
Our main result here follows. Theorem 5.2 If W(A) = {\Ail)\2> I +oo,
ifAeAaUAp,^ eise,
A 7 = {A e M 2 * 2 : -yA^ + TA™ = 0 j ,
7
= aorft
then CQW(A, t) = inf < /
W(.F) du(F) : v is a gradient Young measure
with first moment A and / JM"2
V(F) dv(F) = t \ , J
is explicitly given by CQW(A,t)=.f).
1
.. (/32\AW\2 + \AW\2-(at
+
(3(2-t))detA)
if (A, t) is such that i
aP{P{l-t)+at)\A^\
12
i
12
+ (a(l - t) + /ft) U<2>
< (t(l - t)(/3 - a ) 2 + 2a£) det A,
and CQW(A,t)
= +oo
otherwise. Moreover, optimal gradient Young measures in the above inf are Vi,j = SijdAd + (1 - s*,i) ( z
—SAa,t
+ —T^i
Wi,t J .
* + 3,
230 det (Aa>t - Af,>jtt) = 0, ( i
det
t
Ap,i - —
.
Aa,t
1
—
Si i — t ,
—*!•
\
Af,Jit 1 = 0,
The computations involve in this result can be found in [13]. It is interesting to realize that optimal microstructures are second-order rank-one laminates.
6. Concluding remarks We would like to finish our contribution by briefly comment on some issues. It is worthwhile to notice that all functions QW(A) given explicitly before are quasiconvex, and, in particular, the set where they all are finite is a quasiconvex, nonconvex set, in the sense that weak limits of gradients taking values on it, will also take values within the same set. In fact, our computations amount to realizing that the polyconvexification and the rank-one convexification coincide as is typical in explicit results of this nature (see for instance [7], [8] among others). It can actually be shown that the set T can also be determined by using the G-closure of {al,/91} (1 is the identity matrix). In fact, it is true r = (A € M 2 x 2 : there exists 7 e G[al,/31] such that -yAw + TA{2) = o} . This relationship was noticed in [17] where the idea of introducing a potential to replace the differential constraint was indeed used and indicated. Notice how CQW(A, i) is polyconvex in A and convex on t. Joint convexity, however, is much more than these two separate convexities. If we further introduce the function
h{A) = (a + /3)det,4-a^U (1) | 2 - U<2>|2 then the integrand in Theorem 5.2 can also be given by CQW(A,t)
=
*_
(p2 \A^\2
+ |^|
2
- (at + 0(2 - t ) ) det A
if g(A) > 0,
h{A) > 0,
n(A)
One main objective after having the explicit computation of the previous section is the numerical approximation of optimal structures via relaxation, i. e. approximate the optimal solutions of m i n i f CQW(VU(x),t(x))
dx : U G -ff^fi), Um = u0 on dtt, 0
t(x)dx = \ \ ,
231 and use the local information coming from Theorem 5.2 to build a global picture of optimal structures all over Q. We hope to address this in the near future ([1]). Acknowledgements I would like to thank the organizers of the Fourth European Conference on Elliptic and Parabolic Problems for inviting me to contribute to this event.
References [1] Aranda, E., Donoso, A., Pedregal, P., in preparation. [2] Bellido, J. C , and Pedregal, P., in preparation. [3] Cherkaev, A., 2000 Variational Methods for Structural Optimization, Springer-Verlag, New York. [4] Dacorogna, B. 1989 Direct methods in the Calculus of Variations, Springer. [5] Fonseca, I., Kinderlehrer, D., Pedregal, P. 1994 Energy functonals depending on elastic strain and chemical composition, Calc. Var., 2, 283-313. [6] Grabovsky, Y., 2001 Optimal design problems for two-phase conducting composites with weakly discontinuous objective functionals, to appear in Advan. Appl. Math. [7] Kohn, R. 1991 The relaxation of a double-well energy, Cont. Mech. Thermodyn., 3, 193-236. [8] Kohn, R. V. and Strang, G. 1986 Optimal design and relaxation of variational problems, I, II and III, CPAM, 39, 113-137, 139-182 and 353-377. [9] Le Dret, H., and Raoult, A., 2000 Variational convergence for nonlinear shell models with directors and related semicontinuity and relaxation results, Arch. Rat. Mech. Anal., 154, 101-134. [10] Lipton, R. and Velo, A., 2000 Optimal design of gradient fields with applications to electrostatics, in Nonlinear Partial Differential Equations and Their Applications, College de France Seminar, D. Cioranescu, F. Murat, and J.L. Lions eds, Chapman and Hall/CRC Research Notes in Mathematics. [11] Murat, F. 1977 Contre-exemples pour divers problemes ou le controle intervient dans les coefficients, Ann. Mat. Pura ed Appl., Serie 4, 112, 49-68. [12] Pedregal, P. 1997 Parametrized Measures and Variational Principles, Birkhauser, Basel. [13] Pedregal, P. 2001 Constrained quasiconvexification of the square of the gradient of the state in optimal design, submitted. [14] Pedregal, P. 2000 Optimal design and constrained quasiconvexity, SIAM J. Math. Anal., 32, 854-869. [15] Pedregal, P. 2001 Fully explicit quasiconvexification of the mean-square deviation of the gradient of the state in optimal design, ERA-AMS, 7, 2001, 72-78. [16] Pedregal, P. 2000 Constrained quasiconvexity and structural optimization, Arch. Rat. Mech. Anal., 154, 325-342. [17] Sverak, V., 1994 Lower semicontinuity of variational integrals and compensated compactness, in S. D. Chatterji, ed., Proc. ICM, vol. 2, Birkhauser, 1153-1158.
232 [18] Tartar, L. 1994 Remarks on optimal design problems, in Calculus of Variations, Homogenization and Continuum Mechanics, G. Buttazzo, G. Bouchitte and P. Suquet, eds., World Scientific, Singapore, 279-296. [19] Tartar, L. 2000 An introduction to the homogenization method in optimal design, Springer Lecture Notes in Math., 1740, 47-156.
A regularity criterion for the angular velocity component in the case of axisymmetric Navier-Stokes equations Milan Pokorny* Math. Institute of Charles University, Sokolovska 83, 186 75 Praha 8, Czech Republic Email: [email protected]
Abstract We study the instationary Navier-Stokes equations in the entire three-dimensional space under the assumption that the data are axisymmetric. We improve the regularity criterion for axisymmetric weak solutions given in [10].
1
Introduction
Let us consider the Navier-Stokes equations in the entire three-dimensional space, i.e. the system of PDE's Su dt
3
• u - V u - i / A u + Vp = o | i n ( 0 ) r ) (1.1)
divu = 0
J
u(0,x) = u 0 (x)
inE ,
3
3
where u : (0,T) x R i-»- K is the velocity field, p : (0,T) x K3 t-> 1R is the pressure, 0 < T < oo, v is the viscosity coefficient, u 0 is the initial velocity and the forcing term is, for the sake of simplicity, considered to be zero. The question of smoothness and uniqueness of weak solutions to (1.1) is one of the most chalenging problems in the theory of PDE's. The solution is known to be unique (in the class of all weak solutions satisfying the energy inequality) if it belongs to the class LT'S(QT) with ^ + | < 1, r € [2,+oo], s G [3,+oo] (see [12], [11]). Moreover, if the weak solution belongs to Lr-S(QT) with - + - < 1, r e [2,+oo], s e (3,+00] and the input data are "smooth enough" then the solution is smooth. (See [2], [4]). In the case of the planar flow the weak solution is known to be unique and smooth as the data of the problem allow (see [8], [5]). Thus a natural question, namely what 'Supported by the Grant Agency of the Czech Republic (grant No. 201/00/0768) and by the Council of the Czech Government (project No. 113200007)
233
234 can be said about the axisymmetric flow, appears. The first results in this direction were obtained in the late sixties for ug = 0 (see [6], [14]) and recently also in [7]. The case ug ^ 0, including the 2-axis, was for the first time considered in the paper [10] where for ur e Lr<s(QT) with f + f < 1, r e [2, +oo], s e (3, +co] or for ue £ Lr>s(QT) with =- + § < £ , r e [f ,+oo], s e [6,+co] and f + | < l _ A, r e [10,+oo], s e ( f ,6) the smoothness and thus also the uniqueness in the class of weak solutions satisfying the energy inequality was obtained. See also [1] where the authors give some more smoothness criteria for the vorticity components. Note that the criterion on ur probably cannot be improved by the present technique while the regularity criteria on ug are not optimal from the scaling argument. Here we would like to improve the regularity criteria on ug in such a way that it (almost) undergoes the correct scaling. Our main result is as follows T h e o r e m 1 Let u be a weak solution to problem (1.1) satisfying the energy inequality with u 0 e W 2 ' 2 (R 3 ) so that Vu 0 € L1(R3) and (u0)gr € L°°(R3). Let u 0 be axisymmetric. Suppose further that the angular component ug of u belongs to I/'"(QT) for some r € (2,+oo], s G (4,+oo], 2 + | < 1. Then (u,p), where p is the corresponding pressure, is an axisymmetric strong solution to problem (1.1) which is unique in the class of all weak solutions satisfying the energy inequality. Note that under a axisymmetric solution we understand a pair (u,p) such that in cylindrical coordinates (r, 0, z), r e [0, oo), 9 e [0,27r) and z £ f , uT1 ug and uz, considered in cylindrical coordinates, are independent of 6, and p, written in cylindrical coordinates, is also independent of 9.
2
Preliminaries
Denote by (ur,ue,uz) the cylindrical coordinates of the vector field u and by (uir,uig,uiz) the cylindrical coordinates of curl u, i.e. u)r = — ^jjf-, w« = ^ — ^jf- and uz = \-§^(rwo) for u an axisymmetric field. Moreover, let w = (wi, w2,w3) denote the cartesian coordinates of curl u. We will use the standard notation for the Lebesgue spaces Z^R 3 ) equipped with the standard norm || • ||PiR3 and the Sobolev spaces Wk'p(Mi) equipped with the standard norm || • ||fclP;H3. By Lr's(QT), QT = (0, T) x R3 we denote the anisotropic Lebesgue space Lr(0,T; L S (R 3 )). If no confusion can arise we skip writing R3 and QT, respectively. Vector-valued functions are printed boldfaced. Nonetheless, we do not distinguish between L«(R3)3 and L«(R3). In order to keep a simple notation, all generic constants will be denoted by C; thus C can have different values from term to term, even in the same formula. By Du we mean the gradient of u expressed in the cartesian coordinates, while Vu r denotes the derivatives with respect to r and z only (u is axisymmetric). Similarly for ue and uz. We will use the following inequalities (for the proofs of Lemmas 1-3 see [10])
235 Lemma 1 Let u be a sufficiently smooth vector field. Then there exists a constant C(p) > 0, independent of u, such that for 1 < p < oo ||Z>u||,
(2.1)
Lemma 2 Let u be a sufficiently smooth divergence-free axisymmetric vector field. Then there exist constants Ci(p), i = 1,2 and Cj, j = 3 , . . . , 7 such that for 1 < p < oo
llVtirllp+pJ
(2.2)
ii r lip
I 9 /wr ) | | < 0.||Z)2u||,, \dr\rJ\\p/ lip HVu,||p + p | |
|£(?)|,<<*l|2*% C 2 (p)||.D 2 u||p<||^|| + p | + ||Vu,r||p + II r lip II r lip ||Vw,||, + ||Vw,|| p
(2.3) (2.4)
(2.5) (2.6)
Lemma 3 Let (u, p) be an axisymmetric smooth solution to the Navier-Stokes equations such that (u0)er e L°°(R3). Then also uer G L°°(QT)Lemma 4 Let u be a sufficiently smooth axisymmetric vector field. Then to every e G (0,1] and 1 < p < oo there exists C(e), independent o/u s«cft i/iai ||rp+2-E
(2.8)
i VII r lip II VI r I / 112/ Proof. Due to the Hardy inequality (see [3]) we have rmPr^~^(r)dr
f ° | | - ( | ^ | f V ( 0 ) |V + e rfr 70 lor VI r I /I
for t/)(r) a positive cut-off function equal to 1 near r = 0 and equal to 0 for r > 2. Inequality (2.8) is thus shown. • We will also use the following regularity criterion proved in [10]. Lemma 5 Let u be a weak solution to problem (1.1) satisfying the energy inequality with Uo £ W2'2, axisymmetric and divergence-free. Suppose that ur 6 Lr,s, r e [2,oo), s e (3, oo), - + - < 1. Then (u,p), where p is the corresponding pressure, is an axisymmetric strong solution to problem (1.1) which is unique in the class of all weak solutions satisfying the energy inequality. Remark 1 In fact, due to the absence of the right-hand side, u € C°°([5, oo) x R3) for any 6 > 0.
236 The improvement of the results from [10] is based on Corollary 1 below. We will use the following result concerning the Muckenhoupt classes (see e.g. [13] for the definition and other properties) Lemma 6 Let r denote the distance of a point in R3 from the z-axis. Then rq 6 Ap for any 1 < p < oo if —2 < q < 0.1 Recall that for divergence-free functions the gradient can be computed from the vorticity by means of a singular integral operator with properties analogical to the Riesz transform. For such operators (see e.g. [13]) we have the LP weighted estimates exactly for the weights from Av classes, 1 < p < oo. Thus we get from Lemma 6 Corollary 1 Let 1 < p < oo and 0 < q < -. Then there exists a constant C(p,q) such that llr1+9|lp_
y
'\\ri\\P
v
'
Finally, we will need the following result on the integrability of gradients of u with some p's less than 2 (see [9] for the proof) Lemma 7 Let moreover Du0 6 L 1 . Then the weak solution to (1.1) also satisfies Du e L00'1 and D2u e IS'1, 1 < p < 2.
3
Proof of Theorem 1
The proof is based on the continuation argument. We have the following (more or less standard) result Lemma 8 Let Uo £ W2'2. Then there exists to > 0 and (u,p), a weak solution to system (1.1), which is a strong solution on the time interval (0,io)- Moreover, if Uo is axisymmetric then also the strong solution is axisymmetric. Now let u 0 be as in Lemma 8 (axisymmetric). We define: t* = sup < t > 0; there exists an axisymmetric strong solution to (1.1) on (0,t) [ It follows from Lemma 8 that t* > 0. Now let (u,p) be a weak solution to the Navier-Stokes system as in Theorem 1. Due to the uniqueness property, it coincides with the strong solution from Lemma 8 on any compact subinterval of [Q,t*). There are two possibilities. Either i* = oo and we have the global-in-time regular solution, or t* < oo. However, we will exclude this possibility by showing that ur satisfies on (0,£*) the assumptions of Lemma 5. To this aim we will essentially use both the information about the better regularity of one velocity component and the fact that the solution is axisymmetric. There are also some positive values of q for which rq e Ap but we do not need it here.
237 Now, let 0 < t < t*. Then on (0,1) ( u , p ) is in fact a strong solution t o t h e NavierStokes system. It is convenient t o write t h e Navier-Stokes system in t h e cylindrical coordinates for our purpose. T h u s ur, ue, uz a n d p satisfy in ( 0 , t ) x R 3 t h e system dur
^ du ourr i __ du durr r z at dr dz adue due_ due r z "at dr dz duz duz r z dt dr i
11 r
f[1 l 9o , . du durr^ lr dr dr 1 [1 d due. T r Ir dr dr ^ £ , ^ P _ r 1 d , duz. dz dz lr dr dr du.1 ur duz 0 V dr + -r + ^ dz = 22 e
, dp dp dr
:
2 d6' uTur u„,rr~\, dz2 r2l 2 d ue ue~\ _ dz2 r2\ 2 d uz-i _ dz2 J
Moreover, t h e vorticity components satisfy in (0, t) x M.3 dujr
dui dui ou>rr ou)rT z dt dr dz due , dwe due — - + ur-— + uz— dt dr dz
^£
+ u
^£
+ u
du ourr dr ur
^i_
du rl ourr r d /( du> "''Mr\ dz Lr dr\ dr J 2 rl d ( due\ , we + -uoUr - v —5-I r-r+ r r lr or \ dr 1 dUz
^
^'u -J-^-f
^!£l)
d2wr dz2 2 d wg -5-5 dz* 9
ur --k\
V]
=0
ue~\ 5- = 0 rlJ =o
z dt dr dz dz dr lr dr\ dr J dz2 i T h e main idea of the proof of Theorem 1 is very similar t o t h e idea presented in [10]; we will namely combine t h e estimates of uje in L°°'p with t h e estimates of ^ in L°°'q with the idea t o get an estimate for uT which is in t h e range * + | < 1 (i.e. p > §). Let us take p e (1,2) a n d multiply t h e equation for we by ,J%-P a n d integrate (with respect t o t h e measure rdrdz). We get (in what follows, f f denotes f_oof0 frdrdz)
+! |v |! a >*+»/^ W '"' ' a " n ,„ /ro2 due we P
P
/>r + / I..
IB
,
Ug_
r " dz
(3-1)
\we\2-v'
(Note t h a t all terms are finite because u)e(t) G L1 n L2.) Next, let us multiply t h e equation for u>e by tp(r)\iM-\q~2^-r=r, 5 > 0 and ip(r) a cut-off function equal t o zero near r = 0. Now we integrate t h e equality over M.3 then pass first with ip(r) t o t h e identity function a n d finally with S to zero. Note t h a t we cannot take directly S = 0 as some integrals cannot be controlled, cf. [7]. We get 1 dnugw
4(g-l)
[\
(\we\l\\2
I [2
3a,
%
ll
.
To prove Theorem 1 we can now proceed in the following way. First, we sum (3.1) and (3.2) and estimate the first term on the right-hand side of (3.1) by (3.1) and (3.2) with g = | p (see Proposition 1 below). Next, the other two terms on the right-hand sides of
238 (3.1) and (3.2) can be estimated by the left-hand sides and by another terms containing some powers of u$ and r. All these terms can be again estimated testing the equation for ue by an approprietly chosen function and we will be finally left with estimating of ue \zfur r
/ ' y/r\
(Actually, up to some e > 0.) This last term will be possible to estimate under the assumption that UQ € Ll's for | + | < 1. For s > ™ we immediately obtain p > f; thus ur is bounded in L°°'3+s for some 5 positive and we finish the proof using Lemma 5. For 4 < s < Y w e w m S e * a n estimate for some p e (|, §); nevertheless, if we test again the equation for wo by \ue\"^+5LUo for some 6 > 0 chosen apropriately we will be able to estimate directly all terms on the right-hand side of the inequality corresponding to (3.1). Unfortunately, the case 3 < s < 4 remains open. Proposition 1 Let | < p < 2. Then for any S > 0 there exists C(8) such that
with q = | p . Proof. Take a = ^ ,
p =^ . .
Then £(p-i)'
Using the interpolations ||W0||
n w e i i i - f ||W9||I
II r IL ~ II r IIfP II r II |p II^II^D^INIlllKllp^ where the first one holds true for 1 < p < 3 and the other one for | < p < 2, we get
^^MIIWI^ITIIJITI?
(3.3)
<S\\
Now, the first term on the right-hand side of (3.3) can be included into the left-hand side of (3.2) while the other term is estimated using the Gronwall inequality. Next we want to estimate the other term on the right-hand side of (3.1). However, this term is of lower order and thus, using the fact that ugr G L°°(QT) (see Lemma 3) it is not difficult to estimate it using similar arguments as for I3 below. Further, let us look at the last term coming from (3.2), namely r I
r r dz\ r '
239 Recall that q = | p . Choose e > 0. Then d I
uj
dz I r2-ns-23=l We require %& = 4 - 2a and %& - 2 = 4 - 2/? + 2e - ^ z i l a n d get 12-5p " = —g 5p(l + e) + 12 12 - 5p lOp 6 ' thus, using Lemma 4
\r\
[\v(\u,\*"M\
lhl c
f\U8\ip+ +
- {Jr[\v\
Ml
f
" i f\v(
+
)\ J\v\
+
J 7i^^ l
MlP
\
r\7&w)
Combining all the estimates from above and using the Gronwall inequality we get
(''"•HI?l> + /7( |V |!+ ^ + KI?l*')f)
,,
In order to estimate the last two terms, we test the equation for ua by ^"'J1'' "". We get d f Mi' 5pcd i / rfp(i+o
+
(fp)2-(|P)2(I + ^
(b) 2
+
/• "J
4(fp-l) /•• / |U,|JP vi (|p)2 y I V r AP(i+ £ );l l^ft"
=
r |p(i+^)+2
/_ I
i+£\ /• Mb ur 2 ) J rfp(i+<0 r
i.e. together with (3.4)
( M H I ^'sP'
kl§P re
M*F r|p(l+E)+2
Denote by 74 the last integral on the right-hand side. Take /3 = § and a = 1 — f- — | and choose a £ (1,2). (The aim will be to take a as close as possible to 2.) Then we can estimate 74 as follows h<
3p I "HI 2 II l"«l
« e l | p II".,
I r ILII r|p(i+e) Hi II r|p(i+£)+2 l l J ' ^ ^ d - ^ ) - !
240 Using the fact that o < 2 we get (see Corollary 1)
II —II —II
II T \\a
and employing the interpolation inequality 5p-2o
II r \\a
M r iijfp
6q-5p
II r II| P
we get, using the Young inequality and substituting back the values of a and , r II fp . 2°Pa s6o-5p r^
Thus if ue e £
M
llrfj>(i+0+2||i /lli^llip
II biolfP
VII Ml r/ III? II gP
lI lI yI -eI^P C H - O I I I /
with 20pa :: r-r T7T- , Wpa — 5pas — 18a + 15p '
t = —
s =
10pa 6a — 5p
(note that \ + f = 1 - §) we get
/'ll,, ||P i II ^ H * " i II l ^ l 1 " II W vIWI?
+ ||-|| ip + ||^pi^)||JW ^ / 7 ( | V ( W l ) P + ^ + |v(|^|AP)|2
(3.5)
+| V V^(T^)J| + r|p(i+E)+2J ^ C ( u °) with | < p < 2. Now, as a < 2, we get that for s > y there exists p > | such that s = g ^ . Thus, if uB e Ll's with f + f < l a n d s > f then ue e L°°'p for some p > § and u r G L°°>PI for some pi > 3. Using Lemma 5 we see that the weak solution is in fact globally smooth. If 4 < s < j we can only dispose with (3.5). Note that p e ( | , | ) in this case. Nevertheless, we show that using (3.5) we can finally get an estimate of uig in L°°'? with V]aj0|* in L2'2 (which implies w« 6 i J ' z ) . To this aim, let us test the equation for we by \UQ\~IUJO. Then 2 d
n nl ,- /fM ^ M + ,
3S
l
'„
+
^/|v(W!)| 2
U. I f 2 u dug. . _ 2W . 2.3 I — «I. \=h + hr N \ +\ J/ r~ 0^—N az r
To estimate Is recall that we know that there exists r]0 > 0 such that ^ is bounded in L00-1"^ and in Li+».3(i+i) for 0 < ?7 < 7y0. Thus
/5
II r Il3(l+lj)
ii,,
II
|K||je±sl||W,||J l+2l
3
- C |lTll3(l + ,) l | W , ' l l f
2
2tj
2n
l|Wr 12
'
241 and therefore we can estimate I5 using the Gronwall inequality. Finally, to estimate I6, we I l1+r> will use the fact that there exists r?i > 0 such that for all 0 < r\ < T)\, p £ is bounded in L°°'2 and its gradient in L 2,2 . 2 Now 1+
T ^ 11 IIIIIVT/I Uu»e \\l+ni\\\ \\\
sll
\\Jr\ s/r\
61
III uue III <>I ' "'''!!?
/IkllU/r /Il2lll v ^'l
Hi
and thus 76 can be estimated by means of the Gronwall inequality. The proof of Theorem 1 is finished.
References 1] Chae D., Lee J.: On the Regularity of Axisymmetric Solutions to the Navier-Stokes Equations, (preprint). 2] Galdi G.P.: An Introduction to the Navier-Stokes Initial Boundary Value Problem, In: Fundamental directions in Mathematical Fluid Mechanics, editors G.P. Galdi, J.G. Heywood, & R. Ranacher, Birkhauser Verlag (2000). 3] Hardy G.H., Littlewood J.E., Polya G.: Inequalities, Cambridge University Press (1952). 4] Kozono H., Sohr H.: Remark on uniqueness of weak solutions to the Navier-Stokes equations, Analysis 16 (1996) 255-271. 5] Ladyzhenskaya O.A.: The mathematical theory of viscous incompressible flow, Gordon and Breach, New York-London-Paris (1969). 6] Ladyzhenskaya O.A.: On the unique global solvability of the Cauchy problem for the Navier-Stokes equations in the presence of the axial symmetry, Zap. Nauch. Sem. LOMI 7 (1968) 155-177 (in Russian). 7] Leonardi S., Malek J., Necas J., Pokorny M.: On axially symmetric flows in R3, ZAA 18 (1999) 639-649. 8] Leray J.: Etude de diverses equations integrates non lineaires et de quelques problemes que pose I'hydrodynamique, J. Math. Pures Appl. IX. Ser. 12 (1933) 1-82. 9] Lions, P.-L.: Mathematical Topics in Fluid Mechanics, Volume I: Incompressible Models, Clarendon Press, Oxford (1996). [10] Neustupa J., Pokorny, M.: Axisymmetric flow of Navier-Stokes fluid in the whole space with non-zero angular velocity component, Math. Boh. 126, No. 2 (2001). 2
Actually, we have 1"'J',, , 6 L00,2 and its gradient belongs to L2,2 for some e > 0; but this information is stronger since the main problems are near r = 0.
242 [11] Prodi G.: Un teorema di unicita per el equazioni di Navier-Stokes, Ann. Mat. Pura Appl. 48 (1959) 173-182. [12] Serrin J.: The initial boundary value problem for the Navier-Stokes equations, In: Nonlinear Problems, ed. Langer R.E., University of Wisconsin Press (1963). [13] Turesson B.O.: Nonlinear Potential Theory and Weighted Sobolev Spaces, Springer Verlag, Lecture Notes in Mathematics (2000). [14] Uchovskii M.R., Yudovich B.I.: Axially symmetric flows of an ideal and viscous fluid, J. Appl. Math. Mech. 32 (1968) 52-61.
A Comparison Principle for the p-Laplacian Arkady Poliakovsky and Itai Shafrir Department of Mathematics Technion-I.I.T 32000 Haifa, Israel
1
Introduction
The purpose of this note is to present a quite general comparison principle between sub and super solutions for singular equations involving the p—Laplacian. We are motivated by two such results in the case p = 2. The first is the following theorem of Agmon [2, Theorem 2.7] that we give in a slightly modified form. Theorem 1.1 (Agmon). Let a(x) €_L]0C(R.N \{0}), R>0 andu,v e with QR := {\x\ > R} and v > 0 on QR, satisfy (in the weak sense), —At; — a(x)v > 0 >
—AM
— a(x)u
C(nR)nHlc(nR),
in QR.
Suppose also that v(x) > u(x)
on {\x\ = R},
and that for some a > 1 we have, liminfiT 2 / K
^°°
u2 = 0.
(1.1)
J{K<\x\
Then, v > u in QR. Adapting Agmon's method to a different setting, Marcus, Mizel and Pinchover proved in [5, Lemma 8] the following result (again we slightly modify the statement), Theorem 1.2 (Marcus-Mizel-Pinchover). Let Q be a proper subdomain ofRN with compact boundary and a(x) e L}0C{Q,). Set 8{x) = dist(x, dQ) and assume that /3 > 0 is such that S j := { i 6 fl : 5(x) = 6} / 0. Suppose that u,v e C^p U E^) n H}0C{Q,^), with Qp := {x e CI : 6(x) < /?} and v > 0 on ftp U E^, satisfy (in the weak sense), —Av — a(x)v > 0 > —Au — a(x)u
in Up.
Suppose also that v(x) > u(x)
on Ejj,
and that for some a > 1 we have, liminfr--2 / r_+0
u2 = 0. J{xm-.r<S(x)<ar}
Then, v > u in Qp.
243
(1.2)
244 In [9, Proposition A.l] Shafrir proved a generalization of Theorem 1.2 for any 1 < p < oo. The purpose of this note is to present a comparison principle which unifies and generalizes all the above mentioned results for any 1 < p < oo. We basically follow the strategy of [9], but we make a new key observation about a choice of test functions which enables us to improve the known results even in the case p = 2. In particular, this improvement shows that in both (1.1) and (1.2) it is enough to require that the corresponding limits are finite (i.e. not necessarily zero) in order for the conclusions to hold. We also present some applications for uniqueness and Liouville type results. We begin with some notations and definitions. In the sequel let Q be a proper subdomain of R^ whose boundary dfl is a disjoint union of two compact sets T0 and IV We assume that I \ ^ 0, but To may be empty. When To 7^ 0 we define the distance function 5a(x) = dist(x, r 0 ) , Va; G M.N. In case T0 = 0 it will be convenient to set 50 = 0. For p e (1,00) we denote by Apu = div(|Vu\ p ~ 2 Vu) the p-Laplacian of u. Given a(x) 6 L|oc(Q) we shall say that a function v G W^(Q) nC(Q) is a super solution for the equation, - A p u - o ( x ) | u | p " 2 u = 0, inQ, (1.3) if f \Vv\"-2VvV4> - a{x)\v\<>-2v(t> > 0,
for all 0 < <j> e CC°°(Q).
(1.4)
The notions of subsolution and solution are defined similarly by replacing ">" in (1.4) by "<" and "=" respectively. For 0 < a < b we denote, Q K 6 ) = { i £ f i : 60(x) e (a,b)}
and Q w > ) = {x e « : \x\ 6 (a, 6)}.
Consider first p € (1,2] and a function u e W^{Q). We shall say that u satisfies condition (Co) if for some 0 < a < 1 there exist for each n > 2, n positive numbers {r
.,
r}+i^ffrj
(n)
-,
„
1
.
J= l,...,n-l,
i™^i:/
i^)'=o.
with lim rj™ = 0 and n—>oo
(1.5)
For p > 2 we replace in the definition (1.5) by
m jff-isjL^®"-' Condition (C0) is a growth condition for v near To (it is trivially satisfied when F 0 = 0). We now define analogously a growth condition at infinity (Coo), in the case of unbounded Q.. As above, consider first the case p G (1,2] and a function u € W lo f(0). We shall say that u satisfies condition (Co) if for some 0 < r < 1 there exist for each n > 2, n positive numbers {p^'}™=1 such that, (n)
^
p)+i<rp)',
in)
-,
j=
_
1
l,...,n-l,
245 with lim p„ = oo and n—»oo
limi^/
w
,„, ( M y = o .
(i.7)
For p > 2 we replace in the definition (1.7) by
J5S, i g /^M', (R)' + ^ g /0<^M->>'v"r (R) 2 = °"
(L8)
Our main result is the following. Theorem 1.3. Consider ft as above, p 6 (l,oo) and a(x) e L^Q). Suppose that u, v e C(ft U Ti) n W,^?(ft) are such that v is a super solution for (1.3), v > 0 onQuTi, u is a subsolution for (1.3) satisfying condition (C0), j / r 0 ^ 0, and condition (Coo), */ft is unbounded. Assume also that v>u onTi. (1.9) X/ien, ii > u
2
in
ft.
(1.10)
Proof of t h e main result
For the proof we use two basic tools that we shall now recall. The first is a Picone identity for the p-Laplacian, which is due to Allegretto and Huang [3, Theorem 1.1]. We summarize it in the next lemma. Lemma 2.1. Let G be a domain in RN. For u, v e C(G) n W^(G) v > 0 in G, denote
such that u > 0 and
L{u, v) = IVu\p + (p- 1)—IVv\" - p—rVwlVv\"-2Vv VP v? R(u, v) = |Vu| p - v ( - ^ — ) \Vv\"-2Vv .
, (2.1)
Then L{u,v) = R(u, v) a.e. in G. Moreover, L(u, v) > 0 a.e. in G, and L(u, v) = 0 a.e in G if and only if u = kv for some constant k. The second tool is given by the following simple inequalities from [9, Lemma A.4]. Lemma 2.2. (i) If p > 2 then \z1 + z2\"-\z1\',-p\z1r2z1-z2 (ii) Ifp<2
< ^ ^ ( k i | + \z2\)p-2\z2\2,
Vzuz2 € R*.
(2.2)
then there exists a constant jp > 0 such that \zi+za?-\z1Jf-p\z1JT2z1-Z2<'yP\zt^,
VZl,z2eRN..
(2.3)
246
Proof of Theorem 1.3. Assume by negation that (1.10) does not hold. Then, the open set Q+ = {x e Q : u(x) > v(x)} is nonempty. Put K := sup{log(u(a;)) - log(u(i)) : i 6 ( l , u(x) > v(x)} 6 (0,oo],
(2.4)
1
and fix a positive constant 6 such that 56 < K. Let S s C (R) be a non-decreasing function such that 9(i) = 0 for t < 26,
9{t) = 1 for t > 56 and 6'{t) > 0 for 36 < t < 46.
(2.5)
Let £ = X[u>v] & (log(u/i>))
(2.6)
in O.
If T0 ^ 0 then for any n > 2 we define, for r > r\.W
1 log(r/r< n ) )-log
,' . /»M = < ( l - ^ + ^^St^ 1-
j/n
0
(„)
(„)
.
for crr„- ' < r < rl- , 1 < 7 < n for r ^ < r < ffrf >, 1 < j < n - 1, for r < or)? ,
(2.7)
where {ry'}^=1 and a are given by condition (C0). If r 0 = 0, simply set /„ = 1. Similarly, if Q. is unbounded, using {py }™=1 and r given by condition (Coo), we define, for p > P W >
0 9n{r) = <
j/nj/n 1
/p J) ' o g ( rn|logr| r ~'° g T
for r p f < r < p f , forp^^r^Tp^, for r < rp.,(<•)
l<j
(2.8)
In case fi is bounded, simply set gn = 1. Clearly, both /„ and gn are Lipschitz functions on R. There exists n 0 such that: (i) 9 0 C B oo (0) for n> n 0 . (ii) If T0 + 0 then dist(r x , T0) > r[ n) for n>n0. For every n > rc0 we define a Lipschitz function ipn on RN by ^ ( * ) = /n(«o(x))-ft,(N).
(2.9)
Using (1.9) and our assumptions on r and ipn we get that the nonnegative function
has a compact support in Q + . Therefore, w € W01,!>(fi) and we can use it as a test function in the inequality satisfied by v to get [ i\Vv\*-2Vv-v(^^j n+
> f a^u"in+
[il%up\V\ogv\p-2VlagvVZ. n+
(2.10)
247 By (2.10) we infer that 0 < f L{ijnu,v)Z= Jn+
f RWnU,v)£ Jn+ Jn+ < [
Jn+ \V(4>nu)n - f
+ I
VW i J o^JuPf
^ p u p |Vlog«;| p - 2 Vlogi>- Vf,
where L(u,v) and R(u, v) are defined in (2.1). Similarly, the nonnegative function has compact support in fi+ and therefore z 6 W0'P(Q). Testing the inequality satisfied by u against z yields (-div(|V«| p - 2 Vu) - a|M|p-2u)i/<>£
0> /
(2.12) = / £|Vu| p - 2 Vw • V « « ) + / V > | V u r 2 V u - V £ - / < # > p £ . Jn+ Vn+ ./«+ Subtracting (2.12) from (2.11) we finally obtain that f
VE« p [|Vlogu|' , - 2 Vlogu- |Vlogu| p - 2 Vlogi;] • V£ < [|V(^«)|p-|VWr2V«-V(^»]e.
[
(2.13)
Next we note that |V(V>n«)|P-|V«r2VM-V(fr) = |^„V« + «V^„| P -p\^nVu\"-2(^nVu)
• {uV^n) - |V>nVu|p.
Therefore from Lemma 2.2 we infer that (7pWp|VVn|p |V(^ n u)| - \Vu\"- Vu • V ( ^ » < I {C(u"\ViPn\" + HVV„|) 2 |V«| p - 2 ) p
ifpe(i,2],
2
if p > 2. (2.14)
Assume first that p e (1,2]. Using (2.13) and (2.14) we infer, f VS« P [|Vlogu| p - 2 Vlogu-|Vlogu| p - 2 Vlogt;] • Vf
/•
/|u|y>
C A
f
(\u\y
n
248 where we used (1.5) and (1.7). The computation in the case p > 2 is almost identical; we need only to use (1.6) and (1.8) instead of (1.5) and (1.7). Hence in all cases we obtain that limsup /
ipy[\Vlogu\p-2Vlogu-
|Vlogv| p - 2 Vlogu] • V£ < 0.
(2.15)
Since ipn(x) —¥ 1 a.e. in Q we get from (2.15) and Fatou Lemma that /
w p [|Vlog«| p ~ 2 Vlog«- |Vlog*;| p - 2 Vlogu] • Vf < 0.
(2.16)
Note that in 0+ V£ = 6'(\og(u/v)) ( V l o g u - Vlogu), and that {\Zl\"-2Zl - \z2f~2z2)
• ( 2l - z2) > 0,
Vzi,
z2eRN,
with equality if and only if z\ = z2 (by the strict convexity of the function f(z) = \z\p). Therefore equality holds in (2.16) and consequently, from (2.5) we deduce that u = CiV on each component Ei of the set {x €fi;\og(u(x)/v(x)) € (36,46)}, for some positive constant c;. But this is clearly a contradiction since the image of the function log(u/v) contains the interval (36,46) by (1.9) and (2.4). D
3
Some applications of the comparison principle
This section is devoted to some consequences of Theorem 1.3. We first show how it can be used to deduce improvements of Theorem 1.2, Theorem 1.1 and [9, Proposition A.l]. We start with a simple lemma. Lemma 3.1. Let 0 with boundary 9 0 = r 0 U 1^ be as in the previous section and let u e W^(Q). If for some a e (0,1), liminf/
(^) P
j/p 6(1,2],
then u satisfies condition (Co). Similarly, if for some r € (0,1), liminf/"
(My
if p
e
(1,2],
liminf/ (M)P + | V u r 2 ( M ) 2 < 0 0 , l>^°° JiHrp,p) \\X\J \\X\J then u satisfies condition (Coo).
ifpe(2,oo),
249 Proof. It is enough to prove the first assertion for p 6 (1,2]; all the other cases are similar. By assumption we can find a sequence {rn}'^Ll such that rn \ 0, r n + 1 < arn, Vn and
Vn.
, V 5o I
Jn,
Clearly lim,,-^ ^ S?=i an+j — 0> s o setting rj™ = r„+J- for all n > 1 and 1 < J < n we see that (1.5) is satisfied. • An immediate consequence is the following improvement of Theorem 1.1. Corollary 3.1. Let a(x) G iJ 0C (R N \ {0}), R > 0, QR := {|x| > R} and let u,v e C(QR) n W^f(flfl) be respectively a sub and super solutions for (1.3) on 0 = f2R. Assume that v > 0 on QR, U satisfies condition (C^), and i/ta£ «(x) > u(x)
on {\x\ = R}.
Then, v > u in QR. In particular, the result holds for u which satisfies for some r G (0,1), liminf [ e^™
liminf/
(M)P
0-PrY < o o ,
J{TP<\X\
+
if pe (1,2}
M*K
|vur=(M)2<0O)
,/p6(2,oo)
Proof. The assertions follow easily from Theorem 1.3 and Lemma 3.1 applied with Q. = QR, r 1 = {|a;| =R} and T0 = 0. D In the next proposition we present some common situations in which condition (Co) is satisfied. Proposition 3.1. The function u satisfies condition (Co) if either of the following holds: (i) T0 is a compact Lipschitz hypersurface and for some 0 < r < dist(r0,Ti) we have: u e W l j , ( n r ) with Qr := {x G O, : S0(x) < r} and for some * G C°°(Q) such that $ s 0 on Q\ QT/2 and f s i on Or/3, we have \&u G W0'p(ilr). (ii) To = {^o} and for some 0 < r < dist(x0,Ti), and \P G Cf(BT(xo)) such that "9 = 1 on BT/2{x0), we have ^u G W0'p(BT(x0) \ {x0}). Proof, (i) Since Hardy inequality holds for domains with Lipschitz boundary (see [7]) we have, for some constant c > 0, f |V(*u)|" > c f (|*u|A>r.
(3.3)
It follows from (3.3) that U/5Q is an V function in a neighborhood of T0. In case p G (1,2] this clearly implies (3.1) (in this case the limit in (3.1) is actually 0). In case p > 2, we have by Holder inequality,
/
|V«r2(H/50)2 < ( f
|Vu|') ¥ -(/" (\u\/50)»)\
250 and (3.1) follows in this case too. By Lemma 3.1 we deduce that condition (Co) holds. (ii) Without loss of generality we can assume that x0 = 0, hence 50(x) = \x\. Put w = \PM and wi = w(l - * ) on Br(0). Assume first that p ^ N. Since w G Wo1,p(iJr(0)) we get from Hardy inequality (see [7]), /
\Vw\" > |1 - N/p\" f
JBT(0)
(\w\/\x\)",
(3.4)
JBT(0)
so that w/\x\ e U(Br{Q)). Since clearly wx/\x\ € U(Br{Q)) it follows that u/\x\ £ LP(Br{Q)). But we have also Vu € V{Br{Q)). We deduce thus as above that (3.1) holds, hence condition (Co) is satisfied. In case p = N we argue as above, but with (3.4) replaced by the following improved Hardy inequality due to Adimurthi, Chaudhuri and Ramaswamy [1, Theorem 1.1], /
\Vw\'>C
JBrV(0) (0)
[
(\w\/\x\r(\ogR/\x\)-",
(3.5)
JBr(0)
with R > re 2 / p . It follows then that, u \x\\og(R/\x\)
e L"(Br(0)).
(3.6)
In order to show that condition (C0) is satisfied we take a = 1/2 and set,
rf = (l/2)" + '~\
j = l,...,n,
for n > — j 5 ^ . Since we always assume that p > 1 we have p = N > 2. By Young inequality, ( ^ )
2
| V
U
r < ( 2 /
P
) ^
+
(
l-2/p)|V<.
Therefore,
i E/
(M/M)'+ i
^ / W
V{2- 2 "<|:c|<2-»}
(H/lxD' + c /
iv«r 2 (i«i/ix|) 2 <
E/
^{2-2"<|x|<2-»}
|Vur:=/{") + ^").
Since Vu e I^(B r (0)) it is clear that liirin-Kx,/|n) = 0. Further, / j n ) < — (logjR + 2nlog2) p f It follows that condition (Co) is satisfied.
(. ,, *"[,., , . V -> 0 (by (3.6)). •
Remark 3.1. A comparison lemma related to (ii) of Proposition 3.1 was proved by Sreenadh [10].
251 Next we prove a uniqueness result for global solutions. Theorem 3.1. Let a(x) € L]0C(RN \ {0}) and suppose that u G C(RN \ {0}) n WJ£(RN \ {0}) is a positive solution of the equation (1.3) inQ. — RN \ {0} which satisfies condition (3.2), and for some a € (0,1), lim inf f r >0
-
lim inf r->0
f
(\u\/\x\)" < oo,
if p€ {1,2),
'{°r<\x\
,
(M/My + lVur^M/liD^oo,
3 ?
.
tfPe(2,oo),
J {
or more generally, assume u satisfies conditions (Cx,) and (Co) with T0 = {0}. Let v e C(RN \ {0}) n WJ£(RN \ {0}) be another positive solution to (1.3). Then, v = cu for some constant c > 0. (3.8) Proof. Consider any xr G RN \ {0} and set a := vixj/uixi). Put Qi = RN \ {0,X!} and apply Theorem 1.3 for n = Qi and Ti = {x^} to deduce that i)(a;)/M(a;) > «i = D(:TI)/U(:EI) for every x ^ 0. Since x t and x are arbitrary, we deduce that u/v = c. • Remark 3.2. Theorem 3.1 generalizes Theorem 1.3 of [8] which treated the special case a(x) = |1 — N/p\p\x\~p in which a positive solution to (1.3) is given explicitly by u(x) = x
l-N/p
We close with a uniqueness result in bounded domains. Theorem 3.2. Let O be a bounded domain with boundary T0. Let a(x) e L}0C(Q) and suppose that u e C(O) n W^(Q) is a positive solution of (1.3) which satisfies (3.1), or more generally, condition (Co). Let v € C(Q) PI W^(Q) be another positive solution to (1.3). Then, v = cu for some constant c > 0. Proof. Simply repeat the argument of Theorem 3.1, this time applying Theorem 1.3 to the domain £2 \ { i i } , whose boundary is the disjoint union of To = dQ and r x = {^i}, for any X\ € Q. • Remark 3.3. Let Q be a bounded domain of class C2 and let r\ G C(Q) satisfy r\ > 0 in CI and rj = 0 on dQ. In [6, Theroem 1.2] it was shown that for a certain critical value A*, there exists a positive solution ut to the equation,
-Aput = (yV + (l-l/pr)l-^p±,
(3.9)
which satisfies the growth condition, C-161-1'"
for some C > 0,
(3.10)
where we denoted 5(x) = dist(x,dil). It is easy to show that (3.10) implies (3.1) (with 80 = 8). Indeed, in case p £ (1,2] this is immediate, while for p > 2 we need only to remark that by [9, Lemma A.3], (3.10) implies that |Vw,| < C8~l^v in Q,, and the conclusion follows as well. Therefore we infer from Theorem 3.2 that ut is actually the unique positive solution to (3.9), up to a multiplicative factor.
252
References [1] Adimurthi, N. Chaudhuri and M. Ramaswamy, An improved Hardy-Sobolev inequality and its application, Proc. Amer. Math. Soc. 130 (2002), 489-505. [2] S. Agmon, Bounds on exponential decay of eigenfunctions, in "Schrodinger Operators", ed. S. Gram, Lecture Notes in Math., Vol. 1159, Springer-Verlag, Berlin, 1985, 1-38. [3] W. Allegretto and Y. X. Huang, A Picone's identity for the p—Laplacian and applications, Nonlinear Analysis TMA 32 (1998), 819-830. [4] H. Brezis and M. Marcus, Hardy's inequality revisited, Ann. Sc. Norm. Pisa. 25 (1997), 217-237. [5] M. Marcus, V. J. Mizel and Y. Pinchover, On the best constant for Hardy's inequality in R™, Trans. A.M.S. 350 (1998), 3237-3255. [6] M. Marcus and I. Shafrir, An eigenvalue problem related to Hardy's IP inequality, Ann. Sc. Norm. Pisa. 29 (2000), 581-604. [7] B. Opic, A. Kufner, "Hardy-type Inequalities", Pitman Research Notes in Math., Vol. 219, Longman 1990. [8] A. Poliakovsky and I. Shafrir, Uniqueness of positive solution for singular equations involving the p-Laplacian, preprint. [9] I. Shafrir, Asymptotic behavior of minimizing sequences for Hardy's inequality, Commun. Contemp. Math., 2 (2000), 151-189. [10] K. Sreenadh, On the Fucik spectrum of Hardy-Sobolev operator, to appear in Nonlinear Analysis TMA..
Supercritical variational problems with unique critical points Wolfgang Reichel Mathematisches Institut der Universitat Basel, Rheinsprung 21 CH-4051 Basel, Switzerland Email : [email protected] Abstract We study how a simple class of one-parameter transformation-groups interacts with functionals jC[u] = jnL(x,u, Vu) dx. We show how one can understand Pohozaev's identity as a formula for the rate-of-change of £ under the action of the transformation group. Uniqueness of the critical point of L will follow provided the transformation group reduces the values of C. As an application we find a class of conformally contractible domains, which includes the class of star-shaped domains, and on which Pohozaev's uniqueness result remains valid.
1
Transformation groups
We begin our study with one-parameter transformation groups. Let Q C R n be a bounded smooth domain with exterior unit normal v{x) for x 6 dil. We consider a one-parameter family of maps G = {ge : dom£ C fl x 1 -1 ( 1 x 1 } enjoying the group-property 5ei ° &2 = &1+E2!ffo= W. A particularly nice example of such a group arises when ge is defined as the flow-map at time e of the dynamical system x = £(x,u),
u = (f>(x,u),
with smooth functions £, (/>. To be precise, let (xe(x0,u0),ipe(xo,uo)) of (1) at time e with initial conditions (x0,uo). Then a one-parameter setting ge(x0,u0) = (x£(a:o,«o)^e(^o, Wj), for every (x0,u0) e S l x R , publication [5], we investigate transformation groups arising from the (1), together with their interactions with variational problems.
(1) denote the solution group G arises by In a forthcoming dynamical system
In this paper, we study a simple type of transformation group arising from the following dynamical system: x = £(x), u = a(x)u. (2) The system (2) is a special case of (1) and can be integrated semi-explicitly. Since the two equations in (2) can be decoupled, we denote by Xc{x) the solution of the first equation at time e, starting with the initial condition x at time e = 0. Then ipe(x,u) =
253
254 exp(/0£ a(xT(x))d,T)u solves the second equation, with the initial condition u at time e = 0. Let us now take an initial function u(x). We can insert its graph (x,u(x)) into the flow and obtain after time e the transformed graph (xe{x),i>l{x,u{x)). For sufficiently small e, this will represent the graph of a new function (x,u(x)) with new independent variable x = Xe(x)- Expressing x = X-e{x) we find the following explicit formula for the transformed function u(x)
expf / a(xr-e{x))dTJ
u(x-€(x))
exp f /
u(x-€(x)),
(3) a(xr(x))dTJ
where x belongs to Qe = Xe(0). We will write geu for the new function u(x). In the following, we will use summation convention, i.e., double indices will automatically be summed from 1 to n. Proposition 1.1 Let G = {o be a transformation group generated by (2). Then for every i = 1 , . . . , n and every x € Oe, the following holds
am . «,(/ *,,,»*)gfe^gw +
f°
J
da
( XT
_ dxj
dx" ' ^~dxZ^dTeXP[j
Proof. (4) follows directly from (3).
( f°
a
(XTW>dT)
\
-
u
(X-e{i))-
D
In order to state the following Proposition, let us introduce the space CQ{D) of continuously differentiable functions on D with zero boundary values and the space C 0 ' (D) of uniformly Lipschitz-continuous functions on D with zero boundary values. Proposition 1.2 Suppose £(x) • v(x) < 0 for all x € 80.. Then for e > 0 we have fie C fi. Extending u by zero outside fi, the transformed function geu is zero in Q \ Qe. Moreover, if the initial function u belongs to the space Cg(Q) or C0' (fi) then for small e > 0 we have geu € CQ(Q€),C0' (f2e), respectively. In both cases, we have geu e C 0 ' (fl) for small e>0. A transformation half-group {g€}e>o arising from a flow of type (2) and with the property £(x) • v(x) < 0 on 80. will be called a contracting transformation half-group.
2
Rate of change of a functional under a transformation group
We consider a functional C[u] = JQ L(x, u, Vujdx, where L : Q x K" +1 —» R is a continuously differentiable Lagrangian. Let us suppose w.l.o.g that L(x,0,0) = 0, for all x 6 flit is simple to see that £ is Frechet-differentiable on the space CQ'1(Q).
255 Our goal is to find a formula for the rate of change of C under the action of a contracting transformation group. The theory of interaction of functionals with transformation groups is well developed, see Olver [2]. In order to keep our point of view accessible for nonexperts, we decided not just to quote the following theorem from the literature, but to include a proof for simple transformation groups generated by (2). Theorem 2.1 Let G = {ge}e>o be a contracting transformation half-group generated by (2). Suppose the initial function u belongs to CQ(0,). Then the rate of change of the functional £ under the transformation group can be computed as follows: —C[gM\^o f dL _. + dh
= Ja ^
+ dL
(da
Tu^ ^ fe
M + fldu
d& du\
&T ft^&J
,, r ,. dX +M
^ '
(5)
where the Lagrangian L and its derivatives are evaluated at (i,u(i),Vti(i)), and the functions a, £, u and their derivatives are evaluated at x. Remark 2.2 It turns out that a formula similar to (5) holds for more general transformation groups generated by (1). The proof requires more in-depth analysis of transformation groups and can be found in Olver [2]. Proof. Since L(x, 0,0) = 0 and since supp(geM) c Oe, we may write C[geu] = /
L(x,u,Vxu)dx.
If we consider x = Xe(x) as a change of variables, we obtain C[g€u] = f L(x e (x),u(x,(x)),(V 4 u)(x,(i)))det(£> Xe (s))dx. Jet Differentiation w.r.t. e at e = 0 yields je£[geu}\€=0 f 8Lri
dL
dL d (du ,
. ,,\ i
T
,.
AJ
(6)
where we have used that ^ det Dxe{x) = div£(xe(x)) det Dxe(x) and xo = Id. It remains to express du/dxi in terms of Xe{x) by the help of (4), and to differentiate the result w.r.t. e. Doing so we obtain first by (4)
J|(X.(*)) = exp ( / W ) ) * ) -g(*)^(Xe(*)) f° da +
/
dxJ
faTMT+e{x))-Q^;{Xt{x))dTexp
f fe ( /
\ Xr(x)dr
\ u{x),
°i It is important to have u continuously differentiable. For Lipschitz functions, the formula does not hold in general.
256 and after differentiation w.r.t. e at e = 0 and using \o = Id, dxb/dxi = 6tj we get
At)(JJL{z)
+
pf{x^fjM\
ctey \ axi dxidxi The last relation simplifies since d2xb/dxidxi back into (6). •
W*w„w.
(7)
de IE=O/ cte* = 0. The result follows by inserting (7)
Corollary 2.3 Let G = {ge}c>o be a contracting transformation half-group generated by (2). Suppose that the initial function u belongs to C2(fi)nCQ(f2). Then the rate of change of the functional under the transformation group satisfies ±CMU
= Ja(au - * • Vu)£[u]dx
+
JJ
.U)(L-
£
J £ ) da,
(8)
where £[u] = " " ^ ( f f O + f^ is the Euler-operator associated to the functional C. As before, the Lagrangian L and its derivatives are evaluated at (x, u(x), Vu(i)), the functions a, u, £ and their derivatives at x. Proof. The proof hinges on a formula of E. Noether, which she used in the derivation of her famous theorem on symmetry and conservation laws, cf. Olver [2], Section 4.4. Noether's formula expresses the integrand in (5) as a multiple of the Euler-equation plus a divergence-term, i.e. dL dxi
<9L du
dL_(da_ dui \dxi
^ _ ^ _ ^ _ \ dxt dxidxjj
= {au - £ • Vu)£[u] + 4~ (?L +{au-£aXi \
TA-C
(9) Vu)f^ oUi
Formula (9) is verified by explicit calculation. If we recall that u = 0on dQ. and hence Vu = vdu/dv, an integration by parts results in (8) and completes the proof. Remark 2.4 For solutions of the Euler-equation £[u] = 0, the identity between the volume integral (5) in Theorem 2.1 and the surface integral (8) in Corollary 2.3 is known as Pohozaev's identity, cf. Pohozaev [3] and Pucci and Serrin [4]. It has long been noted, cf. Pucci, Serrin [4] and Van der Vorst [7], that some kind of connection between one-parameter transformation groups, Noether's formula and Pohozaev's identity exists. For yet another variational approach for deriving Pohozaev's identity, which is especially useful for free-boundary problems, see Wagner [8].
3
Variational sub-symmetries and uniqueness of critical points
We can now begin to formulate our main result. Loosely speaking it states under mild assumptions on the Lagrangian that a contracting transformation half-group, which reduces the values of L, implies the uniqueness of the critical point of C.
257 Definition 3.1 Let the functional £[u] = JnL(x,u,Vu)dx be defined on C%A(Q). A contracting transformation half-group G ~_{g€}€Z0 generated by (2) is called a variational sub-symmetry for £ if for all u € C^l(Q) there exists e0(u) > 0 such that £[geu] < £[u] for all e E (0, e0(u)). Proposition 3.2 Let G = {ge}€>o be a contracting transformation half-group generated by (2). Then the condition ~C[g£u}\e=o < 0, for all u E C^U),
(10)
is necessary and sufficient for G to be a variational sub-symmetry. If (10) holds with strict inequality for all u E Cg(fi) unless u = 0 then G is called a strict variational sub-symmetry w.r.t 0. The importance of the proposition lies in the fact that one can use the explicit formula for ^£[geu]\e=0 given in (5) to verify that G is a variational sub-symmetry. We are now ready to state our main uniqueness result, which follows directly from Corollary 2.3. Theorem 3.3 Suppose G is a variational sub-symmetry generated by (2) for the functional £[u] = ffriL(x,u, Vu)dx on the space CQ'1 (ft). Suppose also that L(x, 0, p) is convex in p € Mn for all x e dQ. Then either of the following two conditions implies that u — 0 is the only critical point of £: (i) G is a strict variational sub-symmetry w.r.t. 0, (ii) if u € C2(fi) n CQ(Q) is a solution of £[u] = 0 and (£ • u)(L(x, 0, Vu) - ! ^ ! ^ ( x , 0 , V«)) = 0, on dft, then necessarily u = 0. Remark 3.4
1.
2. For a strict variational sub-symmetry the theorem can be understood heuristically: since a strict variational sub-symmetry strictly reduces the energy £\u] everywhere except at u = 0 it is only compatible with a (unique) global minimizer at u = 0. For example, a strict variational sub-symmetry applied to a mountain-pass reduces the energy everywhere along a minimizing path, which contradicts the definition of a mountain-pass. 3. A verification of condition (ii) is usually achieved via a unique continuation principle, cf. Section J^.l.
258
4
Applications
4.1
A generalization of Pohozaev's result
In 1965 S.I. Pohozaev proved a remarkable identity [3], which allowed him to conclude that the problem Au + \u\p'lu = 0, in ft, M = 0on<9ft, (11) has only the trivial solution if ft C R", n > 3 is a bounded, star-shaped domain and p > jj^§. Theorem 3.3 recovers this result if we consider the transformation group generated by n-2 x = —x, u = —-—u, i.e., geu(x) = e~f~cu(eex) with x € e_6ft and if we assume w.l.o.g that ft is starshaped w.r.t. 0. It turns out that G = {p<J£>o is a variational sub-symmetry for £[u] = /o(ol^ u l +rl u l p+1 )^ a; - The functional verifies the conditions of Theorem 3.3. In particular property (ii) is fulfilled, since there exists a subset of 9ft of positive measure with x • v{x) > 0 and therefore the vanishing of x • v(x)\Vu\2 on 0ft as assumed in (ii) implies that |Vu| 2 = 0 on a subset of 3ft of positive measure. By the unique continuations theorem u = 0 follows. In case p > T~ we could argue without the unique continuation theorem, since in that case G is a strict variational sub-symmetry. Now we want to investigate more systematically what kind of vector-fields £(x) and what kind of functions a(x) can generate a variational sub-symmetry for the functional £[u] via x = t;(x), u = a(x)u. To verify the infinitesimal criterion of Proposition 3.2, we use (5) and calculate d
Je
£[ffew]|e=o = /
Ja
-a\u\p+1 + uVa • Vu + a\Wu\2 -
VuD^Vudx
+/i)<.iv{(i|v„p-;riTi„r)*
+ J
\Vu\2 fa + ^ H
-
VuDiVudx,
where the integral over div(Vau 2 /2) vanishes. In Pohozaev's original result with £ = —x the term VuD^Vu turned out to be just — |Vw|2 which equals ^ | V u | 2 . In order to obtain similar best possible results for more general £ we are interested in those vector fields £ such that bD£b = ^
|b| 2 , for all b 6 K".
(13)
259 Lemma 4.1 In R", n > 3 all smooth vector-fields £ which satisfy (13) are (up to a constant shift) given as linear combinations of the vector-fields Ytj z i
= XjCi — xtej, for 1 < i < j < n, = (xi-Y,rfix2j)ei + 2Y,rtixixjej, fori = l . . . n ,
where e; denotes the unit-vector in the xt coordinate-direction. The linear combinations of these vector-fields are called conformal vector-fields. The proof is obtained by showing first that all component-functions of £ are secondorder polynomials. It is then easy to find the precise form of £. In dimension n = 2 one can show that the conformal vector-fields are exactly those vector-fields £ = (f *, £2) such that ^(x,y) + i£2(x,y) is holomorphic in x + iy. Applications of this observation are discussed in Reichel [5]. We use such a conformal vector-field £ and return to (12) to find
| « u = /n(-fl-^i)i«r-^ In order to achieve ^C[geu]\e=0 < 0 it is helpful to choose a = ^ f div : £. If we note that all our conformal vector-fields satisfy Adiv : £ = 0 we finally obtain
S£MI- -/n(^-p-Tl)wP+ld^Hence the sub-symmetry criterion of Proposition 3.2 will be fulfilled provided div£ < 0 and p > ^ | 2 . Uniqueness of the solution « = 0 o f ( l l ) now follows from Theorem 3.3. We sum up our results: . Definition 4.2 A bounded domain Q C M", n > 3 is called conformally contractible if there exists a conformal vector-field £ such that £(x) • v(x) < 0 on dQ with strict inequality on a subset of 9 0 of positive measure. The vector-field £ is called an associated vectorfield to fi. Theorem 4.3 (i) Letp > ^~. IfQ, C R n , n > 3 is a bounded conformally contractible domain with associated vector-field £ and if additionally div£ < 0 in fi then (11) has only the zero-solution. (ii) In the special case p = ^ | the additional assumption div£ < 0 is not needed. Remark 4.4 Part (i) of the theorem holds if we replace the non-linearity \u\p~lu in (11) by f(u) and assume that F(tu)/t"-i is non-decreasing in t e R for all u e R, where F(s) = ^f(T)dT. °* The alternative div : £ > 0 and p < ^ | must be discarded for bounded domains since we also require £ • v < 0 on d£l. It does however make sense if Q, is the complement of a bounded domain.
260 We give two examples of domains in R3, which are not star-shaped but conformally contractible, and the associated vector-field £ satisfies div£ < 0. Example 4.5 We take £ = (—x + y, — y — x, —z) with div£ = —3. The vector-field £ generates a composition of a dilation and a rotation in the x, y-plane. We construct a conformally contractible domain by extending a Sd-domain cylindrically in the z-direction. Both the 8d-cut and the 3d-domain are shown in Figure 1. In the 2d-domain the trajectories of the flow {x, y) = (—a; + y, —y — x) starting from the boundary are shown. Clearly the 2d-domain is positively-invariant under the flow, i.e. our transformation group is contracting. By the cylindrical extension this remains true for the Sd-domain
Figure 1: 2d-cut and 3d domain Example 4.6 We take £ = (—2xz, — 2yz, —z2 + x2 + y2) with div : £ = —6z. The vectorfield £ is the infinitesimal generator of a one-parameter group of conformal maps involving inversions. We construct a conformally contractible domain by rotating a planar domain around the z-axis. The flow (x,y,z) = £(x,y,z) is also rotationally symmetric around the z-axis. In Figure 2 the 2d-cut and the trajectories starting from the boundary are visualized. Again the 2d-domain is positively-invariant, and due to the rotation-symmetry this remains true for the 3d-domain. Hence our transformation group is contracting, and since the domain lies in the region z > 0 we have div : £ < 0.
Figure 2: 2d-domain and trajectories In Figure 3 the 3d-domain is visualized from above and from below.
261
Figure 3: 3d-domain from above and below
4.2
A problem of Brezis and Nirenberg
In a famous paper [1], Brezis and Nirenberg showed that the problem Au + \u + u 5 = 0 in Bi(0) C l 3 ,
u = 0 on 9Bi(0),
has a positive solution precisely for A £ (n2/i,n2). The fact that no positive solution exists for A < 0 follows from Theorem 4.3. The non-existence for A 6 [0,7r2/4] was found by Brezis and Nirenberg via a Pohozaev-type identity. We indicate how this result follows by the method of transformation groups. Since all positive solutions are radially symmetric, a corresponding functional C[u] = J0 L(r, u, u') dr has Lagrangian L(r, u, u') = {\\u'\2 — \u2 - |u 6 )r 2 . It turns out that for A > 0 the dynamical system r = - sin(2v/Ar),
u = I S "^
T
' - 7Acos(2v/Ar) 1 u,
generates a transformation-half group which can be shown to be a variational sub-symmetry for the functional C by using Proposition 3.2 and (5). However, the half-group is contracting if and only if f < 0, i.e. if and only if sin(2\/Ar) > 0, which restricts A to the interval [0, TT2/4]. Uniqueness of the zero-solution for such values of A is then obtained by Theorem 3.3.
4.3
Further applications
In our forthcoming work [5] we have included many applications of the interaction of one-parameter transformations groups and variational problems. We end this paper with a list of interesting applications which can be treated by our method. (1) For vector variational problems §aL{x,ul,... ,u , Vul,Vuk)dx 1 fc hold provided L(x, 0 , . . . , 0, p , . . . , p ) is rank-one convex.
Theorem 3.3 will
(2) For m-Laplace problems div : (\Vu\m~2Vu) + |u| p - 1 « = 0 in fi with u = 0 on dQ. and p > - ^ 1 uniqueness of the zero-solution will hold for conformally contractible domains, where up to a constant shift the associated vector-field f is a linear combination of vector-fields X and Y* from Lemma 4.1. In particular the domain of Example 1 is included.
262 (3) For the fourth-order problem A 2 u = |u| p_1 u in Q with u = Vu = 0 on dfl the same class of conformally contractible domains as in Theorem 4.3 gives rise to uniqueness of the zero-solution provided p > ~^. (4) Schaaf [6] proved uniqueness results in a different class of domains for exponents P ^ Pc, where pc is larger then the critical Sobolev-exponent.
References [1] H. Brezis and L. Nirenberg, Positive solutions of nonlinear elliptic equations involving critical Sobolev exponents. Comm. Pure Appl. Math. 36 (1983), 437-477. [2] P.J. Olver, Applications of Lie Groups to Differential Equations. Graduate Texts in Mathematics 107, Springer-Verlag, Berlin, 1986. [3] S.I. Pohozaev, Eigenfunctions of the equation Au + \f(u) = 0. Soviet Math. Dokl 6 (1965), 1408-1411. English translation, Dokl. Akad. Nauk. SSSR 165 (1965), 33-36. [4] P. Pucci and J. Serrin, A general variational identity. Indiana Univ. Math. J. 35 (1986), 681-703. [5] W. Reichel, Uniqueness theorems for variational problems by the method of transformation groups. In preparation. [6] R. Schaaf, Uniqueness for semilinear elliptic problems - supercritical growth and domain geometry. Advances in Differential Equations 5 (2000), 1201-1220. [7] R.C.A.M. Van der Vorst, Variational identities and applications to differential systems. Arch. Rational Mech. Anal. 116 (1992), 375-398. [8] A. Wagner, Pohozaev's identity from a variational viewpoint. Z. Anal. Anwendungen, to appear.
Structure of the solution flow for steady-state problems of one-dimensional Fremond models of shape memory alloys Ken Shirakawa Department of Mathematics, Faculty of Education, Chiba University, 1-33 Yayoi-cho, Inage-ku, Chiba, 263-8522, Japan. Email: [email protected] Abstract In this paper, we consider a steady-state problem of one-dimensional Fremond model of shape memory alloys. Recently, the detailed structure of steady-state solutions has been studied in [12]. According to the result, we see a sudden change of solution sets at the critical temperature. The main objective of this paper is to give more detailed information for the variation of solution sets with respect to (steadystate) temperature. More precisely, we shall characterize the variation of solution sets by Hausdorff distance between any steady-state solution at a temperature and any solution set at another temperature. 1
Introduction
This work is motivated by the following one-dimensional mathematical model of shape memory alloys, of the form: ' M
- LXi)t
+ ((<»(*) - 0°?W))X2Ux)t
~ htixx = Oi{d)X2UxU a.e. in Q := (0, +00) x (0,1),
a = ux + /3a(i9)x2, ox = 0, a.e. in Q,
e (xi\ _
^ <
dt\X2j Mx{t,
,a 2 / » V fl, ,
A
7T~2 dx2\x2j
. (HP-o)\
+ 9 M x i > X2) 2>
•
n
, a.e. in Q, \-a{ti)uxJ
0) - k(d(t, 0) - $) = hdx{t, 1) + k{d{t, 1) - 1?) = 0, a.e. t > 0,
u(t,Q) = 0, a(t, 1) = Pg(t),
a.e. t > 0,
(Xi)x(t, 0) = (Xi)x(t, 1) = 0, a.e. t > 0, i = 1,2, . 0(0, •) = ^o, Xi(0, •) = xtfi, « = 1,2, a.e. x G (0,1).
263
(L1)
264 (1.1) is proposed by Fremond (cf. [4, 7, 8, 9]), and the well-posedness for solutions of (1.1) is studied by many mathematicians from various viewpoints (cf. [1, 2, 6, 10, 11]). In this model, the phenomena of shape memory alloys are interpreted as a solid-solid phase transition among three phases: the austenite and two kinds of martensites. In the context, there are five variables: the absolute temperature $, the one-dimensional longitudinal displacement u, the stress a, and two order parameters xi a n d X2 relative to local proportions of martensites. More precisely, xi and Xi a r e defined by the sum of local proportions of two martensites and difference between them, respectively. Then, the pair (xi(t,x),X2(t, x)) is always on the following closed triangle K in K2, of the form: K := { fori) € E 2 | 0 < £ < 1, -( < rj < £ } ;
(1.2)
for any (t,x) G Q. IK is the indicator function on the closed triangle K, and dIK is the subdifferential of IK, which is a maximal monotone in R2. i9* > 0 is the critical temperature, a : [0, +oo) —y R is a smooth function with the derivative a! such that: for a certain large constant dc > d*, a > 0, a' < 0 on [0, dc) and a = a' = 0 on [i9c, +00). g is a smooth function with respect to t such that g(t) —• 0 as t —> +00. $, c0, L, h, k, j3, (j,, A and I are positive constants. Supposing that a is very small and &o, X;,o G H1^, 1), i = 1, 2, the large time behavior for solutions of (1.1) was studied by Colli-Laurengot-Stefanelli (cf. [5]). According to their result, i?(t) -> fl and a(t) —• 0 in appropriate senses as t —> +00, respectively. Therefore, putting "{($) := /3a(^) 2 , the steady-state problem for (1.1) is essentially a boundary value problem of the following vector-valued equation, of the form:
[ (xf)x(0) = (xf),(l) = 0, i = l,2; since the displacement «* in the steady-state can be calculated by ud(x) = -f)a{ti) f xt(y) dy for all x € (0,1). Jo Recently, the detailed structure of solutions {X11X2} (steady-state solutions) of (Py is studied in [12]. According to the result, solution sets in cases of 0 < $ < $ * , $ = i9*, •d* < $ < §c and i5c < d have quite different profiles. It implies that the physical situation changes dramatically at two characteristic temperatures 1?* and $ c . Now we would like to observe the variation of solution sets with respect to $, more precisely. In other words, our interest is the structure of the flow of solution sets with respect to $. In this paper, we shall characterize the variation by a new concept, named "arcwise-connectable". Namely, a steady-state solution at a temperature tf is called
265 arcwise-connectable, if the Hausdorff distance between the solution and any solution set at $ / d converges to 0 as d -» d. By making use of the concept, we shall discuss about the structure of the flow of solution sets with respect to $. Consequently, more detailed information of the variation of physical situations with respect to steady-state temperature will be observed. Notation 1.1 (Hausdorff distance) Let X be a real Banach space. In this paper we denote by | • \x the norm of X, and define the distance between any point a £ X and any subset B c X b y distx(a, B) := inf \x — a\xxEB
2
Structure result for steady-state solutions
Let $ be any (fixed) positive constant. In this section, we recall the structure result for solutions of (-P)'', which is obtained in [12]. Definition 2.1 A pair ( x^ ) of functions xt, i = 1J 2, on [0,1] is called a solution of {P)®, if xt e H2(0,1), (xf)x(O) = (xf)x(l) = 0, i = 1,2, and there exist two functions £f € L 2 (0,1), i = l,2, such that
OM*!,xt)»d-4(«f) + (J) - ( • £ , > i MM). In order to describe the structure result smoothly, we introduce the following notation. Definition 2.2 Let T := [T~*,T*] be any compact interval in R with r„ < r*, n > 0, and r, < r < T*. For any nonnegative integer n, we define a subclass Sr(ft,r;n) in H2(0,1) as follows. (I) ST(K,T;0)
:=
{n,T,r*}.
(II) For n > 1, x € ST(K, T; n), if and only if there exist a constant cx e R, a partition —S/KIT
= : x X i 0 < xXli
< ••• < xx,k
< ••• < xXiH
< xXiJl+1
:= 1
such that: putting JX:k := (xx,k, xXik + \/K7I"), A; = 1, • • •, n, (i) |c x | < m i n { T * - r , r - T » } ; (ii) xXtk + i/K7r < a;x>t+i for k = 0,1, • • •, n; (iii) if |c x | < min{r* — T,T — T»}, then x^.^ + I/KK = xx,k+i, fc = 0,1, • • •, n; in other words, if xx,k + \[KK < xXtk+\ for some 0 < k < n, then cx = min{r* — T,T — T*}, and \ = T* or x = T* o n the corresponding interval [xfc + x/K7r,a;t+i]; (iv) x{x) = (-l) f c _ 1 c x cos (
-j^-
J + T for z e Jx,fc, k = 1, • • •, n.
266 Remark 2.1 Since \Jx,k\ — S/KK for any k, we notice that n*(«) := sup { n \ ST(K, r; n) ^ 0 } < +oo.
(2.1)
"*(") , In fact, since n*(K)y/Kir = V^ \Jxk\ < 1) we have 0 < n*(/c) < K_7r < +oo.
fe? '
Let us define
^
. {r*}, if T
:= •
\J
ST(K,r;n),
if r . < r < r*,
(2.2)
71=0
{r,}, if T > r * . Then, according to results of Chen-Elliott [3], x G "ST(K, T) if and only if x is a solution of the following boundary value problem: -K-XXX + 9IT{x)
3 X-T
a.e. in (0,1),
Xx(0) = x*(l) = 0, where dlr is the subdifferential of the indicator function on the closed interval T. Now the structure result for solutions of (P)* is stated as follows. Proposition 2.1 (Structure of solutions of (P)*J Let Kn, K\, K+ and K_ be subsets of the boundary dK of the triangle K as in (1.2), defined by: K0 := {(0,0), (1,1), (1, - 1 ) } , K, ~ { (£, r,) G dK | £ = 1 } , K+:={(t,rj)edK\Z = Ti} and K. := { (£,,,) G dK | £ = - , , } ,
l
'
j
respectively. Then, a pair ( * l ) is a solution of (P)*, if and onty if; (si^ fi/ie case ofQ
(x*, X*) e ^ i ° " [°> *]> "«meli/ x? = 1 on [0,1], and
X2 is in the class £[-i,i] I -A-,0 I; (s2) (the case of' d = $*,) xi = const, on [0,1], and xt (sS) (the case of •&* < ti < tfc) (x?,xf) e K+UK_
Z5
«n the class S[-i,i] I - V , 0 I;
on [0,1], namely |xf | = x ! on [0,1],
and x? is in the class Sin n I -^4-, ' ~. ' 1 : in particular, if
~_ ' > 1, i/ien x? = X? = 0 o n [0> 1]/
(tyj (the case of $ > tic) xf = x ! = 0 on [0,1].
267 3
Statement of the main result L 2 (0,1)
Throughout this paper, let us denote by W the product space S[d) := |
(x\)
e W
(xtt\
x
i 2 (0,l)
. Here we define
is a solution of {P)9 j for any tf > 0.
Then we see from Proposition 2.1 that S(d) is nonempty for any # > 0, and profiles of S(d) in cases of 0 < d < •d*, d = •d*,tf*< 0 < tfc and tf > tfc_are quite different Now, the objective of this section is to investigate the variation of S{d) with respect to t? more precisely. In order to characterize the variation of S(d), we introduce the following concept. Definition 3.1 (Arcwise-connectable) Let d be any positive constant, and x'0 '•— ( Xl- ] be any solution of (P)*. Let distw(-, •) be the Hausdorff distance as in Notation 1.1 with X = W. Then: (i) xd i s called above (resp. below) arcwise-connectable at d, if dist w (x 5 , S{f)) ->• 0 as 0 \ d (resp. tf />)?); (ii) x"* is called arcwise-connectable at tf, if it is above arcwise-connectable and below arcwise-connectable; (iii) x* is called arcwise-disconnectable at &, if it is neither above arcwise-connectable nor below arcwise-connectable. Remark 3.1 Based on the concept as the above definition, we have some informations concerned with the variation of solution sets. For example, if S(fl) includes a lot of arcwise-disconnectable solutions, then we can imagine a sudden change of solution sets at d > 0. On the other hand, if there are many arcwise-connectable solutions in S{5) at •§ > 0, then it implies that S(-) does not vary so suddenly in a neighborhood of d. Remark 3.2 Let us assume that i9* < d < ?9C with Mz^l
> l, or ?9 > i?c. Then, we
7(«)
easily see from Proposition 2.1 that all solutions are arcwise-connectable, because they are constants on [0,1]. Therefore, in the rest of this paper, we consider only the case of
0 < tf < tfc with ^ p
< 1.
Now, our main result is stated as follows. Theorem 3.1 Let K0, K\, K+ and K- be the same as in (2.3). (A) A solution I x\ ) € S{d) is above arcwise-connectable, if and only if: (al) (the case ofO < d < •&*) (supp (xf)z) c + 0; (a2) (the case of & = 0*) (xf.xf) ^ K0 on [0,1];
268 (aS) (the case of 0 < l-&4^ < 1 ) xf = const, on [0,1] or 0 ± (supp (xf)*)c C (xf)" 1 (0); (WJ (the case of\<
^ T
1
< 1)
xf = const, on [0,1] or 0 / (supp (Xf)*)c c (xf )- l (l)fB) 4 solution I Xl-\ e S(d) is below arcwise-connectable, if and only if: (bl) (the case o/ 0 < i9 < ?9* j Xf = const, on [0,1] or (xf ^ ( l ) U (xf)" 1 (-1) ± 0; f&2j fifce case o/tf = •&*) (xf,xf) £ Ki
on
[°> *] s«cfc iftat
1
Xf = const, on [0,1] or (xf)" !!) U (xf)" 1 !-!) ^ 0/ (63J (Ue case o/O < ^ y ^ < ± J X? = const, on [0,1] or (supp (Xf)*)c C (xf)- 1 (0) * 0; (H) (tte case of \ < l-&4^- < 1 ) xf = const on [0,1] or (supp (xf )*)c C CJcf)_1(l) # 0Remark 3.3 Noting the case of $ = i?*, we see that all steady-state solutions are arcwisedisconnectable except for constant solutions. It implies that the physical situation changes dramatically at the critical temperature $*. 4
Proof of the main theorem
In this section, we give the proof of Theorem 3.1. Let 5r(«, r) be the class as in (2.2) for any compact interval T := [i"*,''"*], K > 0 and T 6 K, and let distL2(0]i)(-, •) be the Hausdorff distance as in Notation 1.1 with X = £ 2 (0,1). Here, for the proof of Theorem 3.1, we prepare some lemmas. Lemma 4.1 Let d > 0, x £ L2(0,1), T := [T,,T*] be any compact interval in R and rt < f < T* . Let © C (0, +00) be an open interval including $, and k : O —> (0, +00) be a continuous and strictly increasing function. Then: (i) distL2(0jl)(x, ST(K(I9), T)) —> 0 as $ \ d if and only if xeST(k,(i)),f)
such that (supp x*) c ^ 0;
(4.1)
(ii) dist£2(01)(x, Sr(/«($),f)) —> 0 as 1? /* $ if and only if xeST(k{d),f)
suchthat
x=const.
on[Q,l\orx~l{r*)\Jx~1{Tt)j=%.
(4.2)
269 Proof. First, let us assume that dist/,2(01)(x, ST(K,(IL)),T)) —> 0 as d \ •& (resp. •$ /• •&). Then we find sequences {pj\ C 0 and Xj £ | ST(K($ ) ),T), j > 1, such that inf K(I?J) > 0, fy \ tf (resp. ^ /> tf), K ( ^ ) \
K(I?) (resp. «($,) /> K(I?)),
(4.3)
3>i
and Xj —• X m £ 2 (0,1) as j -> + o o .
(4.4)
By (4.3) and Remark 2.1, n* := sup n*(«(#,)) < +oo, where n*(/c(i9,)) is the number as in (2.1) with K = K($J). Therefore taking a subsequence if necessary, we find an integer 0 < n < n* such that Xj £ ST(K(I?J), f; n) for all j > 1. If n = 0, then it follows from (I) of Definition 2.2, (4.3) and (4.4) that x £ ST(K(^), f I °)> namely (4.1) (resp. (4.2)) holds. If n > 1, then we see from (II) of Definition 2.2 that for any j > 1, (-l)*c X j . +r for 2 G [z Xj ,fc+ v /«;(^-)' I 'i a ; xj,*+i]' & = 0 , 1 , • • • , n , (-l) f c _ 1 c X j . cos I — - 7 = = - I + T for x G JXjtk, k = 1, • • •, n,
y/W)
where cx., xx-,k and JXtk are notations c x , :rXifc and JXifc as in (II) of Definition 2.2 with X = Xj: respectively. Also, taking a subsequence again if necessary, we may assume t h a t there are a constant c < min{r* — f , f — T,} a n d n - t h points xk G [0,1], k = 1, • • •, n, such t h a t c x . —> c and xXjtk —> £*, as j —» + o o , k = 1, • • •, n, since T a n d [0,1] are compact in R. Moreover, putting £ 0 : = — \fk(&)n and x ft+1 := 1, ni/k(d)
< 1 and xk + ^ K ^ ) ^ < x^+i forfc= 0,1, • • •, n.
Here let us put J& := (xk, %k + \A(i?)7r), k = 1, • • •, n, and (-l) fc c + f for x G [xk + s/k(i))n, xk+i], k = 0,1, • • •, n,
xix)
-l)k
1
ccos (
;- ,_^
/«(*)
] + T for x G Jjt,fc= 1, • • • ,n.
Then, since n
J2
Y, I J*l = " \ / S ( * ) T < n y ^ ) 7 r = ^ resp. ^
| J x .,,| < 1, j > 1
\JXiA = n y K(I?J)7T < n\Jh{'d)ir = ^
|Jfc| < 1, j > 1
we see that c = min{r* - f, f — r»} and x G ST{H(d),f;n)
satisfying (4.1) (resp. (4.2)).
270 Moreover, it follows from (4.3) and Lebesgue's convergence theorem that \j —> X m L 2 (0,1) as j —¥ +oo. So by (4.4) and the uniqueness of the limit, x = XSecondly, let us assume (4.1) (resp. (4.2)). Since the constant case of x is trivial, we consider the nonconstant case of x- Here let cx, x%,k and J%,k be notations cx, xXtk and JXik as in (II) of Definition 2.2 with x = X- Then, since functions cos(-4j(-)) vary continuously with respect to K > 0, it is not so difficult to find an open interval 6 C (#, +oo) n 9 (resp. (-co, $) D 0 ) and a sequence {xtf | i 5 e 8 } such that X* € S T (K(<9), f) for any i? G 0 and x° ->• X in i 2 (0,1) as 0 \
d (resp. -d /• Q).
For example, in the case of x G Sr(ft()?),r; 1), f = T'*T*, c^ = T*~T*, and 2^1 = 0, we can construct them by the following way: © := (d, •& + S) n 0 (resp. (# — 5,fl)n 0 ) with 5 > 0 satisfying yK($ + 6) • it < 1 (resp. tf — 8 > 0), and
-»/
N
™ ~
J cxcos T,-„,
("T^TI
+f
'
ifxe
[°>
V^WT).
\v«(#)/
if a: e [ V ^ J T T , ! ] ,
for any i ) £ 0 . Also, in other cases of x, we can construct the positive number 6 and the sequence {x*} by a similar way. Thus we obtain that dist L . ( o,i)(x,Sr(/s(0),f)) < |x* - xb(o,i) ->• 0 as 0 \ 0 (resp. 0 / 0). • Lemma 4.2 Lei 0, x, T, 0 and K ie the same as in Lemma 4-1- Let r 0 := T j" 7 "*, and f : 0 —• (T*! 7 - *) ^e a continuous and strictly increasing function. (I) dist L j( 0 ,i)(x,5r(«(i?),f(iJ))) -> 0 as 0 \ 0" if and only if: (i) (the case of r* < f(0) < r 0 ) x = const, on [0,1] or%^ (supp Xi) c C X _1 ( T *)/ fnj (t/ie case o/r 0 < f(tf) < T") X = const on [0,1] or%^ (supp Xx)c C X 1(7"*)(II) dist L 2 (0|1) (x, ST{h(d),f(ti))
-* Q asti/* d if and only if:
(i) (the case ofO < f(d) < r0 ) % = const, on [0,1] or (supp Xx)c C X _1 ( r *) ^ 0/ (wj (the case of To < f($) < r* ) x = const on [0,1] or (supp Xx)c C X _1 ( r *) 7^ $• Proof. By Definition 2.1, we notice the following fact: for any nonconstant function X € ST{K, T) with /c > 0 and r, < r < r*, 0C ( s u p p ^ C
f x _1 (n)> i f r » < r < ^ f2* S \ / .+T t
(4.5)
Noting (4.5), we can show this lemma by a similar way as in the proof of Lemma 4.1. •
271 Now by making use of Lemmas 4.1 and 4.2, we prove our main theorem. Let $ > 0 and I Xl- ) be any solution of (P)*. In the proof of Theorem 3.1, the following fact is important: i
1
7(0 ' M-)2
is positive, continuous and strictly increasing on [0, $ c ).
(4.6)
Proofs of ( a l ) and ( b l ) : In this case, since x? = 1 on [0,1], we immediately see that dtatw ( ( * | ) > 5 W ) = d i s t a l ) (xf,S[-i,i] (w)'°)) (4.6) and Lemma 4.1 with % = xi< have assertions (al) and (bl). •
T
for an
_
y 0 < ^ < P. Therefore, by
= [ 1> 1]> ^ W = ^m
an
^ ^ = 0, we immediately
Proofs of (a2) and (b2): (b2) is obtained just as in (bl). Now we show (a2). If ( x ' . ) is above arcwise-connectable, then there are sequences {#j} C ($*,$ c ) and ( xl'. J € S{dj), j > 1, such that •dj \ d* and xfj -> X?* in i 2 (0,1) as j -»• +oo. Here, by (s3) of Proposition 2.1, |x2 J | = Xi , j > 1- So, taking a subsequence if necessary, we have |xfl = x f a . e . o n [ 0 ) i ] . (4.7) Combing (4.7) and (s2) of Proposition 2.1, we conclude (xfiXiT) e Ka o n [0,1]Conversely, if ( X i ' . x D e -^o on [0,1], then it follows from (s3) of Proposition 2.1 that dis% ((xi'.),S{ti))
=0foranytf* < •& < -dc. m
Proofs of (a3) and (b3): If ( Xlf ) is above (resp. below) arcwise-connectable, then d i s t a l ) (xf, Sm
( ^ j , ^ P ) ) -»• 0 as 0 \ 9 (resp. 0 S $)•
So, by (4.6) and (i) of (I) (resp. (II)) in Lemma 4.2 with and f ($) =
X
= X?, T = [0,1], K(I3) =
^ ,
( ^ , we obtain
f y? = const, on [0,11 or l J J Al c 1 0 / (supp ( X f),) C (xf)- x (0) (resp. (supp (X?)*)c C (x?)" 1 ^) * «)•
(4.8)
Conversely, if (4.8) holds, then applying (i) of (I) (resp. (II)) Lemma 4.2 again, we find an open interval 0 O C (#,i?c) (resp. ($*,^)) and a sequence {xf | i? G 0o} such that X? 6 S p , ! ] ^ , ^ ) for d e m a n d Xf -»• xf in £ 2 (0,1) as tf \ # (resp. tf /> #).
272 Here since \x%\ = xi L 2 (0,1) such that
on
[0,1], it is not so difficult to find a sequence {xt \ fl G Oo} C
J, e S{§) for •d e 6 0 and xl -»• X2 i n £ 2 (°- 1) as 1? \ 1? (resp. •d /> 1?). Thus we conclude that ( x* ) is above (below) arcwise-connectable. • Proofs of (a4) and (b4): (a4) (resp. (b4)) is shown just as in the proof of (a3) (resp. (b3)) by applying statements (ii) of (I) (resp. (II)) with x = xt,T = [0,1], re(tf) = ^
References [1] E. Bonetti, Global solution to a Fremond model for shape memory alloys with thermal memory, to appear in Nonlinear Analysis. [2] N. Chemetov, Uniqueness results for the full Fremond model of shape memory alloys, Z. Anal. Anwendungen, 17 (4), 877-892, 1998. [3] X. Chen and C. M. Elliott, Asymptotics for a parabolic double obstacle problem, Royal Soc. London, Proc. Math. Phys. Sci. Ser. A444 (1994), 429-445. [4] P. Colli, M. Fremond and A. Visintin, Thermo-mechanical evolution of shape memory alloys, Quart. Appl. Math., 48, 31-47, 1990. [5] P. Colli, P. Laurengot and U. Stefanelli, Long-time behavior for the full onedimensional Fremond model of shape memory alloys, Contin. Mech. Thermodyn. 12 (6), 423-433, 2000. [6] P. Colli and J. Sprekels, Global solution to the full one-dimensional Fremond model for shape memory alloys, Math. Methods Appl. Sci., 18, 371-385, 1995. [7] M. Fremond, Materiaux a memoire de forme, C. R. Acad. Sci. Paris Ser. II Mec. Phys. Chim. Sci. Univers Sci. Terre, 304, 239-244, 1987. [8] M. Fremond, Shape memory alloys. A thermomechanical model, Free Boundary Problems: Theory and Applications edited by K. H. Hoffmann and J. Sprekels, editors, Number 185 in Pitman Research Notes Math. Ser., pp 295-306, Longman, London, 1990. [9] M. Fremond and A. Visintin, Dissipation dans le changement de phase. Surfusion. Changement de phase irreversible, C. R. Acad. Sci. Paris Ser. II Mec. Phys. Chim. Sci. Univers Sci. Terre, 301, 1265-1268, 1985. [10] K. H. Hoffmann, M. Niezgodka and S. Zheng, Existence and uniqueness to an extended model of the dynamical developments in shape memory alloys, Nonlinear Analysis, 15, 977-990, 1990.
273 [11] N. Shemetov, Existence result for the full one-dimensional Premond model of shape memory alloys, Adv. Math. Sci. Appl., No. 8, Vol. 1, 157-172, 1998. [12] K. Shirakawa and U. Stefanelli, Structure of steady-state solutions for onedimensional Premond models of shape memory alloys, in preparation.
Lagrangian coordinates in free boundary problems for multidimensional parabolic equations Sergei I. Shmarev* Departamento de Matematicas, University of Oviedo, 33007 Oviedo, SPAIN shmarev@orion. ciencias. uniovi.es
Abstract We study the Cauchy problem for nonlinear degenerate parabolic equation p(x)ut = d i v ( u V / ( u ) + uq(u)) +uh(u),
(x,t) €RN
x (0,oo),
in the space dimension N > 1. It is assumed that /(0) = 0, f(s) > 0 and f'(s) > 0 for s > 0, and that p(x) is strictly positive on every compact subset of K.N. Equations of this class possess the property of finite speed of propagation of disturbances from the initial data which gives rise to the interfaces (free boundaries) separating the regions where the solution is positive from those where it is equal to zero. We describe and justify a non-local coordinate transformation that renders stationary the support of the nonnegative solution of the Cauchy problem. We present the explicit formulas which define the solution of the original Cauchy problem and its interface through the solution of the auxiliary problem posed in the time independent domain.
1
Introduction
The aim of this note is to justify the use of a non-local coordinate transformation in the study of free-boundary problems for degenerate parabolic equations. We consider the Cauchy problem for the degenerate parabolic equation p(x)ut
= d i v ( t i V / ( u ) + t i q ( n ) ) + uh(u)
(1.1)
under the assumptions / e C(R+),
/ ( 0 ) = 0,
f(s)
> 0 and f'(s)
> 0 for s > 0.
It is assumed t h a t •The author was partially supported by DGICYT Project BFM2000-1324, Spain
274
(1.2)
275 p(x) is strictly positive on every compact subset of R^.
(1.3) N
Let the initial function u0(x) be nonnegative and compactly supported in TSt . Under convenient assumptions on q and h the weak continuous solution of the Cauchy problem for equation (1.1) is also nonnegative and its support is compact for every t > 0. Thus, the solution exhibits free boundaries or interfaces that separate the regions where the solution is positive from those where it is identically zero. The study of the free boundaries lends itself to various methods. In this note, we describe and justify the method of Lagrangian coordinates which allows one to substitute the original free-boundary problem for a single equation by a system of nonlinear equations posed in a time independent domain. The solution of the original free-boundary problem is then explicitly represented through the solution of the auxiliary one. In particular, the interface T(t) = <9{suppu(:r,£)} is obtained as a one-to-one mapping from T(0) = 9{suppit 0 (a;)}, which gives certain advantage in the study of its properties. The idea of using Lagrangian coordinates in free-boundary problems for evolution equations was independently proposed in [1, 2, 3] and then developed by several authors. A survey of results obtained via this method can be found in the monograph [4] and papers [6, 7].
2
Lagrangian coordinates
In this section we introduce a system of Lagrangian coordinates generated by a solution of a parabolic equation. Let u(x,t) be a solution of equation (1.1). Throughout the section we assume that the solution is as smooth as is needed to perform all the requested transformations. It is convenient to rewrite equation (1.1) in the form dt + div < d
Vp + w(p)
= ug(p)
(2.1)
under the notation d = pu,
p = /(«),
w(s) = q ( / - 1 { s ) ) ,
and
g(s) = hif'^s)),
s > 0.
Let us start with reformulation of the Cauchy problem for equation (2.1) viewed as the mathematical description of the process of propagation of a polytropic gas in a porous medium. There are two possible ways to describe the motions of continua. The first one, usually referred to as the Euier method, consists in viewing the characteristics of motion (such as velocity, density, etc.) as functions of time t and some Cartesian coordinate system xi,...,xN not connected with the medium. The alternative description is due to Lagrange. In this method all magnitudes describing the motion are considered as functions of time and the initial state of the continuum. Let S7(0) and H(t) C Rw be the domains occupied by a polytropic gas at the moments t = 0 and t > 0. This correspondence defines the mapping
276
x = X((,t),
f€fi(0),
which assigns the position X(£,t) to the particle initially located at the point f G fi(0). Given the velocity field v(x,i), the motion of this particle is controlled by the trajectory equation 'Xt((,t) X(t,0)
=v[X(t,t),t], = £, { 6 f i ( 0 ) .
*>0V l
'
J
Another ingredient of the description is the mass balance law. Let u(t) be an arbitrary gas volume. The mass of uj(t) is controlled by the law XW d(x,t)dx)=[ dt (J wit) J
J wit)
F(x,i)dx,
where d(x,t) denotes the density at the point (x,t), and F(x,i) stands for the density of the mass forces. By convention we shall use capital letters to denote the magnitudes considered as functions of Lagrangian coordinates f. Thus, D(£,t) = d[X{(,,t),t\ will be the density, V(£,t) = v[X(£,t),t] the velocity, and P(£,i) = p[X(t;, t),t] the pressure. Let dX be the Jacobi matrix of the mapping £ H-> X and | J | = det(9X/df). Using the trajectory equation (2.2) and applying the rule of differentiation of determinants, it is easy to verify the validity of the relation - ^ = |J|div x v, called the Cauchy identity . Formally passing to the coordinates £, we have now:
/
F{x,t)dx=±{[
d(x,t)dx\=f
dt
Jw{t)
= [
Uw(t)
[Dt + VxD-v[X(Z,t),t]
)
^-{D\J\}dC Jw(p)
at
+ Ddivxv)\J\dZ
(2.3)
Ja(0)
= [
[Dt + div,(dv)] \J\ d£ = f
Jw{0)
[dt + divx(dv)] dx.
Jui(t)
Since the volume w(t) is assumed arbitrary, the mass balance law in the Euler coordinates is given by the equation dt + divx(dv) = F(x,t) and its Lagrangian counterpart has the form
277
ft(D\J\) = F(X,t)\J\. The mass balance law can be written then in the form {D\J\)' D\J\
F(X,t) D '
whence £>(£, t) \J\ = £>(£, 0) exp [J
^
^
dr)
in J1(0),
t > 0.
(2.4)
Now we are able to state in this framework the Lagrangian analog of the Cauchy problem for equation (2.1). Let us assume that the density of the gas is d = pu, that the gas velocity follows the law
v = - ^ y ( V p + w(p)),
(2.5)
and that the density of the mass forces F depends on P = f(U) and is equal to U h(U) = / - 1 ( P ) g{P). Recalculating the derivatives in x by the rule Vi = (
(>/_1)* is the matrix transposed to J~1,
we obtain the following system of equations: p{X)XM,t)
P{x)r\p)
= - V , p - w ( p ) = - (J" 1 )* V e P + w(P),
\J\ = p(orW,o))exP Q* ^dT)
X(Z,0) = {,
P(f,0) = P(f,0)
inJ7(0);
in
(2.6)
Q = Q ( ° ) x (°>r)> ( 2 - 7 )
P(f,t) = 0 on fln(0) x (0,T).
(2.8)
System (2.6)-(2.8) contains N + 1 equations for defining JV components of the particle trajectory X = (Xi,..., XN) and the pressure P along the trajectory. Let us notice that to formulate problem (2.6)-(2.8) we only used a resemblance between an evolution equation and the mass balance law in the motion of a continuous medium, which is not a rigorous justification of the performed change of the independent variables. The next indispensable step is to answer the question whether the constructed solution of the auxiliary problem (2.6)-(2.8) generates a solution of the original free-boundary problem.
3
The inverse transformation
In this section, we check that under certain conditions the solution of the auxiliary fixedboundary problem (2.6)-(2.8) allows us to recover a weak solution of the original freeboundary problem for equation (1.1).
278 A weak solution of the Cauchy problem for equation (1.1) is understood in the following sense. Definition 3.1 A function u(x,t) is said to be weak solution of the Cauchy problem for equation (1.1) in the strip S — R" x (0,T) if 1. u(x, t) is bounded, non-negative, and continuous in S; 2. for every test-function r](x, t) € C^S) such that TJ(X, t) = 0 when t = T and \x\ > R with some R > 0, the following identity holds:
/ p(x)u(x, 0)rj(x, 0) dx
JmN
+ / (p(x)t]tu-u'Vxr]-Vxp Js
+ u'w(p) •Vr) + unh(u))dxdt
= 0,
p = f(u).
Let the pair (X(£,£),P(£,i)) be a solution of problem (2.6)-(2.8). Let us assume that P(£,£) is non-negative, bounded and continuous in Q, that P = 0 on 9Q(0), and that M/'H-P)) a n d |(^ - 1 )* V ? P + w| are bounded in Q. Define the mapping x = X(£,t)
for(£,i)eQ,
*(£,(>) = £
for£en(0),
and assume that for every t e (0,T) it is a bijection of 17(0) onto Cl(t), and that \J\ ^ 0 in il(t). Set X(£,t)=t-
f {{J-1)' Jo
V ? P + w(P)) dr,
$ e 0(0),
(3.1)
and
P(x,t)
= {p{^ (0
ifx = JC(C,*) with f ei^y, otherwise.
_1
The functions p(x,t) and u = / (p) defined in this way are non-negative because of (2.7), bounded and continuous in R" x [0,T], assume the initial values by continuity and satisfy equation (1.1) in the weak sense. Indeed, given an arbitrary function tp{x,t) 6 C1(RW x [0,T]), vanishing for large |a;| and for t = T, the following relations hold:
279 - /
p(x)u(x,0)r](x,0)dx
JRN
= f f Jn(o) Jo
at
= / — < / p(x)u(x,t)r)(x,t)dx Jo at [JRN
\ dt )
±(p{X)U\J\Ti[X(t,t),t])dtdt
= f f £(D\J\)ridtdt+ Jo Jn(o) al
[T [ D\J\%d£dt at Jo Jrt{o)
(3.3)
-JT/„[^'+»+7--T]DIJI«* r
Y.
f
\g(P)
M
VXP + W ( P )
'
D\J\d£dt
(uh{u)rj + r]tp(x)u — uVxr] • Vxp — u w(p) • V^jy) cfc dt,
where we made use of (2.2) and (2.4). We thus obtain a parametric representation of a weak solution u(x, t) of the Cauchy problem for equation (1.1) via the solution of problem (2.6)-(2.8). The free boundary in this solution is governed by equation (2.6) with £ € 9O(0). Theorem 3.1 Let conditions (1.2)-(1.3) be fulfilled and (X(£,t),P(£,t)) be a solution of problem (2.6)-(2.8). If 1. P eC(Q),
P>0
inQ, | (J - 1 )* V ? P + w(P)| andh{f-x(P))
are bounded in Q,
2. \J\ = det [dX/d£] is separated away from zero and infinity in Q, 3. the mapping £ >->• X((, t) is a bisection from fl(0) to Cl(t) for every t € (0, T), then formulas (3.1)-(3.2) define a weak continuous solution of the Cauchy problem for equation (1.1). Remark 3.1 In this note, we avoid discussing the question of existence of such solutions to problem (2.6)-(2.8). For the special case of the Cauchy problem for the diffusionreaction equation ut = Aum + il)(u),
m>l,
JV = 1,2,3,
the complete study of its Lagrangian counterpart is performed in [5j. It is shown that the m solution p = « m _ 1 and its interface X(t;,t) instantly become infinite differentiable m—1 in the space variables and are real analytic functions of t.
280
4
Correspondence of the boundaries
Let us assume that the solution (X, P) of problem (2.6)-(2.8) is sufficiently regular. To be precise, let us assume that
SUp|Jy| +SUP L« Q
dJa
%
+ sup | (J" 1 )* V e P + w ( P ) | (4.1)
Y, supp 6 ((.7-TV { P + w(P)) 1.1
•-
<M,
**
and « < 1^1 < -
(4.2)
in <2 with some e > 0.
The latter condition guarantees that the coordinate transformation is locally invertible inside Q. Moreover, by virtue of (2.7) in the one-dimensional case this condition means that the function X(£,t) is monotone increasing as a function of £ whence the bijectivity of the mapping f i-> X(£, t). The situation is not that simple in the multidimensional case where the topology of the set Q(t) may change with time. To establish bijectivity of the mapping Q,(0) i-> Q(t) amounts to proving that X(dQ.(0),t) = dQ(i) for every t > 0. The inclusion X(dQ,(0),t) C dQ(t) immediately follows from (3.1)-(3.2) because P(£,i) is strictly positive if £ G O(0) and equals zero for £ G 9fi(0). Let us proceed to prove the inverse inclusion. Take two arbitrary points f,»7 £ 9fi(0), £ ^ T\. The inclusion dfi(i) c X(dQ(0),t) follows if X(£, t) ^ X(r], t) for all t € [0, T*] with some T* not depending on £ and r\. Let us take a point £0 £ ^-N such that |?y — £0| = 1 and cos(£-77,77-f„) = 0. Without loss of generality we may assume that £o = 0. We are going to show that for every t > 0 small enough cos(X(^t)-X(ri,t),X(V,t)).
{X{t,t)-X(j,,t),X(r,,t)) \X({,t)-X{ri,t)\\X{ri,t)\
1 2'
(4.3)
Inequality (4.3) means that the particles initially located at the points £ and 77 do not belong to the same ray and, thus, their trajectories cannot hit one another within the time interval [0,T*]. Let X(£,t)=Z + Y, X(r],t)=ri + Z, with
y
= - / t [ U " 1 ) * V f P + w(P)](£,T)dr, JO
Then for every t > 0
Z=-
r [ ( J - 1 ) * V e P + w(P)]( J ? ,T)dr. ./O
281
(x(£, t) - x{r,, t),x(v, t)) = {t-r, + Y-z,r, + z) = (Z-V,Z)
+ (n,Y-Z)
+
(Y-Z,Z).
If 90(0) is Lipschitz-continuous, every two points £, r\ e 80,(0) can be connected by an arc L c fi(0) of length \L\ < C\£ - n\ with a constant C depending on fi(0). It follows then \Y-Z\-
/ > -
Z)ds
< K0 \L\ < Kx \i - r,\
with constants K0, K\ depending on C and M. Thus, (X&t)
- X(r,,t),X(V,t))
n\,
K2 = const.
On the other hand,
X(£,t) -X(r,,t)
= (f-7?) (
1
- / ^ £ ( [ (
J
" T
V e P + w(P)] ( S ,r)) dsdr^j
whence
|X(f,t)-X(r;)t)|>|e-»?|(l-/ir3t) with a constant K3 depending only on M, C, N, and
\x(v,t)\
r,- [ [ ( J - 1 ) * V 4 P + w(P)](7 ? ,r)dr >l-KAt,
Ki =
Ki(N,M).
Jo
Given M, C, and N, we can fulfill (4.3) claiming that T* is sufficiently small: (X(t,t)-X(V,t),X(ri,t))
cos{x(t,t)-xtn,t),x(Ti,t)) =\X^t)-X(r,,t)\\X(r,,t)\ K2\£-n\(l + K4t) - \^ - n\{l - Ktt) (l-Ktt)
K2(1 + K4T*) ~ (1 - K3T>)(1 - KAT*)
2
Theorem 4.1 Let the solution (X,P) of problem (2.6)-(2.8) satisfy conditions (4.1) — (4.2). If the boundary 8Q(0) is Lipschitz-continuous, then there exists T* depending on N, M, and dO(0) such that (3.1) establishes a one-to-one correspondence between Q(0) and 0(7) fort e [0,T*].
282
References [1] J.G.BERRYMAN Evolution of a stable profile for a class of nonlinear diffusion equations. III. Slow diffusion on the line. J.Math.Phys., 21, (1980), 1326-1331. [2] M . E . G U R T I N ,
R.C.MACCAMY,
E.A.SOCOLOVSKY A coordinate transforma-
tion for the porous media equation that renders the free-boundary stationary. Quart.Appl.Math., 42, (1984), 345-357. [3] A.M.MEIRMANOV, V.V.PUKHNACHOV Lagrangian coordinates in the Stefan problem. Dinamika sploshnoy sredy, Novosibirsk, 47, (1980), 90-111 (in Russian). [4] A . M . M E I R M A N O V , V.V.PUKHNACHOV, S.I.SHMAREV Evolution Equations and
Lagrangian Coordinates, Walter de Gruyter, Berlin, 1997. [5] S.I.SHMAREV Differentiability and analyticity of solutions and interfaces in multidimensional reaction-diffusion equation. Departamento de Matematicas, University of Oviedo, Preprint n.l, (2001), 51 pp. [6] S.I.SHMAREV, J.L.VAZQUEZ Lagrangian coordinates and regularity of interfaces in reaction-diffusion equations. C. R. Acad. Sci. Paris, Ser. I, 321, (1994), 993-998. [7] S.I.SHMAREV, J.L.VAZQUEZ On the regularity of interfaces in solutions of reactiondiffusion equations via Lagrangian coordinates. NoDEA, 3, (1996), 465-497.
The Brezis-Nirenberg problem on H n Existence and Uniqueness of solutions Silke Stapelkamp Mathematisches Institut der Universitat Basel, Rheinsprung 21, CH - 4057 Basel, Switzerland Email: [email protected] Abstract We consider the equation AH»U + \u + un~2 = 0 in a domain D' in hyperbolic space H", n > 3 with Dirichlet boundary conditions. For different values of A we search for positive solutions. Existence holds for A* < A < Ai, where we can compute the value of A* exactly if D1 is a geodesic ball. In particular it turns out that - like in the Euclidean space the case n = 3 is different from the case n > 4 and has to be studied separately.
1
Introduction
We consider the problem A H »« + Au + uT~l = 0 in D' u > 0 in D' u = 0 on dD'
(BN)
where D' is a domain in hyperbolic space H", n > 3, A e R, and 2* = - ^ the critical Sobolovexponent. We want to know for which values of A there exists a solution u € W^2(D'). The same problem for balls in Euclidean space was solved in 1983 by Brezis and Nirenberg [BN] and in the following years a lot of extensions of this problem appeared. In spaces of constant curvature it has been studied by Bandle, Brillard and Flucher [BBF]. The special case of S 3 has been treated in [BB]. Our aim is now to extend the problem to domains in hyperbolic space. It turns out that the results are very similar to the results in the Euclidean case. After a brief introduction in the hyperbolic space we will discuss the existence of nontrivial solutions for the two cases n > 4 (section 3) and n = 3 (section 4). In the special case n = 3 we will make further remarks on properties of solutions.
283
284
2
T h e hyperbolic space
The hyperbolic space H n is denned as a subset of R" +1 by Mn = {xe R" + 1 | x\ + • • • + x2n - x2n+1 = - 1 , xn+1 > 0}. We can use the stereographic projection to map H" into R". This is done by mapping a point P' in H n to a point P € R". P is the intersection of the line between P' and the point ( 0 , . . . , 0, —1) and R™. In particular, the space H n is mapped into B(0,1) C R n .
Figure 1: The stereographic projection from H™ into B(0,1) C R™ , Change of coordinates transforms the line element of H n into ds = p(x)\dx\,
2
with p(x) •• ± -
I^I
The gradient, the Dirichlet integral and the Laplace-Beltrami operator corresponding to this metric are VH»ti =
—
P
Du= [ 1 |V H -M| 2 ds= f | V « | V - 2 dx JD
JD
AH»it = p~n div(p"- 2 Vu) Here is D' c H n and D its stereographic projection into R n . The first eigenvalue of the Laplace-Beltrami operator with Dirichlet boundary conditions will be denoted by \\.
3
T h e case n > 4
The main result of this section is
285 Theorem 1 (Existence of solutions for n > 4) Let D' be a bounded domain in H n , n > 4. Then the following statements are true: i) For A > Ai the problem (BN) has no nontrivial solution. ii) For A < "^"4~ ' and if D' is starshaped, the problem (BN) has no nontrivial solution. Hi) If A 6 (n'n4~2', Ai) there exists a nontrivial solution of the problem {BN). Remarks • Statement i) and ii) of Theorem 1 remain true if n = 3. They can be proved in the same way. • If D' is a geodesic ball in H n we may assume without loss of generality that D' is centered at (0, ...,0,1) G H n . Then the stereographic projection D of D' is a ball in R n , centered at the origin with radius 0 < R < 1. We can illustrate the statements of Theorem 1 in the following picture
Nonexistence
Figure 2: Existence of solutions for n > 4 Proof of Theorem 1 We shall sketch the proof and refer to [St] for more details. Denote by ipi the eigenfunction of - A H - corresponding to the eigenvalue Ai on D' with (pi > 0 in D'. Assume that u is a solution of (BN). Then / A H »u ipi ds + A / u (fi ds +
JD1
JD'
J D1
u 2 * - Vi ds = 0
This is equivalent to
/
( A
JD1
-
AI)
u
- /
JD1
u2'~l (fi ds.
The equality above only holds if u = 0. This completes the proof of the first statement. To show the second claim we assume that u is a nontrivial solution of (BN) and we define
286 v(x) := f)T~ (x) u(x) The function v is a nontrivial solution of A ^ + ( A - " ( n ~ 2 ) ) p 2 ) ; + t ) 2 -- 1 = 0 '
2TU —^
'
v>0
inD ( B N *)
inZ>
v = 0 on dD where D C R" is the stereographic projection of D' into R". Notice that D is also starshaped. We now use the classical Pohozaev inequality. Multiplying the equation (BN*) by xVv we get (-Av)(xVv) = (vT~l + /j,p2v)(xVv) This equation is equivalent to -v(Vv(xVv)
- x1—^- + x(^
+ |pV))
! L ^ ( | V * , | 2 - v2') - £ w V -
x\v2Vp2
Integration over D yields
\L\^{x-v)dS=»!y+>2)v2dx Because D is starshaped, the left hand side of the equation is strictly positiv if v is a nontrivial solution. On the other hand, the right hand side is negativ if A < nin~2) which is a contradiction. We conclude that v = 0 in D and u = 0 in D' and the second statement is proved. Existence of solutions of problem (BN) will be shown by the concentration-compactness alternative ([Lil], [Li2]; for a summary see [B]). We have to prove that there exists a function u e W1'2(D,pdx) quotient ^_JD\Vu\2pn-2dx-\fDu2pndx
so that the value of the
(fDu2>P»dxf2' is smaller then 5* where 5* denotes the best Sobolev constant of the embedding of W0' (D) into L2'(D). As trial functions we choose , s
uAx) =p
_n=2, s 2 (x)
„_.
287 with (p a smooth function, tp = l near 0 and (p = 0 on dD and estimate the quotient: QxAc)
~ \S* + 0{e) + c ( ^ 2 1 _ X)e\nE
if n = 4
for positive e and with some constant c > 0. We conclude <2A,P(UC) < S* if e is small enough. In view of the concentration-compactness alternative there exists a minimizer of the quotient if A > "'"4~ ' and this minimizer is a solution of problem {BN) if A < Ai. p.
4
T h e case n = 3
It turns out that in this case the value of A* depends on the geometry of D'. We will give a complete picture of existence of solutions for geodesic balls. Without loss of generality we can assume that this ball has his center in (0,0,0,1) G H 3 . Our main result is Theorem 2 (Existence of solutions for n = 3) Let D' be a geodesic ball in H 3 with center at (0,0,0,1), and D = B(0, R) with 0 < R < 1 the stenographic projection of D' into R 3 . Put A* = l +
^-T-. 16 arctanh R
Then the following statements are true: i) For A < A* and A > Ai tie probJem {BN) has only the trivial solution. ii) If X € (A*, AI) there exists a nontriviai solution of the problem {BN). Proof of Theorem 2 Again we sketch the proof and refer to [St] for details. The nonexistence results for A > Ai and A < | can be shown in the same way as in the case n > 4. To show the nonexistence of a nontriviai solution for | < A < A* we use again a Pohozaev argument with special test functions. By the moving plane method we know that if a nontriviai solution exists it is radial. Problem {BN) is equivalent to u" + -v! + p2\u + p2u5 = 0 in (0, R) r « > 0 m{0,R)
u{R) = 0
(BNR)
288
Nonexistence
1 R Figure 3: Existence of solutions for n = 3 Testing the equation (BNR) with r2f(r)u'
where
{
sinh(2x/r^T 5 (r)) • cosh(2 N /T^Ag(r)) g(r)
if f < A < 1 if A = l
sin(2VA^T 5 (r)) • cos^y/X^l g{r)) if 1 < A < A* and g(r) = arctanh r gives us after some computations an integral equality for the solution u which can only be valid if u = 0. So the first statement is proved. To prove the second part of Theorem 2 we must again estimate the quotient Q\>v. Assuming (p is a smooth function, ip(0) = 1, y'(0) = 0,
(e + laf)1^
we get QX,p(Ue) =S* + with
i-R
F(tp, A) = 47r / Jo
fe (2TT 2 ) 1 /3
F{tp,\) + 0{e)
pR
pR
ip'2p dr + 47r / (p2p2 dr - 4TT\ / ip2p3 dr. Jo Jo
Now choose ip(r) =
m{0,R) in (0, R)
tp1(R) = 0, Vi(0) = 0
In particular /o V>'i Pdr + J0
289 and pR
i-R
I
(p2p2 dr>\*
i-R
I Jo
for all admissible functions tp. We deduce F(y, A) > 47r(A* - A) f
psy2 dr
and F(ip, A) < 0 if A > A*. If e is small enough it follows that Q\tV(ue) < S*. Again we use the concentration-compactness alternative to conclude that there exists a minimizer and if A < Ax this minimizer is a solution of our problem (BN). —. Remarks For n = 3 the following properties of nontrivial solutions of the problem (BN) are known: • By the moving plane method it can be shown that all solutions of problem BN are radial and by [KwL] we know that a radially symmetric solution is unique. • Suppose that u\ is a solution of equation (BN) for A £ (A*, Ai). If A tends to Ai the solution u> belonging to A tends to 0 pointwise. If A tends to A* the radially symmetric solution concentrates at the origin. (see [B] for references) • Suppose that D' C H 3 is a geodesic ball with center at c := (0,0,0,1) e R 4 and G is Green's function of An» + A on D' with Dirichlet boundary conditions. After changing to radial symmetric coordinates in Euclidean space we can compute G with singularity in 0. 1
1-r2 I — cos(4VA — 1 arctanh R — 2\/A — 1 arctanh r) + cos(2\/A — 1 arctanh r) 1
Denote the regular part of G by h\(r). h\(r) is a monotone decreasing function in r and is strictly negativ if A < A*. If A > A* the function h\(r) changes sign in (0, R). In particular h\- (0) = 0. This supports the conjecture of Budd and Humphries ([BuH], confirm also [B]) as for balls in S 3 .
References [B]
C. Bandle. Sobolev inequalities and quasilinear elliptic boundary value problems. This volume.
290 [BB]
C. Bandle, R. Benguria. The Brezis-Nirenberg Problem on S 3 . to appear in J. Diff. Equations.
[BBF] C. Bandle, A. Brillard, M. Flucher. Greens's function, harmonic transplantation, and best Sobolev constant in spaces of constant curvature. Transactions of the AMS, Vol. 350, no. 3 (1998), 1103 - 1128. [Br]
H. Brezis. Elliptic Equations with Limiting Sobolev Exponents - The Impact of Topology. Comm. on Pure and Applied Math., Vol. XXXIX (1986), S17 - S39.
[BN]
H. Brezis, L. Nirenberg. Positive Solutions of Nonlinear Elliptic Equations Involving Critical Sobolev Exponents. Comm. Pure Appl. Math., Vol. XXXVI (1983), 437 - 477.
[BuH] C.J. Budd, A.R. Humphries. Numerical and analytical estimates of existence regions for semi-linear elliptic equations with critical Sobolev exponents in cuboid and cylindrical domains. Mathematics Prepint, University of Bath 99/22 (1999). [KwL] Man Nam Kwong, Yi Li. Uniqueness of radial solutions of semilinear elliptic equations. Transactions of the AMS, Vol. 333, no. 1 (1992), 339-363. [Lil]
P.-L. Lions. The concentration-compactness principle in the calculus of variations. The locally compact case I. Ann. Inst. H. Poincare Anal. Non Lineaire, 1 (1984), no.2, 109145.
[Li2]
P.-L. Lions. The concentration-compactness principle in the calculus of variations. The locally compact case II. Ann. Inst. H. Poincare Anal. Non Lineaire, 1 (1984), no.4, 223283.
[St]
S. Stapelkamp. The Brezis-Nirenberg Problem on H". PhD thesis, in preparation.
Solvability of general elliptic problems in Holder spaces 1
V. Volpert1, A. Volpert2 Laboratoire de Mathematiques Appliquees, UMR 5585 CNRS, Universite Lyon 1 69622 Villeurbanne, Prance 2 Department of Mathematics, Technion, 32000 Haifa, Israel Email : [email protected]
Abstract The paper is devoted to general elliptic operators in Holder spaces in bounded or unbounded domains. We discuss the Fredholm property of linear operators and properness of nonlinear operators. We construct a topological degree for Fredholm and proper operators of index 0.
1
N o r m a l solvability of linear problems
Normal solvability of linear elliptic problems in bounded domains is well studied. In the case of unbounded domains it is studied for some classes of operators in [11], [21] for the whole R", in [8], [16] for conical and in [12], [19] for cylindrical at infinity domains (see also [9]). In this section we consider general elliptic problems in arbitrary domains with some natural regularity conditions. We generalize the notion of limiting problems introduced already in the previous works, and formulate conditions of normal solvability through solvability of the limiting problems.
1.1
Operators and spaces
Let j3 = (/?!, ...,/3„) be a multi-index, pt be nonnegative integers, \/3\ = (31 + ... + /?„, D13 = Dxx...Dn ', Di = d/dxi. We consider the following operators: p
Atu = J2 Y,
a
ffc(*)£"V (i = l,...,p), l e i l c R " ,
*=1 \P\
BiU
= Z) Z) ^(x)D/}uk
(i = l,...,r),
xedSl.
* = l l/5|<7i*
Here /3ik, -yik are given numbers. As in [1], we suppose that there exist integers ti,...,tp, u\,..., aT such that Pij < Si + tj, i,j = l,...,p; %j < cn + tj, i = l,...,r,j
291
= l,...,p, st < 0.
si,...,sp,
292 We suppose that the problem is elliptic in the sense of [4], [1], i.e., the system is elliptic and supplementary and complementing (Shapiro-Lopatinskii) conditions are satisfied. Moreover we assume that the system is uniformly elliptic. We consider the space E0 — C i+tl+Q! (fi) x ... x Ct+t'+a(Q) of vector valued functions u(x) = (MI(S), ...,Up(x)), Uj e C' + ^ +a (f2), j = 1, ...,p, where / and a are given numbers, I > max(0, at), 0 < a < 1. The space C,+t'+a,(Q) is a Holder space of functions bounded together with their derivatives up to the order I + tj, and the latter satisfy the Holder condition uniformly in x. The domain Q, is supposed to be of class Cl+x+a, where A = max(—s;, — ai} afj £ C'-Si+a{Q,), bfj £ Cl-ai+a{dQ,). We assume that A > 1.
tj),
Operator A acts from E0 into C'~ Si+a (fi). Denote L± = (Alt ...,AP). Then L\ acts from E0 into E1 = C'-sl+a(Q.) x ... x C'-s"+a(fl). Operator B{ acts from E0 into Cl-ai+a(dQ,). Denote L2 = (B1} ...,Br). Then L2 acts from E0 into E2 = Cl-"l+a{dn) x ... x C'-a-+a{dn). Let E = Exx E2. Then L = (LUL2) acts from E0 into E.
1.2
Limiting domEiins
In this section we define limiting domains for unbounded domains in R n , show their existence and study some of their properties. We consider an unbounded domain Q C K", which satisfies the following condition: Condition 1.1 For each XQ 6 dQ, there exists a neighbourhood U{XQ) such that 1. U{XQ) contains a sphere with the radius S and the center XQ, where 6 is independent of x0; 2. There exists a homeomorphism ip(x; x0) of the neighbourhood U(XQ) on the unit sphere B = {y : \y\ < 1} in R n such that the images of fl f] U(x0) and dQ n U(x0) coincide with B+ = {y : yn > 0, \y\ < 1} and B0 = {y : yn = 0, \y\ < 1} respectively; 3. The function ip(x; x0) and its inverse belong to the Holder space Ck+OL with k = l + X. Their || • ||fc+Q-norms are bounded uniformly in x0. To define convergence of domains we use the following Hausdorff metric space. Let A and B denote two nonempty closed sets in R". Denote q{A, B) = svpa€A p(a, B), <;(B, A) = sup,, eB p(b, A), where p(a, B) denotes the distance from a point a to a set B, and let g{A,B) = max(<;(A,B),<;{B,A)). (1) We denote H a metric space of bounded closed nonempty sets in R" with the distance given by (1). We say that a sequence of domains Q m converges to a domain Q in Hioc if g(Clm n BR, Q n BR) -» 0, m -> oo for any R > 0 and BR = {x : \x\ < R}. Here bar denotes the closure of domains.
293 Definition 1.2 Let Q, C W1 be an unbounded domain, xm 6 H, \xm\ —* oo as m —> oo; x(z) ^e a characteristic function of fi, and f2m fee a shifted domain defined by the characteristic function Xm(x) = x{x + xm)- We say that Q* is a limiting domain of the domain fi if O m —> f2„ in E( oc as m —» oo.
Theorem 1.3 If a domain Q satisfies Condition 1.1, then there exists a function defined in W such that: 1. f{x)&
Ck+a(W)
2.
f(x)>Oiffxen;
f(x)
;
3- |V/(a;)| >l forxedQ
;
4- min(d(x), 1) < |/(x)|, where d(x) is the distance from x to d£l. Theorem 1.4 Let fi be an unbounded domain satisfying oo, and f(x) be the function constructed in Theorem 1.3. xmi and a function ft{x) such that fmi{x) = f(x + xm.) domain fl* = {x : f,(x) > 0} satisfies Condition 1.1, or in Eloc, where Q.mi = {x : fmi(x) > 0}. 1.3
Condition 1.1, xm e fi, \xm\ —> Then there exists a subsequence —> /»(a) in Ckoc(Rn), and the fi, = R n . Moreover Qm. —> fi„
Limiting problems
In the previous section we have introduced limiting domains. Here we will define the corresponding limiting problems. Let Q be a domain satisfying Condition 1.1 and x(^) be its characteristic function. Consider a sequence xm € Q, \xm\ —• oo and the shifted domains Qm defined by the shifted characteristic functions Xm{x) = x{x + xm)- We suppose that the sequence of domains Qm converge in S;oc to some limiting domain f2». Definition 1.5 Let um(x) 6 Ck(Q,m),m = 1,2,... . We say that um(x) converges to a limiting function u„(x) G Ck(Q.t) in Ckoc(Q.m —> fi,) if there exists an extension vm(x) € Cfc(E") of um(x),m = 1,2,... and an extension v,(x) g Cfc(Rn) of u,(x) such that vm —> v,inCtMn)Definition 1.6 Let um(x) e Ck(dQm), m = 1,2,... . We say that um(x) converges to a limiting function u,(x) £ Ck(dQ,t) in Ck0C(dQm ~> dQ.*) if there exists an extension vm{x) e Ck(M.n) of urn(x),m = 1,2,... and an extension v,(x) e Cfe(E") of u„(x) such thatvm^vtinCic(Rn)Remark 1.7 In these definitions ut(x) does not depend on the choice of the extensions vm(x) and vt(x). Indeed, in Definition 1.5 for any point x G f2» there exists a sequence %m € fim suh that xm —> x. Therefore u,(x) = v,(x) = lim vm(xm) = lim Similarly it can be checked for Definition 1.6.
um(xm).
294 Theorem 1.8 Let um e Ck+a(Qm), ||um||cj:+o < M, where the constant M is independent of m, 0 < k < I + A. Then there exists a function u, £ Ck+a(Q,*), ||w»||Cfc+Q < M and a subsequence umj such that umj —> u* in Ckoc(flmj —» fi,). Let um € Ck+a(dQ.m), ||um||Ck+a < M. Then there exist a function u, G Ck+a(dO,t), \\ut\\Ck+a < M and a subsequence umj such that umj —» ut in Ckoc{dQ,mj —> 9fi*). We recall that the operator L consists of a pair of operators, L = (L 1; L2), where the operator Lj acts inside the domain and L2 is a boundary operator. So we can represent the boundary problem as Liu = / i , L2u = f2, (2) where u e E0(Q), h e -Ei(ft), f2 e £ 2 (#n), £ = £ i x E2. The coefficients a y (a;) of the operator Li are defined in Q and the coefficients bij(x) of L2 in 9fi. The shifted coefficients a^ix + xm) and bij(x + xm) satisfy conditions of Theorem 1.3. Therefore we can define the limiting problem Lxu = f, L2u = g, where u 6 E0(Qt), f € Ei(£lt), g € E2(d£lt), L\ and L2 are operators with limiting coefficients a^(x) € C'- 5i+Q (0»), b^{x) € C-ai+a{d9.t). The operator L = {LUL2) : E0(Qt) —* E(Q,t) will be called limiting operator. We note that for a given problem (2) there can exist a set of limiting problems corresponding to different sequences xm and to different converging subsequences of coefficients of the operators.
1.4
Normal solvability
We introduce limiting domains and limiting operators defined above. Condition 1.9 For any limiting domain £2, and any limiting operator L the problem Lu = 0, u G E0(£lt) has only zero solution. Theorem 1.10 Let Condition 1.9 be satisfied. Then the operator L is normally solvable and its kernel is finite dimensional. We suppose now that a little more smoothness of operators and domains is required: 4k e C'-Si+6(ty,
b?k e Cl-"i+s(dQ) and the domain Q is of class Cl+X+S,
(3)
with a < 6 < 1. Theorem 1.11 Let the operator L be normally solvable and its kernel be finite dimensional. Then Condition 1.9 is satisfied.
295
1.5
Dual spaces. Invertibility of limiting operators
We consider now the space E = E(Q) defined above and the space E°, which consists of functions u e E converging to 0 at infinity in the norm E, i.e., ||w||s(nn{|x|>Ar}) ~* 0 as N —• oo. We say that un —» u0 in 25ioc(Q) if this convergence holds in fi n {\x\ < N} for any N. Lemma 1.12 Let
<j>= lim
(4)
n—*oo
It does not depend on the sequence un. Ifun —> 0 in Eioc, then
Vu 6 £ ° .
(5)
Thus we have the following theorem. Theorem 1.13 The dual space E* is a direct sum of the extension E of (E0)* on E and of the subspace E consisting of all functionals satisfying (5). Lemma 1.14 Suppose that the operator L : EQ —» E is normally solvable with a finite dimensional kernel, and the problem Lu = / , / € E is solvable if and only ifipi(f) — 0, i = 1,..., N, where ipi are linearly independent functionals in E*. Then ipi € E. These results allow us to prove invertibility of limiting operators (cf. Theorem 1.4). Theorem 1.15 If the operator L is Fredholm, then any its limiting operator is invertible. Corollary 1.16 // an operator L coincides with its limiting operator, and it is Fredholm, then it is invertible. The last result shows in particular that the spectrum of operators with constant, periodic or quasi-periodic coefficients in unbounded cylinders does not contain eigenvalues and consists only of points of the essential spectrum.
296
2
Properness and topological degree
It is known that topological degree can be constructed for Fredholm and proper operators with index 0. The construction is based essentially on two properties: density of regular values in the image of the operator [14], [17] and invariance of the orientation in homotopy classes of invertible operators. Different approaches to define the orientation determine some difference in constructions and in conditions on operators and spaces. The degree is constructed for abstract operators in Banach spaces (see [2], [5], [6], [7], [10] and references therein), and applied for elliptic operators of the second order in the one-dimensional spatial case [3], [18], for cylindrical domains [19], in R" [15]. In this work we construct the degree for general elliptic problems. For this we use conditions of normal solvability obtained for general linear elliptic problems in unbounded domains (Section 1); we introduce weighted spaces, without which elliptic operators are not generically proper; the orientation is adapted to elliptic problems. This allows us to remove the limitations in application of abstract degree constructions.
2.1
Operators and spaces
We consider general nonlinear elliptic operators Fi(x,D/3«u1,...,DISirup)
= 0 , i = l,...,p, r e f i
(6)
with nonlinear boundary operators ...,D 7 "u p ) = 0, j = 1, ...,r, x e dQ
Gj(x,D^uu
(7)
in a domain fl C K", which can be bounded or unbounded. Here D^^Uk is a vector with the components Dauk — d'a'uk/dx"1...dx°n where the multi-index a = (ai,...,a„) takes all values such that 0 < \a\ — ax + ... + an < f3ik, /3ik are given integers. The vectors DJihuk are defined similarly. The regularity of the functions Fi, Gj, u = (ui, ...,up), and of the domain Q is determined by j3ik, fjk, i,k = 1,..., p, j = 1,..., r (see below). In what follows we will use also the notations Fi{x,Vi) = Fi{x,D^uu
...,D^up),
Gjix^)
= Gj(x,D^Ul,
...,D^up).
The corresponding linear operators are p
^';
=
E
<^)Davk,
E
i = l,...p, xeQ
(8)
j = l,...r, xedQ,
(9)
*=i M
^
=
E
E
b
%(x)Davk,
*=1 M<7jk
where
**w -
drj?k .
W -
dC%
.
297 rji and (^ are the vector with the components nfk and C°j., respectively, ordered in the same way as the derivatives in (6), (7). The system (6), (7) is called elliptic if the corresponding system (8), (9) is elliptic in the sense of [1] for all values of parameters r^, £.. When we mention the Shapiro-Lopatinskii condition for operators (6), (7) we mean the corresponding condition for operators (8), (9) for any ^ e Rni and C3- € K m '. We introduce Banach spaces where operators Ft and G, act. Denote by E0ili a space of vector valued functions u(x) = (ui(x), ...,up(x)), Uj e C/J+t;'+c'(fl), j = 1, ...,p, where I and a are given numbers, I > max(0,
(10)
We suppose that the weight function p, is a positive infinitely differentiable function defined for all x £ K n , /j(a;) —> oo as |x| —> oo, t e f ! and | i l » ^ | - + 0 as \x\-> oo ( x S f i , |/?| > 0 ) A4
(11)
for. any multi-index (3. We suppose that Ft (G;) satisfies the following conditions: for any positive number M and for all multi-indices /3 and 7: |/3 + 7| < I - st + 2 (\0 + j \ < I - at + 2), \/3\ < I - s; (\/3\ < I-in) the derivatives D^D^F{(x,ri) (DfDJ'G^x.C)) as functions of x e S l . t j e l " ' , \r}\ < M {x G dU, ( <E R mi , ICI < M) satisfy Holder condition in x uniformly in 77 (Q and Lipschitz condition in 77 (£) uniformly in a; (with constants depending possibly on M). The domain Q. is supposed to be of class Cl+x+a, where A = max(—st, —<Ti,tj), and to satisfy Condition 1.1. Operator Ft acts from EQtll into Cl~Si+a(Q.). Denote F = (i^,..., Fp). Then F acts from E0tji into £i,M = C ^ s i + a ( n ) x ... x G|r S p + a (n). Operator Gi acts from E0i)l into G^ _<Ti+a (9n). Denote G = {Gu ..., G r ). Then G acts from E0:ll into £2,^ = Cl-"1+a{dn) x ... x Cl-ar+a{dtt). Let £„ = Si,M x E2lli. Then (F, G) acts from E0ill into EM.
2.2
Properness
We mean properness of the operator A : E0 —> E in the sense that the restriction of A on any closed bounded subset of E0 is proper. Properness of elliptic operators in bounded domains follows from the Schauder estimate (see [22]). In unbounded domains they are not generically proper. The simplest example is provided by the problem u + F(u) = 0, u(±oo) = 0, which can have a bounded for all x € R solution. Since the solution is invariant with respect to translation in space, the inverse image of 0 is not compact in the Holder space C2+a(R) or in Sobolev spaces. To get properness of elliptic operators in unbounded domains we have introduced weighted spaces.
298 Theorem 2.1 Suppose that the system of operators (6) is uniformly elliptic and for the system of operators (6), (7) Shapiro - Lopatinskii conditions are satisfied. Assume further that all limiting operators for the operator (8), (9) satisfy Condition NS. Then the operator (F, G) : E0ill —• Ep is proper. To prove the homotopy invariance of the topological degree we will consider the operators depending on parameters. Similar to the result on properness in Theorem 2.1 we obtain properness of the operators with respect to two variables.
2.3
Orientation
Let E0, Ei and E2 be Banach spaces. We suppose that E0 C E\. Denote E = E1 x E2. We consider linear operators Ax : E0 —> Elt A2 : E0 —* E2, A = (Ai,A2) : E0 —* E, and the following class of operators. Class O is a class of bounded operators A : E0 —> E satisfying conditions: (i) Operator (Ax + \I, A2) : E0 —> E is Fredholm with index 0 for all A > 0, (ii) Equation A\U = 0, A2u = 0 ( « £ E0) has only zero solution, (iii) There exists A0 = Ao(^4) such that the equation {Ax + \I)u = 0, A2u = 0
(ti€ E0)
(12)
has only zero solution for all A > A0. Here I is the identity operator in E0. Definition 2.2 The number o(A) = (-1)", where v is the sum of multiplicities of all positive eigenvalues A of the problem (12), is called orientation of the operator A. Operators A belonging to Class 0 are called orientable. Definition 2.3 Operators A0 e O and A1 £ 0 are said to be homotopic if there exists an operator A(T) : E0 x [0, l ] - > £ such that A(r) e O for all r € [0,1], A(T) is continuous in the operator norm with respect to r, AO(J4(T)) is bounded, and A(0) = A0, A(l) = A1. Theorem 2.4 If A0 and A1 are homotopic, then o(A°) = o(A1).
2.4
Topological degree
Let E0, Ei, E2 and E = E\ x E2 be Banach spaces as in Section 2.3, G C EQ be an open bounded set. We consider the following classes of linear ($) and nonlinear (F) operators. Class $ is a class of bounded linear opeators A = (A\,A2) tions:
: E0 —> E satisfying condi-
299 (i) Operator (Ai + IX, A2) : E0 -> E is Predholm for all A > 0, (ii) There exists A0 = XQ(A) such that operators (A\ + IX, A?) : E0 —> E have inverse which are uniformly bounded for all A > AoClass F is a class of proper operators / € C1(G, E) such that for any x e G the Frechet derivative f'(x) belongs to $. Class H is a class of proper operators f(x, t) € C1(G x [0,1], E), which belong to class F for any t E [0,1]. Two operators /o(x) : G —> 25 and f\{x) : G —> E are said to be homotopic if there exists f(x,i) £ H such that : /o(a;) = f{x,0), /i(z) = / ( s , 1). In what follows D will denote an open set which belongs to G with its closure D. Let a 6 E, f <E C1(G,E), f(x) j= a (x e 3D), where <9D is the boundary of D. Suppose that the equation f(x) = a (x € D) has finite number of solutions Xi, ...,xm and f'{xk) (fc = 1, ...,m) are invertible operators belonging to the class 4>. Then the orientation o of these operators is defined. Denote m
1(f,D;a)
= Y,o(f'(xk)).
(13)
fc=i
If f(x) / a, x € D, then it is supposed that -y(f, D; a) = 0. Lemma 2.5 Let f(x,t) e H, a£ E be a regular value o//(.,0) and f(.,l). f{x,t)^a{xedD, te[0,l]). rten7(/(.,0),D;a)=7(/(.,l),D;o).
Suppose that
From this lemma we easily obtain the following theorem. Theorem 2.6 £ei / 6 F and B be a ball \\a\\ < r,a e E such that f(x) ^ a (x e dD) for all a 6 B. Then for all regular values a G B, -y{f, D; a) does not depend on a. Using this theorem we can give the following definition of topological degree j(f,
D).
Definition 2.7 Let f e F and f(x) =fi 0 (x e dD). Let B be a ball \\a\\ < r in E such that f(x) 7^ a (x G dD) for all a £ B. Then j(f, D) = 7(/, D; a) for any regular value aeB. Existence of regular values a e B of / follows from Sard-Smale's theorem [17], [14]. Theorem 2.8 (Homotopy invariance). Let f(x,t) D,DCG, f(x,t) ± 0 (x e 3D,t € [0,1]).
e H. Suppose that for an open set Then1(f(;0),D)=1(f(-,l),D).
We note finally that f{x,t) is a Fredholm operator of index 1. To conclude that the set of regular points for it is dense, we need to assume that f(x,t) e C2(G x [0,1], E) [14], [17]. This regularity condition is weakened due to [13] where it is shown how to approximate a C1-mapping by another one for which a is a regular value. The topological degree constructed here is applicable for elliptic operators satisfying the conditions imposed above. In particular it is used to prove existence of travelling wave solutions of reaction-diffusion systems (see [18], [20]). The proofs of results presented above are given in the complete version of the paper which will be published elsewhere.
300
S. Agmon, A. Doughs, L. Nirenberg, Estimates near the boundary for solutions of elliptic partial diffrential equations satisfying general boundary conditions. Comm. Pure Appl. Math., 12, 1959, 623-727; 17,1964, 35-92. P. Benevieri, M. Furi, A simple notion of orientability for Predholm maps of index zero between Banach manifolds and degree. Annales des Sciences Mathematiques du Quebec 22 (1998) 131-148. E.N. Dancer, Boundary-value problems for ordinary differential equations on infinite intervals. Proc. London Math. Soc. (3) 30 (1975) 76-94 A. Doughs, L. Nirenberg, Interior estimates for elliptic systems of partial differential equations. Comm. Pure Appl. Math. 8 (1955), 503-538. K. D. Elworthy, A. J.Tromba, Degree theory on Banach manifolds. Proc. Symp. Pure Math. AMS (1970) 86-94 C.C. Fenske, Extensio gradus ad quasdam applicationes Fredholmii. Mitt. Math. Seminar, Giessen, 121 (1976) 65-70. P. Fitzpatrick, J. Pejsachowicz, Orientation and the Leray-Schauder theory for fully nonlinear boundary value problems. Memoirs of the AMS, 1993, 101, No. 483. V.A. Kondratiev, Boundary value problems for elliptic equations in domains with conical or angular points. Trudy Moskov. Mat. Obshch. 16 (1967), pp. 209 - 292. English transl. in: Trans. Math. Soc. 16 (1967). R.B. Lockhart, Fredholm property of a class of elliptic operators on non-compact manifolds. Duke Math. J., 48 (1981), pp. 289 - 312. J. Mawhin, Topological degree methods in nonlinear boundary value problems. Conference Board of Math. Sciences, AMS, No. 40. E.M. Mukhamadiev, Normal solvability and the noethericity of elliptic operators in spaces of functions on R". Part I, Zap. Nauch. Sem. LOMI, 110 (1981), 120-140. English transl. in J. Soviet Math., 25(1) (1984), 884-901. S.A. Nazarov, K. Pileckas, On the Fredholm property of the Stokes operator in a layer-like domain. J. Anal. Appl., 20 (2001), No. 1, pp. 155 - 182. J. Pejsachowicz, P. Rabier, Degree theory for C 1 Fredholm mapping of index 0. J. Anal. Math. 76 (1998) 289 - 319. F. Quinn, A. Sard, Hausdorff conullity of critical images of Fredholm maps. Amer. J. Math. 94 (1972) 1101 - 1110.
301 [15] P. J. Rabier, C. A. Stuart, C^-Fredholm maps and bifurcation for quasilinear elliptic equations on RN. Recent trends in nonlinear analysis, 249-264, Progr. Nonlinear Differential Equations Appl, 40, Birkhauser, Basel, 2000. [16] V. S. Rabinovich, The Fredholm property of general boundary value problems on noncompact manifolds, and limit operators. (Russian) Dokl. Akad. Nauk 325 (1992), no. 2,237-241 translation in Russian Acad. Sci. Dokl. Math. 46 (1993), no. 1, 53-58 [17] S. Smale, An infinite dimensional version of Sard's theorem. Amer. J. Math. 87 (1965) 861-866. [18] A.I. Volpert, V.A. Volpert, Applications of the rotation theory of vector fields to the study of wave solutions of parabolic equations. Trans. Moscow Math. Soc, 1990, 52, 59-108. [19] V. Volpert, A. Volpert, J.F. Collet, Topological degree for elliptic operators in unbounded cylinders. Adv. Diff. Eq., 1999, 4, No. 6, pp. 777-812. [20] A. Volpert, V. Volpert, Existence of multidimensional travelling waves and systems of waves. Comm. PDE, 2001, Volume 26, Numbers 3-4, pp. 421 - 459. [21] H. F. Walker, A Fredholm theory for a class of first-order elliptic partial differential operators in K". Trans. AMS, 165 (1972), pp. 75 - 86. [22] V. G. Zvyagin, V. T. Dmitrienko, Properness of nonlinear elliptic differential operators in Holder spaces. Lecture Notes Math. 1520, Springer-Verlag, Berlin 261-284
Nonlinear diffusion in irregular domains Ugur G. Abdulla Max-Planck Institute for Mathematics in the Sciences Leipzig 04103, Germany Email : [email protected]
Abstract We investigate the Dirichlet problem for the parabolic equation Ut = Aum, m > 0, in a non-smooth domain Cl C R^* 1 , TV > 2. In a recent paper [1], existence and boundary regularity results were established. In this paper we present uniqueness, comparison and stability theorems.
1
Introduction and statement of main results
Consider the equation Ik = Aum, N
(1) d2 dx
m
where u = u(x,t), x = (xu... ,xN) 6 M , N > 2, t € R+, A = J2?=i / h > °. m ^ 1. In this paper, we continue our study of the Dirichlet problem (DP) for the equation (1) in a general domain Q C R^"1"1. In the recent paper [1] (see also [2]), existence and boundary regularity results were established (see Theorem 2.1 of [1]). The purpose of the present paper is to establish uniqueness, comparison and stability theorems. Let Q be a bounded open subset of R^"1"1, N > 2. Let the boundary dQ of Q consist of the closure of an open domain BQ lying in t = 0, an open domain DQ lying in t = T G (0, oo) and a (not necessarily connected) manifold SQ lying in the strip 0 < t < T. Denote Q(T) = {(x,t) £ O : t = r } and assume that fl(t) ^ 0 for t e (0,T). The set VQ = BQ. U SQ is called a parabolic boundary of Q. Furthermore the class of domains with described structure will be denoted by 2?O,T- Let Q € T>0T be given and ip be an arbitrary continuous nonnegative function defined on VQ- DP consists in finding a solution to equation (1) in Q U DQ, satisfying the initial-boundary condition u = V,
on VQ.
(2)
We shall follow the following notion of weak solutions (super- or subsolutions): Definition 1.1 We shall say that the function u is a solution (resp. super- or subsolution) of DP (l)-(2), if : (a) u is nonnegative and continuous in Q, locally Holder continuous in QUDQ, satisfying (2) (resp. satisfying (2), with = replaced by > or <),
302
303 (b) for any t0, t\ such that 0 < t0 < tx < T and for any domain Ox G 2?t0itl such that f i i C f l U DO, and dBQi, dDQi, SOx being sufficiently smooth manifolds, the following integral identity holds /
ufdx=
J
ufdx+
I (uft + umAf)dxdt-
[
umJ-dxdt,
(3)
(resp. (3) holds with = replaced by > or <), where f G Cj* (Oi) is an arbitrary function (resp. nonnegative function) which is equal to zero on SOi, and v is the outward-directed normal vector to Oi(i) at (x,t) G SO^ We shall use the same notation as in [1] : z = (x,t) = (xi,...,xN)t) G R^-1-1, N >2, N 1 2 2 2 x = (xltx) eR ,x = (x2,...,xN) G R"- , |s| = Zli\*i\ , m = £ t =2 W 2 - For a 1 N+1 point z = (x,t) G R ^ we denote by B(z;S) an open ball in W of radius 6 > 0 and with center being in z. Let Q G 2?0,r be a given domain. Assume that for an arbitrary point zo = {x°, t0) G SQ (or (zo = (z 0 ,0) G SO,) there exist 6 > 0 and a continuous function 4> such that, after a suitable rotation of the rc-axes, we have SO D B(z0,6) = {z G B(ZQ, 8) : XI = (f>(x, t)} and signal — <j)(x,t)) = 1 for z G B(z0,i5) Pi O. Furthermore, we always suppose in this paper that the conditions of Theorem 2.1 of [1] are satisfied. We are now going to formulate another pointwise restriction for the point ZQ = (x°,to) G 5 0 , 0 < t0 < T, which plays a crucial role in the proof of uniqueness of the constructed solution. For an arbitrary sufficiently small <5 > 0, consider a domain Q{8) = { ( x , t ) : | S - z ° | <{8 + t0-t)i,to
and /i > ^~
ifm>
+ \x- S°| 2 ]" for (x, t) G Q{8).
(5)
1.
Furthermore we denote v — /i—1 assuming without loss of generality that v G (—1/2,0) if 0 < m < 1 and v G ( - 1 / (m + 1), 0) if m > 1. Definition 1.3 Let [c, d] C (0, T) be a given segment and S O ^ = SO n {(x, t) : c < t < d}. We say that assumption 1.2 is satisfied uniformly in [c,d], if there exist So > 0 and fi> 0 as in (5) such that for 0 < 8 < 80, (5) is satisfied for all Zo G SO[C](j] with the same Our main theorems read: Theorem 1.4 (Uniqueness) Assume that there exists a finite number of points U, i = l,...,k such that t\ = 0 < t2 < • • • < tk < ifc+1 = T and for the arbitrary compact subsegment [<5i,^2] C (*«,t«+i); i = 1,. • • ,k, assumption 1.2 is uniformly satisfied in [Si, 82]. Then the solution of the DP is unique.
304 T h e o r e m 1.5 (Comparison) Let u be a solution of DP andg be a super-solution (resp. sub-solution) of DP. Suppose that the hypothesis of Theorem 1.4 is satisfied. Then u < (resp. >) g in Q. T h e o r e m 1.6 (Stability or Li-contraction) Let the assumption of Theorem I.4 be satisfied. Letgi andg2 be solutions of DP with initial boundary data ipi andip2 respectively. Ifi>i = $2 on SO., then for arbitrary t 6 [0,T] we have llfll - ff2||t!(n(t)) < fIV>X - $21|L,(BO)-
2
Geometric meaning of t h e assumption
Assumption 1.2 is of geometric nature. To explain its meaning, for simplicity assume that N = 2, d(zp) = 1 and rewrite (5) as follows: x? - x1 < [t - t0 + (x2 - x^)2}11 for (x2,t) G Q{6), where x° = <j)(x2,t0) and xx = (p(x2,t) for (x2,t) G Q(S). Consider the hyperbolic paraboloid x\ = t + x\ (Figure 1) in the x1x2t-space. Let Ms be the piece of it lying in the half-space {t > 0}, between the planes {xi = 0} and {xx = — 51/2} (see Figure 2). The projection of Ms to the plane {xi = 0} is Qo(6), where as Qo(<5) we denote Q(6) with N — 2, x° = 0,i 0 = 0- The surface Ms has the following representation: Xi = (j){x2,t) = -y/xj+t, (X2,t) € Qo{6).
Figure 1: Hyperbolic paraboloid x\ = t + x\. Obviously, the function <> / satisfies (5) with = instead of < in the critical case when fi = 1/2 (we also replace Q(6) with Qo{6) in (5)). Consider the displacements of Ms, while it is moved on the xiX2-plane and shifted along the t-axis.
305
I I = X2
Xl =
-X2
Figure 2: Piece Ms of the hyperbolic paraboloid from Figure 1, lying in the half-space {£ > 0} between the planes {xi = 0} and {xi — d)1/2}. Let us now consider the critical case of the assumption 1.2 with jx = 1/2. Namely, we take 1/2 in (5). Equivalent formulation of this critical assumption may be given as follows: Assume that after the displacement of the above type Ms occupies such a position that its vertex coincides with the point z0 = (x°,t0) G SO,, and for S being positive and sufficiently small it has no common point with Q. Similar geometric reformulation of the assumption 1.2 may be given just modifying the subsurface Ms according to the lower restriction imposed on ft. Thus if 0 < m < 1, then the exterior touching surface is slightly more regular at the vertex point than related subsurface Ms of the hyperbolic paraboloid. Otherwise speaking, it is slightly more regular than Gx\ at the vertex point. When m changes from 1 to +oo, the regularity of Ms increases continuously, for each m being slightly more regular than C£t+1'"*+1 at the vertex point. Another limit position of Ms as m —• +oo (or ji —> 1~) is the upper paraboloid frustrum with vertex at the origin.
3
Proofs of t h e main theorems
Proof of Theorem 1.4 Suppose that 51 and 52 are two solutions of DP. We shall prove the uniqueness result proving that 51 = g2 in Q. n {(x, r) : t,- < r <
tj+1},
l,...,k
(6)
We present the proof of (6) only for the case j = 1. The proof for the cases j = 2 , . . . , k coincides with the proof for the case j = 1. We prove (6) with j = 1, proving that for
306 some limit solution u = limu„, the following inequalities are valid
/
(u(x,t)-gi{x,t))w(x)dx
i = 1,2,
(7)
Jn(t) for every t e (0,t 2 ) and for every u> e Cg°(0(i)) such that |w| < 1. Obviously, from (7) it follows that o,r and smooth positive functions ipn. We make a slight modification to the construction of f2n and ip„. Let * be a nonnegative and continuous function in R^"1"1, which coincides with ip on VQ. This continuation is always possible. Let i/>n be a sequence of smooth functions such that m a x ^ n " 1 ) <>„< (tfm + Cn""")™,™ = 1,2,..., N+1
where C > 1 is a fixed constant. For arbitrary subset G C R
(9)
and p > 0, we define
0,(G)=1JS(*,P). Since g and \]/ are continuous functions in H and g — if) on PO, for arbitrary n there exists pn > 0 such that |5m(.z) - # m (z)| < n~ m for 2 € 0Pn(M)
DQ.
(10)
We then assume that Q,n satisfies the following:
ftn € v0,T, o n c a u £>n, 50 n c o,,„(sn)
(11)
We now formulate assumptions on SQn near its point z„, which are direct implications of the assumption 1.2 at the point zo € SCI. Without loss of generality assume that d(zo) = 1. Assume that S£ln in some neighbourhood of its point zn = (KJ ,s°,i 0 ) is represented by the function x\ =
(12)
Obviously, this is possible in view of the uniform convergence of
307 (x, t) e Q{60) (namely (5) is satisfied with = instead of < ), then for all large n such that 6n < 60 we first choose <j)n as follows: ^x,t)^>
^,to)-K[t-to
+ \x-af>\2]
4>{x,t)
for
for
(x,t)€Q(8n),
(x,t)eQ(S0)\Q{6n).
Obviously, (j>„ satisfies (12) and converges to <j> uniformly in Q(6Q)- Then, we easily construct (f>n by smoothing
Sn = n 1+2- with 0 < e < (1 + v)
1
r
/
vlm
1} \\ 1 +-\ + 7.
(13)
where 7 = 1 if m > 1, while if 0 < m < 1 then 7 is chosen such that i < 7 < - — • m vm Let un be a classical solution to the following problem:
(14)
ut = Aum, in fi„ U £>On
(15)
u = ipn on VSl„.
(16)
This is a nondegenerate parabolic problem and classical theory [3, 4] implies the existence of a unique C2+Q-solution. From the maximum principle and (9) it follows that n~l < un < M in fin, n = 1,2,...
(17)
where M is some constant which does not depend on n and M > max(sup p n ip, sup Pfin ipn). As in [1], we then prove that for some subsequence n', u = linv_ K50 u^ is a solution of DP (l)-(2). Furthermore, without loss of generality, we write n instead of n'. Take an arbitrary sequence of real numbers {a/} such that 0 < a i + i < on < t, a, i 0 as I -> +00. Let
Qln = n„ n {(X,T)
: a,
< T < t},
Q°n = nn
D {(X,T)
(18) : 0 < T < t},
SQln =
Sfln n {(X,T) : at < T < t}, Sfi° = Sfi n n {{X,T) : 0 < T < t}. Since M„ is a classical solution of (15), it satisfies / unfdx= Jnn(t)
[ unfdx + [ (unfT+u™Af)dxdrJnn(ai) Jn'n
[ u™?fdxdT, ov Jsn'„
(19)
for arbitrary / 6 C2^(Q.n) which is equal to zero on SQ.ln, and where v = U(X,T) is the outward-directed normal vector to fi n (r) at (X,T) e SQfn. Since g is the weak solution of (l)-(2), we also have / gfdx= Jnn(t)
[ gfdx+ [ (gfT + gmAf)dxdTJn^at) Jsvn
[ gm^-dxdT. av Jsn'n
(20)
308 Substracting (20) from (19), we derive Ln(t)(u»-9)fdx
= Inn(ai)(un-9)fdx-
Jsn,n(u^ -
+ J n|i (un -gi)[CnfT
+
gm)^dxdr
AnAfldxdr,
where Cn = 1 if m > 1 (accordingly 7 = 1) and Cn = Bn if 0 < m < 1, and An = m 7 f {6uZ + (1 - 9)g^)m''-1de, Bn = 7 / (forf + (1 Jo Jo
6)g^y-1d9.
The functions A , and Bn are Holder continuous in Q„. From (17) and Definition 1.1 we derive n ^ < An < ~A, n^? < Bn < ~B for (a;, r) e H n , (22) where A, B are some positive constants which do not depend on n. To choose the test function / = f(x,r) in (21), consider the following problem: CnfT + AnAf
= 0 in ft° U BQn
(23)
/ = 0 o n S n ° a n d / = w(ai) onfi n (t).
(24)
This is the linear non-degenerate backward-parabolic problem. From the classical parabolic theory ([3, 4]) it follows that there exists a unique classical solution / „ € c£'-1+f,/2<sZ), with some /3 > 0. From the maximum principle it follows that l/n|
inn°.
(25)
By the condition of theorem, assumption 1.2 is satisfied uniformly on every compact subsegment of (0, i\. We prove that for every fixed I (see (18)) there exists a positive constant C(l), which does not depend on n, such that sup \Vfn(z)\
(26)
To prove (26), we use the modification of the method proposed in [1] for proving the boundary regularity of the solution to Dirichlet problem. We use (26), in order to estimate the right-hand side of (21) with / = fn(x, r), which is a solution of the problem (23)-(24). We have f {un - g)iv{x)dx = f (un-g)fdxJnn(t) Jan(ai)
f (u^ - gm)^fdxdr au Jsnl„
=X1+X2.
(27)
Using (9)-(ll), we have |2b|< sup |V/(z)| / l (\iP™-ym\ + \$>m-gm\)dxdT<(C+l)n-m z€sn'„ Jsn n
sup |V/(z)|. (28) z£snln
From (26)-(27), we derive |I 2 | < ( C + l ) C 1 ( 0 n ' < 1 + " > ^ ( m + 7 ) .
(29)
309 where e and 7 are chosen as in (13), (14). Applying (25), we have |2k|< /
\un-g\dx.
(30)
•/n„(<*,)
To estimate the right-hand side, we introduce the function , 1 <„\ _ )
u
"-(x>
a
')>
x
ipn{x,ai),
<= ^ n ( a z )
x e n(a ( )\fi„(ai).
Obviously v!n(x), x 6 fi(aj) is bounded uniformly with respect to n, I. From (30), we have
|2i|
|uJ,-
(31)
Since linin^+ooW^a:) = u(x, at) for 2; € n(a(), from Lebesgue's theorem it follows that lim / \u'n — g\dx = / \u(x,ai) — g(x,ai)\dx. (32) Hence, using (28)-(31) in (27) and passing to the limit as n —» +00 we have / (u — g)ui(x)dx < \u — g\dx. Jn(t) Jn(ai)
(33)
Let
Obviously, [/; is uniformly bounded with respect to I. Hence, from (33) we derive that /
{u - g)u(x)dx <
\Ui(x)\dx + C2meas(n(ai)\BQ,),
(34)
where the constant C 2 does not depend on I. Prom Lebesgue's theorem it follows that lim /
'-t+°° J BO
|[/,(x)|dx = 0.
Hence, passing to the limit as / —• +00, from (34), (7) follows. As it was explained earlier, from (7), (6), with j = 1 follows. Similarly, we prove (6) (step by step) for each j = 2 , . . . , k. Theorem 1.4 is proved. The proofs of the Theorems 1.5 and 1.6 are similar to the given proof.
References [1] U.G. Abdulla, On the Dirichlet problem for the nonlinear diffusion equation in nonsmooth domains. J. Math. Anal. Appl. 260(2) (2001), 384-403. 309
310 [2] U.G. Abdulla, On the Dirichlet problem for reaction-diffusion equations in non-smooth domains. Nonlinear Analysis T.M.A., 47(2) (2001), 765-776. [3] O.A. Ladyzhenskaya, V.A. Solonnikov and N.N. Uralceva, Linear and Quasilinear Equations of Parabolic Type. American Mathematical Society, Providence RI, 1968. [4] G.M.Lieberman, Second Order Parabolic Differential Equations. World Scientific, 1996.
Analysis of radiative transfer equation coupled with nonlinear heat conduction equation F. Asllanaj1'2, G. Jeandel1, J.R. Roche2, D. Schmitt2 1 LEMTA, Faculte des Sciences, B.P. 239, 54506, Vandoeuvre les Nancy, France. 2 IECN, Faculte des Sciences, B.P. 239, 54506, Vandoeuvre les Nancy, France. Email : [email protected], [email protected] Abstract A model for coupled radiative-conductive heat transfer through a semi-transparent medium is considered. The system under consideration has multiple practical applications, especially in thermal insulation. The existence and uniqueness results are established for an anisotropically absorbing, emitting and scattering medium, with axial symmetry and non homogeneous Dirichlet boundary conditions. The well-posedness of the system in the steady case is investigated. The existence of solutions is established applying Schauder's fixed point theorem. The uniqueness of the solution in an appropriate functional space is obtained.
1
Introduction
In fibrous insulators used for domestic heating, the heat propagation is due to the radiation and t h e conduction only and approximately one third of the heat transfer is achieved by radiation. T h e forthcoming system (l)-(3) of radiative-conductive heat transfer equations is a generalisation of t h e system considered by Kelley [6], where t h e steady s t a t e is studied. T h e medium is supposed to be independent of the wavelength (grey), isotropic and t h e heat equation is linear. Our weaker physical assumptions result in new theoretical difficulties which are solved in the paper. Let E be the medium thickness. T h e full system of equations is written as follows : oa fa, A) L° (T(x), A) - ae (M, A) L (x, fi, A)
+±J
V(s,/i,A) € ( 0 , £ ) x ( - 1 , 1 ) x (0,oo),
311
W
312
^Ac(rW)fw)
= Sr(x)
L(0,ii,X) L(E,fi,\) T(0) T(£) where
= L°(T 0 ,A) = L°(TE,X) = T0, = TB,
OO
/
toe(0,£), V(//,A)e(0,l]x(0,co), V(M,A)e[-l,0)x(0,oo),
(2)
/*1
/ L{x,ii,X)ndfid\ x=o J -1
Vxe(0,E),
(3)
5 r (x) = ^(x) Vze(0,£). In the above system, the unknowns are the monochromatic radiation intensity L(X,/J,, X) ax observed at the position x, in the direction having cosine fi with the positive x-axis and at the wavelength A, and the temperature T(x) at the position x. The function L° (T, A), which occurs in (1), (2)2 and (2)3, is the monochromatic intensity of the black body at the temperature T and is given by Planck's law : L
° ^ = A 5[ exp(cW-H'
^
where C\ and C% are positive constants. By definition, the black body absorbs the whole received radiation. LP (T, A) is an increasing function of T G (0, oo), for every A 6 (0, oo). Stefan-Boltzmann's law [8] implies : rrT4
/
L°(T,A)dA = — ,
J i0
(5)
T
where a is Stefan-Boltzmann's constant. The absorption and scattering coefficients aa, aa are positive and bounded : 0
. . V (p, A) € [-1,1] x (0, oo).
{
'
The extinction coefficient is defined as : ae = aa + as. We suppose all these coefficients to be continuous functions with axial symmetry with respect to the direction /i. The function P is the phase function and satisfies :
\J
P(p'-*»,\)dn=l,
V(//,A)e[-l,l]x(0,
(7)
The function ji — i > | P (// —> p, A) is a probability density in [—1,1], [7], [8]. According to [7], the scattering coefficient
1
f1
A ) P ( M ' - A i , A ) d / i ' = ff,(/x,A), V ( / i , , A ) € [ - l , l ] x ( 0 , o o ) .
(8)
313 The medium thermal conductivity AC(T) £ C°° (0, oo) is a positive and increasing function in (0,oo). We investigate the well-posedness of this system in the steady state case. In order to prove the existence of a solution of (l)-(3), we first prove that (T,L) is a solution of (l)-(3) if and only if T is a fixed point of an application T, i.e. T satisfies the equation : T = TIT). The existence of a solution thus reduces to the existence of a fixed point for the application T. This problem is solved applying Schauder's fixed point theorem. The uniqueness of the solution of (l)-(3) is not straightforward and is proved adapting the technique used by Kelley [6].
2
Existence and uniqueness of t h e solution
Let E be the function space given by :
E = 1} ((o,oo); c (nu 0Q.+ u an-)) n c 1 ((o, oo) ; c([-1, i] \ {o})),
(9)
where Q = {0,E) x [-1,1], <9fi+ = {0} x (0,1] and <9fi" = {E} x [-1,0). For every function g belonging to E, we define : oo
\\g(.,.,\)\Ld\,
/ o
where || S (.,.,A)|| 0 0 = sup | 5 (x l A 1 ,A)|. (x,/i)en
(10)
(E, ||.|| 1]00 ) is a Banach space. Let V be the function space C([0,E]) n C 2 (0, E). A couple (T, L) is a solution of (l)-(3) if T is an element of V (i.e. it is a classical solution) and L is in E. Under the above assumptions (4)-(8), we can prove the following Theorem. Theorem 2.1 There exists a unique solution (T,L) e V x E of the system Moreover, T satisfies : min (T0,TE) < T{x) < max (T0, TE),
(l)-(3).
Vz 6 [0, E]
and L satisfies : L° (min (T0, TE) ,\)
(x, /J, A) < L° (max (T0, TE), A),
V ( i , ( i , A ) e i l x (0, oo).
Proof. We provide an outline of the proof of Theorem 2.1 in some steps. The complete proof is included in [2] and [3]. We introduce a mapping T which depends in a parametric way on TQ, TE,
with: D= \u€C([0,E})
: T_
(11)
where T_ = min(T 0 ,T B ) and T+ = max(T 0 ,T E ). The map T is the composition of the four applications :
(D, \\.\U ^ {D, !|.||J x (D, ||.||li00) A {DL, ||.||J 3 , (5, ||.|U & (D, \\.\\J ,
314 where L is the unique solution of (1), (2)2 and (2)3, A(F) is the unique solution of the forthcoming problem (14) and : D = {g(x,fi,X)eLi((0,<x);L°°(Cl)):L0(T_,X)
/ OO
pi
/
/ OO
&a (/^ A) L (x, //, A) dftdX />1
/
/
\c(s)ds o
o-a{n,\)f{x,fi,\)diJ,d\.
(12) We prove that the application T is completely continuous [8] from D to D and using Schauder's fixed point theorem the existence of a solution is established. Now, we give some intermediate results. For the details of the proofs, we refer the reader to [2], [3]. Lemma 2.2 Let E £ (0, oo) and parameters T0, TE, aa, os and P be given and assume that all the above assumptions are satisfied. Let the functions H, <po and <J>E be given in L1 ((0, oo); C (fi)), L1 ((0, oo); C ((0,1])) and L1 ((0, oo); C ([-1,0))), respectively. Then, the following boundary value problem has a unique solution in the space E H—(x,fi,\)
+ ae(iJ.,X)L(x,ii,\) L(x,n,X) L(x,n,\)
=
(AL)(x,n,\)
+
H(x,n,\)
V(x,At,A)enx(0,oo), =
(13)
where A is the linear integral operator defined for all ip 6 L1 ((0, oo); L°° (O)) by: 1 (A
f1
(7,(n',X)P(n'-*fi,X)ip(x,^',X)dfi',
V(i,/i,A)€f!x(0,oo).
Proof of the Lemma Let r be the map from (E, ||-|| l o o ) —» (E, ||.|| 1 O O ) defined as : L = r(
= (A
L(x,n,X) L(x,(i,X)
= 4>o(n,X) = <j>o(fJ;X)
V(x,/i, A) 6 <9£2+ x (0, oo), V (a;, /i, A) e 9f2~ x (0, oo),
For 0 < ^i < 1, we have : L (x, M) A) = - fX exp f- (x - y) ^ t l R \ {Aip + H) (y, p, A) dy A* y o v t1 j
+ exp ( - z ^ ^ ) 0o (^ A).
315 Let tp and ip' in E be such that L = r{ip) and V = r (y/)- Using the expression of the integral operator A we deduce : |L(a:)/i)A)-L'(a:,/x,A)|<^^||V-^||00. ae (fi, A) The same result can be obtained for \i = 0 and — 1 < /i < 0. Moreover, for every (n, A) £ [—1,1] x (0, oo) we have :
^)-l-g'M
c=l-- *»
Thus, we deduce : \\L — L'Hj ^ < c \\
-l
/
then ||L|| li00 < ( l - ^ ^ + )
llvlli.00 + ~ r r
llfflkoo + UGH! < 00
and L belongs to E. Finally, applying Banach's fixed point theorem, it follows that the boundary value problem (13) has a unique solution in E. • Remark 2.3 Under the assumptions of Lemma 2.2 and by recurrence, it is easy to prove that the solution L of the boundary value problem (13) is an increasing function of (j>o, <J>E and H. Proposition 2.4 Let E G (0,00) and parameters T0, TB, aa, as and P be given and satisfying the assumptions (5)-(8). Let T G D be given. Then the problem (1), (2)2 and (2)3 has a unique solution L G E. Proof. This result is a corollary of Lemma 2.2, setting : H(x, n, A) =
A), fo(/i, A) = L°(T0l \),4>E(», A) = L°(TE, A).
Clearly the definition of L° implies that H, 4>o and cj)E belong to L1((0,oo);C(Q)), ^ ( ( 0 , 00); C((0,1])) and ^((0,00); C ( [ - l , 0))) respectively. Then, applying Lemma 2.2, the problem (1), (2)2 and (2)3 has a unique solution in E. • Remark 2.5 A corollary of Proposition 2.4 and of the previous Remark 2.3 is the following property : L°(T_,A) < L{x,n,\)
< L°(T+,\),
V(x,/x,A) e f l x (0,co)
and then L belongs to D_. Proposition 2.6 The map C is well defined and continuous from D - » D x D_.
316 Proof. If T e D thanks to Proposition 2.4 the couple (T,L) belongs to D x D, L the unique solution of the system (1), (2)2 and (2)3, then C is well defined. Let (U, L), {V, L')eDxDbe such that C{U) = {U, L) and C{V) = (V, L')> L being the solutions of (1), (2)2 and (2) 3 . Using Planck's law, we deduce : 11
" 1 - ° ° - l - c i 7 - + i7,-7r11
llo
and
L
°
Thus : \\U - V\\x + \\L - L'\\hoo < cte\\U - VW^ and C is continuous.
•
Thanks to Lebesgue's derivation theorem and to Pubini's theorem we obtain :
/
oo pi j
aa(y,X)L°(T(x),X)dyd\ OO
/ =
rl
/
2TT{4>(X) -
aa(y,X)L(x,y,\)dyd\
F(x)}'
where : 4>{x) =
(MaaL°(T))(x),
hence : \Sr(X)\ < 4TT<7+ { J l T U t , + ||L||l,oo} < OO,
VX £ (0,E)
and Sr 6 i°°(0, oo). Let T be such that T = ip{T). Kirchoff's transformation ip, given by (12)4, is a continuous, strictly increasing and positive function of T. Then ip~l is a well defined, continuous and strictly increasing function of T. Therefore, the conduction equation is equivalent to T = ip~l(T), where T is the solution of the semilinear conduction equation : _ T " = Sr(T) = 2,{^(ff(I)))-F(x)}, T(0) = V(7o), (14) T{E) = i,(TE). Proposition 2.7 The map B is well defined and continuous. Proof. Let (T,L) e D x D and F = B(T,L). If T e D, then T_ < T < T+ and if L e D, we have L°(T_, A) < L(x, y, A) < L°{T+, A), for every A and y. Then : MaaL°{T_) < MaaL < MaaL°{T+) and F(x) = {MaaL)(x) e £ ' . Let F, G e jff and (U,L), (V,L') £ D x D be such that F = £([/, L) and G = B{V, V). Then, for every x e (0, E), we have : /
<xa (M, A) (L (x, y, A) - V (x, y, A)) d/WA
<
2-ai \\L- L'||li<x>
<
cte(||L-L'||1)0O + | | C / - ^ | | o o ) .
D
'
317 Lemma 2.8 Under the previous assumptions, the map B oC is an increasing function of
T0,TE andT. Proof. One has : F(x) = (B o C) (T)(x) = M (aaL) (x). In Remark 2.3, we pointed out that L is a monotone increasing function with respect to T0> TE and T. aa is a positive function of y, and A. Thanks to the positive linear operator M the function B o C is monotonous increasing with respect to T0, TE and T. • Now, we prove that the application .A is well defined and continuous with respect to the function F. Proposition 2.9 Let F 6 £>' be given. Then, the boundary value problem (14) has a unique solution r e f l f l W2'p(0, E) for every 2 < p < +00 and thus the application A is well i Proof. Let :
MaaL\{rl{f^))-iT+-^T_f $(f) = { MaaL° (ip-1 (f))
if?
MaaL° ( r 1 (f + )) + ^
_
~
if ?_ < T < f+,
r
if f > f+.
Thanks to the definition of M, aa and L°(T), we remark that $ is a continuous and strictly increasing function of T. If F G D' then F € 17(0, E), for every p > 1 and : $ (f_) < F(x) < $ (f+) ,
Vx 6 (0, £ ) .
Then, according to Da Prato's result [5], the boundary value problem (14) has a unique solution in W2'p(0, E), for every 2 < p < +00. The monotonicity of $ and the maximun principle imply that T_ < f < f'+ and then f e D n W 2 ' p (0,£). This means that the application .4 is well defined. In order to prove the continuity of the map A with respect to F, let F, G G £>' and £/ = .A(F), V = ^t(G). Then, the difference W = [/ - V is the solution of the following problem : _W" = 2rr {$([/) - $ ( y ) - ( F - G ) } , j ^ ( 0 ) = 0, W(£) 0. Thanks to the monotonicity of <&, we obtain : ||tV'||*<27r J E?||W|| o o ||F-G|| 0 0 . Using Sobolev's continuous embeddings and Poincare's inequality, we obtain if W ^ 0 ||iy|| II
Moo —
"
"°°
•
318 Remark 2.10 Thanks to the monotonicity of<j>, the map A is an increasing function with respect to T0, TE and F. A straightforward result is the monotonicity of T = ip~1oAoBoC, with respect to To, TE and T. The inverse of Kirchoff's transformation is a continuous and increasing function with respect to f. Then ^{T) = T belongs to D n W2'P{Q,E), for every 2 < p < +oo. Moreover :
Thanks to the compact embedding of W1,2{0, E) into C([0,E}), we deduce that T is completely continuous. Using Schauder's fixed point theorem, we have the existence of a fixed point T of T. This proves the existence of a solution (T, L) for the coupled system of equations (l)-(3). Let us prove the uniqueness of the fixed point T in D, which proves that the system (l)-(3) has a unique solution (T,L) i n f l x E . First, we express the function F given by (12)5 explicitly in terms of T. Let Q denote the following linear integral operator defined for every u € L1^—1,1] x (0,00)); L°°(0, E)) and every (x, n, A) € fi x (0,00) by Q = Q1 + Q2, where : (Q1u)(x,^,X)
=
q{x,fj,,y,fj,',X)u(y,fj.',X)dfi'dX, J ay 0. E pO / /
-q{x,iL,y,ii',\)u{y,ii',\)dii'd\,
where the kernel q is given by : q (x, n, y, //, A) = — aa (//, A) P (// -> fj,, A) exp ( - (x - y) ^Jh—
j
V(x, /x, y, //, A) e fi x (0, x) x (0,1) x (0,00). Let K be the linear integral operator similar to the operator Q replacing aa by as. Both are positive operators by definition {Kf > 0 and Qf > 0 if / > 0). Let © be the function defined by : 0(x, 11, A,To,TE) = L°{T°'X)
foo-.O*',
A ) * V - fi, A)exp ^-x^f^-^j
dfJ,'
V(Z,/J,A) e fix (0,00).
Let E be the function similar to Q replacing aa by as. Lemma 2.11 Let E 6 (0,oo) be given. The space L 1 ([-1,1] x (0,oo); L°° (0,E)) is equipped with the norm : 00
1
llsll = / J
\\9(;V,X)\Ldnd\, J
where: \\g{.,fi,X)\\00 = sup x€(0,E)
\g{x,/j,,X)\.
319
Let parameters T0,TE,<Ja,os and P be given satisfying the relations (4)-(8). Then, the operator K is a linear continuous operator from the space L1 ([—1,1] x (0, oo); L°° (0, E)) into itself and I — K has a bounded inverse operator on this space. Proof. Let g £ / ^ ( [ - l , 1] x (0, oo); L°°(0, E)) be different from zero. Then we have : \(Kg)(x,fi,X)\
<
hLh"'^ /
/
—<7.(/i',A)P(/x'->/x,A)
x exp I - (x - y) ^ L R \ \\g (., ^ , A ) | L
J J
dfj!dy
-Jjj°-W,X)PW-+l*,X)
xaxp(-(*-*)*<£*>)
HflO.Ai'.AJILd/i'dj,.
Using Pubini's theorem and integrating with respect to y, we obtain the following inequality : \\(Kg)(x,n,\)\L
1 f1
ij|^y^(M'-M-A)ll5(.,M',A)||00dM'
and then ll(^)(^^A)||00<|y1iP(M'-M,A)||S(.,/i',A)||00dM', where c = 1 — aa~/ (aa++as+) satisfies 0 < c < 1 and is independant of A. Then an integration on (n, A) e [—1,1] x (0, oo) gives : ||K"|| < 1. Thus I - K has a bounded inverse operator which is given by the Neumann series. • Now, we can express Sr explicitly in terms of T. By straightforward calculations, we verify that : ST{x)
= 2n{{MaaL°(T)){x)
- F{x)}
=
tf)~>a£0(T(x),
2TTM{(J - Q(7 -
-Q(I - K)-1^,
A)
n, A, T0, TB) - 6 ( i , p, A,T0, TE)}.
Lemma 2.12 Let g € L 1 (fi x (0, oo)) be a positive function. Then : ||((«- + Q)s)(.,.,A)|| Ll(nx(0iOo)) < llz,1^)' Proof. Using the definition of the operators K, Q and Pubini's theorem, we have, for every A € (0, oo) ||((tf + Q)s)(.,.,A)|| L l ( n ) J ^=0 J z=-oo J y=—ao
P
x l{o<x-y<E}9 (y, p, A)
\
P"
l{0
X1{-B<X- V
5(.,^,A)l {0 < y A) 1{0<„
\{o
V M
'
320 But : J
^ ^ { r ^
A*'
) {\\9(--'fl'^1{0
+ b(;-^X)Mo
> °>
because t h e function to be integrated is positive. Finally, we obtain : ||((^ + C?)fl)(.).,A)||L1(nx(0iOo))<||fl(.1.,A)||Ll(n), which proves t h e desired estimate.
VAe(0,oo)
•
P r o p o s i t i o n 2.13 Under the previous assumptions, T ED.
the map T has a unique fixed point
Proof. We apply t h e same technique as the one used by Kelley [6]. Let (Un)n be the sequence defined by : Un = T (£/„_i), with {To = T_. This sequence is increasing and bounded by T+. Since t h e map T is compact, the sequence (Un)n lies in a compact subset of C([0,E]). Then, there exists a subsequence (U^n)) € C([0,E]), which converges to U e C([Q,Ej) and using the continuity of T we get : U = T{U). Similarly, the sequence Vn = F(Vn-x), with V0 = T+ is decreasing and converges to V = T{V) e C([0,E]). Moreover, if T = T(T) then U < T < V. Let U = ip{U) and V = ip(V) be the solutions of (14). T h e difference W = V - U > 0 satisfies : W" W(0) W(E)
= = =
2nM{(I-(K 0, 0.
Q))(I-K)-1o-a(L°(Tp-1(V))-L°(ip-1(U)))},
+
Since Sr e L°°(0, E) C L1 (0, E) then W" G L 1 (0, E). Hence : W"'(£;) -
W'(Q) l-E
I.
W"{x)dx
0
=
2*JOM{(I-(K
+ Q))(I-
K)"1 aa ( fjf^l^
) } (^)^-
In addition : aa (M, A) (L° (V" 1 ( 7 ) , A) - L° ( r T h e operator ( / — K)~ g=(I-
1
K)-
1
(0-) , A)) > 0,
V(/x, A) e [-1,1] x (0, oo).
is positive because K is positive by definition. Then : aa (M, A) (L° (V" 1 ( F ) , A) - L° ( ^ " 1 ((t7) , A)))
is positive. Thanks to Lemma 2.12, we have W'(E) - W'(0) > 0. Since W > 0 and W^(0) = W{E) = 0, we get : W'{0) > 0 > i y ' ( E ) . Then we have W'{0) = W'{E). This implies :
\\{K + Q)g\\ According t o Lemma 2.12, g must be equal t o zero and U = V. This concludes t h e proof of t h e uniqueness. • Then Propositions 2.4 and 2.13 allow us to conclude t h a t there exists one and only one solution (T, L) of the coupled system (l)-(3) in V x E . Remark 2.5 and Proposition 2.13 conclude the proof of t h e main Theorem 2.1. •
321
References [1] V. Agoshkov, Boundary Value Problems for Transport Equations. Modelling and simulation in Science, Engineering and Technology. Birkhauser, Basel (1998). [2] F. Asllanaj, G. Jeandel, J.R. Roche and D. Schmitt, Existence and Uniqueness of a steady state solution of a coupled radiative-conductive heat transfer problem for a non grey and anisotropically participating medium. Preprint of the Institut Elie Cartan, No 41, Nancy, France, (2001). [3] F. Asllanaj, Etude et analyse numerique des transferts de chaleur couples par rayonnement et conduction dans les milieux semi-transparents: Application aux milieux fibreux. Ph. D. Thesis, Universite Henri Poincare, Nancy, France, December 2001. [4] H. Brezis and W. A. Strauss, Semi-linear second order elliptic equation in L1. J. Math. Soc. Japan. 25(4) (1973). [5] G. Da Prato, Somme d'applications non lineaires. Symposia Mathematica VII, 1st. Naz. Alta Mat. Academic Press, p. 233-268 (1971). [6] C.T. Kelley, Existence and uniqueness of solutions of nonlinear systems of conductiveradiative heat transfer equations. Transport Theory Statist. Phys., 25 (1996), 249260. [7] S.C. Lee, Radiation heat-transfer model for fibers oriented parallel to diffuse boundaries. J. Thermophysics, 2 (1988), 303-308. [8] R.H. Martin, Jr., Nonlinear operators and differential equations in Banach spaces. Wiley-Interscience, New York (1976). [9] M.N. Ozisik, Radiative Transfer and Interactions with Conduction and Convection. Wiley-Interscience, New York (1973). [10] M. Zlamal, A finite element solution of the nonlinear heat equation. RAIRO Numerical Analysis, 17 (1980), 203-216.
Viscosity Lyapunov functions for almost sure stability of degenerate diffusions* Martino Bardi and Annalisa Cesaroni Dipartimento di Matematica P. e A. Universita di Padova via Belzoni 7, 35131 Padova, Italy Email : [email protected] ; [email protected]
Abstract The direct Lyapunov method for stability of dynamical systems is extended to the almost sure stability of Ito stochastic differential equations by means of semicontinuous Lyapunov functions satisfying a suitable system of partial differential inequalities in viscosity sense. The a.s. stabilizability of controlled degenerate diffusion processes is also treated. The key tools are some geometric characterizations of the invariance and viability properties of closed sets for controlled diffusions obtained by PDE-viscosity methods.
1
Introduction
Beginning with the work of P.-L. Lions [21], the theory of viscosity solutions to second order degenerate elliptic P D E s has given many contributions t o the theory of optimal control of degenerate diffusion processes. T h e link between t h e two fields is t h e HamiltonJacobi-Bellman equation associated to the value function of an optimal control problem via t h e Dynamic Programming Principle. T h e books by Fleming and Soner [14] and Yong and Zhou [26] give excellent surveys of the subject, see also [6] for t h e connections of first order Hamilton-Jacobi equations with deterministic control. In this paper we use viscosity methods to study a problem where no optimization nor value functions are involved. T h e issue is the stability of the origin, in a suitable sense, for the iV-dimensional stochastic differential equation (SDE) dXt = f(Xt)dt
+
a(Xt)dBt,
*This research was partially supported by M.U.R.S.T., project "Analysis and control of deterministic and stochastic evolution equations", and by the European Community, TMR Network "Viscosity solutions and their applications".
322
323 where Bt is an M-dimensional Brownian motion. There are many different notions of stochastic stability, see, e.g., [20, 17, 15] and the references therein. Here we consider the almost sure (Lyapunov) stability, a strong property that can be satisfied only by truly degenerate diffusion processes. To explain our result let us first recall the classical direct Lyapunov method for deterministic systems. Basically, a Lyapunov function V for the ODE X[ = f(Xt) is a positive definite and proper function decreasing along the trajectories of the ODE. If V is differentiable the last condition is equivalent to the partial differential inequality
DV(x)-f{x)<0
inR^UO}.
The existence of a smooth Lyapunov function implies the stability of the origin for the ODE, but it is not a necessary condition (see, e.g., [16, 27]). Necessary and sufficient conditions can be obtained by considering merely continuous, or even lower semicontinuous, Lyapunov functions satisfying the differential inequality in a suitable weak sense. Similar results were obtained also for the stabilizability of controlled deterministic systems (or, more generally, differential inclusions), i.e., the problem of finding at least one trajectory that remains in a neighborhood of the origin for any given small initial position. If at is the control function in the system X't = f(Xt, at), the appropriate differential inequality involves now a Hamilton-Jacobi-Bellman operator:
mmDV{x)-f(x,a)<0
inR N \{0}.
a
There is a large literature on this, making use of different tools of nonsmooth analysis. Let us mention, for instance, [27, 1, 23, 5] using generalized Dini directional derivatives (also known as contingent derivatives), [24, 19, 25] employing viscosity solutions, and [12, 22] using proximal derivatives, and see also the references therein. The main result of Section 1 is that the origin is almost surely stable for the SDE if there is a lower semicontinuous Lyapunov function satisfying, again, a differential inequality, which now is -DV(x)
• f(x) - ]-Tr [a{x)a{x)TD2V(x)]
> 0 in RN \ {0},
(1)
and, in addition, the system of equations aT{x)DV{x)
= 0 inlw\{0}.
(2)
Both (1) and (2) are interpreted in the viscosity sense and no ellipticity assumption is made on the matrix aaT. Note that the last condition means that there is diffusion only in the directions tangential to the level sets of V. The proof boils down to showing that the sublevel sets of the Lyapunov function V are (a.s.) invariant for the SDE. This is obtained by the stability properties of viscosity supersolutions and by a characterization of invariant sets for SDE's due to the first author and Goatin [7]. The necessary and sufficient condition for the invariance of an arbitrary closed set K in [7] has a simple geometric formulation in terms of the second order normal cone A/J(x) to K at a; this set has a purely deterministic definition, in contrast to the stochastic contingent set introduced by
324 Aubin and Da Prato [2, 1] for earlier characterizations of invariance. Also the proof of the invariance theorem in [7] makes use of PDE methods only (see also [8] for another proof using just the most basic facts of the viscosity theory). A sufficient condition for a.s. stability was obtained a few years ago by Aubin and Da Prato [4] by means of Lyapunov functions satisfying an inequality involving a suitable second order stochastic derivative of V, instead of (1) and (2). Presumably our condition is equivalent to theirs, but we believe it is easier to check and more readily readable, at least for people in the PDE community. Moreover, our methods extend to the problem of a.s. stabilizability of controlled diffusion processes dXt = f(Xt, a)dt + u(Xt, a)dBt. The result for this case is presented in Section 2. The inequality (1) must be replaced by the second order Hamilton-Jacobi-Bellman inequality max l-DV(x)
• f(x, a) - ]-Tr [aaT(x, a)D2V(x)] 1 > 0 in RN \ {0},
whereas the system (2) takes a slightly more technical form, see Definition 3.2. Now one needs a viability theorem for arbitrary closed sets K, namely, a condition ensuring that at least one trajectory of the controlled SDE remains forever in K a.s. (this property is also called controlled invariance). This problem was studied by Aubin and Da Prato [3] by stochastic methods and by Buckdahn et al. [11] using viscosity solutions. We use instead a characterization of viable sets obtained very recently by the first author and Jensen [8] in terms of the second order normal cone NK(X) by means of PDE-viscosity methods. Although we do not include in the present paper the full proof of the stabilizability theorem, we report the results of [8] that are involved in it.
2
Lyapunov m e t h o d for almost sure stability
Consider the Ito stochastic differential equation in MN (SDE)
| |
dXt
= /(Xt)
XQ
= x.
*
+ a X dBt
(J <
* > °'
where Bt is an M-dimensional Brownian motion. Assume that / and a are continuous functions defined in RN, taking values, respectively, in R-^ and in the space of N x M matrices, and satisfying \cr{x)-a{y)\
Vz.j/eR",
(3)
\m-f(y)\
Vx,yeRn,
(4)
The diffusion process described by (SDE) can be degenerate, because the matrix a(x) := is merely positive semidefinite.
-a(x)a(x)T,
325 Definition 2.1 (a.s. stability). The null state X = 0 is almost surely stable for (SDE) if for every e > 0 there exists S > 0 such that if\x\<Sthe corresponding trajectory X. of (SDE) starting at x verifies \Xt\ < e for allt > 0 almost surely. R e m a r k 2.2. This notion of stability for a stochastic system is considerably stronger than the usual concepts of stochastic stability (see the monographs [20, 17]). Here we require that, if the initial value is close enough to 0, the sample paths of the process issuing from a; remain forever within a certain neighborhood of the origin with probability 1, not just in the mean, or with probability increasing to 1 as |a;| decreases to 0 (stability in probability). Definition 2.3 (Lyapunov function). A function V : R w —• K + is a Lyapunov function for (SDE) if it satisfies the following conditions: (i) it is lower semicontinuous; (ii) it is positive definite, i.e., V(0) = 0 and V(x) > 0 for all x ^ 0; (Hi) it is proper, i.e., lim|.E|_>00 V(x) = oo or, equivalently, the sets {x\V(x) < fi} are bounded for every /x G M + ; (iv) it is a viscosity supersolution of the equation -DV(x)
• f(x) - Tr [a(x)D2V(x)] = 0
inRN\
{0};
(5)
(v) it is a bilateral viscosity supersolution of the equations <7i(x)-DV(x) = Q inM.N\{Q),
i = l,...,M,
(6)
where Oi(x) is the i-th column of the matrix a(x). We recall that a bilateral viscosity supersolution, or lower semicontinuous viscosity solutions as defined by Barron and Jensen (see [10, 6]), is a l.s.c. function such that a-t(x) • p = 0 for all p in the subdifferential D~V(x). The definition of viscosity supersolution of second order degenerate elliptic equations such as (5) can be found, for instance, in [13]. Remark 2.4. To motivate this definition, let us assume that V is twice continuously differentiable and look for a sufficient condition on V to be nonincreasing along the trajectories of the stochastic system (SDE), i.e., V(Xt) < V(XS)
for t > s almost surely.
By Ito's formula the inequality dV(Xt)/dt
< 0 becomes
[DV(Xt) • f(Xt) + Tr (a(Xt) • D2V(Xt))}
dt + [aT(Xt)DV(Xt)}
dWt < 0,
and by the properties of the Brownian motion one is led to the conditions DV(Xt)-f(Xt)+Tr(a(Xt)D2V(Xt)) aT(Xt)DV(Xt)
< 0, = 0.
326 Remark 2.5. From conditions (i), (ii), and (Hi) in Definition 2.3 it is standard to infer the following property of Lyapunov functions : for every e > 0 there exists 9 > 0 such that, ifV(x) < 6, then \x\ < e. If, in addition, V is continuous at 0, then the sublevel sets of V form a basis of neighborhoods ofO. This property allows us to deduce the a.s. stability of the Ito stochastic equation (SDE) from the invariance of the sublevel sets of a Lyapunov function, whenever it exists and is continuous at 0. In view of the last remark we can prove the Lyapunov stability in 0 of the system by showing that the sublevel sets of a suitable l.s.c. function are invariant for (SDE). In [7] it is proved the following Invariance Theorem, which states the equivalence between the invariance of a closed set and a Nagumo-type geometric condition involving the second order normal cone. Definition 2.6. The 2nd order normal cone is defined as: jvl(x)
:= {(p, Y) e RN x S(N) : for y -* x y e K p-(y-x) + \(y-x)Y(y - x) > o(\y -
x\2)},
where S(N) is the set of symmetric N x N matrices. Theorem 2.7 (Invariance Theorem [7]). The trajectories X. of (SDE) starting in a closed set K satisfy Xt e K for allt > 0 almost surely if and only if f(x) • p + trace [a(x)Y] > 0
Vz e dK, V(p, Y) € A/£(x).
(7)
Now we can prove the following stability theorem generalizing the direct Lyapunov method to nonsmooth Lyapunov functions and to stochastic dynamical systems. Theorem 2.8 (a.s. stability). Let V be a Lyapunov function for (SDE) (Definition 2.3) continuous at 0. Then the null state X = 0 is almost surely stable for (SDE). Proof. Consider for \i > 0 the sublevel set of the function V: K = {x\V(x) < / i } . We want to show that it is invariant, so we must check condition (7). We consider the function U = —V: U is an upper semicontinuous function which satisfies - p • f(x) - trace \a(x) • Y] < 0,
(8)
for all (p, Y) contained in the second-order superjet of U at x J2'+U(x)
=
{(jp,Y) eRNx
S(N) :
U(y) < U{x) +p-{y-x)
fory^x + \{y-x)-
Y(y - x) + o(\y - x\2)}
and o-i(x)-p = 0 i = l,...,M, (9) for all p contained in the superdifferential D+U(x) of U at x. The set K can be written as {x | U(x) > —[J,}. To prove the invariance of K we need the following lemma on the change of unknown for second order partial differential equations:
327 Lemma 2.9. Let u be a viscosity subsolution of equation (8) and a bilateral subsolution of equations (9). Let <j> e C2(2R) be strictly increasing with cf>' > 0. Then the map w = 4>ou is a viscosity subsolution of equation (8) too. Proof of the lemma. It is easy to check that, if (p,Y) e J2,+w(x), (il)'(w(x))p,il)'(w(x))Y + $"{w(x))p®p)
e
then
J2-+u(x),
where ip is the inverse of
then p <= D+w{x) and ip\w(x))p e D+u(x), so by (9)
trace \a{x) • ip"{w{x))p ®p]= . ; , / ; '.' V[«r4(x) • ^'{w{x))pf
= 0.
Therefore i>'(w{x)) [-p • f(x) - trace (a(x) • Y)] < 0, which concludes the proof.
•
We define now for every A > 0 a nondecreasing continuous real function
0
t>-n,
X(t + ju) -/j, - - < t <
i>x(t)
x
-1
t < -
M
-fi,
1 - r
We claim that the function i/jx o U is a viscosity subsolution of equation (8) for every A. It is actually sufficient to choose a sequence ipn of smooth real maps with ip'n > 0 and converging uniformly on compact sets to ipx. Then for every n the map ipn°U is a viscosity subsolution of equation (8) by Lemma 2.9. By the stability of viscosity subsolutions with respect to uniform convergence (see, e.g., [13, 6]) we get the claim. The sequence 4>\°U decreases as A —* +oo to the indicator function
«"-{ 0 -.:t£ Since the sequence is decreasing and all the maps are upper semicontinuous, the pointwise limit I coincides with the upper weak limit (or half relaxed semilimit) of the sequence defined as limsup*i/'A ° U(x) := inf s u p j ^ o U(y)\\x — y\ < S, 0 < — < <5}, A-.00
6
>°
A
328 see [9, 13, 6]. It is well known that viscosity subsolutions are stable with respect to the upper weak limit [9, 13, 6]. Therefore the indicator function of K is a viscosity subsolution of equation (8). It is easy to check by the definitions that for x € K : J2'+I(x) for x 6 dK : J^+I(x) = Nx(x). Therefore we have proved that Vz G dK, V(p, Y) eNl{x):
and that
f(x) • p + trace \a(x) • Y] > 0,
and by the Invariance Theorem the closed set K is invariant.
3
= J^+I(x),
•
Viability and a.s. stabilizability of controlled systems
In this section we consider a controlled Ito stochastic differential equation: ^
dXt = a(Xt, at)dBt + f{Xt, at)dt, 1 X 0 = *.
t > 0,
We assume that at takes values in a given compact metric space A, f, a are continuous functions defined in RN x A, taking values, respectively, in R^ and in the space of N x M matrices, and satisfying ||cr(a;,Q;) — a(y,a)\\ < C\x — y\, for all x,y e Cl and all a e A, \f{x,a) - f{y,a)\
< C\x-y\,
(10)
for all x,y € fi and a l i a £ A.
(11)
For the precise definition of the set of admissible control functions A, or admissible systems, we refer to [21, 14, 18]. Definition 3.1 (almost sure stabilizability). The null state X = 0 is almost surely stabilizable for (CS) if for every e > 0 there exists S > 0 such that for every x with \x\ < S there exists an admissible control function a. e A with the property that the corresponding trajectory X. starting at x verifies \Xt\ < s for allt>0 almost surely. We define a(x,a) := ^a(x,a)a(x,a)T. The appropriate definition of a Lyapunov function for almost sure stabilizability is the following. Definition 3.2 (control Lyapunov function). A function V : RN —• R + is a control Lyapunov function for (CS) if it satisfies the conditions (i), (ii) and (Hi) of Definition
2.3 and (iv ') it is a viscosity supersolution of the equation max {-DV(x)
• f(x, a) - trace [a(x, a)D2V(x)] } = 0 in RN \ {0},
(v ') for every (p, Y) £ J2'~V(x)
and i = 1,... , M
ai(x,a) • p = 0 Va e argmax{—p • f(x,a)
— trace[a(x,a)Y]}
.
329 If we want to follow the same approach as in the previous section, now we must prove that the sublevel sets of the Lyapunov function are viable, or controlled invariant, for (CS), in the following sense. Definition 3.3. A closed set K C RN is viable for the stochastic system (CS) if for all initial points x £ K there exists an admissible control a, € A such that the corresponding trajectory X, satisfies Xt 6 K for allt > 0 almost surely. This property was studied by Aubin and Da Prato [3], Buckdahn et al. [11], and, more recently, by Bardi and Jensen [8]. The last authors introduced a weaker notion named e-viability: Definition 3.4. The closed set K is e-viable for (CS) if for all bounded uniformly continuous functions I > 0, I = 0 in K, and all T > 0, inf E 7 CC.&A
where X, = Xa
l(Xt)dt = 0,
MxeK,
JQ
is the solution of (CS) starting at x.
The main result of [8] is the following equivalence between the e-viability of a closed set K and a Nagumo-type geometric condition. Theorem 3.5 (e-viability theorem [8]). Given a closed set K C MN the following properties are equivalent: (i) K is e-viable for (CS); (ii) for all A > 0 and all bounded uniformly continuous functions I > 0, I = 0 in K : f+OO
inf E I a.€A
l(Xt)e-udt
= 0
Vxeif;
J0
(Hi) the geometric condition \/x E dK, V(p, y ) e A / J ( i ) , Ba<=A:f(x,a)-p
+ trace [a(x, a)Y] > 0
(12)
holds. This theorem combined with an existence result for optimal controls in [18] has the following consequence. Corollary 3.6 (Viability theorem [8]). Assume in addition {(a(x,a), f(x,a))
: a e A}
is convex for all x e K'.
(13)
Then K is viable for (CS) if and only if (12) holds. By means of these results we can prove a Lyapunov theorem on the a.s. stabilizability of controlled degenerate diffusion. Its complete proof will appear in a forthcoming paper.
330 Theorem 3.7 (a.s. stabilizability). Assume that there exists a control Lyapunov function (Definition 3.2) continuous at 0 and the controlled system satisfies the convexity condition (13). Then the null state X = 0 is almost surely stabilizable for (CS). Remark 3.8. If the convexity condition (13) does not hold, we can use relaxed controls as defined, e.g., by Haussmann and Lepeltier [18]. The Nagumo-type geometric condition (12) turns out to be equivalent to the viability of K by means of relaxed controls, see Corollary 2 in [8]. Therefore we can show that the existence of a control Lyapunov function continuous at 0 implies that the null state X = 0 is almost surely stabilizable by means of relaxed controls. All the details, extensions and variants of this result will appear in a forthcoming paper.
References [1] J.P. Aubin, Viability theory. Birkhauser, Basel, 1991. [2] J.P. Aubin, G. Da Prato, Stochastic viability and invariance. Ann. Sc. Norm. Sup. Pisa (IV) 17 (1990), 595-613. [3] J.P. Aubin, G. Da Prato, The viability theorem for stochastic differential inclusions. Stochastic Anal. Appl. 16 (1998), 1-15. [4] J.P. Aubin, G. Da Prato, Stochastic Lyapunov method. NoDEA Nonlinear Differential Equations Appl. 2 (1995), 511-525. [5] A. Bacciotti, L. Rosier, Liapunov functions and stability in control theory. Lecture Notes in Control and Information Sciences 267, Springer-Verlag, London, 2001. [6] M. Bardi, I. Capuzzo-Dolcetta, Optimal control and viscosity solutions of HamiltonJacobi-Bellman equations. Birkhauser, Basel, 1997. [7] M. Bardi, P. Goatin, Invariant sets for controlled degenerate diffusions: a viscosity solutions approach. In "Stochastic Analysis, Control, Optimization and Applications: A Volume in Honor of W.H. Fleming", W.M. McEneaney, G.G. Yin and Q. Zhang eds., pp. 191-208, Birkhauser, Basel, 1999. [8] M. Bardi, R. Jensen, A geometric characterization of viable sets for controlled degenerate diffusions. Preprint 15, Univ. di Padova, 2001. [9] G. Barles, B. Perthame, Discontinuous solutions of deterministic optimal stopping time problems. RAIRO Model. Math. Anal. Numer. 21 (1987), 557-579. [10] E.N. Barron, R. Jensen, Semicontinuous viscosity solutions of Hamilton Jacobi equations with convex Hamiltonians. Comm. Partial Differential Equations, 15 (1990), 1713-1742. [11] R. Buckdahn, S. Peng, M. Quincampoix, C. Rainer, Existence of stochastic control under state constraints. C. R. Acad. Sci. Paris Ser. I Math. 327 (1998), 17-22.
331 [12] F.H. Clarke, Yu. Ledyaev, R.J. Stern, RR. Wolenski, Nonsmooth analysis and control theory. Springer-Verlag, New York, 1998. [13] M.C. Crandall, H. Ishii, P.L. Lions, User's guide to viscosity solutions of second order partial differential equations. Bull. Amer. Math. Soc. 27 (1992), 1-67. [14] W.H. Fleming, H.M. Soner, Controlled Markov Process and Viscosity Solutions. Springer-Verlag, New York, 1993. [15] P. Florchinger, Feedback stabilization of affine in the control stochastic differential systems by the control Lyapunov function method. SIAM J. Control Optim. 35 (1997), 500-511. [16] W. Hahn, Stability of motion. Springer-Verlag, New York, 1967. [17] R.Z. Has'minskii, Stochastic stability of differential equations. Sjithoff and Noordhoff International Publishers, 1980. [18] U.G. Haussmann and J.P. Lepeltier, On the existence of optimal controls, SIAM J. Control Optim. 28 (1990), 851-902. [19] M. Kocan, P. Soravia, Lyapunov functions for infinite-dimensional systems. J. Funct. Anal, to appear. [20] H.J. Kushner, Stochastic stability and control. Academic Press, New York, 1967. [21] P.-L. Lions, Optimal control of diffusion processes and Hamilton-Jacobi-Bellman equations. Part 1: The dynamic programming principle and applications, Part 2: Viscosity solutions and uniqueness. Comm. Partial Differential Equations 8 (1983), 1101-1174 and 1229-1276. [22] E.D. Sontag, Stability and stabilization: discontinuities and the effect of disturbances. In "Nonlinear analysis, differential equations and control" (Montreal, QC, 1998), F.H. Clarke and R.J. Stern eds., pp. 551-598, Kluwer Acad. Publ., Dordrecht, 1999. [23] E.D. Sontag, H.J. Sussmann, Non smooth control Lyapunov functions. Proc. IEEE Conf. Decision and Control, New Orleans, Dec 1995, IEE Publications, 1995. [24] P. Soravia, Stability of dynamical systems with competitive controls: the degenerate case. J. Math. Anal. Appl. 191 (1995), 428-449. [25] P. Soravia, Feedback stabilization and H-infinity control of nonlinear systems affected by disturbances. In Proceedings of the Third Nonlinear Control Network Workshop "Dynamics, Bifurcations and Control", Springer Lecture Notes in Control and Information Sciences, to appear. [26] J. Yong, X.Y. Zhou, Stochastic controls, Springer-Verlag, New York, 1999. [27] J. Yorke, Differential inequalities and non-Lipschitz scalar functions. Math. System Theory 4 (1970), 140-153.
Mass transport through charged membranes Diether Bothe 1 and Jan Prtiss 2 Fachbereich Mathematik und Informatik Universitat-GH Paderborn War burger Str. 100 D-33098 Paderborn, Germany 2 Fachbereich Mathematik und Informatik Martin-Luther-Universitat Halle-Wittenberg Theodor-Lieser Str. 5, D-60120 Halle Email : [email protected] ; anokd@volterr a. mathematik. uni-halle. de 1
Abstract A modern technique for desalination or softening of water is the so-called nanofiltration by means of membranes which carry a fixed electric charge. For the mathematical modeling of such processes, two features are particularly important. Firstly, the distribution of ionic species generates an electric field which in turn affects the fluxes of these ions. Therefore, besides diffusion and convection, electromigration has to be taken into account. This also leads to a strong coupling of the concentrations of all charged species, which can often be adequately incorporated into the model via the assumption of electroneutrality. Secondly, electrical double-layers (Donnan potential) build up at the surface of the membrane, which cause discontinuities in the ionic concentration profiles. We deduce a mathematical model for such a nanofiltration process. This leads to a strongly coupled quasilinear parabolic system with nonlinear transmission and dynamical boundary conditions. By means of degree theory we obtain existence of stationary solutions, while L^-maximal regularity is employed to get local strong wellposedness of this model.
1
Introduction
Nanofiltration is a pressure driven separation process t h a t employs specific membranes consisting of a polymeric matrix which carries fixed ionic groups. If the void space of such a membrane is filled in with an aqueous solution, these groups dissociate by separation of mobile ions like protons or hydroxide ions. Since the latter enter t h e aqueous phase, a fixed electric charge inside t h e polymeric matrix remains. Due t o t h e fact t h a t ionic solutions are electrically neutral on the macroscopic scale, this leads to a considerable change of t h e composition inside the membrane, compared t o the adjacent bulk phases.
332
333 This effect forms the heart of nanofiltration processes and allows for the selective removal of ionic species with multiple charges like calcium, magnesium, sulfate, phosphate etc. For technically relevant nitration processes sufficiently high flow rates across the membrane are necessary which require a large surface of the membrane. A common way to achieve this within a reasonable volume is to pile up several membranes in so-called disctube modules. Such modules are constructed in such a way that the feed solution flows tangentially along one membrane after the other. A certain amount of the solution is pressed through the membrane to yield the desired permeate, while the remaining fluid collects the rejected ions and leaves the system as so-called retentate with higher concentrations. By means of this cross-flow operation, blockage of the membrane due to precipitation is avoided. Still, a thin film of microorganism will settle down on the surface of the membrane a problem present in any aqueous system. This fouling slows down mass transport and leads to increased concentrations of ions inside the permeate phase. In fact, the rejection rate then decreases for large pressure differences. Concerning the intensification of such nanofiltration processes, a fundamental understanding based on mathematical models is essential. For more detailed information concerning the technical background see [6] and in particular [5].
2
Mathematical modeling
We develop a mathematical model that incorporates the relevant mass transport phenomena mentioned above, while appropriate simplifications concerning the hydrodynamics are introduced. To be specific, we suppose the bulk phases containing retentate or permeate to be ideally mixed and assume stationary hydrodynamics. The latter implies V' = VR + VP, where V' denotes the volumetric flow rate for j = f (feed), R (retentate) or P (permeate), respectively. We consider a single planar membrane carrying a feed-sided film inside of which the species have reduced mobility. The latter represents a combination of the biofilm mentioned above and a hydrodynamic boundary layer (the so-called Nernst boundary layer). Since the concentrations inside ideally mixed bulk phases are homogeneous, mass transport across the membrane is solely oriented in the normal direction as illustrated by Figure 1. Consequently, the mathematical model involves only one spatial dimension. Mass
334 permeate]
retentate
M
C
retentate bulk phase
in
«-"l
biofilni .mil bouml.i^ I'-iM'i'
acti\i' membrane
permeate bulk phase
feed
0
^
s,'M
Figure 1: balances for all species implies retentate : biofilm :
^-{VRcf)
= Vfc{ - VRcR -
• div Jf M
membrane :
c f + div Jt
permeate :
(Vc?) = -
AFJf
t>0,
0
t >0,
-SF < x < 0,
= 0
t > 0,
0 < x < <5A
( +
A"JM=SM
t > 0.
(1) Here we use the following notation: subscripts denote the species number, while superscripts indicate the corresponding phase. Hence c^ means concentration of species i inside phase j , V^ are the bulk volumes and Ai is the interfacial area between phase j and the adjacent bulk phase. So, in the sequel all the hydrodynamic quantities like the volume fluxes Vj, the bulk volumes V\ the interfacial areas Aj, as well as the film and membrane diameters Sj are positive constants. In electrochemical systems, mass fluxes of charges species are not only due to convection and diffusion. The distribution of charges creates an electrical field which causes additional electromigration of ions. A commonly accepted constitutive equation for the
335 fluxes (see [7]) is given by
Jl = -Vt%
+
J4-^M%
U = FlM),
(2)
j
where D\ > 0 denote the diffusivities, v the convective speed of the solvent, z* the charge numbers and <jP the electric potential. Furthermore F is the Faraday constant, R the universal gas constant and T the absolute temperature. Let us note in passing that the Nernst-Einstein equation, relating the mobility of ions to their diffusivity, has already been incorporated into (2) for technical simplicity. To obtain a closed model, an equation for the electric potential has to be added. If electrodynamical effects are neglected, a reasonable choice would be the Poisson equation F "
A0»' = —e-5>c*, where e is the permittivity of the electrolyte solution. Unfortunately, accurate boundary conditions for <jp are hard to get and e depends on the composition of the electrolyte in a complicated, often unknown way. On the other hand, on the macroscopic scale the solution is electrically neutral to a high degree of accuracy inside bulk phases. Mathematically this corresponds to the fact that F/e has a very large value. Therefore we add the condition of electroneutrality, i.e. the algebraic constraint
Equation (3) refers to membranes with fixed negative charges in which case eft1 denotes an equivalent concentration of negative ions with charge number -1. Finally, the model has to be complemented by boundary and transmission conditions. These are continuity of the fluxes and continuity of the electrochemical potentials. While the former directly translates into Jf(0) = j f ( 0 ) forz = l , . . . , n ,
(4)
the latter needs some more explanation. The electrochemical potential fa of a species i is given by
fi^rf
+ RTlnai + ZiFfr
where fi° is the chemical standard potential and a{ is the activity. For dilute solutions, <jj can be replaced by the concentration c* which will be done in the sequel. At interfaces where <j> is continuous, this implies continuity of the concentrations, i.e. cf = cf(~SF)
fori = l , . . . , n .
(5)
The same cannot be true at the surfaces of the membrane, since this would not be consistent with electroneutrality. At the film-sided interface, continuity of the electrochemical potential yields In cf (0) + {fti^iO)
= In c f (0) +
^i
336 hence c f (0) = cf (O)e^
with a = ^ ( ^ ( 0 ) - ^ M (0)).
Electroneutrality inside the membrane yields
f > c f (0)e<-= cf, which uniquely defines a for all compositions such that cf (0) > 0 for some i with Zi ^ 0; notice that tp(a) = Yl"=i ZiCieaZi is strictly increasing such that lima-,±oo il>(a) = ±°oObserve also that a ^ 0 in case c f / 0 which is physically comprehensible: due to the fixed negative charge density inside the membrane, an electrical double layer builds up at the surface. On the macroscopic scale this leads to a discontinuity of the electric potential which causes a jump in the concentration profiles of charged species. The difference (j>F(0) — >M(0) is the so-called Donnan potential. Since the same arguments apply at the interface between membrane and permeate, we obtain the conditions c f (0) = cf (0) gf (cF (0)),
c f (6M) = cf gf(cp)
(6)
with function gf ,gf implicitly given by gf(c) = e"" O J2 z^e"Zi
= coM-
9f(c) = ef"> *• £
zlCie0» = cf.
(7)
In summary, the complete model has the form of a partial differential-algebraic system consisting of the balance equations (1) with fluxes given by (2), the transmission and boundary condition (4), (5), (6) and the algebraic condition (3) of electroneutrality. There are only few papers about electrochemical systems in the mathematical literature. While results for related problems from electrolysis can be found in [1], [2], [8] and [9], rigorous investigations of membrane processes including the appearance of Donnan potentials are apparently missing.
3
The stationary problem
In the stationary case the transport equations in (1) reduce to div j{ = 0, hence by continuity of the mass fluxes and (1), they satisfy = vpcf
jf{x) p
with v
P
M
:= V /A .
{-SF<x<0),
JtM(x)=vpcf
{0<x<SM)
Then (2) yields
_^+!M_^JL^ = ^ r
{1-FM]
A [J P m dx+ Dj RTdx Dj - ' >Multiplication with z4 and summation over all species, accounting for electroneutrality, gives
F fd4?\ A
2
, A
v^4-vpcf
,
337 This allows for elimination of the electrical field from the mass fluxes. Together with the remaining retentate balance as well as (5) and (6), we obtain the following boundary value problem. For clarity, we use the abbreviations c
' = (ci>-">4)i
D_j
=diag(l/dj,...,l/4),
z = (zu ... ,zn)
Z = diag(*i,... ,zn).
With this condensed notation, the stationary problem reads as
r)cM
D M cM
(h
Ti~M(i)McM -
-h = - ^ -^)- ' t,W
vpcp\\
ZcM
(8)
°
<X<6M
'
with boundary and transmission conditions QF( — g„)
c"(0) cM(6M) yfcf
=
QR
= c^(0)-g^(0)), = cp.gf(c^ =
vRcR
+
AM
(9)
~Vpcp,
where a dot stands for componentwise multiplication. Theorem 3.1 For a given nonnegative and electroneutral feed composition c?, the stationary problem (8), (9) has a nonnegative classical solution. Proof. If the feed contains no charged species, then the problem is a purely diffusional one. This case is much simpler and will not be discussed here. Hence we assume that q > 0 for some index i with Zj ^ 0. We reduce to a fixed-point problem, where it is favorable to solve the differential equations backwards, since c p is involved in the boundary condition and appears on the right-hand sides in (8). Given any permeate composition cp > 0 such that (z,c p ) = 0 and cf > 0 for some i with Zi ^ 0, the permeate-sided membrane concentration CM(<5M) is uniquely determined by (9). Notice that cM(<5M) > 0 depends continuously on cp with cf (6M) > 0 iff cf > 0. For this initial value, the second ODE-system in (8) has a unique backward solution up to x = 0. Indeed, the right-hand side is locally Lipschitz continuous with at most linear growth in c M . Hence a unique local solution exists as long as (z, Z c M ) > 0. The latter holds on all of [Q,6M] if the solution stays nonnegative, since (z,c M (fi M )) = cf together with — (z, c M ) = 0 implies ox
M Co
= £ ^ f <£Ncf<£^, M
338 The solution stays in fact nonnegative, since cM(x0) > 0 and cf{x0) = 0 imply gcM
P P
The same argument yields cf*(0) > 0 iff cf > 0. To pass to the film, we rewrite the transmission condition at x = 0 as cf
(0) = CM{0)e-aZi
with a such that ^
Zicf
(0)e-aZi
= 0.
Therefore cp > 0 continuously determines a unique c F (0) > 0 such that (z, 0^(0)} = 0 and cf (0) > 0 iff cf > 0. A similar reasoning as given above shows that the first ODEsystem in (8) with this initial value has a unique backward solution up to x = — 6p- Here l{x) — X)"=i c f (x) satisfies
hence the solution exists up to x = —8p and cF(—6p) > 0 iff cf > 0. Let F(cp) denote the vector cR = CF(—SF) determined by cp this way. To sum up,
F:D{F) = En{ceWl:{z,Zc)>0}-*Enm.l
with E = {c £ R" : (z, c) = 0} (10)
is continuous with Fj(c) > 0 iff c, > 0. Moreover, the above constructed solutions evidently solve the stationary problem (8), (9) iff c.p satisfies p c
^(tCf-^F{cp)). AF \vp Vp I
(11)
We shall apply degree theory to obtain a solution of this fixed-point problem, where we work in the Banach space E (with E"-norm). Let AMVf , c = F P-.— cA V
„, , G(c) = c ^'
AMVR T-
AFVP
F(c)
and fi = {c e E : 0 < Ci < cit y ^ cfc > rj}, where r) > 0 is to be chosen later. Evidently O is open bounded convex and G is continuous on H. To obtain existence of a fixed-point of G we shall use homotopy invariance of the degree (cf. [3]) to pass to a simpler problem. For this purpose we define a homotopy H(cF,\) as follows. Replace D>byDt(\) = (l-\) + XDi ij = F,M), p v> by v>{\) = (1 - X)v + \v> (j = F, M),
339 Given A € [0,1] and cp > 0, repetition of the above arguments yields a corresponding c which gives F(cp, A) := cR. We then define the homotopy H by R
AMVR H(cp,\)=c--JF^F(cp,X). Obviously H(-, •) is continuous due to the continuous dependence of the solutions of (8) on the parameters and initial values. We claim that H(-, A) has no fixed-point on dfl if 77 > 0 is sufficiently small. Assume that H(cp, A) = cF for some A € [0,1] and cp 6 0. If c p £ dQ then there are three possible cases. (i) There exists some i such that cp = 0. Then Fi(cp, A) = 0 yields the contradiction Hi(cp,X)
=Ci>
cP.
(ii) There exists some i such that c[ = ct. Then Fi(cp, A) > 0 yields the contradiction
Hi(cp,\)=ct-Fi(cp,\)
(iii) The remaining case 2 Sfc #o ck breviations z£ = max{zk,0},
=
V needs some preparation. Below we use the ab-
zk = max{-zk,0},
z+ax = max* zk, z^K = ma,xk(~zk),
vm*x = max{vF,vM,vp},
d? = min{l,D{,...
,D{};
observe that D\ (A) > d? holds for any A £ [0,1]. Let Vf X^ P = ^ >
VR+n
f
AFVP
n
Ci > 0,
<7 =
n
— > 0,
AMVR
notice that
X c * = X F* (°p >A) = p -a X ck = p - ar> and recall (z,c B ) = 0. Hence
P - <"? = X Ck< X Z*C* + X C* ^ X ^ Zfc5^0
Zfc>0
z fc <0
+ Zm«)
z fc <0
shows that there exists some i such that * < 0 and cf = cf(-tf F ) >
P ail
' _
.
(12)
We fix this index i and use differential inequalities to get the contradiction cf > 77 for
340 small r\ > 0. Due t o zt < 0, the ODEs (8) imply 84
vi{\)4-vpcf
-z4
f
zkc?k
zk4
> owi-*"* _ ^ ( A ) E - < ° ^ ( A ) - ^ ,
>
ZjCfo j (A) _ ?> P (1 + Z + a x )
^"(A)4
~ -D^'(A)
*
d?>
^max^max j _ " P ( l + 4 a x ) v
-
d?
{
&
V
'
hence ^
> _a»c| - b ^ with o? =
max msx
,
, V=
I 1 +,*•»"•'.
Consequently, cf(x) > c[(-6F)e-aF^+s^ M
aMx
c, (z) > cf{Q)e-
- 7) bF/aF M
M
- r) b /a
on [SF, 0] on [0,8 M ].
By means of (12) we obtain cf(0)>e-°F^
P aTI
~
-77—
hence there is ??0 > 0 such that -aFSF
whenever 0 < r\ < rj0. To pass to the membrane, we need an estimate from above for a. Exploitation of (6), (3) and c ^ > eg* (A) as well as a > 0 yields n
n
c0M > Ylz^k(0)eaZk
= Ylcl(0)^(eazt
- 1) > cf(0)zt(e°*'
- 1)for any I.
To obtain an appropriate index Z, notice that cf (0) < £ *fc-cf (0) = J2 4 c f ( 0 ) < n ^ m a x c f (0). Zk<0
zk>0
Thus there is an index I such that z; > 0 and cf (0) > cf (Oj/nz+^j. Therefore
(14)
341 due to (14), hence if we start with n < n0 then eQ < K by (13) with some K > 0 independent of A G [0,1]. Consequently, cf (0) = cf (OK" > cf (0)K->™* >
pe a 2n
~ "_ K - - . ( l + 2max)
Since (5 > 0, this finally implies V>c?>
cf(SM)
>
F
zn^i -t- z max j
K->™ - r,~M for all 0 < r, < a
%,
a contradiction for sufficiently small n > 0. Therefore the claim holds, hence
d(/-G,n,o) = d(/-if(-,i),n,o) = d(7-if(-,o),n,o). The case A = 0 corresponds to D\ — 1, iP = vp and c ^ = 0, where the solution is given by F P M P R P I C = C , C = C , C" = C ( =
A :
V :
y. CT).
v
AFVP + AMVR This yields H(c . 0) = c /a, hence (eventually after adjustment of rj) p
p
d(I - H(-,0), ft, 0) = d((l + - ) / - c, ft, 0) = 1. a Consequently there is at least one fixed-point of G, i.e. a solution of the stationary problem. • Although uniqueness is to be expected from the physical viewpoint, this has not yet been proven mathematically.
4
The instationary problem
Electroneutrality again allows for elimination of the electrical field which leads to a strongly coupled quasilinear parabolic system with nonlinear transmission and dynamical boundary conditions. Of course we now also assume inital conditions c«(0)=c«
c F (0,-) = 4 o n [ - 5 F , 0 ] ,
c M (0, •) = c0M on [0,5 M ],
< = » = cp.
(15)
Theorem 4.1 Fix p > 3/2. Suppose that all initial values are strictly positive satisfying CF
g
W2-2/P([_SFI
„]; R B ) J
C
Me
W ^ - 2 / p ( [ 0 i SM].
Rn)t
the electroneutrality conditions (z,c£) =
(z,cF) = 0 a.e. on [SF,0],
(z,c^) = c^ a.e. on [0,SM],
342 as well as the compatibility conditions corresponding to (4), (5) and (6). Then there is a unique solution {cR,cF ,cM ,cp) of the instationary problem (1), (3), (4), (5), (6) on a time interval [0,T] for some T > 0. The solution is continuous with values in the phase space R" x W%-2/p([-6F,0];Rn)
x Wp2-2/p([0,<5M];Rn) * »",
stays electroneutral and depends continuously on the data. The differential equations are satisfied in the strong IP-sense and the boundary as well as transmission conditions hold pointwise. The proof of this result is based on I^-maximal regularity (cf. [4]) of an appropriately linearized system and will be given in a forthcoming paper. Acknowledgement. The subject of this paper was brought to our attention by D. Jakobs and H.-J. Warnecke, University of Paderborn. We appreciate many helpful discussions and are grateful for the continuing cooperation.
References [1] H. Amann, M. Renardy, Reaction-Diffusion problems in electrolysis. NoDEA 1 (1994), 91-117. [2] Y.S. Choi, R. Lui, Multi-dimensional electrochemistry model. Arch. Rational Mech. Anal. 130 (1995), 315-342. [3] K. Deimling, Nonlinear Functional Analysis. Springer-Verlag, Berlin, 1985. [4] R. Denk, M. Hieber, J. Prtiss, iJ-Boundedness, Fourier Multipliers and Problems of Elliptic and Parabolic Type. Reports of the Institute of Analysis, Univ. HalleWittenberg 2001. [5] D. Jakobs, Stofftransport durch Nanofiltrationsmembranen unter Berflcksichtigung von Biofilmen. PhD-Dissertation Univ. Paderborn 2001. [6] M. Mulder, Basic Principles of Membrane Technology, 2 nd edition. Kluwer, Dordrecht, 1997. [7] J.S. Newman, Electrochemical Systems, 2 nd edition. Prentice Hall 1991. [8] S.I. Pohozaev, On the reaction-diffusion electrolysis nonlinear elliptic equations. Lect. Notes Pure Appl. Math. 194, 255-288 (1998). [9] J. Wiedmann, An electrolysis model and its solutions. PhD-Dissertation Univ. Zurich 1997.
The Hopf solution of Hamilton - Jacobi equations* Italo Capuzzo Dolcetta Dipartimento di Matematica, Universita di Roma - La Sapienza, P.le A. Moro 2, 00185 Roma, Italy Email : [email protected]
1
Introduction
The Hopf formula u(x,t) = inf
«,(„) + « r ( ^ ) ]
(i.i)
provides a simple representation for the solution of the Hamilton - Jacobi equation ut + H(Dxu) = 0 , (x,t) eJRN x (0,+oo)
(1.2)
equipped with the initial condition u(x,0)=g{x) . l e R " .
(1.3)
Indeed, formula (1.1) expresses the solution u of the Cauchy problem (1.2), (1.3) as the optimal value of an unconstrained minimization problem on R " , parametrized by (x,t), which involves the initial datum g and the Legendre - Fenchel transform H* of the given (convex) Hamiltonian H. The first part of this paper comprises a quick review of selected well - known results about the interpretation of the Hopf function as a global generalized solution of (1.2), (1.3), either in the weak sense of Kruzkhov as well as a viscosity solution a la Crandall - Lions. The first section is completed by some remarks concerning the asymptotic behaviour of the classical integral representation formula for the solution of a standard parabolic regularization of the Cauchy problem above: using the Varadhan's Large Deviations Principle (see [16]), the Hopf function is interpreted as the limit of the solutions of the regularized problem as the diffusion coefficient vanishes. The second part is devoted to the presentation of a new result, due to the collaboration with H. Ishii [7], which shows that the Hopf representation formula still holds for state dependent Hamiltonians H = H(x,p), provided H has an associated stationary equation 'Work partially supported by the TMR Network Viscosity Solutions and Applications
343
344 which is of eikonal type and, more precisely, which exhibits a metric character in terms either of a Riemannian distance on R.^ or, when some degeneracies in the x - dependence are present, in terms of a Carnot - Caratheodory type distance. For the sake of simplicity the results will not be reported at the highest possible level of generality; the interested reader is rather referred to the bibliography (see [1], [2], [7], [8], [9], [11]).
2
T h e Hopf function as a generalized solution
Two preliminary simple observations are that for affine initial datum g(x) = q • x + c the smooth solution of (1.2), (1.3) is
v(x,t) =
g{x)-tH(Dg(x))
and that, for general g, the functions v>*(x,t)=g(y)
+
q-(x-y)-tH(q)
N
solve (1.2) for any choice of (y, q) e TR x IR" but do not satisfy (1.3). It is not hard to realize that the envelope procedure proposed by E. Hopf [9], namely to take inf sup vv'q(x,t) , yem." ,6]Rw which defines indeed the function u in (1.1), preserves, at least at points of differentiability of u, the fact that each vy
is convex
,. H(p) hm -+Y- = +oo
|p|-»+o°
there exists
\p\
G > 0 : \g{x) — g(y)\ < G\x — y\ for all x, y.
(2-1) , s (2.2) (2.3)
Then, the Hopf function
u(x,t)=MN{g(y)
+
tH<(^l)
(2.4)
is Lipschitz continuous on IRJV x (0,+oo), satisfies the equation (1.2) almost everywhere and lim u(x,t) = g(x) at any x € IR" .
345 By a well - known result of S. N. Kruzkhov [10], uniqueness for problem (1.2), (1.3) holds in the class of Lipschitz continuous functions which are semiconcave, that is u(x + h,t)-
+ A \h\2
2u{x,t) + u{x -h,t)<(c
(2.5)
holds for some C > 0 and any x, t, h. It is easy to check that if the initial datum satisfies g(x + h)-2g(x)
g(x-h)
+
for some constant C > 0 , then the Hopf function satisfies condition (2.5) uniformly in t > 0 and, consequently, is the unique weak solution of (1.2), (1.3) in the sense of Kruzkhov. An alternative condition for (2.5) to hold is the uniform convexity of H (see ([8])The analysis of the Hopf function took later a new impulse with the work of P.L. Lions [11] and M. Bardi - L.C. Evans [2]. The use of the notion of viscosity solution which makes sense even for merely continuous functions and of the comparison results available in that theory allow to interpret the Hopf function as the unique global solution of (1.2), (1.3) in a larger class of functions. A typical result in this direction (note that condition (2.2) is not assumed there) is the following one taken from [2]: T h e o r e m 2.2 Assume (2.1) and (2.3). Then, the Hop} function (2.4) is the unique uniformly continuous function on ]RN x (0, +oo) which satisfies equation (1.2) in the viscosity sense and lim u(x,t) = g(x) at any x € TRN . A classical method to construct solutions of first order fully nonlinear partial differential equations is through parabolic regularization. Let us discuss briefly this issue in the special case H(p) = ||p| 2 . Assume g bounded and consider the parabolic regularization of the Cauchy problem (1.2), (1.3), that is u\ - eAxu< + i | I > x u f = 0 , ue(x, 0) = g(x) ,
(2.6)
where e is a positive parameter. The Hopf - Cole transform of ue, namely we = e~^i satisfies then the linear heat problem w\ - eAxwe = 0 , wc{x, 0) = e" 2 ^ 1 . By classical linear theory, wc has the integral representation w (x, t) = (4iret)
2
e
4
<< e
2
< ay ;
(2.7)
346
hence, «£(x, t) = -2e log ((47ret)-f J
e-^SV^dy)
(2.8)
is a solution of the quasilinear problem (2.6). It is natural to expect that the solutions ue of (2.6) should converge, as e —> 0 + , to the Hopf function of problem (1.2), (1.3) which in the present case is given by u(x, t) = inf
9(y)
(2.9)
it
This can indeed be proved as a simple application of a general large deviations result by S.N. Varadhan. Making reference to the notations in [16] Theorem 2.2, it is enough to consider the family of probability measures Px ( defined on Borel subsets of IR^ by
P^t{B) = (tort)"* JB /
e-^dy
and the rate function
i*Av) = l-^f • The above mentioned theorem states in fact that
lime log (f £-+0+
e^dPZ/yj)
Vfft"
= sup [F(y) - I(y)\ /
ycTuN
for any bounded continuous function F; for F = — § the above gives (2.9). For a direct asymptotic analysis (i.e. not making use of the explicit representation (2.8)) see [11], [3]. A further remark about the approach described above is that the Hopf - Cole transform can be also used in a similar way to deal with the parabolic regularization of more general equations such as ut + -\a(x)Dxu\2
=0
where a is a M x N given matrix, provided the regularizing second order operator is chosen appropriately. Indeed, if one looks at the regularized problem u\ - e div (a*{x)a(x)Dxu€) + -\a(x)Dxue\2
= 0,
then the Hopf - Cole transform we — e~ ar solves the linear equation w\ — e div (a*(x)a(x)Dxwe)
= 0 .
This observation will be developed at the purpose of asymptotic analysis of Hamilton - Jacobi equations in a forthcoming work.
347
3
A n Hopf t y p e formula for state dependent Hamiltonians
In this section we present a new Hopf type formula, obtained in collaboration with H. Ishii (see [7]), for the viscosity solution of the state - dependent Cauchy problem iH + H(x, Dxu) = 0 , (x,t)&JRN N
u{x,0) = g(x) , x€B.
x (0, +oo)
(3.1)
.
(3.2)
It is easy to realize that the Hopf envelope method does not work if H depends on x. However, as we shall show below, an Hopf type formula can be proved even in this more general case under the basic structural assumption that the Hamiltonian H : JR2N i-> 1R is of the form H(x,p) = 9(H„{z,p)) (3.3) where H0 is a continuous function on JR2JV satisfying the following conditions p t-> H0(x,p) convex , H0(x, \p) = \H0(x,p)
, (3.4)
H0{x,p) > 0 , \H0(x,p) - H0(y,p)\ < u(\x - »|(1 + |p|)) for all x, y, p, for all A > 0 and for some modulus u such that lims_>0+ w(s) = 0 . Concerning function <& we assume $ : [0, +oo) —> [0, +oo) , convex , non decreasing , $(0) = 0 .
(3.5)
The next result shows that the validity of an Hopf type formula for the solution of problem (3.1), (3.2) is guaranteed if the associated stationary eikonal problems H0{x,Dxd) = l,xeJRN\{y} d{y) = 0
(3.6) (3.7)
have a solution d(x) = d(x; y) for any value of the parameter y e IR". We shall describe below a setting in which this condition can be enforced. Theorem 3.1 Assume (3.3), (3.4), (3.5) and g lower semicontinuous , g(x) > —C(l + \x\) for some C > 0 .
(3.8)
Assume also that for each y G IR^ problem (3.6), (3.7) has a unique nonnegative continuous viscosity solution d(x) = d(x;y). Then, the function u(x, t) = inf sent"
g(y) +
tVd{x
(3.9)
is the unique lower semicontinuous viscosity solution of (3.1) which is bounded below by a function of linear growth and such that l a n u t l Z } L (y> )
= 9{x) .
(3.10)
348 In order to understand why the Hopf function (3.9) solves (3.1), let us proceed heuristically by assuming that (3.6), (3.7) has a smooth solution d(x) and look for special solutions of the form vy(x,t)=g(y)
+ ty(^±)
(3.11)
where y e IR" plays the role of a parameter and $ is a smooth function to be appropriately selected. Set T = ^y' > 0 and compute vy
= *(r) + < * ' ( r ) ~ = *(r) r Dxvy = W(T)^
=
r^'M
V(T)Dxd.
Imposing that vy solves (3.1) gives $ ( r ) - TV(T) + $(H0{x, V'(T)Dxd)) = 0 . By choosing a strictly increasing \&, the positive homogeneity of H0 yields 0 = ^(T)-T^'(T)+^'{T)H0(x,Dxd)) = tf(r) - rtf'(r) + * ( W ( T ) ) ,
(3.12)
using the fact that d solves the eikonal equation. Since the solution of the Clairaut's differential equation (3.12) is * = $*, the above heuristics lead then to formula (3.9). The rigorous proof of Theorem 3.1 is made up of three basic steps. The first one is to show that the functions vv defined in (3.11) are indeed viscosity solutions of (3.1) for each y, even when d is assumed to be just a continuous viscosity solution of (3.6), (3.7) and $ is a general convex nondecreasing function. This requires, in particular, to work with regular approximations of $, namely
* e ( S ) = $( S ) + | « a and the use of the standard reciprocity formula
M(*:)'( a ))+*:( S ) = S(*e7oo. The other fundamental tool in the proof is the use of the stability properties of semicontinuous viscosity solutions with respect to inf and sup operations in order to show that u is a lower semicontinuous viscosity solution of (3.1) in the sense of Barron - Jensen [4]. The second step is to check the initial condition in the weak sense (3.10) and the fact that u is bounded below by a function of linear growth; this and the final step concerning uniqueness are performed by suitable adaptations of the methods of [1], [4]. We refer to [7] for details. The assumption that the eikonal equation has a unique continuous viscosity solution made in Theorem 3.1 is trivially satisfied when H0(x,p) = \p\.
349 In this case, the viscosity solution of (3.6), (3.7) is d(x;y) = \x — y\ and the Hopf formula becomes x u(x,t) = inf g{y) + t$* '\ ~y\. t a simple argument using the monotonicity of $ shows that
*.(l£^)=(*oji.)-(]£^l and our formula (3.9) reduces then to the classical one (2.4). It is easy to check that the simplest state - dependent case covered by Theorem 3.1 is H0{x,p) = h0{x)\p\ , / i o ( z ) > 7 > 0 . More generally, one can deal with homogeneous Hamiltonians of the form H0{x,p) = \A(x)p\ where A{x) is a symmetric positive definite N x N matrix. The associated eikonal equations are solved in this case by Riemannian metrics (see below and also [11], [15] for previous results of this kind). In all the cases mentioned above, the coercivity condition lim HJx,p) = +oo
(3.13)
|p|->+oo
obviously holds true. Let us discuss now the issue of finding sufficient conditions for the validity of the eikonal assumption in Theorem 3.1 even for degenerate situations when (3.13) fails. We will show next that a solution d of (3.6), (3.7) can indeed be constructed in a quite general setting by a control theory approach. By standard convex analysis arguments it is seen that the mapping x^dHo{x,0) is upper semicontinuous with closed convex values. Consider now the differential inclusion X(i) e dHo{X{i),0)
(3.14)
and, for i , y 6 K " , the set Fx>y of all trajectories X(-) of (3.14) such that X(0) = x, X(T) = y for some T = T (X{-)) > 0 . Define then d{x;y)=
inf
T(X(-))
.
(3.15)
and assume that there exist e > 0 and, for each x G IR^, a n M x J V (with M < N) matrix a(x) 6 C°°(fl w ) such that a(x) satisfies the Chow - Hormander rank condition of order k ,
(3.16)
350 and a{x)([-e,e)M)
CdHo{x,0)
.
(3.17)
Conditions (3.16), (3.17) guarantee by the well - known Chow's Theorem that Fx
(see [13]). It follows that, locally, x —¥ d(x;y) is £ Holder continuous and is a natural candidate to be the required solution of the eikonal equation. Because of the multivalued character of the control system (3.14), several technical refinements to the standard dynamic programming argument (see, for example, [3]) are needed in order to show that d is a viscosity solution of the eikonal equation (3.6). We refer to [7] for a detailed proof. Let us only observe here that in the present setting the function d is not, in general, differentiable almost everywhere; the notion of viscosity solution seems therefore to be essential to interpret d as a solution of (3.6). A large class of examples, which is relevant both from a PDE and control theory point of view, to which our generalized Hopf formula
u(x,t)
=MN[9(y)+t*>(^l
(3.18)
applies is provided by homogeneous Hamiltonians H0 of the form H0(x,p) = \a(x)p\ where a(x) satisfies conditions (3.16), (3.17) above. An interesting particular case (here N = 3 to simplify notations) is Ho(x,P) = J fa - f
P3)
2
+ fa + f p 3 ) 2 ,
arising in connection with the second order subelliptic operator A#i (the Kohn Laplacian on the Heisenberg group, see [14]); note that the differential inclusion (3.14) reduces in this case to the symmetric system in the well - known Brockett's example in nonlinear control theory, see [6]. Our Hopf formula (3.9) coincides in this case with the one recently found for this example by Manfredi - Stroffolini [12].
References [1] O. ALVAREZ, E.N. BARRON, H. ISHII, Hopf formulas for semicontinuous data, Indiana University Mathematical Journal, Vol. 48, No. 3 (1999).
351 [2] M. BARDI, L.C. EVANS, On Hopf's formulas for solutions of Hamilton - Jacobi equations, Nonlinear Anal., TMA 8 (1984). [3] M. BARDI, I. CAPUZZO DOLCETTA, Optimal Control and Viscosity Solutions of Hamilton - Jacobi - Bellman Equations, Systems & Control: Foundations and Applications, Birkhauser Verlag (1997). [4] E.N. BARRON, R. JENSEN, Semicontinuous viscosity solutions for Hamilton - Jacobi equations with convex Hamiltonians, Comm. PDE 15 (1990). [5] A. BELLAICHE, The tangent space in sub - Riemannian geometry, in Sub-Riemannian geometry. Edited by A. Bellaiche and J.-J. Risler. Progress in Mathematics, 144, Birkhauser Verlag, Basel (1996). [6] R. W. BROCKETT, Control theory and singular Riemannian geometry, in New directions in applied mathematics, P. J. Hilton and G. J. Young eds., Springer - Verlag (1982). [7] I. CAPUZZO DOLCETTA, H. ISHII, Hopf formulas for state dependent Hamilton - Jacobi equations, to appear. [8] L.C. EVANS, Partial Differential Equations, Graduate Studies in Mathematics 19, American Mathematical Society, Providence RI (1988). [9] E. HOPF, Generalized solutions of nonlinear equations of first order, J. Math. Mech. 14 (1965). [10] S.N. KRUZKHOV, Generalized solution of the Hamilton - Jacobi equations of eikonal type, Math. USSR Sbornik 27 (1975). [11] P. - L. LIONS, Generalized solutions of Hamilton - Jacobi equations, Research Notes in Mathematics 69, Pitman (1982). [12] J. J. MANFREDI, B. STROFFOLINI, A version of the Hopf - Lax formula in the Heisenberg group, to appear in Comm. PDE. [13] A. NAGEL, E.M. STEIN, S. WEINGER, Balls and metrics defined by vectorfields I: basic properties, Acta Math. 155 (1985). [14] L.P. ROTHSCHILD, E.M. STEIN, Hypoelliptic differential operators and nilpotent groups, Acta Math. 137 247 - 320 (1976). [15] A. SICONOLFI, Metric aspects of Hamilton - Jacobi equations, to appear in Transactions of the AMS. [16] S.R.S. VARADHAN, Large Deviations and Applications, Society for Industrial and Applied Mathematics, Philadelfia PA (1984).
Conservation of critical groups in a weaker topology for functionals associated with quasilinear elliptic equations Jean-Noel Corvellec and Hamid Douik Laboratoire MANO, Universite de Perpignan F-66860 Perpignan cedex, France and Lehrstuhl C fur Mathematik, RWTH Aachen D-52056 Aachen, Germany Email : [email protected] ; [email protected] Abstract We prove that for a class of continuous functionals on HQ(CT) associated to quasilinear elliptic equations (!) C R", a bounded domain with smooth boundary), the critical groups at the origin coincide with those of their restriction to HQ (Cl) n L°°(Q).
1
Introduction
Let H be an open bounded subset of R n with smooth boundary (n > 3), and let / HQ (fi) - » R b e a functional of the form: /(«) = I f \Du\2dx - f
2 Jn
Jn
G{x,u)dx.
In [6, 7], Chang has proved that under some natural conditions on G, so that in particular / is of class C 2 , then the critical groups of / at an isolated critical point agree with those of its restriction to CQ(0). Such a result can be used to study the existence of weak solutions of the semilinear elliptic problem associated with / , by combining the sub- and super-solution method with the variational one; see also [3, 1] for the particular case of local minima. We intend to yield a similar result concerning the critical groups at the origin for functionals / : HQ (Q) —> R of the form f(u) = - / \^ aij(x,u)DiuDjudx
352
—/
G{x,u)dx,
353 associated with quasilinear elliptic equations. We note that if, under natural assumptions on the dij's and G, such a functional is continuous, it fails to be differentiable, or even locally Lipschitzian, see [4, 5]. However, we may use the methods of the nonsmooth critical point theory initiated in [14, 11, 8], in order to follow the lines of proof of [6], combining the "regularity" of appropriate deformations of a neighborhood of the origin with an appropriate version of the Morse splitting lemma. In this paper, we provide a first step in our study (to be further developed in [12]) by establishing that the critical groups of our functional at the origin coincide in the topologies of HQ(Q,) and H^itl) fl L°°(Q). Our main result is in section 4, the proof of which is based on the Morse lemma that has been recently established by Lancelotti [17], and that we recall in section 3, and on the existence of appropriate "regular deformations", which are constructed in section 2 in a fairly general setting.
2
Regular deformations
In this section, we recall from [5] some features concerning a general class of functionals of the Calculus of variations, and we prove a deformation result for this class. The technique of proof follows the lines of that of the abstract deformation theorems [11, Theorem (2.8)] and [9, Theorem 2], dealing with arbitrary continuous functionals on complete metric spaces, the novelty here being that the deformation obtained is regular, by which we mean assertion (d) of Theorem 2.3 below. Let fibea bounded open subset of R n (n > 3), and let / : H&{Q) -> R be defined by: f(u) := / L(x,u,
Ja
Vu)dx,
where the function L : !] x 1 x R° -> R is C^-Caratheodory, i.e. x — i > L(x,s,^) is measurable for every (s, £) e R x R n , (s, £) i-+ L(x, s, £), is of class C 1 for a.e x e ft, and satisfies the following growth condition : |L(a:, s,£)| < a0(x) + foo(IsI":r5 + l£|2), f° r a - e x 6 H and every (s,£) £ 1 x I " , where ao 6 L1(i7) and b0 > 0. Under these conditions, the functional / is well-defined and continuous on HQ(Q). Assume further that there exists ai G Lloc(fl) and &i 6 L™C(Q) such that |D.L(a,a,OI \V(L(x,s,0\
< «i(*) + h(x)(\s\^ + |£|2), < a ^ + M x X I s l ^ + KI2).
Then, for every u e HQ(Q.) we have that DsL(x,u,Vu) LlJP^W1), so that : -div(V ? L(x,w, V«)) + DsL{x,u,Vu)
e L\oc{£l), V^L{x,u, Vu) G € V'{Q).
Definition 2.1 We say that u is a weak solution of , . W
J -div(Vi;L{x,u,S7u)) \« = 0
+ DsL{x,u,Vu)
= 0 in ft on dQ,
354 ifue H^(n) and -div (V(L(x,u, VM)) + DsL(x,u, Vu) = 0, in V(U). and w e H£(Q) n L°°{Q), we have: f'(u)(w) := Mm A ! i ± H j l i M
=
f [V€L(x,u,Vu)-Vw
For u € H&(Sl)
+ D„L(x,u,Vu)w]dx
eR.
Clearly, w H-> f'(u)(w) is linear, while u >-> }'{u)(w) is continuous, as follows from (1). In particular, we see that u e HQ(Q) is a weak solution of'(*) if and only if f'(u) = 0 in tte distribution sense. Let now V and W be (closed) linear subspaces of HQ(9.) such that V C L°°(fi) and fl^fi) = V © W (so that, also, ^ ( f i ) n L°°(n) = f e ( l f n £°°(fi))), and let Pv : i?o(n) —> V be the projection onto V. We denote by ||«||i,2 := ||Vu||2 the norm in HQ(Q), where ||-||2 is the norm of L?(Q), and by ||u|| := | H | 1 2 + ||u|l<» the norm in #o(n) 0 L°°(Q.), where H-]^ is the norm of L°°(fi). The open ball, closed ball, and sphere in i?Q(f2), centered at u e i?Q (f2) with radius 8 > 0 will respectively de denoted by B(u; 8), B(u; 8), and dB(u; 8). li X C H£(Q), X~ will be the closure of X, and we let Xx := XnL°°(Q), which we shall consider as endowed with the topology ofHQ(Q)<~)L°°(Q). Set: \dwf\{u) := suV{f'(u)(w) :weWn L°°(fi), ||u/||i.2 < 1}. According to the above mentioned properties of f'(u)(w),
we have:
Proposition 2.2 Let u e HQ(0,). Then, \dwf\(u) € [0,+oo], and if \dwf\{u) there exist w G W H L°°(Q,) with ||w||i,2 = 1 and 8 > 0 such that \fv e B{u;8),Vt 6 [0,8] : f(v + tw) < Given X C H^tt)
> a > 0,
f(v)-at.
and a continuous function c : X —> K, we set: fc := {u e X : /(«) < c(u)}.
T h e o r e m 2.3 Let H^(Q) = V ® W with V C L°°(Q) and f : H^(Q) -> R as aiow. Assume £/ia£ /(0) = 0, and that there exist r, p > 0, such that: VveVn
5(0; r), VUJ e W n dB(0; p) : /(v + w) > 0.
(2)
Set X := [(F n B ( 0 ; r ) ) + ((ffl 5(0; p))\ D / ° , and fe£ further a : X -> ] - oo, 0] 6e a continuous function with a(0) = 0, SMC/I iftai: Vu, v! E X : pv(u) = Pv(w') =>• a(u) = a(u'), a
inf |aV/| > 0, for every neighborhood U of f .
(3) (4)
Then, for Y := X or Y := X \ {0}, and for any open neighborhood U of f DY in Y, there exists a continuous map n : Y x [0,1] —> Y such that (a) 7?(u,0) =u; (b) uefa => n(u,t)=u; (c) n{u,l)eU; (d) 77(^00 x [0,1]) C Yx, and n : Yoo x [0,1] —> Y^ is continuous.
355 Proof. We first consider the case of Y = X. Let U be an open neighborhood of / ° ; we may of course assume that X \ U ^ 0. Let g : X \U —>]0, +oo[ be defined by g(u) := f(u) — a(u), and : 7 := inf g, 0 < n < inix\u \dwf\- Let u £ X \ U be such that <7(u) > 7. According to Proposition 2.2 and (2), we find w e Hg(Q) C\ L°°(Q) with ||w||i,2 = 1 and S > 0 such that: Vt€[0,6]:u
+ tweX\U
and
f(u + tw) < f{u) - fit.
Letting
lVg|M:=limsuP^-y)+ denote the strong slope of g at u [13], we thus have, taking (3) into account: i „ ,, N^ ,. q(u) — q(u + tw) ,. f(u)J — f(u + tw) \Vg\{u) > limsup = ^ — — - = limsup -^-t—^ ^ > /j, It follows that -1(7) 7^ 0, according to Ekeland's variational principle [16]: see [2, Section 2] for more details. We conclude that 7 > 0; thus, considering 0 < ry1 < 7, and setting ai(u) := a(u) + 7 ^ a(u) := a(u) + 7 for u 6 X, we obtain: M 6 X, f(u) < a(u) ==>ueU.
(5)
Let now a > 0 be such that \dwf\{u) > c, for every » 6 Z : = I \ / a i . For u e Z , let wu e W n L°°(n) with | K | | i , 2 = 1 and Su > 0 with B(«; 2<5„) C #<}(fi) \ / Q 1 such that Vv € B(u; Su),\ft e [0, Su] : f{v + twu) < f(v) - at, according to Proposition 2.2, so that
v« e B(u; <$„) n x , Vt e [o,
if U G V I , A
_
if U ^ U A S A I V U ,
356 ft > 2. If u € Vh,u for some ft > 2 and ^ 6 A;,, then
HW^M - u||li2 < I £ £ M«) * < n(«) < \sh,»This shows that Hh-i(u,t)
€ B(uhtll\8htli),
whence Hh is well-defined and
f(Hh(u, t)) < /(«) - * ( £ £
9
^ u ) ) *•
Since the family {Vj^} is locally finite, for every u € Z there exist an open neighborhood TV of w and hu e N, such that Tih(v, t) = Hhu(v, t) for v 6 A/' and ft > ftu. We can thus define r}l : Z x [0, +oo[—> Z by »h(u,t) ~ lim Wft(i;,mm{t,r1(M)}). For v e M, we have ||»7i(i>,t) - *>||i,2 < min{£, Ti(V)}, while if u 6 Z^,, then Th(/M) € Zoo for i; € J\fr\L°°(fl), and |l»71(i'1 <) — v\\ < min{t, Ti(v)} maxi<j
if 0 < i < rfe_! (u) r h _i(u))).
Clearly, each r/, is continuous, and so are the r)h's and their restrictions from Z ^ x [0, +oo[ to Zoo. For each u £ Z and each ft € N, we have Pv(Vh(u> *)) = ?v( M ) f° r every t > 0, and: 0 < i < Th(u) = • /fa h (u, <)) < /(u) - at, (6) ll»//i+l(Wi1"h+l(w)) ~ Vhiu,Th{u))\\l,2
< Tl{rih(u,Th{u)))
= Tfc+^U) - Tfc(u).
(7)
We infer from (6) that T/,(U) < (f(u) — a(u))/a for each ft € N, so that there exists r(u) := lim^ rh(u) € ]0, {f(u)~a(u))/cr], and it is readily seen that r is lower semicontinous on F . We then obtain from (7) that (%(«, T^(U))) C (p^(u) + TV) n Z is a Cauchy sequence. Hence, letting u := limfcr]h(u, Th{u)), we see that f(u) •=• «i(u) = Qi(u) — for, otherwise, we have u € Z while Ti(u) = 0, which is not true. Consequently, we can define 6 : Z -> N U {0} by: e{u)
''
.= ( ° if /(«) < "(w) \ min{ft 6 N :/(?7fe(u,T/,(u))) < a(w)} otherwise
and clearly, 6 is upper semicontinuous with 6{u) < T(U) for every u S Z. Let then /? : Z ~+]0, +oo[ be continuous with 6 < (3 < T, and define r\ : Z x [0,1] —> Z by: ij{u,t) := lim r)h(u,t/3(u)). h—»oo
357 For each u e Z, there exists hu e N such that rhu(u) > (3{u), hence there exist an open neighborhood JV of u such that Thu{v) > P{v) > t(3(v) for all v e Af, so that f)(v,t) = rjh^(v,tp(v)) for such u. This shows that 77 is well defined, continuous, and so is its restriction from Z^ x [0,1] to Z^. Finally, let 7x < 7 2 < 7, set a 2 (u) := a(u) + 7 2 (u e X), and let 1? : X —• [0,1] continuous, such that $(u) = 0 if f(u) < a2(u), i?(u) = 1 if f{u) > a(u). Define r]: X x [0,1]-> X by:
•q{u,t) := J
7?(u, tf(u)) if u e Z •u otherwise.
Then, 77 has the required properties: we just verify that (d) holds. Indeed, if f(u) < a(u), then f(rj(u, 1)) < f(u) < a(u) = a(r](u, 1)), so that ?7(«, 1) G [/, according to (5). If f{u) > a ( u ) i then i9(u) = 1, it follows that ri{u,l) = T!(U, 1) = lim ijh(u,f3(u)) =
r)h(u,0(u))
for some h u > 0(u) (since 6(u) < j3{u) < T(U)), and f(r,(u, 1)) = /(77k>,/?(«))) < /(»/,(„,(«,»(«))) < a(u) = 0(77(1*, 1)), so that r)(u, 1) e [/ again. We now consider the case when Y := X \ {0}. We first observe that since a(0) = 0 and (3) holds, we have: ueX\r=^pv(u)^Q.
(8)
Let U be an open neighborhood of f \ {0} in X \ {0}. For d e N large enough so that B/, := B(0; l/h) n / ° C X, say h > h 0 , we have that f/ U Bh is an open neighborhood of f in X. As in the beginning of the proof, we thus find a nonincreasing sequence (7/,)h>/i0 of positive real numbers such that u£X\(UuBh)=*f(u)>a(u)+jh,
(9)
and we may assume (without loss of generality, indeed) that 7 h —• 0. Define a : pv(X) { 0 } ^ ] 0 , + c o [ by a W W
. = /7fc. iff^B(0;l/M ' \ T M - I if « e B (0; l/h) \ B (0; l / h + 1),
\
h>h0.
Since a is lower semicontinuous, we find a continuous function a : Pv{X) \ {0} —> ]0, +oo[ such that a < 5, and we define a continuous function a : Y —> R by: ^ „ \ ._ / a ( u ) + S (Pv (")) if Pv (u) 7^ 0 a W - \ 0 ifpv(«) = 0. Then, a(u) = a(u') whenever pv(u) = pv(u'), and f(u) > a(u) > a(u) for u € Y" \ U, as follows from (8), the fact that ||u||i i2 > ||pv(«)||i,2, (9), and the definition of a. Considering ai : Y —> R continuous such that a(u) < «i(u) < a(u) whenever
358 Pv(u) 7^ 0 and ai(u) = ai(u') whenever py(tt) = py(u'), we find, according to (4), another nonincreasing sequence (crfc)h>ho of positive reals converging to 0, such that inf
\dwf\ > ah,
Y\(f°lUBh)
and we proceed in a similar fashion as for the definition of a, in order to define a continuous a : Y —» [0, +oo[ such that a{u) = cr(u') whenever py(u) = pv(u'), and \dwf\(u)
> a{u) > 0, for every u e Y \
fai.
Then, we can proceed in a similar way as before (replacing a with <J{U)) in order to construct the required deformation r). •
3
A generalized Morse splitting lemma
In this section, we consider a functional on Hg(Q) of a more specific form than that considered in the previous section, and for which we recall a generalized Morse lemma from [17]. Let fl be a bounded open subset of R™ (n > 3) with d£l of class C 1 ' 1 . For every 1 < i, j < n, let dij : S] x R -» R be such that: (a.l) dij is of class C 1 , and the function s i-+ a.ij(x,s) is of class C 2 for a.e. x € Q.; (a.2) ctij = aji for every 1 < i,j < n; (a.3) there exists v > 0 such that for a.e. x € fi, for all s £ R, and for all £ 6 R™ n
n
^a„(x,s)^ > v^£2; (a.4) there exists C > 0 such that for a.e. a; 6 fi, for all s e R and all i,j, k \a-ij{x,s)\
\Dsai:j(x,s)\ < C,
\D
s)\ < C,
\D2ssaij{x,s)\
(a.5) for a.e. x € SI, for all s € R, and for all £ € R"
Let also p : Q x R —> R be such that (g.l) for every s e R, x — i > j ( s , s) is mesurable; for a.e. x e O, s i-> p(x, s) is of class C 1 ; <7(z,0)=0; (g.2) there exist b e R and 0 < p < ^
such that for a.e. x e ii and for all s e R,
|£3(MI<6(l + |s|p).
359 Let / : HQ(Q) —> K be the continuous functional defined by: I y_]aij(,xiu)DiuDjudx
f(u):=-
— I G(x,u)dx,
(10)
where G(x, s) := f*g(x, t)dt. Observe that u 6 HQ(Q) is a weak solution of the problem (*) associated with / if 71
1
— \] Dj{aij{x,u)Diu)
n
+ - 2_\ Dsai:j(x,u)DiuDjU
= g(x,u)
in P'(fi),
and that since g(x,Q) = 0, this problem has the trivial solution 0 € HQ(£1). fl^fi) - • R be defined by Q(u) := / y ^ aij(x,0)DiuDjudx Jnij=1
—/ Jn
Dsg(x,0)u2dx.
Let Q :
(11)
Due to (a.4) and (g.2), Q is a well defined continuous quadratic form. We can write Q(u) — (Au,u), where n
Au := — 2_, Dj(aij(x,0)Diu)
— D3g(x,Q)u,
and, taking also (a.3) into account, the elliptic operator (with O-Dirichlet boundary condition) A possesses a nondecreasing sequence (A^) of eigenvalues (repeated according to multiplicity) diverging to +oo. Then, m(f; 0) := min{j e N : Xj > 0} - 1 is equal to the strict Morse index of Q, namely, the supremum of the dimensions of the subspaces of HQ(Q) where Q is negative definite, and m*(/; 0) := min{j 6 N : Xj > 0} - 1 is equal to the large Morse index of Q, namely, the infimum of the codimensions of the subspaces of HQ(Q) where Q is positive definite. We thus have the decomposition tfi(fi) = V 0 W, where V C ff0x(fi) D L o c (n), d i m F = m*(/;0) < oo, and Wv £V, \tw€W:
/ V^ a,ij(x,0)DivDjWdx — / Dsg(x,0)vwdx Jn ;T"i
3// > 0, Viu G W :
Jn
= / vwdx = 0, •> n
Q(w) > n\\w\\ia.
Under the above assumptions on the o^'s and on g, the following generalized form of the Morse splitting lemma is proved in [17, Theorem 6.17, Lemma 6.15]:
360 T h e o r e m 3.1 There exist r > 0, p > 0, and a map $ : F n 5(0; r) -> Wx n 5(0; p), which is Lipschitzian as a map from HQ(£1) D L°°(n) into itself, with $(0) = 0 ; and such that: (a) V t ) G F n S ( 0 ; r ) , Vw e W H d_B{0;p) : f(v + w)> 0; (b) VveVD 5(0; r ) , V m e I V n B ( 0 ; p ) . . i u ^ $(v) : /(u + w) > /(v + $(u)); (c) for every v € V C\ -0(0; r) and every w €W !1 5(0; p), we have: (VzeWn
L°°(Q) : f'{v + w){z) = 0) <=> w; = $(v).
Let ip(v) := f(v + ${v)), v eVH Vv€Vn
5(0, r). Then, ip is of class C1 onVH 5(0; r) with
5(0; r), V z e V : f'(v)(z)
= /> +
$(v))(z),
4
Conservation of critical groups in t h e HQ(CI) r\L°°(Q,) topology
In this section, we consider the functional / : Hg(fi) —> M defined by (10), assuming that (a.l)-(a.5) and (g.l), (g.2) hold. We shall denote by f^ the restriction of / to H%(n)nL°°(n). Definition 4.1 Let Z be a metric space and g : Z —> R be a continuous function. For u e Z, c := g(u), U a neighborhood of u, and q 6 Z+, set Cq(g; u) := H\g* n U, (gc \ {u}) n U), where Hq(A, 5 ) denotes the q-th relative Alexander-Spanier cohomology group of the pair (A,B), with coefficients in R (say; see [18]), and gc := {v 6 Z : g(v) < c}. The vector space Cq(g; u) is called the q-th critical group of / at u. Due to the excision property of the Alexander-Spanier cohomology, its definition does not depend on the particular choice of the neighborhood U. It is easy to see, as in [8, Proposition 3.4], that if / is the functional given by (10), then \d„^)f\{u) > 0 =>• Cq(f;u) = {0} for all q. Note that \dHi(fl)f\(u) = 0 if and only if u is a weak solution of the problem (*) associated with / . Abusing a little the terminology adopted in [14, 11, 8] when dealing with general continuous functionals defined on arbitrary metric spaces, we might say that such a is a "critical point" of / , which thus accounts for the related notion of critical group. T h e o r e m 4.2 Let f : H^(Q) -> R be defined by (10), such that (a.l) - (a.5) and {g.l), (g.2) hold, and let /«, denote the restriction of f to Hg(£l) D L°°(f2). Then, Cq{f;0)=Cq(/«,;
0) for every
qeZ+.
361 Proof. Let Q be the quadratic form defined by (11). If m*(/;0) = 0, that is, if Q is a nondegenerate, positive quadratic form, then the result follows from those of [10] (where indeed singular homology was used instead of Alexander-Spanier cohomology, but this clearly makes no difference here). More precisely, [10, Theorem 5.13] shows that Cq(f;0) = (5,0-R for every q (where 6tj is the Kronecker symbol), and from [10, Lemma 5.9] we deduce that 0 is a local minimum of f (arguing in a similar way as in [6, Theorem 1.4-6]), hence a local minimum of fx, so that finally, C^f^O) = fi9o'R for every q. Therefore, we now assume that -ffg(fi) = V © W, where Q is positive definite on W and V C L°°(fi) with 0 < dim V = m*(f; 0) < oo, and we show that for every 9 £ Z + : C,(/;0) = C,( V ;0) = C,(/ oo ;0), where tp is given by Theorem 3.1 (d). Indeed, the fact that Cq(f;0) = Cq(tp;0) was already proved in [17, Theorem (6.4)], combining Theorem 3.1 with an abstract deformation theorem. In fact, Theorem 2.3 could be used instead of the latter for that purpose, and our point here is to establish that, through the same argument but using specifically Theorem 2.3 (d), we also have Cg(/oo;0) = C,(?;0) (q being fixed in the sequel). Withr, p > 0 as in Theorem 3.1, let X := [{VnB{0;r)) + (Wr\B{0; p))]nf° and define a : X —>] — oo,0] by a{u) := f(pv{u) + $(pv(u)))- Then, a(u) < f(u) for every u € X, according to Theorem 3.1 (b), and it is clear that all the assumptions of Theorem 2.3 are satisfied, but for (4): the argument is given in the proof of [17, Theorem 6.20], we reproduce it for the sake of clarity. Let (u^) C X \ fa be such that e/, := \dwf\{u^) —> 0. We thus have: l / ' K ) CO I < eh||*Hi,2 for every z£WC\ LX{Q). (12) Let {?!,..., (p*m} be an I?'-orthonormal basis in V, where m* := m*(/, 0). Then,
Pv(z) = 5Z ( / fkzdx\ Vk, and writing (12) for z — Pv(z), we obtain /
I^
J a \ ^ ^
= 1
a,ij(x, uh)DiUhDjZ + - V*
2 ^tj=i
- /
(g{x, uh) +
Dsg(x, uh)DiUhDjUhz 1 dx /
u
dx
Y^=1f'( h){>Pk)'Pkz) <£h\\z\\i,2,
with (f'(uh)(ifik))h bounded for every k = l,...,m*, and Eh •= £h\\Id - Pv\\ —* 0. Hence, f'(uh) € iJ"_1(fi) (more precisely, f'{uh) can be extended to a continuous linear form on HQ(£1)) and, up to a subsequence, f'(uh) is strongly convergent in i? _ 1 (fi). It follows from [5, Theorem 2.2.4] that, up to a further subsequence, (u^) is strongly convergent to u € X, and passing to the limit in (12), we have f'(u)(z) = 0, for every z € W n L°°(Q). According to Theorem 3.1 (c), it follows that u = v + $(?;) for some v G V D B(0; r), so that f(u) = a(u), i.e., u € /". This establishes (4), upon arguing by contradiction. Now, set: M:={v + $(v) :veVr\B{0;r), f(v + ${v)) < 0} C H£{Q.) I~l L°°(fi),
362 and observe that : (fa,fa \ {0}) = {M,M\ {0}) = (/£,,/£, \ {0}). It follows from Theorem 2.3 and the properties of Alexander-Spanier cohomology that the inclusions /£, —> Xao and /£, \ {0} —> Xoo \ {0} induce isomorphisms in Alexander-Spanier cohomology: see [15, Lemma 2.6]; hence, from the Five lemma, so does the inclusion (/£,, /£, \ {0}) —> (XoojXoo \ {0}), and iue have Cg(f«,; 0) := ^ ( X c , ^
\ {0}) = # » ( £ , , / £ \ {0}) = H"{M, M \ {0}).
On the other hand, from the properties of $, the pair (M, M \ {0}) is clearly homeomorphic to the pair {ip°,
Jt.3
1.
(a) In [12], we improve Theorem 4-2 in two directions: firstly, by considering the restriction of f to Cg(n) instead of f^, secondly, by considering critical groups defined through singular homology, which is possible using an extension of the so-called Second deformation lemma of [8, 9] in place of Theorem 2.3. (b) As consequences of Theorem 3.1 mainly, it is shown in [17, Corollaries 6.5, 6.6] that C g (/;0) = 0 if q £ [wi(/;0),m*(/;0)], and that ifO is an isolated weak solution of(*), thenCq(f;0) has finite dimension for every q. See [10, 17] for an application of nonsmooth Morse theory to the existence of weak solutions of (*) for functionals f of the type (10).
References [1] A. Ambrosetti, H. Brezis, and G. Cerami, Combined effects of concave and convex nonlinearities in some elliptic problems. J. Punct. Anal. 122 (1994), 519-543. [2] D. Aze, J.-N. Corvellec, and R. E. Lucchetti, Variational pairs and applications to stability in nonsmooth analysis, Nonlinear Anal. Theory Methods Appl. 49 (2002), 643-670. [3] H. Brezis and L. Nirenberg, H1 versus C 1 local minimizers, C. R. Acad. Sci. Paris, Ser. A 317 (1993), 465-472. [4] A. Canino, Multiplicity of solutions for quasilinear elliptic equations, Topol. Methods Nonlinear Anal. 6 (1995), 357-370. [5] A. Canino and M. Degiovanni, Nonsmooth critical point theory and quasilinear elliptic equations. Topological Methods in Differential Equations and Inclusions (Montreal, 1994), A. Granas and M. Prigon Eds., 1-50, NATO ASI Series, C 472, Kluwer, Dordrecht, 1995. [6] K. C. Chang, Infinite Dimensional Morse Theory and Multiple Solution Problems. Progress in Nonlinear Differential Equations and Their Applications, 6, Birkhauser, Boston, 1993.
363 [7] K. C. Chang, H1 versus C 1 isolated critical points. C. R. Acad. Sci. Paris, Ser. A 319 (1994), 441-446. [8] J.-N. Corvellec, Morse theory for continuous functionals. J. Math. Anal. Appl. 196 (1995), 1050-1072. [9] J.-N. Corvellec, On the Second Deformation Lemma. Topol. Methods Nonlinear Anal. 17 (2001), 55-66. [10] J.-N. Corvellec and M. Degiovanni, Nontrivial solutions of quasilinear equations via nonsmooth Morse theory. J. Diff. Equations 136 (1997), 268-293. [11] J.-N. Corvellec, M. Degiovanni, and M. Marzocchi, Deformation properties for continuous functionals and critical point theory. Topol. Methods Nonlinear Anal. 1 (1993), 151-171. [12] J.-N. Corvellec and H. Douik. In preparation. [13] E. De Giorgi, A. Marino, and M. Tosques, Problemi di evoluzione in spazi metrici e curve di massima pendenza. Atti Accad. Naz. Lincei Rend. CI. Sci. Fis. Mat. Natur. 68 (1980), 180-187. [14] M. Degiovanni and M. Marzocchi, A critical point theory for nonsmooth functionals. Ann. Mat. Pura Appl. (4) 167 (1994), 73-100. [15] M. Degiovanni and L. Morbini, Closed geodesies with Lipschitz obstacle. J. Math. Anal. Appl. 233 (1999), 767-789. [16] I. Ekeland, Nonconvex minimization problems. Bull. Amer. Math. Soc. 1 (1979), 443-474. [17] S. Lancelotti, Morse index estimates for continuous functionals associated with quasilinear elliptic equations. Adv. Differential Equations 7 (2002), 99-128. [18] E. H. Spanier, Algebraic Topology. McGraw-Hill, New York, 1966.
Approximating exterior flows by flows on truncated exterior domains: piecewise polygonal artificial boundaries Paul Deuring Universite du Littoral, Centre Universitaire de la Mi-Voix, Laboratoire de Mathematiques Pures et Appliquees, B.P. 699, 62228 Calais Cedex, France Abstract The article deals with a mixed finite element method for discretizing the Stokes system in truncated exterior domains. This method yields approximate solutions of the Stokes equations on exterior domains, as is shown by error estimates. An implementation of the method is discussed, and the results of test computations are presented. In particular, it is indicated how a truncated exterior domain with spherical outer boundary may be triangulated in a suitable way. Similar mixed finite element methods work for other exterior problems which may be written as mixed variational problems.
1.
Introduction
In this article, we discuss how exterior flows may be approximated by solutions of mixed variational problems in truncated exterior domains. In particular we will be concerned with truncated exterior domains which are polyhedral and thus may easily be decomposed into grids suitable for finite element methods. T h e model problem we consider is the Stokes system in the exterior domain fl := K 3 \fi, with a Dirichlet boundary condition on dQ and a decay condition near infinity: - A U + VTT = / ,
/
2
div« = 0 x
H z ) | dox = 0(BT )
in I f ,
for
u\dQ = b,
(1.1)
R^oo,
JdBi
where Q. C M3 is either the empty set, or a bounded open polyhedron with Lipschitz boundary and with 0 G Q. We further suppose b G H1/2(dQf,f G L 2 (R 3 ) 3 , and we assume there is a e (4, oo), 5 , 7 G (0, oo) with IT C Bs, \}{x) \ < 7-|z|~ < r for x G W?\BS. T h e quantities a, S, 7 and the functions b and / will be kept fixed throughout. Here and in the following, for e G (0, 00), we denote by Bt the open ball in M3 with radius e > 0 and center in the origin.
364
365 Under the preceding assumptions, problem (1.1) admits a unique solution (u,7r) with the properties u 6 #, 2 oc (n c ) 3 , TT 6 ^ ( i f ) n L 2 (ff), Vu e L ^ j f ) 9 ("exterior Stokes flow"); see [6, Theorem V.2.1,_V.1.1]. Now take R e [S, oo) with f2 C BR, and denote by OR that part of the exterior domain Q which lies inside the ball BR, that is, QR := BR\Q. Then we consider the boundary value problem -Av
+ VQ = f\nR, divu = 0 iuUR, CR(BR)(v, Q){X) = 0 for x £ dBR,
v\dQ = b,
(1.2)
where CR is defined as follows: 3
jCflMCiD,*),•(*) := 3 • tuJ-(a;)/(2 • R) + £ (£>*«;,• - 1 ^ / 2
- <S# • a){x) • n[A\x)
*= i
for j e {1, 2, 3}, R e (0, oo), a; £ dA, A C E3 a Lipschitz domain, n'-4' its outward unit normal, w a vector-valued function and a a scalar function with the property that w, Vw and a are defined on dA. The symbols D% and Dj denote partial derivatives. This choice of CR is based on two considerations: first, it allows to write problem (1.2) in mixed variational form, and second, the difference between the exterior Stokes flow and the solution of (1.2) ("truncation error"), measured in some suitable norm, decays with an optimal rate. More precisely, setting for v, w £ ^(QR)3, Q e L2(QR), aR(v,w):=
3/(2 • Ri) • [ v • w dox + [ £ Jen JnR ^
- (1/2) • DjVk(x) • DkWj(x)\ dx,
v
(DjVk-DjWk
(1.3)
1
pR{y,Q):=-l
Ayvv-gdx, JUR
we could show (see [1, Theorem 4.2]) there is a uniquely determined pair of functions
(wH,te)e#1(n*)3xi/1(nfl) with aR(vR,w)
+ fiR{w, gR) = I
f-wdx
for we H^QR)3
with w\ 8Q. = 0, (1.4)
JSIR
div vR = 0,
VR
I dQ = b.
Note that problem (1.4) is a weak form of (1.2); see [1, Lemma 4.2]. The equation divVR = 0 may be replaced by the relation fiR(vR,<j) = 0 for a £ L2(QR), so problem (1.4) is a mixed variational problem. It is further shown in [1] that || (u - vR) I O s ||2 + || V(u | nR - vR) h + R1/2 • II (« - VR) \ dBR ||2 (1.5) + \\x\nR-QR\\2 < C - (|| / I « s ||a + 7 + ||6||i/2,2)foranyi?e[5,oo), with a constant C > 0 which only depends on Q, S and a ([1, Theorem 5.1, Lemma 5.1]). Actually, only the case b = 0 is considered in [1], but a technique for extending this result to the case 6 ^ 0 may be found in [5]. The important point of this extension is that the constant C in (1.5) remains independent of R. We further remark that the results in [1] were inspired by a similar theory in [8], [9] pertaining to the Poisson equation.
366 An artificial boundary condition for the Stokes system different from the one in (1.2) is presented in [10]. Similar estimates as in (1.5) are proved in [5] for Oseen flows. Due to its mixed variational form, problem (1.4) may be discretized by finite element methods. A finite element solution to (1.4) may then be considered an approximation of the exterior Stokes flow (U,TT). The corresponding error consists of the sum of the truncation and the discretization error, the latter one arising due to the transition from (1.4) to a finite-dimensional problem. In [1, Theorem 6.3, Corollary 6.2], we estimated this sum, under the assumption that the Babushka-Brezzi condition [1, (6.9)] is valid with a constant independent of R. This assumption is satisfied by the Mini element ([3, Theorem 4.1]). When we performed test computations ([2], [4]), we did not actually decompose QR. Instead we used tetrahedral grids of such a type that the union of the tetrahedrons of each grid yields a polyhedron P, depending on the respective grid, with P C £lR and 0 R \ P a slender set near the sphere dBR. The artificial boundary condition was then imposed on the outer part dP\dQ. of the boundary dP of P. It is the purpose of this paper to give a precise description of the transition from QR to these polyhedra, and to quantify the error related to this transition. To this end, we fix some index set I. For i g l , let n; e No with n,- > 3, and put Ri := 2 ni • S. (The parameter S was introduced at the beginning of this section and refers to the size of Q and the decay behaviour of /.) Assume that T, = (Kf ) 1 < i < T ., with Ti € N, is a family of open tetrahedrons in B ^ \ Q . We will write K\ instead of K\ since the dependence of Ki on i should be obvious. Denote by Pt the interior of the set U{ Ki : 1 < I < ^ }. We suppose that Pi C Q,Ri and 0,R\Pi is a remainder set near dB^. Let hi denote an upper bound of diamKi for tetrahedrons Ki near Q. Following Goldstein [8], [9], we suppose the mesh size of T; to augment from this value ht near Q, to hi- Ri/S near dB^. These and other properties of Ti will be made precise in Section 2. For each i e l , the polyhedron Pi is used for setting up a mixed variational problem. In fact, define the bilinear forms a,- : Hl(Pif x H 1(Pif i->- K, ft : H1^)3 x L2(Pi) M-1 in the same way as respectively aR and fiR in (1.3), but with R = Rt, and with the domains of integration QR and dBR replaced by Pi and dPi\dQ, respectively. Then we consider the boundary value problem -Avi + Vft = / | Pi, divvi = 0 in Pi, Vi\dQ = b, CR, {Pi U Q) (vi, Qi) (x) = 0 for x e dPi\dtt, which admits a weak solution for any i el
with hi small enough:
Theorem 1.1 There is S0 G (0, S) such that for i £ I with h{ 6 (0, S0], a uniquely determined pair of functions (vi,gi) exists in H1^)3 x L2(Pi) with ai(vi,w) + PAw, Qi) = / f-wdx
forweWi,
divvi = 0,
vt I <9f2 = 6,(1.6)
J Pi
where Wi := { w G H1(Pi)3
: w \ d£l = 0 }. Moreover, for i as before, we have
|| (u - v^ | n s ||3 + || v(« | Pi - v^ ||2 + R;1/2 • H (u - v^ | dPi\dQ \\2 + Iklfi-ftll2
(1.7)
367 <
C -I$2
.((Vi-
V^2
+ (Vt + V^2
• (V( • hi)1'2 + Vi + (V~ + Vi) • hi
+ T>i), with a constant C > 0 independent ofi, and with the terms Vt, T>i and T>i defined by
A:=sup{|£ Ri (S ft )(w,7r)(s)| :
xedB^},
Vi := sup{ | Fi(t • x) | : x e dB^, t e [(1 - hj/S2)1'2, 1] }, Vt := sup{ | Fi(x) -Fi(t-x)\ : x e OB^, t € [(1 - h2/S2f/2, 1] }, where Fi(x) := (R^1
• u(x), V M ( I ) , TT(X) ) for x e fi .
This theorem, which will be proved in Section 2, yields an analogue of inequality (1.5) once we are able to find upper bounds for Vi, Vi and Vi in terms of ||/ | fis||2, 7 and ||6||1/2,2- In other words, we still have to establish pointwise estimates of the exterior Stokes flow (w, 7r) in the annular domain Bni\BR..(l_h2/S2y/2. Such estimates follow from our assumptions on / . For lack of space, we can only give the result here, which says there is a constant C > 0 only depending on Q, S and a such that Wi
Vi
Vi
+ \n(S/hi))-r
(1.8)
for i el, where T := ||/ | Bs||2 + 7+||6||i/2,2- A proof of (1.8) may be based on an integral representation of u and 7r similar to the one in [1, (3.7), 3.8)]. In fact, the term X>, is evaluated in this way in [1, p. 259], under the assumption 6 = 0. Combining (1.7) and (1.8) finally yields an estimate of the truncation error which corresponds to one given in (1.5) for the case of spherical truncating surfaces, and which states that the left-hand side of (1.7) is bounded by a constant times a sum of terms of the form /i" • R[b • (1+ln(S//ij)) c , with a + b > 3/2, c > 0. Also for lack of space, we cannot establish error estimates associated with finite element discretisations of the mixed variational problem (1.6). We only mention that the error estimates in [1, Theorem 6.3, Corollary 6.2], which pertain to finite element discretisations of (1.4), must be modified by adding to their right-hand side some additional terms which may be traced back to (1.8) and (1.7). The proof of this result does not differ much from that of [1, loc. cit.]. The main changes concern [3, Theorem 4.1], which yields the Babushka-Brezzi condition [1, (6.9)] for the Mini element, with a constant independent of Ri. In fact, [3, Theorem 4.1] pertains to domains Q^; if we consider Pi instead of fij^, we have to replace [3, Corollary 3.1] by Theorem 3.1 below, and [3, Theorem 2.2] by Theorem 3.5 below. But otherwise, the transition from Q^ to Pi simplifies the proof of [3, Theorem 4.1]. In particular, we may drop the smallness condition on hi required in [3, loc.cit].
2.
Partitionings of truncated exterior domains
We begin by describing our grids Ti in a more precise way. First we characterize the set ( A l ) Let i £ I. Then Ki C B^Q for I e {1, ..., TS}; the set Pt is a connected polyhedron; any vertex of Pi is either a vertex of Q, or it is located on dB^.
368 We do not admit hanging nodes, and we require that our grids do not degenerate: (A2) Ifiel, k,l G {1, ..., Ti} with k^l, and if F := ~Kk n~K~t / 0, then F is either a common face or side or vertex of K^ and Ki. (A3) Put Br(x) := { y G ffi3 : \y - x\ < r } for x G K3, r > 0; Br := B r (0) for r > 0; U0 := Bs\Ti;
U5 := Bv.s^&pTs
for j G N.
Then there is a constant ay > 0 and a family {hi)iez in (0, S/4) such that diamKi < V • hit sup{ r G (0, oo) : Br(x) C Kt for some x G Kt} fori&X,
> o~\ • V • ht
Z G {1, ..., T^, j G {0, ..., n j with Uj n # , ^ 0.
Note that although the mesh size of Ti increases with growing distance from fl, as already indicated in Section 1, our assumptions in (A3) imply that the family (Tj) i(S j is nondegenerate in the sense that the ratio of the smallest circumscribing sphere to the largest inscribing sphere of any element Kf' is bounded independently of i G I and / G {1, ..., T;}. In the three-dimensional situation we are considering, it might happen that all four vertices of a tetrahedron are situated on the sphere dB^. We want to exclude this case, so we require (A4) For i G X, I G {1, ..., Ti}, at least one vertex of Ki is located in the open set Pi. Finally we prescribe a condition which allows us to project the points of 8BR onto dPi\dU and vice versa. (A5) For i G X, the ray { r • (cos g • cos $, sin g • cos i9, sin 0 ) : r e ( 0 , oo) } has precisely one point of intersection with 3Pi\dQ, for any g, d G R. Let i G I. By (A4) and (Al), we may assume there is some K, G {1, ..., Tt — 1} with the properties to follow: For / G {1, ..., «,}, at most two vertices of Kt are located on dB^. If I G {^ + 1, ..., Ti}, three vertices of Ki are situated on dB^. For I G {1, ..., Ti}, let a'1' = a^(i,l), ..., a'4' = a' 4 '(i,I) denote the four vertices of Ki. If I > Ki, then, by the choice of Kt, we may suppose that aW, a' 2 ), a'3' G dB^ and oW G Pi. We note a well known consequence of assumption (A3): Lemma 2.1 There is a constant 02 G (0,1) such that I a « - a'3' I"1 • I a<2> - a<3> |~x • (a™ - a<3>) • (a™ - a<3>) < a2 fori el, l£ {1, ..., T{}, where a.W := a
= U{T, : Ki + 1 < I < n}
for i G X.
(2.1)
369 We introduce a two- and a three-dimensional standard tetrahedron K^ tively, by setting K{2) :={x€ (0,1) 2 : x2 < 1 - xx } , ^ ( 3 ) := {x € (0, l) 3 : x2 < 1 - T i , a 3 < 1 -xi
-
and A^3', respec-
x2}.
For i e l , / G {1, ..., Ti}, we put A, := Af» := ( a « - oW, a(2) - a(4), a'3' - a<4>) e K3x3, where the vectors a^ — a'4' = aW (i, I) — a'4' (i, I) are to be understood as columns (v G {1, 2, 3}). We further set Ai(x) := A?\x)
:= Af • x + o (4) (i,0
for i G X, 1 < Z < 7y, X G M3.
Denote by n^Pi\ n' flK ;' the outward unit normal to Pt and B ^ , respectively. Define the functions tpt := yf : K™ ^ QP^dQ, Vi := # : ^ ( 2 ) ^ 9B «i> ft : = &W : ^ ( 2 ) ^ E, i/( := v,W : if' 2 ) H l b y
:=
( | a (D - a(3)|2 . |a(2) _ a(3)|2 _
vi{rj) := {det{DjMv)
•
( ( a (l)
_ a (3) } . (ffl(2) _ a ( 3 ) )
)3y/^
DkipiCn))^^^)
for 77 G K{2\ where a ^ := a^(i, I), j G {1, 2, 3}, m + 1 < I < TV, i G X. Put F, : = i?W — {ipf\ri) •
v e
K{2) } for i G X, K{ + 1 < I < T{.
We note that ipi is a local parameter of dPi, with range (ipi) = XJ, and gi is the corresponding element of surface area. The function Ai \ K^ is the usual affine mapping from K^ onto Ki, and Fi is defined by a radial projection of the triangle T( on dB^. Due to (A5), ipi is one-to-one. We shall show below (Corollary 2.3) that if hi is small enough, ipi is regular and hence a local parameter of dB^. The related element of surface area is given by the function V[. As a consequence of (A5) and (2.1), we note dBfl, = V{¥i /
fdoy
: Ki + 1 < I < n}, =
£
/ fdoy
=
(2.2) •£
LM(f°
(2.3)
for i G X, / G Lx(dPi\dQ). In order to show that ipi is a regular mapping, and to estimate the difference between integrals over dPi\Q and dBn, we begin by recalling a result from
[3]: Lemma 2.2 ([3, Lemma 4.2]) We have B
,
2
u/^I^cPj
fori
el.
We note that the quantity ft/4 in [3] corresponds to hi in the present context. This explains why a factor h2 • S~2 arises here instead of h2 • (4 • S)~2 as in [3, Lemma 4.2]. By Lemma 2.2 and (A3) we obtain
370 Corollary 2.1 Fori el,
xe dPj\dtt, we have \x\ > Ri-{1
- h2 • S~2 ) 1 / 2 .
Another technical result which will be useful is Lemma 2.3 Let i el, | MvT1
I e {K{ + 1, ..., r,}, j e {1, 2}, rj G K<-2). Then | < diamKt • (8/%/l5) • Rr1 • | D i W fa) |.
• (w(i?) • Dm(ri))
Proof: Abbreviate a
\{Wi{v)\-'•
Vi{v) ~
\aM+aW\-1-{a®+aW)).Dm{T))\
(3)
\.
4 • | (1/2) • ( a « + o ) - Mv) I • I a (j) + a (3) l _1 • | Dm(V) 3
But | (1/2) • (a«> + a< >) - (pi{q) | < diamKi, and by Corollary 2.1 (1/2) • | aP + a™ | > Ri-(1
h2-S-2f'\
-
Since /ij < 5/4 (see (A3)), we obtain Lemma 2.3 by combining the preceding estimates, o Lemma 2.3 allows us to estimate n^ °
« « : = |o«(<)0-aW(*,Z)|- -(a05(i>Z)-oW(t,0)
an
=
d for j € {1, 2},
I ^ W r ' - ^ ^ W -
3
Let n denote the unit vector in K satisfying two properties: first, n-aW = 0 for j e {1, 2}, and second, there is some e > 0 with t-n+ipi(rj) $ Ki for t G (0, e). Then v = n^BRi>{ipi{r])) and n = n^Pi\tpi{rj)), and there are S\, 82, S3 e K with u = <5i • a' 1 ' + S2 • a' 2 ' + S3-n. We have <53 = n • v > 0, \v - n\ = y/2 • (1 - 5z)112, and 1 _ 5\ = |JX . a (D + S2 • a<2>|2 <
=
(61 • a^ + 52 . a W ) . „
Cl^xl -I- |<52j) • 8 - (Vl5 • S-)-1 - ^ ,
where the last inequality follows with Lemma 2.3 and (A3). We further find with Lemma 2.1: 1 = \v\2 > S2 + 82-2-\51-S2\-a2
+ 82 > (l-ff2)-(<5i + (52)72 + (52.
Lemma 2.4 follows from the preceding estimates. Lemma 2.5 Let i el,
I G {K* + 1, ..., r j , r) G K^.
o Assume in addition that
hi < (y/Tb/16) • S • (1 - alfl2, with 02 from Lemma 2.1. Then I Ml) ~ 9i(v) I < (163/152) • 9l(v) • h2 • S-2 • (1 - a 2 )" 1 .
(2.4)
371 Proof: Abbreviate 7 := |
=
l^iWI" 1 • Ri • ( " W - (7 • a 0 ) ) -7)
for j e {1, 2, 3}. Put
for j £ {1, 2}.
Using this equation, we get after some computations «4fo)2 = . l w f o ) r 4 - t f - ( f t f a ) 2 - * f o ) ) , where 17(77) (2.5):
ls
1
(2.5) 2
2
1
2
defined by a{rj) := | (7 • a' ') • a' ) — (7 • a' ') • a' ) | . We conclude with
lflifa)-nfa)l
=
~ (Ri/\fim)2
9i(v)-\l
(2.6) 2
+ W l w f a ) l ) - ( l - (1 "
2 1/2
Dn the other hand, we have by Lemma 2.1 p,(„)2 = | a « | 2 > ( 2 > | 2 . [ l >
(|a^r1-|aWr1-(o(1)-a(2)))2]
|aW|2.|a(2)|2-(l-*2),
md we may deduce from Lemma 2.3 and (A3) a{rj) < (16 2 /15)-i?- 2 -(diamA-,) 2 -|a ( 1 ) | 2 -|a ( 2 ) r < (162/15) • (hi/S)2 • \a^\2 • |a<2>|2. Therefore assumption (2.4) on ft, implies (1 - aW-gtiv)-2)1'2 > 1 - a(V) • gi(v)-2 > 1 - (1 - a2)'1 • (162/15) • hi • S~2 > 0. Jsing this estimate, recalling that \ipi(77)|2/ R2 > 1 — h2 • S~2 > 15/16 (see Corollary l.l), and noting that Ri/\ tp(rf) | > 1, we may now deduce from (2.6): < ftfo) • max{ (1 - h2 • S" 2 )" 1 - 1,
I 9i{v) ~ Ml) I
(1 - h\ • 5- 2 )" 1 • (1 - a 2 )" 1 • (162/15) • h\ • S-2 } 2
< gi(r,) •/i • S" • max{ 16/15, (163/152) • (1 - a2)'1 } <
P((r?)-(16
2
3
/152)-/l2-5-2.(l-a2)-1.
^et us state some consequences of the preceding lemma. Corollary 2.2 Put S0 := (15/128) • S • (1 - af) 1 / 2 . _ Then, if i € 1 with ht < S0, I 6 {K; + 1, ..., T;}, 77 € A"'2', we have »(«?) < 2-^C?)-
Ui{rf)/2 <
Uorollary 2.2 yields vi{rj) > gi(r])/2 > 0. This, in turn, implies Dorollary 2.3 In the situation of Corollary 2.2, the vectors Diipiir)), D2ipi(i]), with tpi := />, , are linearly independent.
372 Thus ipi is a local parameter of dB^, for I € {K; 4-1, ..., r;}, i e 1, hence by (2.2),
/
hdoy =
JT J hdoy =
for i e X with h, < S0, h e
£
Lm(h°
(2.7)
V-idB^).
Corollary 2.4 There is a constant C > 0 with the ensuing properties: Let G e Hlffi), H e Hlffi)3, £,£,£e ( 0 , o o ) , t e l with hi<S0,
\G{t-x)\ + \H{t-x)\<£,
\G{x) + H{x)-n{BRi)\<£
| G{x) - G{t • x) \ + \ H{x) - H(t • x) | < £, t e [(1 - h2 • S" 2 ) 1 / 2 , 1].
forx£dBRt,
Abbreviate at := G \ dB^ + (H | dB^) • n^BR>\ Then
I |k|| 2 - Whh I < C-((£-1)1'2
h:=G\
dP^dQ + {H \ dP^dQ)•
n<«>.
+ (£ + ~£)1'2 • (£ • h)1'2 +£ + (£ + £). ^) • Ri-
Proof: Let I e {K{ + 1, ..., r j , n G K&\ and put t := |,(r)) e dPi\dQ, Corollary 2.1 yields t e [ ( l - /i 2 /S 2 ) 1 / 2 , 1]. Moreover, for 9, g e E, /i, /i, n, n e E3, we have \(g + h-n)2-(g + h-n)2\ < 2 • \g + h • n\ • \g + h • n - g - h • n\ + (g + h-n-g~h-
n) 2 .
Starting from these observations, we may deduce Corollary 2.4 from (2.3), (2.7), Lemma 2.4, Lemma 2.5 and Corollary 2.2. o
3.
P r o o f of T h e o r e m 1.1
Let i e l The space W* defined in Theorem 1.1 is a Hilbert space with respect to the inner product 3
(u,v){i) := /
Y\
Djiik-DjVkdx + ft1 • / DiUk • DjVk dx + R71 •
u-vdox
(u,veWt).
We denote the corresponding norm by | |W, that is,
I"I w : = ( /
£
(•D^*)2 da; +
R
f -INdpi\dalis)V2
f r
o « e Wi.
Theorem 3.1 For A C R3, Zef 77i(A) denote the set of polynomials over A with degree less than or equal to 1, and put |w|o,2 := HHI2 f0T w e ^{Af, |w|i,2 := (H-Di^H2. + 2 2 1/2 X 3 \\D2w\\ + \\D3w\\ 2 ) for w e tf (A)' . Set K : = { » e C ° ( H ) 3 : wlif, e ^ i ( i f , ) /or 1 < I < n } for
iel.
373 Suppose there is C\ > 0 and for each i e l , a linear operator IIj : i7 x (Pj) 3 i-» Vt such that (^(diamKi)2^-
\(w -ni(w))\Kl\l2)1/2
< Cx • |«,|1>2
(3.1)
1=1
/or u> e iJ^Pj) 3 , A; e {0,1}. Then there is a constant C2 > 0 with R~1/2 -\\(w-
Uiiw)) | dPi\dQ ||2 < C-\w | l j 2
for i eT, we
Hl{Pif.
Proof: Let i e I , u; e .H" 1 ^) 3 , I e {«; + 1, ..., n } . We shall write C for constants which do not depend on i, w or I. For brevity, we set v := w — IT,(iy). Recalling that .4((?7, 1 — r]i — 772) =
C • (dmmKt)2
<
•[
\v(iPl(V))\2
dV
C-(diamJif,) 2 -|l« 0 ^i|^ ( 3 > ll? 1 2.
where the last inequality follows by a standard trace theorem on K^. Using the transformation rule and recalling the standard estimate detAf 1 < C • (diamAT; )~ 3 , we conclude f~m\v(
• 9i(v) dr)
< C-diamXr^((diamA'()"2-k(2/)|2
+
\Vv{y)\2)dy.
But diumKi < h( • Ri/S < Ri by (A3), so Theorem 3.1 is implied by (2.3) and (3.1). o We need some further technical results related to the outer part <9Pj\9Q of the boundary of P^ Theorem 3.2 There is a constant C > 0 with || w I dBs ||2 < C • I w | (i)
for i el
with ht < S0, we
Hl{Pif,
where S0 was introduced in Corollary 2.2. Proof: Take i € K. with h( < So, we ^(Pi). Recall the definition of the patches Fi = Fi and T; = T, on dB^ and 9P,\9f2, respectively, and of their local parameters fi =
-y)eTtC
dPj\dtt
for y e R;1 • Ft and for l 6 {m + 1, ..., n}.
For such I, we set r , := {r • y : y e R;1 • Ft, r e K with S < r < \ ( W o ^ f 1 ) ^ • j,) | } . Then we have by (A5)
u{r7 : Ki + i
= Pi\Bs,
r,nr fc = 0 for k,ie {«* + !,..., r j ,
k^i.
374 Thus, in view of (2.2) and (2.3), it suffices to show for / G {«i + 1, ..., r,} : / \w\2 dox < C-( f \Vw\2dx + R'1 • f \w\2 dox). (3.2) Jis/RtyF, Wr, JT, I Here and in the rest of this proof, the letter C denotes constants only depending on S. In order to prove (3.2), we fix I e {/c* + 1, ..., TJ}. Then we find for y € R~l • Ft, with the abbreviation 7 := | ((pi o •0i~1)(i?i • y) | : \w(S-y)\
= \w(S-y)
- w(j • y) + w(-y-y)\
= | / Vw(r • y) • y dr + w(^ • y) |
< (J\-2dr)1/2-(f\2-\Vw(r-y)\2dr)1/2
+
\w^-y)\,
hence S2-\w(S-y)\2doy
/ <
r2-|V™(r-y)|2d0ydr
(S-T7-1
(S/Ri)3-A),
+
with A := /
«/(I (w o ^l){Ri
flM
• y) I • y) | 2 A y
It follows
|H 2 *>*
/"
<
J{S/Ri)-F,
c - ( /" |v™| 2
R;1-A).
•
>
The mapping R^1 • ipi is a local parameter of .R,"1 • Ft, with i?t~2 • vt as element of surface area. Thus, taking into account (2.3), Corollary 2.2 and the definition of ipi, we get A
=
/~,J™(<M f ?))| 2 -^('?)*?
JKW
= 2- J
<
2
- L,J™(
\w(y)\2doy.
This estimate completes the proof of (3.2), and hence of Theorem 3.2.
o 1
3
Theorem 3.3 There is C > 0 such that for i € X with hi < So, w € H ^/^) ,
we have
RJ112 -\\W\ dP^dQ ||2 < C • (|| Vw ||2 + i?r 1 / 2 • I M dBn, || 2 ). Proof: Theorem 3.3 follows by arguments analogous to those used in the proof of Theorem 3.2. We omit the details. o We note two consequences of the preceding theorems. Theorem 3.4 There is C > 0 with || w | B s \ ^ l l 2 < C • \ w \{i} So, w e Wi.
for i e 1 with h{ <
Proof: According to [1, Lemma 4.1], there is some C > 0 with IMI2 < C • (|| Vt; ||2 + S-1/2 • || v I dBs || 2 )
for v e ^ ( B ^ H ) 3 with v\dQ = 0.
Theorem 3.4 follows from this result and Theorem 3.2.
o
375 Theorem 3.5 There is a constant C > 0 and for any i € I with ht < So, an operator Vi-. L2{Pi) ^W{ such that divV^-K) = v, |X>i(7r) | (i) < C • ||TT ||2 for TT e L2{P{). Proof: According to [1, Theorem 4.1], there is a constant C > 0 and for any i G I, an operator Vi : L 2 (Q ft ) >-> H1^^)3 such that we have for g G L 2 (Q^) :
div^(g) = Q, Vi(Q) | an = o, || VV^Q) ||2 + R'l/2 • || Ha) I OB* \\2 < c-\\e||2. Thus, if we set Vi{g) := T>i{g) | Pi for i G I, g G L2(Pi), where g denotes the trivial extension of g to Q^, then Theorem 3.5 readily follows from Theorem 3.3. o Now we are in a position to carry out the Proof of Theorem 1.1: Take i e I with ht < So- Let C > 0 denote constants in (0, oo) which only depend on Q, 5 and the parameter a2 from Lemma 2.1. We have ( M ( i ) ) 2 < C-ai(v,v),
\a{v,w)\
< C-M«-Mw
for v, we Hl {Pi)3.
(3.3)
Moreover, by Theorem 3.5, the pair of spaces (Wi,Z/2(P;)) satisfies the Babushka-Brezzi condition, uniformly in i G X. Due to these results, we may apply the theory of mixed variational problems, as presented in [7, p. 57-61]. It follows there is a pair (vi,gi) € 77 1 (P i ) 3 x L2(Pj) satisfying (1.6). The first inequality in (3.3) and a partial integration which exploits (1.1) yield || V(« | Pi - v)\\2 + R]12 • || (u - v) | dPi\dtt ||2 <
C-P t
1/2
(3.4)
-||£R j (P i un)(«,7r)|| 2 ;
compare the proof of [1, Theorem 5.1]. Using Theorem 3.5, we may deduce from (3.4) that the term || 7r | Pt — g ||2 is also bounded by the right-hand side of (3.4). Details of this argument can again be found in the proof of [1, Theorem 5.1]. Due to (3.4) and Theorem 3.4, the term ||(u — D) | Q s ||2, too, may be seen to be bounded by the right-hand side of (3.4). Now inequality (1.7) follows with Corollary 2.4. o
References [1] Deuring, P.: Finite element methods for the Stokes system in three-dimensional exterior domains. Math. Methods Appl. Sci. 20 (1997), 245-269. [2] Deuring, P.: Numerical tests of an artificial boundary condition for exterior Stokes flow. In: Feistauer, M., Kozel, K., Rannacher, R. (eds.): Proceedings of the 3rd Summer Conference on Numerical Modelling in Continuum Mechanics, Prague, 1997. Matfyspress, Univerzity Karlovy, Prague, S. 227-235 (1997). [3] Deuring, P.: A stable mixed finite element method on truncated exterior domains. RAIRO Math. Model. Anal. Numer. 32 (1998), 283-305. [4] Deuring, P.: Calculating Stokes flow around a sphere: comparison of artificial boundary conditions. In: Proceedings of the 7th Conference on Navier-Stokes and Related Equations, Ferrara, 1999. Ann. Univ. Ferrara - Sez. VII - Sc. Mat. 46, 1-9 (2000).
376 [5] Deuring, P., Kracmar, S.: Artificial boundary conditions for the Oseen system in 3D exterior domains. Analysis 20 (2000), 65-90 [6] Galdi, G. P.: An introduction to the mathematical theory of Navier-Stokes equations. I. Linearized steady problems (Rev. Ed.). Springer, New York e.a., 1998. [7] Girault, V., Raviart, P.-A.: Finite element methods for Navier-Stokes equations. Springer, Berlin e.a., 1986. [8] Goldstein, C. I.: The finite element method with nonuniform mesh sizes for unbounded domains. Math. Comp. 36 (1981), 387-404. [9] Goldstein, G. I.: Multigrid methods for elliptic problems in unbounded domains. SIAM J. Numer. Anal. 30 (1993), 159-183. [10] Nazarov, S. A., Specovius-Neugebauer, M.: Approximation of exterior boundary value problems for the Stokes system. Asymptotic Anal. 14 (1997), 233-255.
On the Stefan problem with surface tension Joachim Escher1, Jan Priiss2, Gieri Simonett3 1 Institute for Applied Mathematics University of Hannover, D-30167 Hannover, Germany 2 Fachbereich Mathematik und Informatik Martin-Luther-Universitat Halle-Wittenberg D-60120 Halle, Germany 3 Department of Mathematics, Vanderbilt University Nashville, TN 37240, USA Email : [email protected] ; [email protected] ; [email protected] Abstract We present results on the uniqueness and the existence of smooth solutions to the one and two-phase Stefan problem with surface tension.
1
Introduction
The classical Stefan problem is a model for phase transitions in solid-liquid systems and accounts for heat diffusion and exchange of latent heat in a homogeneous medium. The strong formulation of this model corresponds to a moving boundary problem involving a parabolic diffusion equation for each phase and a transmission condition prescribed at the interface separating the phases. Molecular considerations attempting to explain supercooling and dendritic growth of crystals suggest to also include surface tension on the interface separating the solid from the liquid region. In order to formulate the Stefan problem, we introduce the following notations. Let f2 be a smooth bounded domain in R n , whose boundary dQ consists of two disjoint components, an 'interior' part J 1 and an 'exterior' part J 2 . We think of O as a homogeneous medium which is occupied by a liquid and a solid phase, say water and ice, that initially occupy the regions Qj a n ( i ^o> a n d ^ n a t ^o an< i ^o a r e separated by a sharp interface r 0 . More precisely, we assume that r 0 C fi is a compact closed hypersurface, that fij and fip are disjoint open sets such that H = fij U f2§ and such that d£T0 = J' U F 0 for i = 1,2. For the sake of definiteness, we consider the open set fij as the region occupied by the liquid phase. Consequently, the component J 1 is in contact with the liquid phase and J2 is in contact with the solid phase. The boundaries J 1 and J 2 , corresponding for
377
378 instance to the walls of a container, are fixed, whereas r 0 will change as time evolves, due to solidification or liquidation of the two different phases. Given t > 0, let T(t) be the position of r 0 at time i, and let V(-,t) and «(-, t) be the normal velocity and the mean curvature of T(t). Moreover, let f21(i) and fi2(£) be the two regions in Q separated by T(t). According to our assumption, Qx(i) is the region occupied by the liquid phase, and F(t) is a sharp interface which separates the liquid from the solid phase. Let v(-,t) be the outer unit normal field on T(t) with respect to 0 1 (i). We shall use the convention that the normal velocity is positive if Q 1 ^) is expanding, and that the mean curvature is positive if the intersection of H1(t) with a small ball centered at T(t) is convex. Consequently, the normal velocity is positive if the liquid region is growing, v points into the solid phase, and K is positive for a water ball surrounded by ice, and negative for an ice ball surrounded by water. Let To and u'0 : Q'0 —> R be given, where uj and u2, denote the initial temperatures of the liquid and solid phase, respectively. The strong formulation of the two-phase Stefan problem with surface tension consists of finding a family T := {F(t); t > 0} of hypersurfaces and functions ul : Ut>o (Q'{t) x {t}) —> R, satisfying dt^ - Au* =
0
in Q* (t),
BV = V
on J', on T (t), on F (t), infft,
u1
=
OK
[dvu] =
IV
rf(0)
= <4
(1)
r(o) = r0l where I > 0 is the latent heat per unit volume absorbed or released for melting or solidifying, and a > 0 is the surface tension. Moreover, [dvu] := dvu2 - dvu\
(2)
denotes the jump of the normal derivatives of u1 and u2 across the boundary T(i). Finally, B' denotes the Dirichlet or the Neumann boundary operator on the fixed boundary J', and b1 are given functions, where i = 1,2. For simplicity, we assume the conductivity and the diffusion coefficients for the different phases to be the same, which we then normalize to one. Using distinct constants for the solid and liquid phase does not alter the mathematics in a significant way. The condition u' = an on the free interface is usually called the Gibbs-Thomson condition, see [8, 10, 29, 30, 32, 38, 44, 46, 53]. If either u1 or u2 is replaced by 0, while all the other aspects of the problem are left unchanged, then the modified problem is called the one-phase Stefan model with surface tension. If the Gibbs-Thomson condition on the free interface T(t) is replaced by u' — 0, then (1) is called the (classical) Stefan model. It should be emphasized that the Stefan problem with Gibbs-Thomson correction (1) differs from the classical Stefan problem in a much more fundamental way than just in
379 the modification of an interface condition. This becomes evident, for instance, by the fact that in the classical Stefan model, the temperature completely determines the phases, that is, the liquid region can be characterized by the condition u > 0, whereas u < 0 characterizes the solid region, where u = 0 is the melting temperature. The inclusion of surface tension will no longer allow to determine the phases merely by the sign of u. The main reason for introducing the Gibbs-Thomson correction u = an stems from the need to account for so-called supercooling, in which a fluid supports temperatures below its freezing point, or superheating, the analogous phenomena for solids; or dendrite formation, in which simple shapes evolve into complicated tree-like structures. The effect of supercooling can be in the order of hundreds of degrees for certain materials and is required for nucleation, namely the forming of a new phase in a set previously occupied by the parental phase, see [10, Chapter 1] and [52]. However, (1) does not account for nucleation effects and can only be applied in the presence of impurities or of some other mechanism inducing nucleation. Supercooling is an equilibrium phenomenon and is not merely a transient effect. A simple argument for appreciating these equilibrium effects as a first-order correction to the continuum mechanics is given in [8, pp. 209-210]: Suppose that u = 0 is the equilibrium temperature between a sohd and liquid separated by a planar interface. Then a certain amount of energy is required for an atom at the surface to overcome the binding energy of the crystal lattice and become part of the the liquid with lower binding energy. The amount of energy required to produce this transition depends on the number of nearest neighbors in the crystal structure and on the number of nearest neighbors of an atom on the surface. If the interface between the solid and liquid is curved, then an atom on the interface has fewer nearest neighbors and one expects that it will require less energy to produce a phase transition. If one considers a solid with constant mean curvature (an ice ball) in equilibrium with its melt, then one expects the prevailing temperature to be lower, and the Gibbs-Thomson condition indeed predicts that u = —o/R, where R is the radius of the ball. Hence, the fluid is supercooled even at equilibrium. A more detailed account of this argument, which is well-known to scientists in solid state physics and material science, can be found in [10, 32]. A more satisfactory argument leading to the same conclusion (and further generalizations) may be obtained from statistical mechanics see [9, 33, 44]. In [29, 30], a theoretical framework is developed starting from general thermodynamical laws which are appropriate to a continuum and which include interfacial contributions for both energy and entropy. The Stefan problem has been studied in the mathematical literature for over a century, see [49, 42] and [53, pp. 117-120] for a historic account, and has attracted the attention of many prominent mathematicians. The (classical) Stefan problem is known to admit unique global weak solutions, provided the given functions u'0 and b' have the physically correct signs; see for instance [25, 26, 34] and [37, pp. 496-503]. It is important to point out that the existence of weak solutions is closely tied to the maximum principle, see for instance the proofs in [25]. If the natural sign conditions for uj and b' are obstructed, then the Stefan problem becomes ill-posed, see [16] for instance.
380 As soon as surface tension is included in the model, the phase regions Q'(t) are in general no longer coupled to a sign condition for the temperature distributions, and one cannot resort to a comparison principle. As a consequence, many of the methods which were successfully applied for the classical Stefan problem are not available for the Stefan problem with surface tension. Results concerning the regularity of the free boundary for weak solutions of the multidimensional one-phase Stefan problem (without surface tension) were established in [4, 5, 27, 35, 36], and continuity of the temperature was proved in [7]. The regularity results were derived by formulating the Stefan problem as a parabolic variational inequality, see [18, 27]. In order to obtain the smoothness results, the authors in [27, 36] had to impose restrictive geometric assumptions on the initial data which assure that the melting is rapid and free from breaking off. Under fairly weak assumptions on the data it was shown in [40] that any weak solution eventually becomes smooth and that T(t) approaches the shape of a (growing) sphere. If the data are sufficiently smooth and satisfy high order (up to order 23!) compatibility conditions, classical solutions were obtained in [31]. The approach relies on the Nash-Moser implicit function theorem. Continuity of the temperature for weak solutions of the multidimensional two-phase Stefan problem (without surface tension) was obtained in [6, 14, 15, 54]. More recently, the regularity of the free boundary for weak (viscosity) solutions was studied in [1, 2, 47] under a non-degeneracy condition. Local existence of classical solutions in a small time interval was proved in [41], provided that the initial data satisfy high order compatibility conditions. The continuity of the temperature distribution for an m-phase Stefan model with m > 2 has been studied in [17]. Although the Stefan problem with surface tension (1) has been around for many decades, only few analytical results concerning existence and the regularity of solutions are known, see [19, 28, 39, 41]. In [28], the authors consider system (1) with small surface tension 0 < < r C l and linearize the problem about a = 0. Assuming the existence of smooth solutions for the case a = 0, that is, for the classical Stefan problem, the authors prove existence and uniqueness of a weak solution for the linearized problem and then investigate the effect of small surface tension on the shape of T(t). Existence of global weak solutions for the two-phase problem (1) is established in [39], using a discretized problem and a capacity-type estimate for approximating solutions. The weak solutions obtained in [39] have a sharp interface, but are highly non-unique. In [41], the way in which a spherical ball of ice in a supercooled fluid might melt down is investigated. A proof for the existence of classical solutions for (1), assuming restrictive high order compatibility conditions for the initial data, is sketched in [48]. In [19], the existence and uniqueness of analytic solutions in case that F(i) is the graph of a function over R" _1 is obtained. If the diffusion equation dtu'—£yu? = 0 is replaced by the elliptic equation Au* = 0, and the initial condition for ul is dropped, then the resulting problem is the quasi-stationary Stefan problem with surface tension, which has also been termed the Mullins-Sekerka model (or the Hele-Shaw model with surface tension). Existence, uniqueness, and regularity of solutions for the quasi-stationary approximation has recently been investigated in [3, 11, 12, 21, 22, 23, 24].
381 Weak solutions to the modified Stefan problem with surface tension and kinetic undercooling, that is, with the condition ul = <JK — 0V on the free interface r(t), were obtained by Visintin [50]. For classical solutions for the Stefan problem with the condition u* =
i = l,2,
(3)
1
where v is the outward pointing normal on <9fi. The case S = 0 corresponds to a Dirichlet boundary condition on the fixed boundary J*, whereas 5Z = 1 results in a Neumann (that is, no-flux) boundary condition on J'. In the following, we denote by W£ the Sobolev-Slobodeskii spaces of fractional order, where s > 0 and p 6 (1, oo). Let I = [0,T] be a given interval. In order to deal with the inhomogeneous boundary conditions Blu% = W, we introduce the function spaces G'(J) := W^V^1to(I,Lp(Ji))nLp(I,W-ti)-1/''(Ji)), Theorem 1.1 Letp£
(n+2, oo). Suppose that{U\,YQ,V) a/p
i = 1,2.
(4)
satisfy the regularity conditions
3
K,r0,&') e w?- (n&) x w^ '" x G I (I) and the compatibility conditions
ui\ro = *K0,
B'4 = bi(o), [cUo] e % 2 - 6/p (r 0 ).
(5)
Then there exists a positive number To G / such that the Stefan problem (1) has a unique solution on [0,To] in the above-defined class (6). The manifold M. := Ute(o,T0)(r(i) x {£}) is real analytic and
«jeC"
(J ($T (t) u r(t)) x {t} J. \t£(0,T 0 )
/
Proof. This follows from Theorem and from parabolic regularity theory.
•
Remark 1.2 (a) Suppose that (u'0,r0,bl) satisfy the regularity conditions
(uj.ro.b*) e c3(W0) x c 4 x (tfiiMJ))
nc(/,cV)))
and tte compatibility conditions u 0 | r o = <XK0,
B*t4 = 6*(0).
Then the conclusions of Theorem 1.1 hold. (b) The first two compatibility conditions in (5) arise naturally for classical solutions. Besides, observe that [<9„HO] e W1~3/p(Yo) since u0 belongs to W2~2/p(fl'0). Hence the third compatibility condition in (5) is a mild assumption on the spatial regularity of the initial velocity of [t — t * T(t)] in normal direction. (c) Theorem 1.1 specifically asserts that T(t) is an analytic hypersurface in Rn for eacht e (0,T 0 ).
382
2
T h e transformed equations
A major difficulty in the mathematical treatment of the Stefan problem (1) comes from the fact that the boundary T(t), and thus the geometry, is unknown and ever changing. A widely used method to overcome this inherent difficulty is to choose a fixed reference surface £ and then to introduce an unknown diffeomorphism 6{-,t) : £ —> F(i) which tracks the position and the regularity of the moving surface T(i). In a next step one can try to extend this diffeomorphism to a diffeomorphism of fixed reference regions D' into the unknown domains £}*(£). This approach is usually called the direct mapping method. The treatment of the moving boundary problem (1) then proceeds by transforming the equations into a new system of equations defined on the fixed domain D1 U D2 from which both the solution and the diffeomorphism 8 have to be determined. In the context of the Stefan problem this method has first been used by Hanzawa [31]. In order to implement this approach we choose a closed compact analytic hypersurface S c i l which is close to To- Such a surface always exists, see [20]. We can assume that E divides fl into two disjoint regions Dl and D2 such that D1 is enclosed by E, and such that dDl = J1 U E for i = 1,2. Clearly, we could choose E = r 0 . However, since our main objective is to impose only minimal regularity and compatibility conditions for the initial data and then to prove that solutions regularize, the choice E = To is not appropriate. Let p, be the outer unit normal field on E with respect to D1. Then we can find a number ao > 0 such that /Ex(-ao,«o)
-+
V
| >
(p.n
-
R-
\
p+w ) /
is a smooth diffeomorphism onto its image H :— im(X), that is,
XeDifP°(Ex
(-a0,a0),n).
It will be convenient to decompose the inverse of X into X - 1 = (P, A) such that
fer(K,E),
AeC°°(TZ,(-a0,aQ)).
Of course, 1Z, is the set of those points with distance less than ao to E. For further use, we introduce the set of admissible functions 21 := 2l0 := {p e C 2 (£); HPHCHE) < a }i where a e (0, a0) is a fixed number. Given p G 21 we define the mapping p
>-> P +
(7)
P{P)KP)
and we set Fp = im(^). Then Fp is a C2-hypersurface and 6P : E —v Fp is a C 2 diffeomorphism between the manifolds E and Tp. Moreover, if a0 is small enough then Fp is contained in fl and it separates 0 into two domains Oj and Q2 just as E does. In particular, diYp = J1 U Tp for i = 1,2. We will now introduce an extension Qp : R n —> R" of 6P, with the property that e p GDiff 2 (R",R n ) and
e , e Diff2^, a) n Diff2(Di, %),
e p | E = ep.
(8)
383 Let a g (0, oo/4) be given and let tp be a C^-cutoff function with the properties that 0 <
'
+ ¥>(A(z))p(P(s))) if xGTl, if a; g ft.
\ x
The function [r i-» r + ^ W P W ] i s strictly increasing for any p e 21 and any p C E , since |
(9)
We now consider time-dependent functions p e C-([0,T],C(Y:)) n C([0,T],2l) with T > 0 a fixed number. Each p defines a family T := {T(i) = r p ( t ) ; t 6 [0,T]} of hypersurfaces. Note that T(t) can also be described as the zero-level set of the function •R x [0, T] ->
(x,t)
R
(10)
.-» A(x)-p(P(x),t)
that is r(<) = 4>p(-,t) 1(0), for £ G [0,T]. Consequently, the unit normal field v{-,i) on T(t) at the point x = X{p, p(p, £)) can be expressed in coordinates of E as v{p, t) = V«MM) IW„(a:,i)|
,
pGS
(11)
x=X(p,p(p,t))
where V denotes the gradient with respect to the x-variable. Moreover, the normal velocity V of T at time t and at the point x = X(p,p(p,t)) e T(t) is given by
v(p,t)= | V dtp{p 't] ^(x,i)|
,
pGE.
(12)
x=X(p,p(j,,t))
We can now readily infer from (11)-(12) that the Stefan condition [d„v\ = IV in (1) translates into ldtP{p,t)={Wu2{x,t)-Vu\x,t)\W
peE,
(13)
where (• | •) denotes the Euclidean inner product of ". In the following, we will assume that
pec1([o,nc(E))nc,([o,r|)ai)
(14)
where 21:= {g G C 2 (E); HsHc^s) < °} with a e (0,a 0 /4). For each p we define the cylindrical domains DT := (D1 U D 2 ) x [0,T],
n„, T := ( J (fi^, U ft2(t)) x {t}. te[o,T]
384 Every p which satisfies (14) induces a diffeomorphism * p : DT -> nPiT,
§„(y, T) := ( 0 p M ( y ) , r).
We can now consider the pull-back operator $* given by $*pu := u o $„,
for
u G BUC (Q„,T).
Analogously, the push-forward operator $£ is defined by <$>y :=Vo $ - ! ,
for
v G BUC(DT).
For a given function p which satisfies (14), we consider the transformed differential operators A(p)v := SJ (A (*>«)), C (p) „ := <^ 7 2 (V (*£«») I V0„) - - ^ y (V (fcjt;1) | V ^ ) , for ?; G C((0,T],C 2 ((D 1 U -D2)), where u' := t>|z>x[o,T], and where 7* denotes the trace operator from D' onto E. As before, V stands for the gradient with respect to the space variables. The transformed mean curvature operator is given by H(p):=%Krp,
(16)
where « r (t) denotes the mean curvature of Tp(t) for t e [0,T]. Lastly, we consider the mapping R(v,p) given by R(v,p)(y,t)
:= { f(My))dtP(P(y),t)
(%(V(^v))(y,t)\p(P(y)))
y e K,
(1?)
It follows from (9) and from (12)-(17) that the two-phase Stefan problem with surface tension (1) can be transformed into the following system of equations for v := $*« dtv Bv v ldtp v (0) p{0)
= = = = = =
A(p)v + R(v,p) 6 aH{p) C(p)v v0 po,
in DT, on JT, onET, onE r , in D,
{
,.R) '
where JT := (J 1 U J 2 ) x [0,T], E T := E x [0,T], and where (B«)Ux[o,n ~ B V ,
6| J i x | 0 ,T]=6 i ,
t = l,2.
Theorem 2.1 ILet p G (n + 2,00). Suppose that the functions {vo,po,b) satisfy the regularity assumptions (v0,Po,b) e Wl-**{D)
x W^p)
x (\I)
x 2 (/))
385 and the compatibility conditions C(p0)v0eW^p(D).
«b|E = ff#(A>), Bvo = b(0),
Then there exists a positive number T0 6 I such that the coupled system of transformed equations (18) admits a unique solution (v,p) on I0 := [0,T0] with v e W}{I0, LP(D)) n LP(IQ, W2(D)), p € W P 3/2 - 1/2p (I 0 , L P (E)) n W}-1,2p(I0, % 2 (E)) n L p (/ 0 , W P 4 " 1/p (E)). Moreover, the solution (v, p) satisfies PeC-(Ex(0,T0)),
v* G C^dD* U E) x (0,T0)) n C^V* w/tere V : = { i e ( ] ; dist(a;, E) < a} and V* =
x (0,T 0 )),
Vnl>.
Proof. Based on [19], the proof of this result will be given in [20].
References [1] I. Athanasopoulos, L. Caffarelli, S. Salsa, Regularity of the free boundary in parabolic phase-transition problems. Acta Math. 176 (1996), 245-282. [2] I. Athanasopoulos, L. Caffarelli, S. Salsa, Phase transition problems of parabolic type: flat free boundaries are smooth. Comm. Pure Appl. Math LI (1998), 77-112. [3] B.V. Bazalii, Stefan problem for the Laplace equation with regard to the curvature of the free boundary. Ukrainian Math. J. 49 (1997), 1465-1484. [4] L.A. Caffarelli, The regularity of free boundaries in higher dimensions. Acta Math. 139 (1977), 155-184. [5] L.A. Caffarelli, Some aspects of the one-phase Stefan problem. Indiana Univ. Math. J. 27 (1978), 73-77. [6] L.A. Caffarelli, L.C. Evans, Continuity of the temperature in the two-phase Stefan problem. Arch. Rational Mech. Anal. 81 (1983), 199-220. [7] L.A. Caffarelli, A. Friedman, Continuity of the temperature in the Stefan problem. Indiana Univ. Math. J. 28 (1979), 53-70. [8] G. Caginalp, An analysis of a phase field model of a free boundary. Arch. Rational Mech. Anal. 92 (1986), 205-245. [9] J.W. Cahn, D.W. Hoffman, A vector thermodynamics for anisotropic surfaces II. Curved and faceted surfaces. Acta Metallurgica 22 (1974), 1205-1214. [10] B. Chalmers, Principles of solidification. Krieger, Huntington, N.Y., 1977.
386 [11] X. Chen, The Hele-Shaw problem and area-preserving curve-shortening motion, Arch. Rational Mech. Anal. 123 (1993), 117-151. [12] X. Chen, J. Hong, F. Yi, Existence, uniqueness, and regularity of classical solutions of the Mullins-Sekerka problem. Comm. Partial Differential Equations 21 (1996), 1705-1727. [13] X. Chen, F. Reitich, Local existence and uniqueness of solutions of the Stefan problem with surface tension and kinetic undercooling. J. Math. Anal. Appl. 164 (1992), 350362. [14] E. DiBenedetto, Regularity properties of the solution of an n-dimensional two-phase Stefan problem. Boll. Un. Mat. Ital. Suppl. (1980), 129-152. [15] E. DiBenedetto, Continuity of weak solutions to certain singular parabolic equations. Ann. Mat. Pura Appl. 130(4) (1982), 131-176. [16] E. DiBenedetto, A. Friedman, The ill-posed Hele-Shaw model and the Stefan problem for supercooled water. Trans. Amer. Math. Soc. 282 (1984), 183-204. [17] E. DiBenedetto, V. Vespri, On the singular equation (5{u)t = Au. Arch. Rational Mech. Anal. 132 (1995), 247-309. [18] G. Duvaut, Resolution d'un probleme de Stefan (fusion d'un bloc de glace a zero degre). C.R. Acad. Sci. Paris Ser. A-B 276 (1973), 1461-1463. [19] J. Escher, J. Priiss, G. Simonett, Analytic solutions for a Stefan problem with GibbsThomson correction. Submitted. [20] J. Escher, J. Priiss, G. Simonett, Smooth solutions for the Stefan problem with surface tension. In preparation. [21] J. Escher, G. Simonett, On Hele-Shaw models with surface tension. Math. Res. Lett. 3, 467-474 (1996). [22] J. Escher, G. Simonett, Classical solutions for the quasi-stationary Stefan problem with surface tension. Differential equations, asymptotic analysis, and mathematical physics, (Potsdam, 1996), Math. Res., 100, Akademie Verlag, Berlin, 1997, 98-104. [23] J. Escher, G. Simonett, Classical solutions for Hele-Shaw models with surface tension. Adv. Differential Equations 2 (1997), 619-642. [24] J. Escher, G. Simonett, A center manifold analysis for the Mullins-Sekerka model. J. Diff. Equations 143 (1998), 267-292. [25] A. Friedman, The Stefan problem in several space variables. Trans. Amer. Math. Soc. 133 (1968), 51-87. [26] A. Friedman, Variational Principles and Free-Boundary Problems. Wiley-Interscience, New York, 1982.
387 A. Friedman, D. Kinderlehrer, A one phase Stefan problem. Indiana Univ. Math. J. 24 (1975), 1005-1035. A. Friedman, F. Reitich, The Stefan problem with small surface tension. Trans. Amer. Math. Soc. 328 (1991), 465-515. M.E. Gurtin, On the two phase problem with interfacial energy and entropy. Arch. Rational Mech. Anal. 96 (1986), 199-241. M.E. Gurtin, Multiphase thermomechanics with interfacial structure. Arch. Rational Mech. Anal. 104 (1988), 195-221. E.I. Hanzawa, Classical solutions of the Stefan problem. Tohoku Math. Jour. 33 (1981), 297-335. P. Hartman, Crystal growth: An introduction. North-Holland, Amsterdam, 1973. D.W. Hoffman, J.W. Cahn, A vector thermodynamics for anisotropic surfaces I. Fundamentals and Application to Plane Surface junctions. Surface Science 31 (1972), 368-388. S.L. Kamenomostskaja, On Stefan's problem. Math. Sbornik 53 (1965), 485-514. D. Kinderlehrer, L. Nirenberg, Regularity in free boundary problems. Ann. Scuola Norm. Sup. Pisa 4(4) (1977), 373-391. D. Kinderlehrer, L. Nirenberg, The smoothness of the free boundary in the one phase Stefan problem. Comm. Pure Appl. Math. XXXI (1978), 257-282. O.A. Ladyzenskaja, V.A. Solonnikov, N.N. Ural'ceva, Linear and quasilinear equations of parabolic type. Translations of Mathematical Monographs, Vol. 23, American Mathematical Society, Providence, R.I. 1968. J.S. Langer, Instabilities and pattern formation in crystal growth. Rev. Mod. Phys. 52 (1980), 1-28. S. Luckhaus, Solutions for the two-dimensional Stefan problem with the GibbsThomson law for melting temperature, Europ. J. Appl. Math. 1 (1990), 101-111. H. Matano, Asymptotic behavior of the free boundaries arising in one phase Stefan problems in multi-dimensional spaces. Lecture Notes in Num. Appl. Anal. 5, Kinokuniya, Tokyo, 1982, 133-151. A.M. Meirmanov, On the classical solution of the multidimensional Stefan problem for quasilinear parabolic equations. Math. Sb. 112 (1980), 170-192. A.M. Meirmanov, The Stefan Problem. De Gruyter, Berlin, 1992. A.M. Meirmanov, The Stefan problem with surface tension in the three dimensional case with spherical symmetry: non-existence of the classical solution. European J. Appl. Math. 5 (1994), 1-20.
388 [44] W.W. Muffins, Thermodynamic equilibrium of a crystal sphere in a fluid. J. Chem. Phys. 81 (1984), 1436-1442. [45] W.W. MuUins, R.F. Sekerka, Morphological stability of a particle growing by diffusion or heat flow. J. Appl. Phys. 34 (1963), 323-329. [46] W.W. MuUins, R.F. Sekerka, Stability of a planar interface during solidification of a dilute binary alloy. J. Appl. Phys. 35 (1964), 444-451. [47] R.H. Nochetto, A class of nondegenerate two-phase Stefan problems in several space variables. Comm. Part. Diff. Equations 12 (1987), 21-45. [48] E. Radkevitch, Gibbs-Thomson law and existence of the classical solution of the modified Stefan problem. Soviet Dokl. Acad. Sci. 316 (1991), 1311-1315. [49] L.I. Rubinstein, The Stefan Problem. Translations of Mathematical Monographs, Vol. 27, American Mathematical Society, Providence, R.I. 1971. [50] A. Visintin, Supercooling and superheating effects in phase transitions, IMA J. Appl. Math. 35 (1985), 233-256. [51] A. Visintin, Models for supercooling and superheating effects. Pitman Research Notes in Math. 120, Longman Sci.& Tech., Essex, 1995, 200-207. [52] A. Visintin, Remarks on the Stefan problem with surface tension. In Boundary value problems for partial differential equations and applications. (J.L. Lions and C. Baiocchi, eds.) RMA: Research Notes in Appl. Math. 29, Masson, Paris, 1993. [53] A. Visintin, Models of Phase Transitions. Progress in Nonlinear Differential Equations and Their Applications, Vol. 28, Birkhauser, Boston, 1996. [54] W.P. Ziemer, Interior and boundary continuity of weak solutions of degenerate parabolic equations. Trans. Amer. Math. Soc. 271 (1982), 733-748.
Epidemic Models with Compartmental Diffusion 1
W.E. Fitzgibbon1, M. Langlais2 and J.J. Morgan3 Department of Mathematics, University of Houston Houston, Texas 77204-3476, USA 2 UMR CNRS 5466, case 26 Universite Victor Segalen Bordeaux 2 33076 Bordeaux Cedex, France 3 Department of Mathematics Texas A & M University, College Station, Texas 77843-3368, USA
Abstract Using a basic SI model we undertake the question of assessing the effects of a spatial heterogeneity upon the geographic spread of disease.
1
Introduction
In this note we undertake the question of assessing the effects of a spatial heterogeneity, i. e. a fragmented habitat, upon the geographic spread of disease. Although generalizations to more complex and realistic epidemiological models are possible, we shall consider only the basic SI model for disease transmission. To be more precise we consider a population subdivided into two subclasses susceptible and infective. Susceptibles are individuals who have yet to contract the disease. The infective class consists of those infected with the disease who are capable of transmitting. Interactions between susceptible and infective will produce a loss from the susceptible class and a gain in the infective class. We shall assume that the susceptible and the infective populations disperse throughout a bounded region or habitat and that they remain confined to this region for all time. We employ Fickian diffusion as a model of Brownian dispersion in the standard manner. Heretofore the dynamics and the qualitative behavior of what we have described are well known. What distinguishes the work at hand is that we shall assume existence of a subregion Q* C f2 where the transmission of disease takes place. In the region Q — CI* we assume that the disease spreads only via diffusive transport. The rates of diffusion are different for the susceptible and infective populations and the diffusion rates are distinct in fl* and Q — Q*. To conceptualize this model we may envision S¥ as an urban environment and the surrounding region fl — fi* as a rural environment. Here we are saying that the disease is transmitted and spreads through the urban area f2* and out into the surrounding rural region f2 — Q* where there is only dispersion for the population and no
389
390 transmission of the disease. Building upon this idea, we next consider a more complex habitat with several separated urban areas fJJ, • • • , 0£ surrounded by a rural environment Q — U;=i ^X- Finally, we are lead to the consideration of a complex region consisting of repeated fragmented local structure. Such a scenario would be exhibited by regions found in developped areas like western Europe, Japan or perhaps New England where one encounters a highly heterogeneous landscape consisting of repetitively interspersed fields, forests and small villages. At first glance, it appears as though this is covered by case above. However, from a practical point of view we are more interested in how the complex microstructure of the region affects the general macroscopic progression of the infection through the region. If we formulate this question in mathematiccd terms we can obtain a partial answer by implementing homogenization arguments. Ordinary differential equations have been used to model the circulation of disease through populations begining with the 19th century. Murray [18], and Busenberg and Cooke [3], have excellent monographs detailing the state of the art of this activity. Diffusive epidemic models have become widely structured in the past two decades. The interested reader is referred to [4], [8] and the references contained therein. Differential equations involving diffusion operators with discontinuous coefficients were originally termed diffractive and were used to model neutron transport in reactor engineering. The literature on this subject is incomplete but the reader is referred to [16]; see also [4], [20] [22], [23], [24]. Homogenization techniques were introduced in the late seventies and they have proved useful in the modeling of composite materials, in macroscopic properties of crystaline or polymer structures, and the description of aqueous and hydrocarbons flow through porous materials (cf. [2] and [15]). We set the stage by describing the basic underlying epidemiological kinetics. We use the state variables u and v to denote susceptible and infective classes of individuals. We model the transfer from the susceptibles to the infectives with an incidence term f(u, v) = kuv where k > 0 is a positive constant. The resulting system of differential equations has the form u = —kuv,
v = kuv;
w(0) = u0, v(0) = VQ.
(1)
Standard techniques yield (cf [8]) the following result. T h e o r e m 1.1 Ifu0 and VQ are nonnegative initial data then lim u(t) = 0 and lim v(t) = w0 + v0. t—*oo
t—*oo
Thus we totally answer questions of wellposedness, characterize forward invariant regions and define the long term behavior of solutions. From the viewpoint of qualitative dynamics this is certainly an unsurprising result, and from an epidemiological view point it simply says that the whole population contracts a non fatal disease. There certainly are much more complex models with much more complicated and exotic qualitative behavior. However, our purpose at hand is not to describe complex realistic disease kinetics but rather to describe how spatial considerations may be introduced to epidemiological modeling.
391
2
Diffusive epidemic modeling
Here we describe a population dispersing over but remaining confined to a bounded domain (or habitat) in R2. For technical reasons we assume that Q is a Lipschitz domain with piecewise smooth boundary dQ such that fi lies locally on one side of dQ. Our state variables become population densities u(x, t) and v(x, t) for x e O and t > 0 representing the spatial density of the susceptible and infective populations. The dispersion mechanism is assumed to be the standard Fickian approximation for Brownian motion. However we shall specify distinct diffusivities di and a\ for the susceptible and infective classes. Additionally, we require that mi{di(x),
rfiGC°°(n)nC(n) fori = l o r 2 , x G Q | i = 1 or 2} > d > 0 for some d > 0.
(2)
The following reaction diffusion type system describes the spread of the disease through ft du
-— = V • di(z)Vu - kuv,
i e ! l , t > 0,
— = V • d2(x)Vv + kuv, at
x € Q, t > 0,
§1
(3)
with no-flux boundary conditions f)ii
dl{x)^-=
on
f)n
d2{x)— = 0, Or]
xedtt,
t>o,
(4)
(where rj is the outward unit normal vector to O on dQ) and initial conditions u(x,0) = UQ(X) and Let Tp denote the spatial average Tp = ods established in [11], see also [4] and [8]
v(x,0) = v0(x), mE^s(m
x £
fi.
(5)
Jn tp(x)dx. It readily follows from meth-
T h e o r e m 2.1 Assume u0, v0 e C(S7) and are nonnegative on fi. Then system (3)-(4)-(5) admits a continuously differentiate solution pair (u,v) with u(x,t) > 0 and v(x,t) > 0 for x e 0, t > 0 and u, v € L°°(fi x R+). T/ien h m ^ ^ \\u(.,t) - " ( 4 ) 1 1 ^ = 0, and l i m t ^ o | K . , t ) - « ( t ) l l o o ^ = 0Finally, limt-,,^ u(t) = 0, and limt_,00 v(t) = Co = uo + t>oThus the diffusive model mimics the longtime behavior of the spatially homogeneous one.
3
Models with fragmented habitat
Here we take up the fragmented situation described in the introduction, first with two compartments. The surrounding region Q satisfies the conditions of section 2 as does Q*.
392
Moreover we assume that dQ* n dQ = 0. We assume the diffusivities to be discontinuous across the boundary of dQ*, namely
[ "i2(^J
for x G S2 — £2 *,
where (d^, di2) satisfy (2) on Q* and 0 — Q* respectively. We further assume that the disease may be transfered from infectives to susceptibles only in the region Q*. This can be modeled by introducing an incidence term having the form f(x,u,v) = k(x)uv where k(x) = k\(x),k > 0 and % i s * n e characteristic function of Q*. These considerations produce a system of partial differential equations with possibly discontinuous coefficients du — = V • dAx)Vu - k(x)uv,
x e Q, x 4 dQ*, t > 0,
i%
(6)
-H = V • d2(z)V«; + fc(a:)w, re e ft, a; 0 <9ft*, i > 0. We need to specify what happens at the interface between Q* and the surrounding region. We let [ra]an. denote the saltus, or jump, of a function w on dQ* and require that Man* = Man- = 0, and
du dn
an*
dv drj
• 0.
(7)
We require boundary and initial conditions (4)-(5). We point of that the first part of (7) insures continuity of the state variables across dQ* and the second part of (7) requires continuity of the flux across dQ*. In engineering applications such a condition stipulates that matter is conserved during transport across dQ*. A moment's reflection will convince the reader that a discontinuity of diffusitivities across dQ* produces a discontinuity of the normal derivatives across the interface. This discontinuity is typical of diffractive diffusion and it can be observed even in the computation of eigenvalues associated with the one dimensional steady state case of the differential operator, [14]. By a classical solution to (6)-(7) and (4)-(5) we mean a pair of continuous functions u(x,t), v(x,t) for x 6 fi, t > 0 which are twice continuously differentiable in x and continuously differentiable in t on open subsets of Q* and O — f2* and satisfy the partial differential equations (6) on Q* and f2 — H* and the boundary and interface conditions (7) and (4). Finally we require that
Jim |K,t) - «o|L,n = J™ H-.*) ~ "OIL* = °-
(8)
Theorem 3.1 If the foregoing conditions are satisfied there exists globally defined classical, nonnegative solutions to (6)-(7)-(8) and (4). There exists a constant MQ > 0 which is upon Hwollooft, Ikolloon and0 < d = inf{dj(x) | i = 1 or2,x € Q} such that
8up{|K.,t)IL,n.ll«(-.*)ILfl}<-Mo.
(9)
393 Proof. We begin with the observation that we use the differential operators V • di(x)Vw = 0, x £ Q. — 8Q* together with saltus and boundary conditions similar to (7) and (4), to define generators Ai(p) of analytic semigroups {Tf(t) \ t > 0} in LP(Q), p > 1 and C(Q). Thus we define
^•flw^l;:!
: k(x)uv, ••
—h(x)uv,
and locally solve our problem using the standard variation of parameters formulations [19] «(.,t)=7T(t)«o+ / Jp «(.,*)= I f (t)«o + f
Tf(t-3)F1(u{.,8)v(.,3))d8, T?(t-s)F2(u{.,s)v{.,8))ds.
J 0
The nonnegativity of u(x,t),v(x,t) is established in [14] and [21] and follows from a maximum principle argument. By virtue of the k(x)u(x,t)v(x,t) > 0 and hence du/dt < V • di(x)Vu, we may now apply a maximum principle to observe that l l u ^ ^ U ^ j j < ll^olloofj = ki and hence that dv/dt < V • d2(a:)Vr> + kik(x)v. In addition, we can integrate our equations over Q and sum the results to obtain a uniform L1 (f2) a priori estimate for v(-,t). These latter observations allow us to apply Moser-Alikakos iterative arguments to guarantee the existence of Mo > 0 so that IIKC^H,*, n < Mo. • In [1] these techniques are carefully outlined for standard diffusion; it has been observed in [6] that these arguments immediately carry over to the diffractive case because the interface compatibility conditions insure the contributions resulting from boundary integrals in integration by parts cancel. Given the existence of a priori estimates on u and v our local results can be extended globally via standard analytic semigroup continuation arguments [13]. Theorem 3.2 If (u,v) is the solution pair to (6)-(7)-(8) and (4) then the conclusion of Theorem 2.1 still holds. Proof.
We begin with the first equation. If we integrate on fi x (0,t) we obtain ll u (-. i )lli 0 o+ / Jo Jn
k(x)u{x,t)v{x,t)dxdt=\\u0\\1M.
Use of the existing uniform a priori estimates on ||«(-,*)lloo,n a n d ll,;(-!*)lloo,n yields kuv G Lp((0, oo) x fi), for all p G (1, oo). Multiplying (6) by u and du/dt yields udu/dt = uV • di(x)Vu — k(x)u2v, and
Standard applications of integration by parts and Young's inequalities produce the following |Vu|eL2((0,oo)xn),
f)ij
— eL2((r,oo)xn),
r > 0,
(10)
394 weLoc((0,oo),Z, 2 (n)),
IVuleLooCCr.oo),^^)),
r>0.
(11)
Analogous arguments produce the same results for v,dv/dt and |Vu|. We may now apply Meyers Lemma for parabolic equations with discontinuous coefficients, [17] and [2], to guarantee that u,v€ W1
^'
(—too
^'
A standard application of the Ritz Lemma guarantees that there exists a constant M 3 (obtained from the positive eigenvalues of the stationary diffractive diffusion problem) so that
l|w(.,*)-«(*)lk n <W3|||V«(.,t)||| 2 i n H;t)-v(t)\\2,n<M3\\\Vv(.,t)\\knTherefore there will exist a constant Mp so that \H,t)
-u(t)\\P,n<
Mp\\\Vu(.,t)\\\P,n
IK,<)-^)IU<Mp|||V<,t)|||P]n. The Sobolev embedding theorem now insures the existence of a Cp so that IK,<)-«Wlloo,n
\H,t)-v{t)\\ooV
395 system of partial differential equations of the form (6), (4) and (5) is straightforward. We require the diffusitivities to be strictly positive, i.e. maXx^^d^x) \ i = 1, 2} > d > 0 for some d > 0, but allow them to be distinct and possibly discontinuous across the interface of fi* = fij U • • • U Q*k with the ambient region. For i = 1 or 2 we specify
4(*) = f }°,(l> ^ - ^ u - - - u ^ ) > n
'
(12) y
\ dij(x) x e 0.*,
1
'
1
where di0 and dy £ C (n — (fi* U • • • U f2£) and C (H^) respectively. We allow a more general incidence term f(x,u,v) = k(x)uv, where k(x) is uniformly bounded on Q, and uniformly continuous on Q*j (j = 1 to k) and 0 — (O* U • • • U fi£) and nonnegative on H. The interface compatibility conditions are a straightforward adaptation of (7). Time dependent diffusivities could be treated using the theory of linear evolution equations rather than semigroup theory.
4
Complex Dynamics with repeated microstructure
Many habitats consist of repetitively interspersed fragmented subpatches. In certain cases we can isolate a subdomain 0 , and consider fi as being produced by a periodic reproduction of Q where the structure of © is small by comparison with the size fL Despite the disparity in scale this microstructure can have profound effects upon the long term dynamics of the spread of the disease through Q, [12]. Theoretically, this situation can be described by methods of the preceding section however from a more practical computational view the microstructure of our domain is far too fine to track and it becomes reasonable to implement homogenization technique to in essence average the effects of the microstructure over the whole domain and obtain a system which approximates the global dynamics over Q. We introduce a basic cell 0 = J\2j=1 [0, Y}°] C R2. A function (p : R2 -> R is said to 6 periodic if it admits a period Y? in the direction yj. We let di(y) and (^(y) be functions which satisfy (12) with subregions 0 ! • • • 6fc in 9 . We assume that di(y) and d2(y) satisfy the conditions outlined at the end of section 3. We can also find the appropriate conditions for the coefficient of the incidence term f(y,u,v) = k(y)uv on the fundamental cell 0 . We extend di, d2 and k to all of R2 by periodicity, producing periodic functions dx, d2 and k. We let n* • • • fj* be the distinguished regions of the periodically reproduced copy 0. For e > 0 we define ipe(x) = tp{x/e) for x 6 Q, and ip = d\,d\,k. following system of partial differential equations on Q x (0, oo) 3li
-^ OV,
= V • d^(z)Vu£ €
ke(x)ueve,
For small e the
(13)
c
— = V • d 2(x)Vve + k {x)ueve.
will have rapidly oscillating coefficients. We impose homogeneous Neumann exterior boundary conditions similar to (4) and initial conditions (5).
396 We shall need compatibility conditions on the interfaces between the Oj and 0 and their respective reproductions ©^ (J = 1 • • • k, I = 1 • • • n) we require that Mae*, = MeSj, = 0.
Ki^du/dr,]^
= [^(^dv/dr,]^
= °-
( 14 )
We are guaranted the existence of solutions to (13)-(14), (4) and (5) for sufficiently small e > 0 and the question now becomes what happens as e J. 0. Let M{
Je *P(y)dy-
T h e o r e m 4.1 For each e > 0 and T > 0 there exists a globally defined classical, nonnegative solution of (13)-(14), (4) ".nd (5), {ue,ve). Moreover there exists positive definite, symmetric matrices with constant coefficients Z)'1' and D^ representing homogenizations of di and d2 respectively and a constant k = M(k), such that (uc(.,t),ve(.,t)) converges strongly in L2((0,T)) to a nonnegative classical pair (w,z) solution to ~ = V -D^Vw-kwz, & — = V -D^Vz + kwz, at
xeQ,t>0, ( 15 )
x(=Q,t>0,
with an initial condition (5) and standard no flux boundary conditions
J2^f^«*fa.XJ)
= J2D^^.cos(-ri'**•)
= °'
x€an
''>°-
^
Proof. The well posedness is an immediate consequence of our results for diffractive systems. The convergence result is predicated upon the application of results in [2] and [15]. The crucial step lies in using the result in Theorem 3.1 stating that max t > 0 {||u e (.,i)|| o o r i , ||v e (.,t)|| oot2 } < M0, M0 being a constant independent of e for 0 < e < 1. Then, it follows from integration by parts that for any fixed T > 0 the solution couples {uE, ve) are bounded in the Sobolev space of order one H1 (fi x (0, T)), independently of e for 0 < e < 1. A compactness argument yields that they lie in a relatively compact subset of L2(0, x (0,T)). Hence there exists a sequence [uEi,vEi) converging to some limit (u, v) strongly in L2(Q x (0, T)) and weakly in H1 (fi x (0, T)) as e' —> 0. Next it is well known that for any function
397 A uniqueness argument insures that this subsequential convergence is indeed convergence.
• We point out that computation of the homogenized diffusion operators is complicated. However an algorithm using asymptotic expansions appears in [2] and [15], see also [9], [10]. One can also establish the following result from the methods of Theorem 3.2. Theorem 4.2 Let (w, z) be a solution pair to the homogenized system (15), (16) and (5). Then l i m ^ o I K . , * ) ^ = 0 andlim^oo |k(.,i)|| 00] n = c0.
References [1] N. Alikakos, An application of the invariance principle to reaction diffusion equations. J. Diff. Eq., 33 (1979), 201-255. [2] A. Bensoussan, J.L. Lions and G. Papanicolaou, Asymptotic Analysis for Periodic Structures. North Holland, Amsterdam, 1978. [3] S. Busenberg and K. C. Cooke, Vertically transmitted diseases. Springer-Verlag, New York, 1993. [4] W. Fitzgibbon, M. Langlais, M. Parrot and G. Webb, A diffusive system with age dependence modeling FIV. Nonlinear Analysis T.M.A., 25 (1995), 975-989. [5] W. Fitzgibbon, S. Hollis and J. Morgan., Steady state solutions for balanced reaction diffusion systems on heterogeneous domains. J. Diff. Int. Eq., to appear. [6] W. Fitzgibbon and J. Morgan, Diffractive diffusion systems with locally defined reactions. In : Evolution Equations, J.J. Goldstein et al. eds, M. Dekker, New York, 1994, 177-186. [7] W. Fitzgibbon and J. Morgan, Analysis of a two compartment model reaction diffusion model on a heterogeneous domain. In : Mathematical Models in Medical and Health Sciences, M. Horn, G. Simmonett and G. Webb eds; Vanderbilt University Press, 139-144. [8] W. Fitzgibbon, C. Martin and J. Morgan, A diffusive epidemic model with criss cross dynamics. J. Math. Anal. Appl., 184 (1994), 399-414. [9] W. Fitzgibbon, M. Langlais, and J.Morgan, Martin's Problem for Systems with Compartmental Diffusion. J. Egyptian Mathematical Society, 9 (2001), 59-67. [10] W. Fitzgibbon, M. Langlais, and J.Morgan, A Mathematical Model of the Spread of Feline Leukemia Virus (FeLV) Through a Highly Heterogeneous Spatial Domain. SIAM J. Math. Anal., 33 (2001), 570-588. [11] A. Haraux and M. Kirane, Estimations C 1 pour des problemes paraboliques semilineaires. Ann. Fac. Sci. Toulouse, 5 (1983), 265-280.
398 [12] F. Heiser, Contribution a I'analyse mathematique de deux systemes ecologiques en environnements heterogenes. Dissertation, University of Bordeaux I, 1998. [13] D. Henry, Geometric theory of parabolic equations, Springer-Verlag, Berlin, 1981. [14] W. Horton, Global existence of solutions to reaction diffusion systems heterogeneous domains. Dissertation, Texas A & M University, 1998. [15] V.V. Jikov, S.M. Kozlov and O.A. Oleinik, Homogenization of Differential Operators and Integral Functionals. Springer Verlag, Berlin, 1994. [16] O. Ladyzhenskaya, V. Rivkind and N. Ural'ceva, On the classical solvability of diffraction problems for equations of elliptic and parabolic type. Translation AMS, 23, Providence, Rhode Island, 1968. [17] N.G. Meyers, An Lp estimate for the gradient of solutions of second order elliptic divergence equations. Ann. Sc. Nom. Sup. Pisa, 17 (1963), 189-206. [18] J.D. Murray, Mathematical Biology. Springer-Verlag, Berlin, 1989. [19] A. Pazy, Semigroups of linear operators and applications to partial differential equations,. Springer-Verlag, Berlin, 1983. [20] Z. Seftel, Estimates in Lq of solutions of elliptic equations with discontinuous coefficients and satisfying general boundary conditions and conjugacy conditions. Soviet Math, 4 (1963), 321-324. [21] J. Smoller, Shock waves and reation diffusion equations, Springer-Verlag, Berlin, 1983. [22] H. Stewart, Generation of analytic semigroups by strongly elliptic operators under general boundary conditions. Trans. AMS, 259 (1980), 299-310. [23] H. Stewart, Spectral theory of heterogeneous diffusion systems. J. Math. Anal. Appl., 54 (1976), 59-7. [24] H. Stewart, Generation of analytic semigroups by strongly elliptic operators. Trans. A.M.S., 199 (1974), 141-162.
Algebraic Multigrid for Selected PDE Systems T. Fiillenbach, K. Stiiben Praunhofer Institute for Algorithms and Scientific Computing (SCAI) Schloss Birlinghoven, D-53754 Sankt Augustin, Germany Email: [email protected], klaus.stueben® scai.fraunhofer.de Abstract In this paper, strategies for solving systems of partial differential equations by algebraic multigrid are discussed. In particular, a general framework for so-called point-based strategies is introduced. For a demonstration, we have investigated several industrial applications from semiconductor process and device simulation. It is shown that this framework allows to construct robust and fast algebraic multigrid approaches even for cases, where iterative solvers of the type commonly used in such applications exhibit bad convergence or even fail.
1
Introduction
Classical algebraic multigrid (AMG) [1, 3] is known to provide very efficient and robust solvers or preconditioners for large classes of matrix problems,
Au = f, an important one being the class of (sparse) linear systems with matrices A which are "close" to being M-matrices. Problems like this widely occur in connection with discretized scalar elliptic partial differential equations (PDEs). In such cases, classical AMG is very mature and can handle millions of variables much more efficiently t h a n any onelevel method. Since explicit information on the geometry (such as grid data) is not needed, AMG is especially suited for unstructured grids both in 2D and 3D. In fact, the coarsening process is directly based on the connectivity pattern reflected by the matrix A and interpolation is constructed based on the matrix entries. However, extensions of classical AMG are required to efficiently solve systems of P D E s involving two or more scalar functions (called unknowns in the following). This is because classical AMG realizes a variable-based approach which does not distinguish between different unknowns. Unless the coupling between different unknowns is very week, such an approach cannot work efficiently for systems of PDEs where, in general, the corresponding matrix A is far from being an M-matrix. In the past, several ways to generalize AMG have been investigated and there is still an ongoing rapid development of new AMG and AMG-like approaches. For a review, we
399
400 refer to [2]. However, there is no unique and best approach yet. In fact, none of the known approaches is really satisfactory in dealing with practically relevant problems and many problems cannot be tackled at all yet. All methods seem to have their range of applicability but all of them may fail to be efficient in certain other applications. In this paper, the focus is on extensions of AMG which are direct generalizations of the classical approach. We first want to recall a rather popular AMG approach to solve systems of PDEs, the so-called unknown-based approach, which is very similar to the variable-based approach except that all unknowns are treated separately. To be more specific, let us assume the variables to be ordered by unknowns, that is, Au = f has the form •A[l,l]
•' "
-^[l,nsys]
M[l]
A[nsys,l]
' ' *
™[nsys,nsys]
^[nsys]
(1) J[nsys]
where nsys > 1 denotes the number of unknowns of the given system of PDEs, u[„] denotes the vector of variables corresponding to the n-th unknown and the matrices A[n,m] reflect the couplings between the n-th and the m-th unknown. Using this notation, coarsening the set of variables corresponding to the n-th unknown is strictly based on the connectivity structure reflected by the submatrix A[n^ and interpolation is based on the corresponding matrix entries. In particular, interpolation to any variable i involves only coarse-level variables corresponding to the same unknown as i. The Galerkin matrices, however, are usually computed w.r.t. all unknowns. The unknown-based approach, which has been proposed already in the very early papers on AMG (see [1]), is certainly the simplest approach for solving PDE systems. By now a lot of experience has been gained with this approach which, in practice, works quite efficiently for many applications. Compared to the variable-based approach, the only additional information required is information about the correspondence between variables and unknowns. The unknown-based approach is mainly used for applications where the diagonal matrix blocks A[n,n] are close to being M-matrices. The essential additional condition for the approach to work is that smoothing the individual equations is sufficient to cause the resulting error to be smooth separately for each unknown. One advantage of this approach is that it can easily cope with anisotropies which are different between the different unknowns. Another advantage is that unknowns can virtually be distributed arbitrarily across mesh points. However, this approach will become inefficient, for instance, if the coupling between different unknowns is too strong. In this paper, we focus on applications for which the unknown-based approach does not work, unless we introduce very special modifications. In particular, we consider reactiondiffusion equations from semiconductor process simulation which lead to matrices A for which the submatrices A[n>n] are far from being M-matrices. In fact, off-diagonal entries may be larger than the diagonal entry by orders of magnitude. Hence, the size of matrix entries is no measure any more to decide about the strength of connectivity in the AMG context. In Section II, we outline a flexible framework for constructing new AMG approaches to solve various types of PDE systems. In contrast to the previous approach, all of the new
401 ones operate on the level of grid points rather than variables. Based on this framework, our AMG code "RAMG", described in detail in [3], has been substantially generalized to provide more flexibility in solving PDE systems. Recent results for industrial applications in semiconductor process and device simulation, obtained by the generalized code (called "SAMG"), are presented in Section III, showing that suitable point-based AMG approaches yield efficient solution processes.
2
A general framework for point-based approaches
We talk about a point-based approach if, geometrically speaking, coarsening takes place on the level of points (rather than variables as before) and all unknowns are defined on the same hierarchy. Note that this is different from the unknown-based approach where each unknown is associated with its own hierarchy. Since we have the solution of PDEs in mind, we think of points as being real physical (grid) points in space. However, we want to point out that, from AMG's point of view, it is not important whether "points" really correspond to physical points. Instead, one may think of the nodes of a graph representing the connectivity structure of A. Regarding a point-based approach, it is only relevant for AMG to know whether there are "blocks" of variables (corresponding to different unknowns) which may be treated (coarsened and interpolated) simultaneously. We assume that corresponding information is available to AMG. In all our point-based approaches the coarsening process is performed based on some auxiliary (sparse) (npnts x npnts)-ma,tnx P = (pu), called the primary matrix, with npnts denoting the number of points. The same coarse levels are then assigned to all unknowns. For this process to make sense, the employed primary matrix should reflect the physical connectivity (the general structure as well as the strength of connections) of neighboring variables reasonably well, simultaneously for all unknowns. A special point-based approach, sometimes called "block approach", has already been introduced in the very early paper [1] and has been further investigated, for instance, in [4]. To be more specific, we assume the variables to be ordered pointwise, that is, Au = / has the form A1.1)
'''
A(l,npnts)
^(npnts,l)
'''
^(npnts,npnts)
W(l)
/(i)
^•(npnts)
hnpnts)
(2) where u^) denotes the "block" of variables located at point k and the (nsys x nsys)-matrix A(k,l) represents the "block coupling" between u^) and uy). Block coarsening corresponds to defining the primary matrix P by Pkl = -P(M)II
ik / 0
and
Pkk = -^2Pkl
(3)
with ||.|| denoting a suitable norm. Various different norms have been considered in practice.
402 Depending on the type of application, there are many other possibilities for defining a primary matrix. Often, this can be done automatically as part of AMG's setup phase. In other cases, it may be better to let the user of AMG provide a reasonable matrix himself, based on his knowledge of the underlying physics of the given problem. In such cases, a primary matrix can usually be interpreted as describing the connectivity structure of some auxiliary (scalar) primary unknown. Clearly, this primary unknown should represent the connectivity structure of all "real" unknowns in the given system of PDEs reasonably well. For instance, in simple cases, one may select P = Ayn^ with n being any of the unknowns of the given system of PDEs. Whether or not this makes sense, depends on the application, in particular, whether the connectivity structure of the n-th unknown is also representative for the other unknowns. If anisotropies in a given problem are mainly due to non-uniform mesh spacings, a simple primary matrix might be given by a discretization of the Laplace operator. One can also imagine cases where it makes sense to define a primary matrix based on some natural physical quantity for which there is no reasonable equation contained in the original system of PDEs, an example being the pressure in the context of the Navier-Stokes equations. The original AMG did not exploit any information on the given problem apart from the matrix A itself. In many PDE applications, this unnecessarily limits the possibilities for an efficient coarsening and interpolation. As a matter of fact, geometric information such as the coordinates of grid points, is usually available and can be exploited in AMG's setup phase. Note that this does not restrict the generality of the grid shape in any respect. If we assume coordinates to be known, P may often most easily (and automatically) be defined based on distances of points, leading to coarsening processes which are closely related to geometric coarsening. The most simple definition would be Pki = -1/5*1
(k =£ I) and pkk = - ^Pu
(4)
where 8kt denotes the distance between points k and I. (Clearly, since P has to be sparse, only points in small neighborhoods - corresponding to the non-zero pattern of A - are taken into account here.) R e m a r k : In practice, often, not all unknowns are represented at a point, that is, the number of variables may vary from point to point (cf. the applications discussed in Section III). If this is the case, a reasonable primary matrix is required to "represent" all points. The main purpose of a primary matrix is to define an AMG hierarchy in terms of points. As part of the (recursive) coarsening process, also interpolation operators have to be constructed. In practice, there are various possibilities to generalize the interpolation approaches used in classical AMG. First, the use of block interpolation seems most natural, in particular, if P is defined according to (3). That is, a formula to interpolate the error e^) at a point k is constructed by approximating the block equations e
(k) = Aik\) Y, A(k,mi) l£k
(5)
403 in a way which is completely analogous to the classical approaches to define interpolation. However, this type of interpolation is very expensive. In practice, simpler types of interpolation often lead to more efficient AMG processes. Thus, besides the above block interpolation, we consider (variable-wise defined) interpolation formulas which are either • separate for each unknown, or • the same for each unknown. This classifies interpolation just according to its general structure. Typical ways to define the interpolation weights are based on entries in the original matrix A, based on distances and/or positions of points, or based on entries in the primary matrix P. We cannot go into further details here but just want to mention that "classical" interpolation schemes, as described in [3], such as direct, standard or multi-pass interpolation, can be generalized to this setting in a straightforward way. Also the concept of aggressive coarsening carries over. The general framework outlined above formally allows to define various concrete algorithms. It seems clear that there exists no unique AMG procedure which will work satisfactorily for all systems of PDEs. Instead, major work consists in developing concrete algorithms separately for certain classes of industrial applications.
3
Applications
Due to the complexity of the models and grids used, industrial semiconductor process and device simulation is increasingly recognized as an important and challenging area for numerical simulation. Corresponding PDE systems include stress governing, reactiondiffusion and drift-diffusion equations, all of which exhibit different numerical difficulties. That simple unknown-based AMG is suitable to speed up stress simulations has already been shown in [5]. For reaction-diffusion and drift-diffusion equations, the situation is considerably more complicated. Where classical iterative solvers often converge only slowly (or even break down) and straightforward unknown-based AMG is not sufficient any more, suitable point-based AMG approaches, accelerated by BiCGstab or GMRes, can still cause remarkable speedups. In the following, we present some typical examples. We will see that reaction-diffusion problems can efficiently be solved by using a primary matrix based on geometric distances (4) and an interpolation which is separate for each unknown with weights being also defined based on distances. Drift-diffusion problems, on the other hand, are solved efficiently by selecting the primary matrix based on norms (3) and choosing interpolation to be the same for each unknown with weights being based on the entries of P. 3.1. Process simulation: reaction-diffusion equations Systems of reaction-diffusion equations occur, for instance, in the simulation of annealing steps after ion implantation into a wafer. They consist of a sequence of balance equations of the form ([6, 7]) c)n •
-^-rV-J^Ri
(t = l
JV)
(6)
404 where the J; denote (diffusion and field driven) fluxes given by
Ji = -Di{Vui + y ^ ^ A •
(?)
Here Ui denotes the concentration of the i-th species, TV their number, i\> the electrostatic potential, Rt = Ri{u\,..., u/f) a reaction term (often a polynomial), Dt the diffusivity, q the elementary charge, ks the Boltzmann constant and T the absolute temperature. ILUT<5)
/-v
[V \AMG, std.coars. iGS smoother AMG, aggr.coars LIJ smoother
M
AMG, std.coars. ILU smoother J
' '
•' * • '
20 ^ 30 iter
40
50
Figure 1: Convergence histories for a 3D reaction-diffusion example. All AMG and ILUT variants are accelerated by BiCGstab ("aggr." or "std.coars.": aggressive or standard coarsening, resp.; GS: Gauss-Seidel). After inserting (7) into (6) for each i, the above system consists of typically 30 to 40 equations. By employing some equilibrium assumptions, a reduction to a system of 3 to 6 equations of a similar form as (6) can be performed, with unknowns being the concentration of species such as interstitials, vacancies, Arsenic, Boron, Phosphorus or others. For the potential ip an additional Poisson's equation can be solved, which could be coupled to the above system. We have only investigated the typical, uncoupled case. In common process simulators, an implicit approach is chosen for the time discretization. The spatial discretization is performed by the so called "box method" on Delaunay grids, and the resulting nonlinear system is linearized by a modified Newton-Raphson method. ILU-preconditioned BiCGstab or GMRes are commonly used as solvers for the resulting linear systems. More precisely, modified ILUT [8] or even ILU(0) methods are employed. Of particular interest are the concentration profiles in and near the reaction front, a narrow region, moving from the "implantation" surface of the wafer towards the interior,
405 where fast reactions occur due to large concentration gradients (see for example [9]). In this region, the reaction terms cause very large positive or negative off-diagonal entries in the corresponding rows of the matrices A, leading to serious problems for the standard iterative solvers mentioned above. It can be observed that these solvers are getting less efficient or even stagnate in an unpredictable way. The difficulties often increase during later time steps of a simulation. As an example, Fig. 1 depicts the convergence histories for several ILUT-variants used as preconditioners for BiCGstab. The solvers are applied to a typical system, arising in a 3D-simulation in a particular step within Newton's iteration. Because of the very large off-diagonal elements, not only the full matrices A, but also their submatrices A[ni„] are far from being M-matrices. Therefore a straightforward unknown-based AMG method does not make sense here (see, however, the remark below). As can be seen in Fig. 2, the underlying grids are adaptively refined in the reaction front. Hence, it seems promising to define a point-based AMG method with a primary matrix based on distances and a separate interpolation with weights also based on distances. This combination yields an AMG method which treats both the "pure" diffusion outside the reaction front and the fast reactions inside properly (see Fig. 3). As can be seen from Fig. 1, three AMG variants (with standard or aggressive coarsening, GaussSeidel- or ILU-smoother), accelerated by BiCGstab, all converge much faster than the ILUT-preconditioned ones. Moreover, this behavior is typical for a whole simulation: AMG yields stable and superior convergence rates for residuals and errors in all Newton iterations for each time step, which makes it more robust and efficient than the ILUTpreconditioners. Remark: As mentioned above, due to the large off-diagonals arising from fast reactions, a straightforward unknown-based approach does not work. However, while these reactions take place only inside a narrow area, outside a relatively 'harmless' diffusion problem remains to be solved. Numerical experiments have shown that, by introducing a very simple modification of AMG's coarsening process, one can make the unknown-based AMG work again: Simply do not coarsen at all inside the narrow reaction area. To demonstrate this, we have forced all those variables i to stay in the coarse levels, whose corresponding rows strongly violate diagonal dominance,
53|tJy| > « M •
(8)
Depending on the threshold parameter e > 1, the resulting unknown-based AMG (employed as a preconditioner) converges similar to the point-based approach described before. However, the choice of e is crucial and example-dependent. Clearly, the above criterion is not always suitable to reliably distinguish the reaction-dominant from the slow-diffusion part, and therefore the resulting approach is not robust enough for a use in practice. 3.2. Device simulation: drift-diffusion equations The second class of examples is originating from device simulation, where for instance the behavior of transistors is simulated. The simulation domain usually consists of two parts, E and Q. The subdomain E usually represents the semiconductor region(s) (doped
406
Figure 2: 3D reaction-diffusion example: (part of a) 2D cross-section of the grid which is adaptively refined near an edge of the interface oxide/wafer.
Figure 3: Coarse levels for the grid on the left: the larger the box, the longer the corresponding point stays in the coarse levels.
silicon, the wafer), in which typically three coupled equations, a Poisson-like equation and the electron and hole continuity equations, are solved for the electrostatic potential i/> and the electron and hole carrier concentrations n and p, respectively: -V • (esVip) + q(n-pdn
„
N) = 0 , „
T
q-^-V-Jn-qRn dp _ q
!Ft
+
" ~qRp
(9)
= 0,
(10)
=
(11)
°'
The parts Jn and Jp of the (diffusion and electric field driven) flux J = Jn + Jp are given by Jn = —qiinnVil> + qDnVn , Jp = -q^pPVip - qDpVp .
(12) (13)
es is the dielectric constant of the semiconductor, q the elementary charge, JV = N(x) the net impurity concentration, R„ = Rn{ip,n,p) and Rp = Rp{%j),n,p) the recombinationgeneration terms, fi„ = jj,n{x,Vtp,...) > 0 and (J,P = (iv{x,Vip,..) > 0 the mobilities, Dn = ^^Hn > 0 and Dp = ^-(J,p > 0 the diffusivities, ks the Boltzmann constant and T the device temperature. All of them are given (functions). The second subdomain 17, consisting of at least one region (usually an oxide), is treated as an insulator, so that (nearly) no charge carrier currents can occur. Therefore the above PDE system degenerates to Laplace's equation: -V • (e0VV) = 0 ,
(14)
where to represents the dielectric constants of the corresponding material layers in Q.
407 A detailed discussion of the properties of such systems can be found in [10, 11, 12], for instance. In the following, we concentrate on the steady-state simulation, where the time-derivatives ^ and | | are vanishing. In practice, the system is discretized by a box method, employing special analytical ID solutions along mesh edges (Scharfetter-Gummel approach [11]). The resulting system is linearized by a (modified) Newton method and usually solved by ILU-preconditioned BiCGstab. More precisely, some modified ILU(O)approach is used. For each of the following four different types of devices we have considered one example: • a shallow trench isolated transistor (STI) [13], • an electrically erasable programmable read-only memory cell (EEPROM) [13], • a FinFET (a double-gate MOSFET structure in which a thin, fin-shaped body is straddled by the gate forming two self-aligned channels that run along the sides of the fin) [14], • a power bipolar transistor (PBT) [13]. Details on these problems can be found in Table 1. Layouts of the STI, the FinFET and the PBT example are shown in Fig. 4, 5, and 6, respectively. For each example, a whole simulation series was run (using the commercial device simulator TAURUS by Avant! Corporation) by applying increasing voltages. All matrix equations, produced during the Newton iterations and for each voltage step, were used to test the performance of AMG. Example STI EEPROM FinFET PBT
dim 2D 3D 3D 3D
ns no np 1 7 5516 1 9 10493 2 5 27173 1 3 76714
nv na 9212 125620 15415 310361 42489 987123 149100 2908954
Table 1: Details on the four examples (dim: spatial dimension, np: number of points, ns: number of regions in E, no: number of regions in Q, nv: number of variables, na: number of non-zero matrix elements). Because of the tight coupling between the different PDEs in E, it turns out that unknown-based AMG fails for such applications. Instead, we have chosen a point-based AMG, in which the primary matrix P is defined by norms (3), and the same interpolation is used for all unknowns with weights being based on the entries of P. In Table 2, we compare the convergence behavior of this AMG approach with that of a standard ILU method of the type commonly used in device simulations (both accelerated by BiCGstab). The results represent the typical behavior of the solvers during a whole simulation run. If two exemplary results for a device (FinFET and PBT) are given, they represent the 'best' and the 'worst' case regarding ILU, the first of which occurs usually at an early stage of the simulation, the second near the end. For the 2D simulations, AMG with standard coarsening gives the best convergence rates and fastest timings. Compared to ILU, AMG yields a speedup of approximately 1.5 here. In the 3D cases, it is more favorable to use AMG with aggressive coarsening.
408
Figure 4: STI example: layout and doping profile. Courtesy of Avant! Corp.
Figure 5: FinFET example: layout and doping profile. Courtesy of Avant! Corp.
Figure 6: PBT example: layout and doping profile. Courtesy of Avant! Corp. This reduces memory requirements, which is particularly important in 3D where relevant problems are much larger than in 2D. As can be seen in Table 2, for the largest two examples, ILU is inefficient already in the 'best' case, and in the 'worst' case it virtually stagnates (PBT) or even fails (FinFET). In contrast to this, AMG exhibits a stable convergence behavior in all cases and is always faster than ILU.
4
Conclusions
AMG approaches for solving systems of PDEs were presented and discussed. Especially a general framework for point-based approaches was introduced, which employs a primary matrix to construct a point-based coarsening. Several possibilities for selecting a primary matrix and for the computation of the final interpolation weights were outlined. Recent results for applications in semiconductor process and device simulation were presented, which demonstrate that robust and fast point-based AMG methods can be obtained using this framework.
409 Example STI2D
Preconditioner Cycles AMG 13 44 ILU EEPROM aggr-AMG 12 22 ILU FinFET 14 aggr-AMG ILU (50)1 FinFET aggr-AMG 15 2 ILU PBT aggr-AMG 9 (50)3 ILU PBT aggr-AMG 16 ILU (50)4 1 residual reduction by 10~3 only! 2 method diverges! 3 residual reduction by 10 -4 only! 4 magnitude of residual not reduced!
ARF 0.273 0.670 0.303 0.518 0.303 0.894 0.365 1.05 0.202 0.811 0.407 0.991
Time 9.95 15.74 14.32 17.27 87.58 (111.62)1 136.19 2
203.66 (462.27)3 269.12 (582.80)4
Table 2: Number of cycles, average reduction factors (ARFs) and wall-clock timings (in seconds) for different examples. Accelerator always BiCGstab. Residual reduction by 10~6, except stated otherwise, "aggr" means aggressive coarsening.
References [1] J.W. Ruge and K. Stiiben, Algebraic Multigrid (AMG), in "Multigrid Methods", S.F. McCormick (ed.), Frontiers in Applied Mathematics, Vol. 3, pp. 73-130, SIAM, Philadelphia, 1987. [2] K. Stiiben, A Review of Algebraic Multigrid, Journal of Computational and Applied Mathematics 128, pp. 281-309, 2001. Also available as GMD Report 69, November 1999. [3] K. Stiiben, An Introduction to Algebraic Multigrid, in "Multigrid" by U. Trottenberg, C.W. Oosterlee and A. Schiiller, Academic Press, London, San Diego, pp. 413-532, 2001. Also available as GMD Report 70, November 1999. [4] D. Oehlz, Algebraische Mehrgittermethoden fur Systeme partieller Differentialgleichungen, Diplomarbeit, Rheinische Friedrich-Wilhelms-Universitat, Bonn, April 2001. [5] T. Fullenbach, K. Stiiben and S. Mijalkovic, Application of an algebraic multigrid solver to process simulation problems, Proceedings 2000 International Conference on Simulation of Semiconductor Processes and Devices, IEEE Catalog Number: 00TH8502, ISBN 0-7803-6279-9, pp. 225-228, 2000.
410
[6] A. Hofler and N. Strecker, On the Coupled Diffusion of Dopants and Silicon Point Defects, Technical Report 94/11, Integrated Systems Laboratory, Swiss. Fed. Inst, of Technology (ETH), Zurich, 1994. [7] A. Hofler, Development and Application of a Model Hierarchy for Silicon Process Simulation, PhD thesis, Swiss. Fed. Inst, of Technology (ETH), Zurich, 1997. Also available as Series in Microelectronics, vol. 69, Hartung-Gorre, Konstanz, 1997. [8] Y. Saad, ILUT: A dual threshold incomplete ILU factorization, Numerical Linear Algebra with Applications 1, pp. 387-402, 1994. [9] A. Pomp, O. Schenk and W. Fichtner, An ILU Preconditioner Adapted to Diffusion Processes in Semiconductors, Technical Report 99/11, Integrated Systems Laboratory, Swiss. Fed. Inst, of Technology (ETH), Zurich, 1999. [10] S. Selberherr, Analysis and Simulation of Semiconductor Devices, Springer, Wien, New York, 1984. [11] P.A. Markowich, The Stationary Semiconductor Device Equations, Springer, Wien, New York, 1986. [12] P.A. Markowich, C.A. Ringhofer and C. Schmeiser, Semiconductor Springer, Wien, New York, 1990.
Equations,
[13] S.M.. Sze (ed.), Modern Semiconductor Device Physics, Wiley, New York, Chichester, 1998. [14] X. Huang et al., Sub-50 nm P-Channel FinFET, IEEE Transactions on Electron Devices 48, pp. 880-886, 2001.
Numerical Computation of Electromagnetic Guided Waves in a General Perturbed Stratified Medium D. Gomez Pedreira, Dpto. de Matematica Aplicada, Univ. de Santiago de Compostela, Spain Email: [email protected]
1
Introduction
The computation of electromagnetic waves propagating in perturbed stratified waveguides is important both for the theoretical analysis and many applications in physics and engineering, like integrated optics. An open stratified waveguide is a structure composed by parallel layers of dielectric materials whose electromagnetic behaviour is characterized by a physical property called the refraction index n. The goal in the study is to compute the frequencies for which harmonic electromagnetic waves of finite transverse energy can propagate without attenuation along the waveguide. These waves are called guided waves (or guided modes) and they are particular solutions of the Maxwell's equations in the form E(x,x3,t) H(x,x3,t)
= E(x)ei^t-^\ = B.(x)e^ut-px3\
^'
where u > 0 is the angular frequency of the wave, ft > 0 is the wavenumber (or propagation constant) of the mode, and E(x) = {Ei(x), E2{x), E3(x)) and ffl(x) = (Hi(x), H2(x), H3(x)) are vector functions of x = (11,2:2) describing the distribution of the electromagnetic field in each cross section. Their main property is that their amplitude has finite transverse energy f (|E|2 + |M|2)da; < 00.
(2)
In fact, it is this condition which determines if a mode is guided or not and physically means that the energy of the mode remains practically confined in some bounded region of the cross section. We will assume that the guide is invariant, with respect to both the geometry and its physical characteristics, under any translation in the propagation space direction x3. Thus the refraction index will depend only on the two transverse coordinates (xi, x2). The existence of the guided modes will depend on the refraction index distribution in the cross section of the guide. If the materials composing the different layers are chosen with a suitable refraction index, then the energy of the wave is vertically confined in the
411
412 material with the largest index. Nevertheless, to confine the light laterally —which is essential for the design of these devices— it is necessary to modify the refraction index in a bounded region, here denoted by K, in such a way that if the refraction index in this region is properly chosen, the energy of the wave remains practically confined in a neighborhood of K.. The dimensions of K, are so small in comparison with the stratified medium that, from the mathematical point of view, the cross section of the guide is considered as an unbounded domain. This is why this kind of guides are called open waveguides. We will assume that n{x) = n(x2)
Vi = ( x 1 , i 2 ) ^ K c l 2 ,
where n £ L°°(R) is a function only of the vertical variable which represents the refraction index of the so called reference medium, which is nothing but the stratified medium without the perturbation K. Mathematically, the problem to be solved can be reduced, under the assumption of weak guidance (i.e. large wavenumber and weak variations between the maximun and the minimun values of the refraction index, see Bonnet[l], Vassallo[2] ), to Find w > 0, /3 > 0 and u € L2(R2) ( t i / 0) such that (V)
- A u + /32 u = w2 n2 u, where u(x) = Ei(x), E2{x), Hi(x)
or H2(x).
For P > 0 given, (V) is a scalar eigenvalue problem set in the cross section of the waveguide, where w2 plays the role of eigenvalue of the selfadjoint operator Ag = l/n 2 (—A + /32) of L2(R2) and u, which denotes any of the transverse components of the electric or magnetic field, is the associated eigenvector. We will be interested in the computation of guided modes associated to eigenvalues J1 in the discrete spectrum of Ag, assuming they exist (see Bonnet et al. [3] or G6mez[4] for existence conditions). These are isolated eigenvalues of finite multiplicity which satisfy (cf. Gomez and Joly[5]) P2/n2+ < u? < <je(/3),
with
n+ = sup n(x), xeK2
(3)
and <7e(/3) denotes the greatest lower bound of the essential spectrum of the operator Ag. In the sequel, we shall denote E = {(w,
ffl6R2/U>0,^>0,
H2ln\ <w2< aM
}.
(4)
Remark 1.1 Let us point out that the guided modes, even when they exist, do not exist for just any values of w and j5 but for u and/3 linked by a certain relation UJ = f(/3) called the dispersion relation of the mode. From the numerical point of view, the main difficulties in solving the problem (V) come from the unboundedness and the stratified nature of the domain, which made the numerical methods proposed by other authors to be either non applicable , either "non
413
Figure 1: Sketch of an open stratified waveguide and the model cross section exact" or very costly from the numerical point of view (cf. Joly and Poirier[6] or Mahe
[7])The main idea of the method proposed in this work for the computation of the guided waves consists of a reformulation of the original problem (V) into a new one via the introduction of a pseudo-differential operator K depending on (u, j3) in such a way that solving the problem (V) is reduced to finding the values of (w,/3) for which —1 is an eigenvalue of K(LO, /3). The efficiency of the method relies in an efficient way to compute numerically this operator. The practical evaluation of K(o;, j3) is obtained by a combination on analytical computations —which take into account the unboundedness of the propagation medium— and numerical approximations involving: a domain truncation parameter R, a series truncation parameter N and a space step h for the finite element approximation of an auxiliary problem localized in a neighborhood of K,. The method combines three different techniques: Fourier transform, Fourier series and mixed finite elements. A particular case of this method was introduced and analyzed in Gomez and Joly[5, 8] for the case of a medium composed of three homogeneous layers with a perturbation K, of the refraction index involving only the central one. In these papers, the authors have announced that such a method was generalizable to the case where the interior layer is itself a stratified medium (made up of parallel layers), where the perturbation is not necessarily embedded in one of them. The objective of this work is to present the generalization of the method introduced in Gomez and Joly[5].
2 2.1
The new formulation of the problem A first reduction to the boundary T
Let us going to consider a reference stratified medium composed of three different domains: two exterior homogeneous layers —denoted by fl+ and f2~ — separated by an interior medium Q,, = {(xi,X2) £ R 2 , 0 < i 2 < L} made up of q parallel homogeneous layers. For j = 1 , . . . , q , we denote hj-\ and hj the x2 coordinates corresponding to the lower and upper boundaries respectively of the j t h layer Oj (see Figure 1). Thus, the refraction
414 index of the reference medium is a piecewise constant function of x2 given by if x2 € £l~ = {(xux2) n(x2)
x2
e R2 , x2 < 0},
€ Qj = {(2:1,2:2) G R 2 , < / i j _ i < £ 2 < / i j } ,
,3
if
+
if i 2 G ( l + = { ( i i , i 2 ) e K ! , i 2 > I } ,
(5)
where (5), rij denotes the constant value in the j t h layer and we will assume, without loss of generality, that n+, > n^,. The first step of the method is to formulate a problem whose unknown ip = (ip+,
8Q-, 90+.
For (w, ft) £ E given, we introduce the operator S(ui, P) defined as follows (for the sake of simplicity, we omit to mention the appropriate functional framework and refer the reader to Gomez and Joly[5, 8] for more details): S(w,P)
du(ip)
A
du{ip)
'du(tp)~ 1
r+
(6)
dn
where u(
CPV)
-A u(
in Q = . on T.
\r,
(7)
and n denotes the outer unit normal vector, with respect to the domain Q;, on boundaries T+ orT". Moreover, [q]r = {[q]r+,[q]r-), [?]r+ = (9|nv)|r+ " (9|n+)lr+> Mr- = (9| n j )|r- ~ (9|n.-)lrThe idea is the following: take a function ip defined on T and solve the boundary value problem (V^), (which consists, in fact, of two decoupled problems, one outside the strip Qi and another one inside). By construction, the function u(tp) is continuous. In order that u{if) be a solution of the problem (V), it is enough to ensure the matching of the normal derivatives on the lines x2 = 0 and x2 = L. This means that the jump of the normal derivative of u(tp) across T, namely S(u},/3)ip, must be equal to 0. Then, problem (V) is equivalent to
CPs)
For p > 0, find w > 0 with (w, p) 6 E such that 0 is an eigenvalue of S(uj, /3).
For numerical purposes, we use a decomposition of S{ui,P) into three operators S(u, P) = Si{u, p) + Sp(w, P) - Se(cj, P) which corresponds to a decomposition of u as:
(8)
415 • u | n =uf,
with uf = u*(tp) the unique solution of (Fe}
f - A u ± + (p-n^rf)ut
= 0
I
in S£,
«± = v
on r,
• u| n . = Uj + Up, with Uj = Ui(
- A Mj + (fi2 - n 2 u2)ui = 0 M; = V
in n*, on T,
and up = up{ip) the unique solution, for certain values of a; and p, of —A Up + (P2-n2u2)up = {n2-n2)w2Ui in Qi: Vpl ^ \ uv = 0 on T. Then we define
„ , „, (dui \ du~ \ Se(u,P)
fdw
3M,|
\
(dun I duv I \ on |r+ on I Problems (,Pe) and (T'i) are coercive (see Gomez and Joly[5]) and hence uniquely solvable. In Gomez and Joly [5], it is also shown that problem (Vp) is well-posed if and only if u2 g Gi(P), where Gi(/8) is a finite (possibly empty) set of irregular frequencies (such as those appearing in integral equations) which can be treated and computed separately. Thus, if u)2 £ Gi(fi), Sp, and thus S, are well defined.
2.2
>
Introduction of t h e operator K(w,/3)
We refer to Gomez and Joly[9, 5] for various properties of Si{ui,P), Se{u),P) and Sp{ui,p). In particular, it is explained that S(w,P) is not very easy to handle numerically, because it has a continuous spectrum. That is why we introduce the operator K(w,/3) = ( S i - S e ) - 1 S p ,
(10)
2
which is well defined as a linear operator in L (D since Si — Se is an isomorphism from H ^ r ) into L 2 (r). It can also be proven that K(w, P) is a compact operator whose spectrum is purely discrete and consists of a countable infinite set of real eigenvalues admitting 0 as unique accumulation point (the proof is based on the fact that Sp is a compact selfadjoint operator in L 2 (D, see Gomez and Joly[9]). The relationship between S{CJ, P) and K(u, P) is simply (I denotes the identity operator) S(co,p) = (Si-Se){I
+ K(uJ,p)},
(11)
so it is obvious that the problem (Ps) is equivalent to the following For a given /3, find w > 0 (w2 0 Gj(/3)) with (w,0) e E,
rpK) '
such that - 1 is an eigenvalue of
K(LJ,P).
416
3
C o m p u t a t i o n of the operators Se, Si and Sp
The practical computation of the operator K(OJ,/3) involves the computation of the operators 5 e , Si and Sp. The main difference with the method introduced in Gomez and Joly[5] lies in the computation of the operators Si and Sp. Let us briefly explain in this section how these computations are done (see Gomez and Joly [10] for details). Exploiting the fact that both (Ve) and (Pi) are invariant under translation in the x\ direction, these problems can be solved by partial Fourier transform in xx variable. As a consequence, the operators Se and Si, and then operator (Si — S e )~\ can be analytically computed. Let us denote by k the dual variable of x\. By solving the problem (Pe) in the Fourier domain and applying the definition of the operator Se given in (9), it is easy to check that [Sefw^te] (*) = Me(u, P; k) ftk) where
A
J & = (fc2 + ^ 2 - n + ^ ) i / 2 i
and <
When solving the problem (Vi) we must take into account that the domain Q; is a stratified medium composed by q parallel layers. The equation (in the Fourier domain) -xi- + (fc2 + p2 - n}ui2) Ui = 0,
j = 1, • • • , q
to be satisfiyed in each layer must be related to those of the adjoining layers by the continuity of the solution ut and of its normal derivative. The computations lead to [SiK/3")?] (*) = M{(uj,P; k) ftk). where the entries of the matrix Mi(uj,P;k) associated to the operator Si in the Fourier domain are given by M}1{LJ, P; k) =
6 (e-^L(T12
+ Tn) + T22 + T21)/D
2
Ml (u,P;k)= -2^e~^L/D 21 M (uj,P;k) = -2^e~^L/D M22(w, P; k) = £, (T12 - T n - e - 2 ^ ( T 1 2 with D = T22-T21
2 L
+e-
^ (T12-Tn),
Tu))/D
417 and [T(n](,n=lj2 denotes the elements of the matrix T which results from the product of matrices Tq_! Tq_2 • • • 7\ given by -. ( ehj (fi+l-fj) (1 + JL.)
e-hj
fe+i+fj) (i _ Ji_\
TMatrices Tj, j = 1,... ,q — 1 are called transmission matrices since they derive from the application of the boundary conditions to be met at an interface between two layers. Remark 3.1 The fact that D ^ 0 for any k is not immediate but is a consequence of the fact that problem (Vi) is coercive. Contrary to the operators Se and Si, operator Sp cannot be analytically computed. The definition of this operator involves the resolution of the problem (Vp) which is a boundary value problem posed in the infinite strip 0*. Taking into account that the right hand side has compact support included in /C and the fact that the solution uv satisfies and homogeneous Dirichlet condition, this problem can be numerically handled by using a localized finite element method. This method will allow us to reduce the actual computations to a rectangle Of, surrounding the perturbation with the help of a transparent boundary condition. The method implies introducing two artificial vertical boundaries in the domain Qi: namely S + = {a+} x (0,L) and E~ = { a - } x (0,L), to split Cli into a bounded domain 0(, containing the perturbation and another one Qj \ 0(, where the refraction index is a piecewise constant function of x2 (see Fig. 2). The boundary conditions to be defined on E* involve the introduction of a Neumann to Dirichlet operator T(ui,/3) depending on w and j3 given by oo
[T(w, flip]{x2) = 5 3 C V fl wk(x2)
(13)
*=i
where • £fc(w, P) = (/32 + Ak(w))1''2, which is strictly positive as a consequence of the inequality (3). • Afc(ai) denotes the eigenvalues of the one-dimensional eigenvalue problem -w"(x 2 ) -n2(x2)u}2w(x2) = Aw(i 2 ), w(0) = w(L) = 0.
w(x2) i= 0, [
• {wk(u},x2)} denotes the set of associated eigenfuntions, which forms an hilbertian basis of L 2 (0,L). • tpk denotes the expansion coefficients of tp in the basis {w^w,:^)}By using the operator T(CJ, ft), we can write an equivalent (in a certain sense) formulation of the problem (Vp) which couples a variational formulation in the bounded domain Qj with a Fourier expansion outside. More precisely, the computation of up has to be done in two steps:
418 One computes up inside Clb by solving numerically the boundary value problem 2
2
2
2 \ 2 ti — _ u u + _!_ (^ fl 2 — —Aup —mn/., oj ) up =
2 2 2 In(n —— m2n \ 2,., ) aii 2 «*
3£\
re)
-T(u,,P)^=
in fi|,,
on rb = r n dnb,
u„ = o up
on E±,
where u denotes the outgoing unit normal vector, with respect to the domain Qb, on boundaries E + or S~. • Knowing up inside Qb, it is analytically computed in the exterior domain Qt \f2(, via the formulas oo
up(xux2)
= ^2u^wk{u},x2)e-ik{xi-a+)
if xi > o+,
t1
(15)
up{x1,x2) = ^2,ulwk{u,x2)e~ik{a
~x,)
if xi < a~,
where u£ and u^ denote the expansion coefficients in the basis {w^} of the trace on E+ and £~ of the solution up of (Vf). Once up has been computed, then it suffices to apply the definition (9) to compute Sp.
4
Numerical approximation of t h e problem (VK)
4.1 The truncation of T The first difficulty one has to face for solving problem (VK) lies in the fact that the unknown function tp is defined on a one dimensional but unbounded domain, namely the boundary T. To avoid this problem, Gomez and Joly[8] propose an approach based on a truncation method that can be applied for a multi-layered structure like the one considered in this work following an analogous procedure. The idea is working with functions defined on the bounded domain r R = T n {(xux2)
I a~ - R < xx < a+ + R},
where R > 0 is an approximation parameter which goes to +oo. Let us consider the orthogonal projector IIR on L 2 (r R ) defined by
n R : L2(r) —• L2(r) tp
—•
nR
(16)
where x R denotes the characteristic function of T R . The idea is to write an equation for TlRtp, the "restriction" of
419
~" R
I Q„ I
R
K
l!I _
S+
series expansion
series expansion
Figure 2: Numerical approximation Error estimates that can be proved thanks to regularity and decay properties of the solution Up of the interior problem (Vp) (cf. Gomez and Joly[8]), allow us to approximate the problem {PK) by {I + K R K = O, ^ e L 2 ( r „ ) ( (17) with
K R = nR(Si - 5 e )- 1 n R 5 p n R ,
(is)
This method has been proven to be exponentially accurate (cf. Gomez and Joly[8]). More precisely, it can be proven that the nonzero eigenvalues of KR converge exponentially to the nonzero eigenvalues of K as R goes to infinity. In particular, the eigenvalue —1 will be approximated with an exponential accuracy.
4.2
The series truncation
In order to solve numerically the problem {Vp) we must truncate both the series in (13) and in (15) at (large) rank JV. Then, the problem (Vp) is approximated by
(Vf)N {
-Att£ + (/?2 - n2w2)w£ = (n2 - n2)uj2 ut in Qb on Tb < = 0 9
T
"? -
u"
on E±
where TN denotes the operator obtained by the truncation at rank N of the series given by (13). Once time we have computed up in Qb, the solution up e fij \ Oj, is computed via a Fourier series development N
*=i J
up (xux2)
u w
(19) u}
x
e (k{a
= '^2 k k( > 2) ~
~~
Xl)
ifa:i
Taking into account the series truncation, the operator Sp is approximated by the operator
420
Sp given by
,s>,l
duN
i
^f):'
Finally, the problem (P^)N is numerically solved via a mixed-hybrid finite element method which involves the introduction of other approximation parameter h corresponding to the step-size of the mesh. This method is particularly suitable since it provides a direct way to approach the normal derivative of up (even if it obliges us to write the boundary condition on E in the way —Tdup/dv = uv). As a consequence of the numerical approximation, the operator K(u>,/3) is approximated by 1 N K& W (W, p) = (n R {Si - &)- n R )„ (n R s^ n R ) ft , (21) and finally the numerical problem to solve is KR
'
J
For a given /3, find w > 0 (w2 g Gi{P)), with (w, $) € E,
1
such that - 1 is an eigenvalue of K^N(u},/3).
Thus, the numerical algorithm is as follows: For a given p, we look for a guided mode u in an interval [wj,a;,] without singular frequencies. To do that, we compute the operators KR'w(a;j, P) and K^N (v,, 0) (which involves the resolution of problems (Vi) and then (Vp) for each basis function on TR) and their corresponding eigenvalues X^NR(uji, 0) and X^NR(UJS, 0). If (A^AfR(wi,0) + l)(A^ R (w s , 0) +1) < 0, then there exists u> € [ui,us] such that \^[NR(co,P) = — 1 and we compute it by using a fixed point procedure. Then we compute an associated eigenfunction ip and we solve problems (Pi) and (Pp) for the Dirichlet condition Ui = ip on T. This allows us to compute uf and finally to represent it on f2;,.
5
Numerical results
The first test corresponds to a step-index optical fiber whose core is a circle of radious 0.45 with refraction index n + = 1.7 and cladding index equal to n K = 1 (see Fig. 3). This device can be considered as an "artificial" stratified medium composed of three layers (with hi = 0.5, Yi2 = 0.75, h3 = 1, see Fig. 1) where the refraction index of all layers is constant and equal to 1. The interest of this test lies in the fact the dispersion relation between (OJ, 0) has an analytical formula in terms of Bessel functions 00
iC(0-45Q_ tfm (0.45/Co)
-4(0-45 K) Jm(0A5K)'
(22)
where Koo = A//3 2 — n^ui2
and
K = \Jn\u
By solving numerically equation (22) we can obtain some reference solutions. Thus, for instance, we obtain that for the value /? = 2 there exists a guided mode associated
421
:m?ty
Figure 3: Cross section for the optical fiber test Table 1: Eigenvalues of the operator K^,iv(1.8534,2). h = 1/33, N = 100. R=l
R=2
R=3
-0.9952 -0.9989 -0.9990
to u! = 1.8534. A first test was to compute the operator KR' (ui,/3) associated to this pair (w,/?) and to check that —1 is an associated eigenvalue. Tables 1, 2 and 3 report the eigenvalue of KR,JV(tj,/9) which is closest to —1 for different values of h, N and R in order to show the influence of these parameters in the numerical procedure. The abscises of the boundaries S + and S~ are located respectively at a + = 0.5 and a~ = —0.5. The second test corresponds to an integrated optics device whose cross section has been sketched in Fig. 4. We have considered that the interior layer fij is a three-layered medium (with boundaries hi = 0.25, h2 = 1.25, ^13 = 1.5) where the refraction index is the same in all layers and equal to 3.38. The refraction index profile is given by 1 3.17 3.44 3.38
n(xi,x2)
if x2 > 1.5, if x2 < 0, if (xi, x2) e (-1,1) x (0.25,1.25), otherwise.
l*»
l-k
mm . jn.=a.3a.
m&mmm^^ Figure 4: Cross section for the integrated optics test In this case, analytical solutions are no available and this is why the solutions have been compared with the ones obtained with the method proposed by Mahe[7]. Following
422 Table 2: Eigenvalues of the operator 1^(1.8534,2). h = 1/33, R = 3. N =3
N = 30
JV = 100
-0.9920
-0.9930
-0.9990
Table 3: Eigenvalues of the operator 1(^(1.8534, 2). TV = 100, R = 3. ft = 1/11 ft =1/33 -0.9985
-0.9990
Mahe's method, we can obtain that there exists a guided mode for the values 8 = 13.5 and w = 4. When computing the eigenvalues of the operator K^,Ar(4,13.5) for the values N = 100, h = 1/20 and R = 1 we obtain -1.000067 as eigenvalue closest to - 1 . The associated guided mode has been represented (restricted to domain fi(,) in Figure 5.
Figure 5: The guided mode for 8 = 13.5, w = 4.
References [1] A. S. Bonnet-Ben Dhia. Analyse mathematique de la propagation de modes guides dans les fibres optiques. Technical Report 229, Ecole Nationale Superieure de Techniques Avancees, 1989. [2] C. Vassallo. Theorie des guides d'ondes electromagnetiques. CNET-ENST, Paris, 1985.
423 [3] A. S. Bonnet-Ben Dhia, G. Caloz, and F. Mahe. Guided modes of integrated optical guides, a mathematical study. IMA Journal of Applied Mathematics, 60:225-261, 1998. [4] M. D. Gomez Pedreira. A numerical method for the computation of guided waves in integrated optics. PhD thesis, Universidad de Santiago de Compostela, 1999. [5] D. Gomez Pedreira and P. Joly. A method for computing guided waves in integrated optics. Part 1: Mathematical analysis. To appear in SIAM J. Numer. Anal. [6] P. Joly and C. Poirier. A numerical method for the computation of electromagnetic modes in optical fibres. Math. Meth in Appl. Sciences, 22:389-447, 1999. [7] F. Mahe. Etude Mathematique et numerique de la propagation d'ondes electromagnetiques dans les microguides de I'optique integree. PhD, Universite de Rennes I, 1993. [8] D. Gomez Pedreira and P. Joly. A method for computing guided waves in integrated optics. Part 2: Numerical approximation and error analysis. To appear in SIAM J. Numer. Anal. [9] D. Gomez Pedreira and P. Joly. Mathematical analysis of a method to compute guided waves in integrated optics. Technical Report 3933, Institut National de Recherche en Informatique et Automatique (I.N.R.I.A), May 2000. [10] D. Gomez Pedreira and P. Joly. In preparation.
A non-stationary model for catalytic converters with cylindrical geometry Jean-David HOERNEL Laboratoire de Mathematiques et Applications Universite de Haute-Alsace, 4 rue des Freres Lumiere F-68093 MULHOUSE Cedex FRANCE Email : [email protected] Abstract We prove some existence and uniqueness results and some qualitative properties for the solution of a system modelling the catalytic conversion in a cylinder. This model couples parabolic partial differential equations posed in a cylindrical domain and on its boundary.
1
Introduction
A gas containing N — 1 different chemical species is flowing through a cylindrical passage with a parabolic speed profile. Chemical species are diffusing in the cylinder and are reacting only on the boundary of the cylinder. We investigate t h e existence, uniqueness and qualitative properties of t h e solution of a non-stationary system of partial differential equations describing t h e evolution of the concentrations of the N — 1 chemical species and of the temperature, both with respect t o t h e time variable and along t h e cylinder. This model of catalytic converters starts with t h e contribution of Ryan, Becke and Zygourakis [5]. However, we have added an axial diffusion term on the boundary in the present model, see [1] for a more detailed description of this model. Due to its internal symmetry, the cylinder may be [0,1[ x ]0,1[, t h e boundary of which is S = {1} x ]0,1[. temperature) inside t h e cylinder fi are named C,f, i = concentrations (resp. the temperature) on the boundary T h e problem is written in a normalized way as t-x
2^ dC*f 3
I
-^{z,t)
*\
Pit
d
( 9Cif\
reduced t o the domain Q = The concentrations (resp. the 1,...,N — 1 (resp. CJV/)- T h e E are named Ci3 (resp. C/v s ).
^
= - 7 i . ^ ( l ) z , t ) + ^(C£ ) ... ) C+.)( Z ) t)
424
(1)
425 for i 6 { 1 , . . . , N} and with C£ (z,t) — sup (C is , 0). The initial and boundary conditions are :
<
Cit(r,0,t) Cif(l,z,t)
= =
«^(l,t)
dci}
Ci0(r), Cis(z,t),
9r
(0,z,t) = C*is(z,0) =
0, Cis0(z)
(2)
. ^ ( o , t ) = o.
= 0,
The functions r*, i e { 1 , . . . , JV}, are supposed to be Lipschitz continuous
\ri[Cis,•••,cNs)
-rt(cls,...,cNs)I
<
fcj2_^|chs-C;
and verify the following hypotheses : (HI) \fx e {R+f
: I, {xlt...
,xN) > 0.
(H2) If at least one of the xt, 1 < i < N, is equal to 0, then Tt(xi,... JV
(H3) For every (x,y) G (R+) -~^26i— i=l
, 0 , . . . , XN) = 0.
N
x (R+)
{rt(xi,...
,xN) -Ti(yu...
,yN))(xi~yi)
> 0.
7is
R e m a r k 1.1 We observe that : 1. Hypotheses (H2) and (H3) imply : N
Vz 6 (K+) N : - ^
0 8p*-r* (xu ... , xN) x{ > 0.
2. We have: Cif (1, z, t) = C is (z, t) ; C,, (r, 0, t) = Ci0 (r) ; Cis (z, 0) = Cis0 (z). In order to ensure the continuity of the concentrations in Q and on the boundary £ at z = 0, we must have at t = 0, C,/ (1,0,0) = CiS (0,0), which implies : Qo (1) = C is0 (0) ;Ci.(0,t) = C«,(l).
2
Existence of t h e solution
We establish the existence of the solution using the diagram :
r Cif
—>
Cif
Cis
Indeed, the proof of the existence is decomposed in two steps
426 1. Existence of a solution in £2 (given the solution on the boundary E) and existence of a solution on the boundary E (given the solution in SI) ; 2. F = \P o $ is a contraction in some appropriate functional space.
2.1
Preliminary results
We have the following continuous embeddings (see [3, p. 103]) L2 (Si) C L2 (SI) c L2r(1_r2) (SI) ; W1'2 (SI) C W?-2 (Si); W1'2 (Si) C < ( f_ r 2 ) (SI). Because -§^-(1, z,t) = j - J0 -gjf-r (I — r2) dr, we can write the problem as
ddif
(1 dCia
at
=
^LL^f
Q
' dz
PiS
d2Cis ' dz2
_lisy» f pad 2 £ n r (l-r3)dr + /?, W o dz
rdr
\
1
dr
(3)
to(C£,...,C+,),
with i g j l , . . . , N}, and the initial or boundary conditions Cif{r,0,t)
=
Cif(l,z,t)
=
^ ( M )
2.2
Ci0(r), Cis(z,t),
= 0,
dCif dr <°'*'*> Ci.(z,0)
o,
^(o,«)
0.
Ciso (z)
Existence in t h e cylinder
Assuming that the Cis are known on the boundary, and performing a change of function in order to have homogeneous boundary conditions at r = 1, we obtain the following problem in which we omitted the time variable t (l-r2)
'
<
OUf
d 1 duf \ ~dz" - * ; dr \ dr J uf(r,0) uf(l,z)
dz '
= u0(r), = 0,
(4)
7?
«. with i € {1,
dus
, N} and
w, = ' ( d , , . . • uf
=
Wf
,
Ctff),
—Us
Us
=
Pf
= diag(j3lf,---
(Cls,
• • • ,
CMS)
I
,(3Nf
We set
Wr =
Le(LU0,i)f\~e(L2r(QA))N},
W0T = l[u€(L2(0A)f\^e(L2r(0,l))N,u(l) and let WQT be the dual space of W 0r .
= 0^
427 Definition 2.1 Assume that u 0 € (L2 (0,1)) N and ^ (z) € (L2 (0,1))^. A function uf is a weak solution of (4) if and only ifuf G L 2 (0,1; Wo), -g^- € L2(0,1; W^) and if for every ip € L 2 (0,1; Wor), we /im;e ;
/:/:(??-K'w:/>^>« J 0 J C
<9z
, ip ) r(l — r2)drdz.
Proposition 2.2 Let ug and 4 ^ be as in the preceding definition. Then there exists at least one weak solution of (4). This solution verifies
HII J c
^(^(^u-^cwr) 8r[
< c,
dz < C, ' (L?(0,1))W dui < C, dz L2(0,1:W'
(5)
where C is a positive constant which only depends on the data of the problem. Proof. We use a Galerkin approximation of uf for which we establish the three above estimates, and pass to the limit in order to prove the Proposition, see [1] for the details. Proposition 2.3 Under the conditions given in the Definition 2.1, the solution of (4) is such that : uf € L 2 (0,1; Wor)
n C (o, 1; (L2 (0,1))") .
Proof. From the embedding W0 C (L2 (0,1)) C Wgr, and because us belongs to 2 L (0,l;WOr) and -^f belongs to L 2 (0, l;W^.), we deduce the result using Proposition 23.23 of [6, p. 422]. * • Remark 2.4 In order to take into account the time variable t, all the above expressions v (.) G H have to be understood as v(.,t) £ l 2 (0,T; H). 2.3
Existence on the boundary
Assuming that -^f is known, we have d2u W~°'~dzT us (z, 0)
du„
e^(o,t) aus %t) dz
= MO-r.> = us0 (z),
= o, = 0,
/ > "
• r2)
dr,
(6)
428 with: r = *{vlt...,TN),
IV =
S
9S
=
(Si,-•
• ,6ff),
=
diag(^,---,^± \Pif PN{ (Ols,
• • • ,6NS)
•
Let {U€(L2(0,lf||e(L2(0,l))N}
ff2'(0,l) =
and let H*(0,1) be the dual space of tf](0,1). Definition 2.5 Suppose that ^
belongs to L2 (o, T; L2 (o, 1; (L2 (0,1))N}\
and that us0
belongs to (I? (0,1)) . A function us is a weak solution of (6) if and only if us € L2(0,T;Hi (0,1)), ^ e L2(0,T;H*z(Q,l)), satisfies u3(z,0) = us0(z), and if for all tp<=L2 (0, T; Hi (0,1)) ; we have :
= 1 1
(6r(u+),tp)dzdt-
f
f
(rf^,^\r(l-r2)drdzdt.
J
Proposition 2.6 Let Uf and us0 as in the preceding definition. Then there exists at least one weak solution us of (6). Proof. We prove the existence of a solution for the linearized weak formulation of the problem using Theorem 2.2 of [4, p. 286] and then we use some fixed point argument. • Proposition 2.7 Under the conditions given in the Definition 2.5, the solution of (6) is such that : 2
1
2
u. € L ( 0 , r ; f f z ( 0 , l ) ) n c ( o , T ; ( L ( 0 , l ) ) " ) .
Proof. Prom the embeddings Hi (0,1) C (L2 (0,1)) N C H*z (0,1) and the fact that 2 us 6 L (0, T; Hi (0,1)), ^ e L2 (0, T; H*z (0,1)), we deduce the result using Proposition 23.23 of [6, p. 422]. • 2.3.1
T = $ o $ is a contraction
Consider the mapping Y : Uf — i > Us —• Uf, and let fi = swp{ (lulP'if) Proposition 2.8 T
/
/"* 3U
/ oio
-^ dz
/inf; 9is.
1. The mapping <& : Uf —> £/, is such that if fia2 < 4, then : 2
fT Z*1 ii i i 2
An dzdt <
* a2(4-fia
sup / / MA] (r,s,t)r(l-r2)drdt. )s€[0,i]J 0J o" "
2
(7)
429 2. The mapping <£ :{/„—> Uf is such that : SUP / / \\Uff(r,s,t)r(l-r2)drdt<»e[0,l] J 0 J 0
f 4J
0
J
f 0
dUs dz
dzdt.
(8)
Define W, =
L2(0,r;L2(0,l;^or))nL2(o,r;C(o,l;(L2(0,l)f)),
Wp =
i2(0,T;#j(0,l))nc(o,T;(L2(0,l)f),
^9
=
{(u,T>)eW)xW r ,,|u(l ) z,t) = «(*,*)}•
Theorem 2.9 Under the hypothesis {HZ) the problem (1) admits a solution if:
This solution belongs to Wg. Proof.
Using (7) and (8), we obtain :
H^1'/- \ / ^ 7 T ^ T N L
; llUfl
ff:=
SUP
f
f\\Ui\\2{r,s,t)r{l-r2)drdt.
The a which minimizes the Lipschitz constant is given by a2 = 2/n, which leads to 11^/11/ < jU-v/e K^f /2- This proves that F : (7/ — i > [// is a contraction if and only if
/i<2/v/i. •
3
Uniqueness of t h e solution
Theorem 3.1 Assuming that the solution of (1) is smooth enough, the system (1) has at most one solution. Proof. We suppose the existence of two couples of solutions {C}j,C}s) and {Cff,Cl).=1 N of (1), and define: Wif = C}f - Cff ; Wls = C}S- Cfs. We multiply the i-th equation of the system verified by Wif by rWif, integrate on [0,1] x [0,1] x [0, T] and use the equation verified by WiS on the boundary. Thanks to {H3), one can deduce that Wif is equal to zero in the cylinder because Wif is equal to zero at the inlet {z = 0) and at the outlet of the cylinder {z = 1) and its partial derivative with respect to r is equal to zero too, which implies, thanks to the equation verified by Wis on the boundary, that the partial derivative of Wif with respect to z is equal to zero in the cylinder. We also deduce that ^gf- is equal to zero on the boundary. Wis is equal to zero at time T and at time 0 because of the initial conditions. Because T is arbitrarily chosen, Wu is equal to zero on the boundary E at any time. •
430
4
Qualitative properties of the solution
4.1
Nonnegativity of the solution
Proposition 4.1 For almost every [r,z,t) m]0,l[x]0, l[x]0,T[, and fori € { 1 , . . . one has : 0 < C;/ (r, z,t), 0 < Cis (z, t).
,N},
Proof. This is proved multiplying the equations of (1) by the non-negative parts of df or Cis, respectively. •
4.2
U p p e r and lower bounds of t h e concentrations
Proposition 4.2 1. Let Si = —1. For almost every (r,z,t) in ]0, l [ x ]0, l [ x ]0,T[, we have : 0 < Cif (r, z,t) < Am ; 0 < Cis (z,t) < Am, with: Ai0 = max
sup Ci0 (r), sup Cis0 (z) Yre[o,il *e[0,i]
2. Let Si = 1. For almost every {r,z,t) Cif (r, z,t) , ai0 < Cis (z,t), with :
in ]0,1[ x ]0,1[ x ]0,T[, we have: ai0 <
ai0 = min I inf Ci0(r), inf Cis0(z) \7-e[o,i] ze[o,i] 3. Let 6i = l. For almost every (r, z, t) in ]0,1] x ]0,1[ x ]0, T[ one has : df (r, z, t) < amext , Cis (z, t) < ai0ext, with : ai0 = max
sup Ci0(r), sup Cis0(z) I ; A = sup kit Yrslo.i] zs[o,i]
ki being the Lipschitz constant of the function r^. 4- Let Si = 1. There exist two positive constants a and b such that for every I in ]0,1[ and for every T > 0 : i I {Cif)2{r,l,t)r{l-r2)drdt J oJ o / (ds)2 (z,T)dz Jo
< aT + b, < aT + b.
Proof. The verification of these qualitative properties of the solution is essentially obtained multiplying the equations of (1) by the appropriate non-negative or non-positive parts of the corresponding test-functions. •
431
5
Numerical simulation for t h e reaction CO + 0 2 —> C02
In the fluid, we use the following discretization method : • If i is different of N (chemical species), we directly solve (l)i using the method of finite differences or that based on finite elements. This requires that the equation with i = N has already been solved in order to put the appropriate values of UN• If i is equal to N (temperature) we evaluate the coefficients at the step before. This possibly requires the use of some fixed point argument. On the boundary, we still use some finite differences method or some finite element method (see [2] for the details). Let us take the following initial and boundary conditions: within the cylinder
CO{r,Q,t) O2(r,0,t) CO 2 (r,0,i) T(r,0,t)
= 0.02 = 0.05 = 0 = 500,
on the boundary
CO (z, 0) O 2 (z,0) CO2(z,0) T(z,0)
= 0.02 = 0.05 = 0 = 490,
We have the following graphs at t = 0.3s, 12s, 24s, 36s and 60s. We observe that the CO and O2 concentrations are decreasing and that the temperature and the CO2 concentration are increasing. - Temperature
4 9 0 I '^TOWMOMasMMqMffWireflagtBtwowgi
39TOt»?C0000C30OtO0eCMflC00Cfi0«»8C
432
CO
o2
co2
433 Remark 5.1 The reaction ends after 54s with the following values at the outlet of the cylinder (z = 1): (70(1,54) 0 2 (1,54) C 0 2 (1,54) T(l,54)
= = = =
0.017777, 0.048912, 0.002355, 500.873497.
References [1] J.-D. Hoernel, Etudes theorique et numerique d'un modele non-stationnaire de catalyseurs a geometrie cylindrique. PhD thesis, Universite de Haute-Alsace (2002). [2] J.-D. Hoernel, Aspects numeriques de l'etude d'un modele non-stationnaire de catalyseurs a passages cylindriques. Actes de la journee CRESPIM du 24 Janvier 2002, Universite de Haute-Alsace (2002). [3] A. Kufner, Weighted Sobolev spaces. Teubner-Texte zur Mathematik, 31. BSB B. G. Teubner Verlagsgesellschaft, Leipzig (1980). [4] J. Malek, J. NeCas, M. Rokyta and M. RuziCka, Weak and measure-valued solutions to evolutionary PDEs. Applied Mathematics and Mathematical Computation 13, Chapman & Hall, London (1996). [5j M.J. Ryan, E.R. Becke, K. Zygourakis, Light-off performance of catalytic converters. The effect of heat/mass transfer characteristics. SAE 910610 (1991). [6] E. Zeidler, Nonlinear functional analysis and its applications. Springer-Verlag, NewYork (1985).
Exact controllability of piezoelectric shells Bernadette Miara Laboratoire de Modelisation et Simulation Numerique Ecole Superieure d'Ingenieurs en Electrotechnique et Electronique 2 boulevard Blaise Pascal, 93160 Noisy-le-Grand, Prance Email : [email protected] Abstract We propose a dynamical model for piezoelectric thin shells and we establish, using the multiplier technique, an observability result. Then, by applying Lions' Hilbert Uniqueness Method, we prove that exact controllability can be achieved with a control vector field acting on the whole boundary of the shell.
1
Introduction
Latin indices and exponents take their values in the set {1,2,3}, Greek ones take their values in the set {1,2}. Einstein's summation rule is used. Boldface symbols represent vectors or vector spaces. We examine first the linearized piezoelectric effect in a three-dimensional body. Let T > 0 and n be a domain in R 3 with C 2 -boundary T. We denote by Q the domain Q = fi x (0,T) and by E its boundary: E = T x (0,T). The equations modelling the time-evolution of the elastic displacement field y = (jk) : Q —• R3, and of the scalar electric potential 9 : Q —> R of a piezoelectric body without any volumic electric charges or mechanical forces are given by: f y" + A(y,0) = \
B(y,8)
OinQ, [l
= OinQ,
>
with the boundary control and the initial conditions y 6 y(0) y'(0)
= = = =
w w y° y1
on E, on E, in 0, infi,
with y" = y"(x,t)
<92v = -g^(x,t),
<9v y(0) = y(x,0), y'(0) = -£{x,0),
434
— x 6 fi, 0 < t < T.
435 These equations are of the divergence type: A = —divT and B = — div<5 where the second Cauchy stress tensor T and the electric displacement S are related through the constitutive equations TV(y,0) = c^klskl{y) + e^dk0, fi*(y,0) = -tPsuW + dVdjO, where (ski) is the linearized deformation tensor skt(y) = (dkyi + diyk)/2. The material is characterized by three time-independent tensors: the symmetric positive-definite tensor of three-dimensional elasticity (cufc'), the symmetric coupling tensor (e,J*) and the symmetric positive-definite dielectric tensor (d y ), C «H =
JM
= cku^ 3 a c > 0
. c^'XijXkl
eHk = eikj^ da = dji^ 3ctd>0.
> acXtjXij:
tfiXiXj
\/Xtj = Xj{ G R,
> adXiXi,
MXi € K.
We have shown in [6] that, for a star-shaped domain fi with respect to x° 6 R 3 , it is possible to build a control (w,w) 6 L 2 (S) x L2(E) acting on the whole boundary E such that system (1) can be driven to rest after time T° = 2max xe jj |x — x°|, i.e. y(t) = y'(t) = 0,t> T°. The aim of this paper is to extend this approach to thin piezoelectric shells. In section 2 we propose a new time-dependent model for Koiter-type piezoelectric shells. In section 3 we study the properties of the homogeneous evolution problem and derive direct and indirect inequalities obtained by the multiplier technique. This leads to an observability condition which allows, in section 4, to establish the exact controllability conditions. Finally, in section 5, we sum up all the results already obtained for three- and two-dimensional structures.
2
A time-dependent model for Koiter-type piezoelectric shells
In order to avoid introducing new notations, we will use the same ones for the twodimensional problem as for the three-dimensional problem. Thus, let Q. be a domain in R2 with boundary T of class C 4 and x = (xi, x2) € fi a system of curvilinear coordinates. The middle surface 5 of the shell is given through an injective mapping ip e C 4 (f2;R 3 ), S =
\x3\ < e}.
where a 3 = -. r is the unit normal at each point of S. |ai x a 2 |
436
2.1
T h e static problem
By an asymptotic analysis, Ch. Haenel has shown in [3] that the limiting two-dimensional model of a piezoelectric shell depends on the geometry of the middle surface S. He established that (as in the pure elastic case) the solution of the three-dimensional problem converges, as the thickness of the shell goes to zero, to the solution of either a "membranedominated" model with a piezoelectric coupling effect or a "flexural-dominated" model, where the elastic and the electric effects are decoupled. Following Koiter's approach for elastic shells, we propose the following static equations with mechanical clamping conditions on the lateral boundary and no electric surface charges. The unknowns are the three covariant components (j/;) : O —> R 3 of the elastic displacement field and the scalar electric potential 6 : fi —• K3. They are, for all applied volume forces f e L 2 (£2), the unique saddle-point in the space [ [ ^ ( Q ) x ffi(n) x Hg(Q.)] x H^(Q)] of the functional
(v,V<)
- / [c (v, v) + 2e (v, ip) — d(ip, ip)] i/adx — / f • v^/adx 2
with CM(y.v) = <*(y,v) = c(y,v)
=
c^-7a/3(y)7,T(v),
e(v,^)
=
e<"% T (v)cW,,
where (7Q/3) is the two-dimensional linearized change of metric tensor, (pa/3) the twodimensional linearized change of curvature tensor and where ca/3°"r, e a/3r and da/3 are twodimensional tensors with the same properties of symmetry and positivity of the analogous three-dimensional tensors. (For a description of these tensors in the case of a Saint VenantKirchhoff material, see Haenel [3]).
2.2
T h e time-dependent problem
Associated with the static problem, we can introduce the time-dependent problem. Let T > 0 and, as for the three-dimensional problem, we note Q = fi x (0, T), E = T x (0, T). The time-dependent covariant components of the displacement field and of the scalar electric potential solve the following system of P.D.E. f y" + A(y,0) = 0 B(y,0) = 0 y = w dnVi = w 9 = Ul y(o) = y° 1 y'(o) = y 1
inQ, inQ, on E, on E, on E, in fi, in SI,
(2)
437 where dn stands for the normal derivative. The control problem, (see the differences in the boundary conditions between the threeand the two-dimensional problems) we are interested in, is therefore: Is it possible to find suitable initial conditions (y^y 1 ), boundary control (w,w),u;) (i.e. in which functional spaces does it belong) and minimal control time T° so that the system can be driven to rest after T° ? As a first step to the study of this problem, we establish useful properties of the related homogeneous problem.
3
T h e homogeneous problem
We consider the homogeneous problem u" + A(u,4>) B(u,<£) u dnu3 4> u(0) u'(0)
= = = = = = =
0 0 0 0 0 u° u1
inQ, inQ, on E, on E, onE, inO, inQ,
(3)
We denote by V s = # 2 (Q) x H2(Q) x tf4(0) and VW = if*(ft) x Hl{U) x Hg(Q) the functional spaces associated with the strong and the weak solutions of (3). Theorem 3.1 Let fi be a domain in M2 with boundary T £ C 4 and T > 0 (i) For (u°, u 1 ) e V s n Vw x Vw
system (3) has a unique strong solution
s
(u, u') e C(0, T; V n Vw x Vw),
4> e C(0, T; H2(Q) n H^(Q)).
(ii) For (ii^u 1 ) e V x L 2 (fi), system (2) has a unique weak solution (u,u') e C ( 0 , T ; V f f x L 2 ( I ! ) ) , <j> e
C(0,T;H*(Q)).
(Hi) The energy E(t) = | J n ( | u ' | 2 + c(u, u ) + d(<j>, <j)))^/adx, associated with a weak solution, is constant over all the trajectories E{t) = E° = \ f [|u x | 2 + c ( u ° , u ° ) + d{<j>°,<j>0)} ,/a~dx, t > T, where f d{4>°,i>)^fa~dx = / e(u°,i/j)^dx, (iv) The weak solution
satisfies Maupertuis'
VV> e
H^(n).
principle:
/ [|u'| 2 - c(u, u ) - d(
Jo
Ja
u°]\fadx.
438 (v) The weak solution satisfies the direct inequality also called "hidden regularity" property: for all multipliers q 6 C1(H)
/.
(vi) For all x° £ K2, there exists a constant a = a(ip, Q), that depends on the geometry of S, such that the weak solution satisfies the indirect inequality established with the multiplier p = x — x°, i f [c(u, u) + 2e(u, 0) - dfa 4>)]p • ny/ZdZ > (1 -
T°)E°,
where T° = 2 maxx6j5 |x — x°|. Proof. The proof of this theorem is obtained for steps (i) and (ii) by application of Hille-Yosida theorem. For steps (iii) and (iv) by multiplying the equilibrium equations by (u, (/>), (u', /) and by integrating on n and Q. For step (v) we make use of the multipliers technique. Indeed, we multiply the equilibrium equations by gAii, qidi(j> and integrate on Q. For step (vi) we follow the same approach, with the new multiplier q = p. • R e m a r k 3.2
1.
2. In step (v), with the multiplier q = n (the unit normal vector) on T, we get [c(u,u) + 2e(u,
we get
(1 -a){T-
T°)E° < C J [c(u,u) + d(<j>, >)]yfaIE.
Hence, when [c(u,u) + d(cj), 4>)]^fadT, € L1(f2), we can define a new norm ||(u°, u 1 )!! 2 = y [c(u, u) + d(0 : 0)]V^dS, equivalent to the energy norm.
439
4
Exact controllability
The exact controllability is established thanks to Lions' Hilbert Uniqueness Method (HUM). Associated with smooth initial conditions (u°, u 1 ), we consider the adjoint backward and the nonhomogeneous problem v" + A(v,V;) = 0 iaQ, B(v,V0 = 0 i n Q ,
(4)
with final conditions: v(T) = 0, v'(T) = 0 in fi, and boundary conditions: v = dnu
on E 0
9nV3
=
9nn«3
v dnv3 ip ip
= 0 = 0 = — dn
°n E0
on E \ E 0 onE\E0 on Ej onE\Ei,
(5)
where E 0 = T0 x (0,T), Ei = Ti x (0,T), T0 and Ti being parts of the boundary dQ. We introduce the operator A: (u^u 1 ) 6 V(Q) x V{Q) -> A(u°,u 1 ) = (v'(0),-v(0)). Let (z,£) be the solution of the homogeneous problem with initial conditions (z^z 1 ) e V(Q) x D(fi). We formally have: f A^u1)-^0^1)^
=
JO.
f
c ( u , z ) d E - / e(z,
J So
+ (
e(u,C)d£+/
«/ So
^
d(C,*)dSSi
Therefore, when E 0 = E : = E, we can introduce, for (ii^u 1 ) e V(Q) x £>(fi), the semi-norm IKu 0 ,^ 1 )!! 2 = / A ( U ° , U 1 ) - ( U ° , U 1 ) ^ = / ( c ( u , u ) + d(0,^))dE. ./n JE Let us denote by F the Hilbert space which is the closure of V(Q) x T>(SV) for the norm IKiAu 1 )!! and F' its dual, let F s = ( V s n Vw) x Vw, Fw=Vwx Fs C F C Fw
L2,
, ( F ^ ) ' C F ' C (F s ))'.
Hence A is uniquely continued in an operator A : F —> F'. By transposition, for (y 1 , —y°) e F' and w £ L 2 (E), w e L2(T,), u e £ 2 (E), there exists a unique weak solution of (4) and we have the final result.
440 T h e o r e m 4.1 We consider a star-shaped domain f2 with C4 boundary and a shell with "shallow" middle surface, i.e. a < 1. There existsT0 > 0, such that for initial conditions (y1,— y°) G F' one can find a control (W,W,LJ) S L 2 (E) X L 2 (S) X L?(Y,), acting on the whole boundary S such that w = 9 n u, w = dnnu3,
LJ = —dn
where (u, <j>) is the unique solution of the evolution problem (3) associated to the initial conditions (u°, u 1 ) 6 F given by: A(u°,u 1 ) = ( y 1 , - y 0 ) This control drives the system to rest for T > T°. Proof. Once the expression of the boundary condition (5) has been estabUshed, the proof is directly adapted from HUM to the system of equations considered here. •
5
Summary
For each case, we recall the direct and indirect inequalities leading to exact controllability (with appropriate definitions of the functional c of the energy E° and of the control time T°). (1) 3d elasticity ([4]): there is a control on part S° C S of the boundary C f
c(u,u)dS < (T+1)E°
, (T-T°)E°
c(u,u)dE.
(2) 2d elastic shell ([5]): there is control on part E° of the boundary, for "shallow" shells C [ c(u, u ) v W E < (T + 1)£°, (1 - a)(T - T°)E°
<
(T+1)E°,
< C
f
[c(u,u) + 4 ^ ) ] d E .
(4) 2d piezoelectric shell: for a star-shaped domain and "shallow" shell, there is a control on the whole boundary C| /
[c(u,u) + 2 e ( u , 0 ) - d ( 0 , 0 ) y o d E | (l-o-){T-T°)E°
<
(T + l)E°,
< C
f
[c(u,u) + d(^,^)]v^dS.
The next step ([7]) of this study is, for the last three cases, to obtain a control on only a part of the boundary S and for ordinary (not shallow) shells.
441
References [1] S. Busse, P.G. Ciarlet and B. Miara, Justification d'un modele lineaire bi-dimensionnel de coques "faiblement courbees" en coordonnees curvilignes. Math. Model. Num. Anal., 31(3), (1997), 409-434. [2] P.G.Ciarlet and B. Miara, Justification of the Two-Dimensional Equations of a Linearly Elastic Shallow Shell. Comm. Pure Appl. Math., 45 (1992), 327-336. [3] Ch. Haenel, Modelisation, analyse et simulation numerique de coques piezoelectriques. Doctoral Dissertation, University Pierre et Marie Curie (2000). [4] J.-L. Lions, Controlabilite exacte, perturbations et stabilisation de systemes distribues. Tome 1. Masson, Paris (1988). [5] B. Miara and V. Valente, Exact Controllability of a Koiter Shell by a Boundary Action. J. Elasticity, 52 (1999), 267-287. [6] B. Miara, Controlabilite d'un corps piezoelectrique. C.R.A.S., Serie I, 333 (2001), 267-270. [7] B. Miara, Carleman estimates for elastic and piezoelectric shells. Work in progress.
Optimal control approach for the fluid-structure interaction problems CM. Murea Universite de Haute-Alsace, Laboratoire de Mathematiques et Applications, 4, rue des Freres Lumiere, 68093 MULHOUSE Cedex, Prance, Email: [email protected]
1 Introduction A fluid-structure interaction problem is studied. We are interested by the displacement of the structure and by the velocity and the pressure of the fluid. The contact surface between fluid and structure is unknown a priori, therefore it is a free boundary like problem. In the classical approaches, the fluid and structure equations are coupled via two boundary conditions: the continuity of the velocity and of the constraint vector at the contact surface. In our approach, the equality of the fluid and structure velocities at the contact surface will be relaxed and treated by the Least Squares Method. We start with a guess for the contact forces. The displacement of the structure can be computed. We suppose that the fluid domain is completely determined by the displacement of the structure. Knowing the actual domain of the fluid and the contact forces, we can compute the velocity and the pressure of the fluid. In this way, the equality of the fluid and structure forces at the contact surface is trivially accomplished. The problem is to find the contact forces such that the equality of the fluid and structure velocities at the contact surface holds. It's a exact controllability problem with Dirichlet boundary control and Dirichlet boundary observation. In order to obtain some existence results, this exact controllability problem will be transformed in an optimal control problem using the Least Squares Method. This mathematical model permits to solve numerically the coupled fluid-structure problem via partitioned procedures (i.e. in a decoupled way, more precisely the fluid and the structure equations are solved separately). The aim of this paper is to present an optimal control approach for a fluid structure interaction problem and some numerical tests.
442
443
2
Notations
We study the flow in the two-dimensional canal of breadth L2 € R2; 0 < xi < L2, -H < x2 < +H} .
Q = {(xux2)
In the interior of the canal there exists a deformable beam fixed at the one of the his extremities (see the Figure 1). X2
b
A \ i
i
K\
* —-
-
c if5r""c'
B i
1 xi!
Figure 1: The flow around a deformable beam In the absence of the fluid, the beam has the parallelepiped shape [ABCD]. coordinates of the vertices are A = ( - r , 0 ) , B = (-r,L1),
C={r,Lx),
The
D = (r,0).
The beam is deformed under the action of the fluid and it will have the shape [AB'C'D]. The deformation of the beam is described using the displacement of the median thread w = (wi,w2) : [0,L!]-)-K 2 . which satisfy the compatibility condition ux (0) = 0, u2 (0) = 0. For instant, we assume that «i = 0. The domain occupied by the beam is nf = {{xux2)
€M 2 ; xx G]0,Li[, \x2 - u2 (Xl)\ < r} .
Consequently, the domain occupied by the fluid is
n« = n\fif. The contact surface between fluid and beam is Fu = \AB'[ U [B'C] U ]CD[ where }AB'[ = {(a;i,a; 2 )GR 2 ; Xl e]0,Li[, x2 = u2 (xx) - r) , [B'C] = {{xux2) e R2; xx = Lu x2 e [-r,r]} , }C'D[ = { ( i 1 , x 2 ) 6 R 2 ; i i e ] 0 , L 1 [ , i 2 = u 2 ( i 1 ) + r - } . The other boundary of the fluid domain is noted IV
444
3
Beam equations
The beam has one end clamped and the other is free. It is deformed under the flexion efforts and we suppose that the longitudinal traction compression forces are negligible. We don't take into account the effect of the transverse shear. We present the mathematical model, following [1]. In view of the Sobolev Embedding Theorem, we have H2QQ,L1\)^C1([0,L1]) and we denote ^ { ^ f l 2 ( ] 0 , i i [ ) ; 0(O) = 0'(O) = O}. Let D2 6 K + be given by the formula D2 = E I x\ dx2dxz is where E is the Young's module and S is the cross section of the beam. We set as • U x U -> R
os( )=i?2
^
!
LS ( a ; i ) S ( a : i ) ^
(1)
Remark 1 As a consequence of the Lax-Milgram Theorem, we have the following result: Let f^ 6 L 2 (]0,Li[) and r)2 £ L 2 (]0,Li[). Then the problem: Find U2 in U such that as{u2,ip)=
77 2 (zi)V(zi)cfai+ / V]0,Li[
fiix^ipix^dxu
Vip 6 U
(2)
*']0,Ll[
has a unique solution. The volume forces (the gravity forces) are included in f2. We denote by rj2 the fluid forces acting on the beam. When the data and the solution are smooth enough the solution u2 verifies the strong formulation given by: <(xi)
= jf
M * i ) + / 2 S (zi)) ,
Vsi € ]0, Li[
U2(0) = u'2(0) = «5(Li) = < ( L X ) = 0. In the particular case f£ = 0 and ^2(^1) = a + fixi + 72:1, ^ - [ 1 5 ( ^ - 4 ^ + 6 1 + (x(-
20xiL{ + 45L?) 7] •
we obtain
445
4
Fluid equations in moving domain
We suppose that the fluid is governed by the two dimensional Stokes equations in the velocity-pressure-vorticity formulation: —F
—F
n
Find the velocity v : Qu —• R , the pressure such that dp du dx\ 8x2 dp dui dxn dxi ov2 dv\ — — ~i- — • dxi dx2
—F
p : U,u —>• R and the vorticity w : fiu —>• R, =
/iF
inO£
=
f?
infif
— 0 in£2£ = 0 inQ£ V = 9 onTi v • n = 0 on r u P = Po on T„ w = 0 on IV On r„, we have the boundary conditions v • n = 0 and p = po- The validity of these boundary conditions was proved using the least squares variational formulation. See ([3, Chap. 8]) for more details. The boundary conditions p = p 0 a n d v • T = 0 were studied in [10] and the slip boundary conditions v • n = 0 and (an) • r = 0, where a is the stress tensor, were studied in [11] and [12], but these boundary conditions aren't appropriate for our approach of the fluid structure interaction. divii
Now, we will present the least squares variational formulation for the problem (3). Let u2 be the solution of the equation (2). We have
flaao,iiD^c1([o,Li]) therefore the domain H£ has a Lipschitz boundary, so that we can define the spaces H1 (fi£), H1'2 (I\) and H1'2 (r„). We recall that dQ^ =TUU TV We denote by n = (ni,n 2 ) the unit outward normal vector and by r = (—ra2,ni) an unit tangential vector
to dVtl. We denote by • the scalar product. Let us consider the following vectorial spaces W
=
Iw = (wu w2) e H1 (fi£) 2 ; w = 0 on Ti and w • n = 0 on Tu\ ,
Q = {qeH1 M
=
(fij) ; q = 0 on Tu) , 1
{ u e f f ( ^ ) ; u = 0onr1}.
Let g € HQ (IV)2 be given, such that fv g • n da = 0. Then there exists v0 e 1
H (n£) , such that divi>0 = 0 in H£, v0 = 0 on Vu and v0 = g on IV Let po be given in H1?2 (Tu). Then there exists a function in H1 (fi„), such that its trace is p0. We denote this function by p0, also.
446 P r o p o s i t i o n 1 For all u2 in U, and fF in L2 (Q„ ) , the problem: Find v — v0 G W, p — p0 6 Q, u> € M, such that ( dp \dxi ( +
did dq dx2'dxi dv2 dvx
8p\ dx2) dw2
( dp \dx2 dwA
dw dq dxi' dx2
dp \ dx\)
{U-dx-1 + dx-2'p-dx:+8X-2)+(dlVV>dlYw)
-{*>& has a unique
(4)
+ '&) + {*•&--£)>
v—,v9eQ,v,eM
solution.
Here, /i > 0 is the dynamic viscosity of the fluid, fF is the external given force per unit volume and (•, •) is t h e inner product of L2 ( £ ^ ) Proof. We first prove t h a t dq_+
dp_
dq
dx\
8x2
8x2
dp dxi
l | V ^ + M2||Vp||^
where ||-|| 0 is t h e standard norm of L2 (QF). Le us consider t h a t q € C1 ( ^ „ ) , q = 0 on r u and p e C2 (ttF), dq
dp
dxi
ox2
dq_
+
dx2
dxi
p = 0 on Ti. We have
;.^dwiv*«,(^g)-*.(i.i)
But using Green's formula, we have (8q_ \dxi'
8p\ _ /8q_ 8^ 8x2) \dx2'dxi
d d P n daj • /f 9 ^P— " l;d a - f q-— 2 x Jenz " 2 Jen? dx±
( [q, d P \ I +( [q, d2P \ 8x2dxxJ \ dx^dx-t
= I[ q(Vp-r) q(VP.r)da da++ (q,-^l(q,-
8x28x1
By assumption of the regularity of p, we have
9x2^1
+
J^-\.
8x \dx2)
<Ja:i0:E2'
Since p = 0 on T i , we obtain t h a t V p • T = 0 on F i and then, by a density argument, the equality from t h e beginning of the proof holds. T h e rest of the proof runs as in [3, Sect. 8.2.2]. •
447
5
Optimal control approach of the teraction problem
fluid-structure
in-
In the classical approaches, the fluid and structure equations are coupled via two boundary conditions: continuity of the velocities and continuity of the forces on the contact surface. We denote by A = (Ai,A2) the forces induced by the beam on the contact surface. Consequently, —A represent the forces induced by the fluid acting to the beam. We denote by S : M -4 U the application which computes the displacement of the beam knowing the forces on the contact surface. This application is linear and continuous. We denote by F : U x M —>WxQ the application which computes the velocity and the pressure of the fluid knowing the displacement of the beam (therefore the domain of the fluid) and the forces on the contact surface. This application is non-linear on U x M.
r
s
u F
v, P
A
Figure 2: The computing scheme We search to find out A, such that ii|r„ = 0. This is a exact controllability problem. In our approach, the target condition will be relaxed. We assume that the forces on the contact surface have the form A = —pa n, where pa is the pressure of the fluid. We consider the following optimal control problem: inf J(ai,a 2 ,ft, / 9 2 ,7i,72) = ^ II" • i"Ho,r„
(5)
subject to: (ai,a 2 ,/0i,j02,7i,72) e K C '. M^i)
=
(6)
[15 (x\ - 4a:iLi + 6Z»i) (OJI - a 2 ) +3 (x\ - lOxii? + 20L?) (A - fa) + (x\ - 20iiL? + 45L}) (71 - 72)]
(7)
(v,p, <j) solution of the Stokes problem (4) with p0 (xux2)
f (ai + jSiii + 71Z?), if (xux2) € ]A, B'\ = < (a2 + fax\ + j2xl), if (xi, x2) e }D, C] { ( I - | i ) P o ( B ' ) + (i + ^ p o ( C ) , if (xux2)£}B',C'{
(8)
It's an optimal control problem with Dirichlet boundary control (p0) and Dirichlet boundary observation (i>|r„)-
448 The relation (6) represents the control constraint. The relation (7) represents the displacement of the beam under the cross forces \2 = (ai + hxi + Ti^i) on ]A, B'[ and A2 = (-»2 - h^\ - 72K?) on ]D, C'[. We assume that the displacement of the beam under the longitudinal forces Ai is negligible. This mathematical model permits to solve numerically the coupled fluid-cable problem via partitioned procedures (i.e. in a decupled way, more precisely the fluid and the cable equations are solved separately). Remark 2 The existence of an optimal control could be find in [5] for a related problem. In [7] it is proved the differentiability of the cost function and it is given the analytic formula for the gradient. Remark 3 An open problem is to find additional conditions in order to obtain zero for the optimal value of the cost function. This is an approximate controllability problem. For a linear model (the domain of the fluid doesn't depend upon the displacement of the structure), we can find approximate controllability results in [4], [8] and [9]. Remark 4 If v • T is constant on Tu, then v is constant on Tu. Using [9, Prop. 3.1], we can prove that (an) • n = —po, where a is the stress tensor. Consequently, solving the beam equations under the action of the surface forces —\=p0n on Tu is reasonable.
6
Numerical tests
The parameters for the simulation are listed below: the geometry L\ = 0.5, L2 = 1, H = 2, r = 0.05, the beam D2 = 5, the fluid fi = 1, fF = 0, g = (0, Vx{) on the left and right parts of T\, g = (0, V) on the bottom of Ti, g = (0,0) on the top parts of T\, V = 0.5. The choice of these parameters induces small displacements of the beam. The fundamental hypothesis in linear elasticity of the beam is that the displacements remain small. The optimal control approach for fluid structure interaction presented in the present paper could be employed also for the large displacements of the structure, but in this case we have to use a well adapted model for the structure. For a guest (a 1; a 2 , A , hi 7i> 72)1 w e compute the displacement of the beam using the formula (7). Now, we know the moving boundary of the fluid and we generate a mesh consisting of triangular elements. Then, we solve the fluid equations (4) with boundary condition (8). We have used the P\ finite element for the velocity, the pressure and the vorticity. The target is to minimize the cost function (5). The numerical tests have been produced using freefem+ (see [2]). The boundary condition v • n = 0 on Tu was replaced by v2 = 0.
449
Figure 3: The computed velocity around the beam The computed velocity isn't a divergence-free field. For a better approximation of the incompressibility condition, we can penalize the term (di\v,divw) in (4). The optimal value of the cost function is J=2.36033e-04 and it was obtained for the penalizing factor 105. In this case ||divu||g is 4.80281e-04. In the Figure 3, we can see the corresponding displacement of the beam and the velocity of the fluid. The velocity of the fluid was multiplied by 2 for a better visualization. We can avoid to generate a new mesh for each evaluation of the cost function by using the dynamic mesh like in [6].
References [1] M. Bernadou, Formulation variationnelle, approximation et implementation de problemes de barres et de poutres bi- et tri-dimensionnelles. Partie B : barres et poutres bidimensionnelles, Rapport Techique no 86, INRIA, 1987 [2] D. Bernardi, F. Hecht, K. Ohtsuka, O. Pironneau, A finite element software for PDE: freefem+, h t t p : //www-rocq. i n r i a . f r / F r e d e r i c . Hecht [3] B.N. Jiang, The least squares finite element method. Theory and applications in computational fluid dynamics and electromagnetism, Springer, 1998 [4] J. L. Lions, E. Zuazua, Approximate controllability of a hydroelastic coupled system, ESAIM: Contr. Optim. Calc. Var., 1 (1995) 1-15 [5] CM. Murea, Y. Maday, Existence of an optimal control for a nonlinear fluid-cable interaction problem, Rapport de recherche CEMRACS 96, Luminy, France, 1996 [6] CM. Murea, Dynamic meshes generation using the relaxation method with applications to fluid-structure interaction problems, An. Univ. Bucuresti Mat., 47 (1998), No. 2, 177-186 [7] CM. Murea, C. Vazquez, Computation of the gradient for a nonlinear fluid-structure interaction problem, in preparation
450 [8] A. Osses, J. P. Puel, Approximate controllability for a linear model of fluid structure interaction, ESAIM: Contr. Optim. Calc. Var., 4 (1999) 497-513 [9] A. Osses, A rotated multiplier applied to the controllability of waves, elasticity and tangential Stokes control, to appear in SIAM J. Control Optim. [10] O. Pironneau, Conditions aux limites sur la pression pour les equations de Stokes et Navier-Stokes, C.R. Acad. Sc. Paris, 303 (1986), Serie I, No. 9, 403-406 [11] R. Verfurth, Finite element approximation of incompressible Navier-Stokes equations with slip boundary condition, Numer. Math., 50 (1987) 697-721 [12] R. Verfurth, Finite element approximation of incompressible Navier-Stokes equations with slip boundary condition II, Numer. Math., 59 (1991) 615-636
Hyperbolic propagation of singularities in a parabolic system of shell theory J. Sanchez-Hubert Mathematiques et Mecanique, Universite de Caen Boulevard Marechal Juin, 14302 Caen, FRANCE E-mail : [email protected] Abstract The mechanical properties of shells are concerned with the membrane system. When including the smoothing terms of flexion, which are small for thin shells, the singularities become narrow layers, it is easily seen that the energy concentrates in these layers. For parabolic surfaces, by choosing appropriate parameters, we may solve the system step by step. Each step equation of order 1 is hyperbolic with simple characteristics coinciding with the generators. Classical hyperbolic propagation of singularities applies. We give the corresponding transport equation of the singularities. At each step the effective data involve the components of the loading and some derivatives of the unknowns considered in the previous steps, so that this process implies singularities of increasing orders.
1
Introduction
We present a class of systems of P D E (issued from shell theory) which are parabolic. Their structure is such t h a t they may be solved by a sequence of step equations. Each step equation is hyperbolic, so t h a t classical propagation of singularities theory applies. Moreover, at each step, t h e effective d a t a involve some derivatives of the unknowns considered in t h e previous steps, so t h a t this process implies singularities of increasing orders along t h e chain. For instance, a singularity of the d a t a f3 along the characteristic y2 = 0 of the form where Y is t h e Heaviside function and F3 denotes a smooth function, implies a singularity of t h e unknown u3 of the form u3 ~ 6'" (y2) U3 (y1) , where C/3 (y1) satisfies genuine transport equation so t h a t it may be different from zero in regions where F3 = 0.
451
452 Such a situation is in evident contrast with the behavior of the "model parabolic heat equation" : du d2u _ where the presence of the time derivative classically implies hypoellipticity (see for instance [8], page 145) i.e. u is of class C°° in an open set whenever / is also C°°, so that singularities cannot propagate. The mechanical behavior of a shell (in the Kirchhoff Love model) in the static case is described by two energy forms (see for instance [1] and [6]) : am (u, v) : membrane bilinear form e2a,f (u, v) : flexion bilinear form. Let Am and E2Af be the operators associated with the two bilinear forms am (u, v) and e2a,f (u, v) respectively. Then A; is elliptic and Am is of the same type as the points of 5. Its characteristics are the asymptotic curves of S. For e > 0, the energy space V is such that am + e2a,f is continuous and coercive on V. In the sequel, we only consider the case when the surface S, along with the kinematic boundary conditions, is geometrically rigid in the linear sense. This amounts to the property that am (v,v) = 0 => v = 0. As the order of differentiation in Af is higher than in Am, a singular perturbation phenomenon appears when e \ 0. Moreover, as the limit process e \ 0 goes from a higher order elliptic problem to a lower order parabolic (or hyperbolic) one, the limit process e \ 0 is non-standard. For E \ 0, the limit problem involves a new energy space Vm for which the bilinear form am is continuous and coercive : 1. Vm is the completion of V for the norm y/am (.,.) (which is norm as a consequence of the above mentionned geometric rigidity). 2. Vm contains functions less smooth than those of V. Consequently, the solutions ue belong to V, but their limit as e \ Junction : ue exhibit boundary layers for small e.
0 is a less smooth
We have the embeddings
VcVm,
VicV,
so that the usual loadings / 6 V are admissible for the variational problem for e > 0, but it may happen and often happens that f £ Vm. It will proved useful to consider the system of PDE occuring in the membrane shell theory (i.e. the limit problem for e = 0) :
f -Dpi** \ -ba0T°e
= =
fa, f3,
(1)
453 where rpa(3
AafBXy.
n (Dxu^ + D^ux 2
bXliu3)
(2) l
are the stress components and u the displacement vector. The variables are y and y2, which describe the middle surface 5 of the shell, Da denotes the covaxiant differentiation on the surface (Throughout this work, the notations are those of surface theory, see for example [6]). The above coefficients bap are smooth given functions. They are indeed the coefficients of the second fundamental form of the surface S. The coefficients Aa0X,x are also smooth given functions, namely the membrane elasticity coefficients. They satisfy the classical symmetry and positivity hypotheses. The unknowns are the stresses T a/3 = T@a and the displacement components Uj (the greek and latin indices run in {1,2} and {1,2,3} respectively); ua are the covariant tangential components and M3 is the normal component to the surface. The given loading / is defined by its contravariant components /* with f3 normal to the surface. The system (l)-(2) contains six equations and six unknowns Uj, T 11 , T 12 and T22; its total order is four. In the sequel, we shall consider the case of an everywhere parabolic middle surface, that is the second fundamental form is degenerated. It is easily checked that system (1)(2) will then be parabolic in the following sense. Denoting DQ (£) the principal symbol of the system, det [Do (£)] = 0 has one root £ of multiplicity 4. More precisely, the corresponding characteristic (with multiplicity 4) coincides with the asymptotic direction of the corresponding points of 5 (which is double in the parabolic case). It is well known in surface theory that the set of surfaces everywhere parabolic coincides with the set of developable surfaces. Then, choosing the parameters y1 and y2, with y1 along the generators, we have bu = b12 = 0, b22 ^ 0.
(3)
1
Moreover, taking the curve y = 0 orthogonal to the generators, the Christoffel symbols are such that
iTx = r £ = 0.
(4)
We have the following criterion in order to determine if a given loading belongs to V^ or not : Theorem 1.1 A necessary and sufficient condition for f to belong to V^ is that there exists TaP G l? (S) satisfying (1) and the boundary conditions Ta^n^ = 0, on the free parts of the boundary. Here ~n is the unit normal to the boundary in the tangent plane. This theorem and its proof are analogous to those of Section 2.2 in [4]. Using this criterion, let us give an example of loading / with a jump (then not so much singular!) such that / does not belong to V^ :
454 Theorem 1.2 Let us consider a shell the middle surface of which is developable and such that b22 7^ 0. Let the loading f defined as /1 = /2 = 0, and with U =/= 0 a piecewise smooth with a jump discontinuity along a part of a generator. Then f does not belong to V . Proof. The limit problem in terms of the stress components write
where l\™ = 0, since the sin-face is developable. The last equation immediately gives h
n22 _
b22(y1,y2Y Then, if we assume, to fix ideas, / 3 (y1, y2) = F (y1) Y (y2), where Y (y2) is the Heaviside function, we get
F(yl)Y(y2)
T22 = — -
b22(y\y2)
and (5) becomes f - D j T 1 1 - D2T12 = 0 dlT-
- 3V22T" = ^ 6
\
(y2) + (T£ + 21-) W i m
°22
^
022
The second equation is a differential equation with respect to y1 with parameter y2 the solution of which is classically
T" =
6(y2)<S>(y\y2)+y(y\y2)+K(y2),
where
*(2/V):= f exp- / V ^ , ! / 2 ) ^ F(n) dn Jo
Jo
and * (y\ y2) := Y (y2) J^
exp (- J^ 3I\22 (£, y2) df) ( i ^ + 2r 2 2 ) ^ d n .
Clearly, * € L2 ((0, J^ x {-l2,I2)), * € L2 ((0, Ji), tf"1 (-I 2 , J2)) and A- € H~l (-l2, l2). As the function $ depends on y1 and K is only a function of y2, the solution T12 cannot belong to L2 (fi) and according to 1.1 the conclusion follows. • We are mainly concerned with the propagation of the singularities of system (l)-(2). We consider the classical sequence of distributions on R with increasing singularities •••,xY(x),Y(x),S(x),6'(x),
•••
(7)
455 where Y and S denote the Heaviside function and the Dirac mass respectively. More precisely, these distributions are considered as singularities at x = 0 whereas their values for x ^ 0 are discarded. For instance Y (x) is considered merely as the unit jump at x = 0. For instance, in order to describe the singularity of T 11 along y2 = 0, we consider expansions of the form (for example) T"~#(vi)A11(y1)
+ 6(vl)B"(y1)
+ ...
(8) 2
where it is understood that the terms • • • are less singular than the previous ones at y = 0. Such kind of expansion is in the framework of discontinuous solutions (see for instance works by Egorov & Shubin [2], Sanchez Palencia [5] and Gerard [3]). We always assume that the geometric data and the coefficients are smooth, so that the sequence (7) is consistent with the singularities of the solutions, provided that the singularities of the data are in that sequence, which covers most of the usual examples.
2
Singularity along a characteristic due to a discontinuity of t h e loading / 3
Let us assume that the parameters belong to the domain O = (0,1) x (0,1) and that the boundary conditions are such that the shell is inhibited, for instance fixed along yl — 0 and free elsewhere : u(0,y2)=0. (9) Denoting by n the unit normal to the boundary in the tangent plane we have on the free part Ti = {y1 = 1} of the boundary the natural boundary conditions Tafinp = 0 which give f ^ ( l . V 2 ) = 0, (10) \ T 1 2 ( l , y 2 ) = 0. Indeed, n//a1 which implies that n\ ^ 0 and «2 = 0. As an example, let us consider the loading
r = o, f
=
Y(y2-c2)dlalM](yi)F(y\y2),
where Y is the Heaviside function, 0[ai,f,i] is the characteristic function of an interval [a'jfc1] C [0,1] and F denotes a smooth function. The jump of / 3 lies along the internal characteristic y2 = c2 and according to 1.2 / does not belong to Vm. Obviously a boundary layer appears along y2 = c2. Equation (6)2 is hyperbolic of order one for T 12 . The main singularity (in the sense of (7)) in the right hand side is 6 (y2 — c2) #[ai,j,i] {y1) Fty1). Consequently, we search T 12 as T*2^6(y2-c2)A12(y')+--Then, A12 satisfies the transport equation ^
+ 2r 2 2J 4 12 = V,M](2/ 1 )*(2/ 1 ),
(11)
456 where $ (y1) = F ^ c 2 ) . The boundary condition T 1 2 (l,y 2 ) = 0 gives A12 (1) = 0. Consequently, the equation (11) completely determines the unknown A12. More precisely, we have for 0 < y1 < a1
ro fy S ( „ ) e x p | T 2r22(£)d£ drj A12
(v1) =
(-2jV^12(Odt
xexp Cexpl
2rf2(0de
for a 1 < y1 < ft1
for b1 < y1 < 1,
where the constant C is such that A12 (y1) is continuous at y1 = 61. We observe that A12 (y1) = 0, for 0 < y1 < a1, but in general A12 (y1) ^ 0 for b1 < y1 < 1, though #[ai (,i| = 0 : this is the phenomenon of propagation of the singularities. T 11 is the solution of the equation (similar to (6)i) :
dtTn + r 2 2 r u = r 2 2 -^ - d2Tn - r 2 2 r 12 , 022
the main singularity of which in the right hand side is in 8' (y2 — c2). Consequently, we search the main singularity as T»c6'(y2-c2)A»(yl), 11
(12)
1
where it appears that A (y ) is solution of the equation dA" + r 2 2 A n = -^ 1 2 , dyi the solution of which is determined as before using the boundary condition A11 (1) = 0. Now, we consider the system (2) yp/3A/i
2 iD\uii + D^ux - 6AMM3)
: T^
At the leading order of singularities in the sense of (7) this gives dlUl = SuuT11 d2u2 - r^ 2 u! - T| 2 u 2 - 622u3 = S 2 2 1 1 T n I 2 n
(13)
(ftui + 9i«2) - r 2 u 2 = 5 1 2 1 1 r ,
where Sapx^ are the comphance coefficients (5 is the inverse matrix of A). Then we search the main singularity of the components as (
Ul
u2 I u3
= S'(y2-c2)U1(y1), 2 2 = 8"(y -c )U2(yl), = 8"'{y2-^)Ui{y1).
(14)
457 Substituting (14) into (13), we immediately obtain, taking into account the boundary condition at y 1 = 0 U1(yl) = [V Jo
Snu(Z,c2)An(Od£.
Then U2 is obtained from (13)3
1
U2boundary {y ) = - condition j T Uxfo) (- 0 j which gives, with the U2exp (0) = Jo \ Jo
f 2T\2 (0 d£drj. /
At last, we obtain U2(y1) U yl)
^
3
=
b22(yWY
Numerical simulations
In this section, we present numerical simulations for the system am (u, v) + e2af (u, v) = (f, v), (see section 1) with e = 10~3. Numerical simulations allow us to exhibit the phenomenon of propagation of singularities, more precisely, thin layers which converge to the singularities for e \ 0 see [7]. The developable surface S is the cone parametrized by ^
/ SI \{y\y2)
-> ^ (y1,(y1-a)
S sin (2Try2),(y1-a)
cos (2Try2)), a> 1
where ft = (0,1) x (0, | ) , the image of the mapping being presented in the following figure : The numerical computations are implemented with Hermite finite elements. The exact numerical integration of the rigidity matrices needs six Gauss points. In this case, the error estimate is of order O (ft3). The meshes for the domain H are generated using the Modulef code. The mesh is obtained in order to perform the refinement of the mesh in the vicinity of the layers (see hereafter) using a distribution function of the points, according to a geometric progression with ratio q.
458
Figure 1: Mapping of the cone : fi = (0,1) x (0,0,5) ; x = y1, y = (y1 + 2) sm2ny2, z = {yl + 2) cos 2ny2
3.1
Characteristic internal layers and propagation of t h e singularities
Let us consider the loading /1 = ( 0 , 0 , - y ( y 1 - l ) K ( y J - i ) ) where {y
'
1 0 if ya < 0.
The numerical results are shown on figure 2 We observe an internal layer along the whole characteristic y2 = 1/4, while the discontinuity of the data is only along the interval 1/2 < y1 < 1.
459
Figure 2: Case of a characteristic internal layer, propagation of the singularity for the loading / j . Plot of u | in the whole domain Another interesting example with propagation of singularities is given in figure 3 The loading is here defined by
£=(o,0,(ion(i'£)xG^ V
I 0 elsewhere.
>
•
)
2
We observe internal layers along the whole characteristics y = 3/16 and y2 = 5/16, while the discontinuity of the loading lies only on 1/4 < y1 < 3/4.
460 0.02 r 0.015 0.01 -
Figure 3: Case of characteristic internal layers, propagation of the singularites for the loading / 2 . Plot of u | in the whole domain.
References [1] Bernadou M., Methodes d'elements finis pour les problemes de coques minces. Masson, Paris (1994). [2] Egorov Yu.V. & Shubin M.A. Editors, Encyclopaedia of Mathematical Sciences, vols 30-31-33 (Partial Differential Equations I, II, IV), Springer Verlag, Berlin (1991). [3] Gerard P., Solutions conormales analytiques d'equations hyperboliques non lineaires. Comm. Part. Diff. Eq., 13 (1988), 345-375. [4] Karamian P., Sanchez-Hubert J., Sanchez Palencia E., A model probem for boundary layers of thin elastic shells. Math. Model. Num. Anal., 34 (2000), 1-30. [5] Sanchez Palencia E., Propagation of singularities along a characteristic boundary for a model problem of shell theory and relation with the boundary layer. Compt. Rend. Acad. Sci., Paris, serie lib, 329 (2001), 249-254. [6] Sanchez-Hubert J. & Sanchez Palencia E., Coques elastiques minces. Proprietes asymptotiques. Masson, Paris (1997). [7] Karamian P. & Sanchez-Hubert J., Boundary layers in thin elastic shells with developable middle surface. Europ. J. Mech. A/Solids (to appear). [8] Schwartz L. Theorie des distributions, Hermann, Paris 1966.
Singular perturbations going out of the energy space. Layers in elliptic and parabolic cases E. Sanchez-Palencia Laboratoire de Modelisation en Mecanique CNRS-Universite Pierre et Marie Curie 4 place Jussieu, F-75252 Paris Cedex 05 France Email : [email protected] Abstract We consider singular perturbation problems depending on a small parameter e. The right hand side is such that the energy does not remain bounded as e —> 0. Boundary layers bearing most of the energy and accounting for non-smoothness of the limit appear. We consider problems which are elliptic for e > 0 in two cases : the limit problem is elliptic and the limit problem is parabolic.
1
Introduction
Classical singular perturbation theory for symmetric variational problems in real Hilbert spaces deals with the following situation : let V be a real Hilbert space and a and b two bilinear symmetric forms on V satisfying : \a(u,w)\
Vu,weV
(1)
|6(u,w)|
\/u,weV
(2)
a(w,w)>0,
VweV
(3)
a(w, w) = 0 =>• w = 0 a(w,w)
+ b(w,w)
> c\\w\\2,
(4)
Vw£V,
(5)
where J.|| denotes the norm of V and c, C are constants. Let / e V be t h e dual of V. For e > 0 we consider t h e variational problems Pe ue eV a{u',w)
J / „ e, E „ „ \ _ If „„\ + e22b{u ,w) = (f,w) >
461
w , „ a T/ Viu e V,
(")
462 which are well-posed in the Lax-Milgram framework, as the form in the left hand side is coercive on V (but the coerciveness constant decreases as e2 when e —• 0). To study the asymptotic behaviour of ue as e —• 0, we note that (3)-(4) imply that (7)
ll-llv. = «(•>•), is a norm on V. Let Va be the completion of V with respect to this norm. Clearly VCVa
,
7'CK,
(8)
with dense and continuous embeddings. Clearly / e V but it does not necessarily belongs to V£. When this happens, the "limit problem" P0 :
f ueVa \a(u,w)
()
= {f,w),
V»eV„,
w
is a Lax-Milgram problem. The main result in this theory ([10], [6]) is given by : T h e o r e m 1.1 Under the previous hypotheses, let ue be the solution of (6). i) Iff
e V i , then ue —> u strongly in Va,
(10)
where u is the solution of 9. ii) If f i V'a, the energy i [ a ( u e , u e ) + e 2 6(« e .« e )] e
of the solution u tends to +oo as e —• 0. Proof. The part i) is classical (see for instance [6]). Its proof is analogous to that of Theorem 2.4 hereafter. The part ii) is a complement (the proof is easy and may be seen in [4]) which shows in particular that the condition / 6 V'a is necessary to keep the energy bounded. • In this paper we consider examples of singular perturbations with / ^ V'a involving problems which are elMptic for e > 0 and either elliptic or parabolic for e = 0, i.e. the limit problem (9) is either elliptic or parabolic, but of course, it does not make sense as a variational problem as / 0 V'a. The physical motivation for studying such kind of problems comes from the thin shell theory. In the case when the middle surface, with the kinematic boundary conditions is geometrically rigid, which furnishes the hypothesis (4). In that case the limit problem (9) is elliptic, parabolic or hyperbolic according to the nature of the points (elliptic, parabolic or hyperbolic) of the middle surface of the shell. Moreover, in shell theory the space Va is "very large" (going in certain cases out of the space of distributions, see for instance [4]) so that its dual V'a is "very small" and "usual loadings / " do not belong to it. We consider here several model examples where / ^ V'a. because of a singular behavior along a curve. It then appears a strong layer phenomenon along that curve : energy concentrates along it and grows without limit as e —> 0. In order to study that phenomenon,
463 we perform a dilation of the normal coordinate to the curve ; the problem Pe becomes a new one, noted Ve in the new coordinate system. The dilation is obviously anisotropic, and even if Pe and Ve are equivalent for s > 0, the asymptotic behaviour of their energies is different : boundedness deals with different terms of the expression of energy. With an appropriate choice of the scaling of the dilation, the "new energy" remains bounded, and we have a "limit problem" Vo of VE. This process, which is specific to each problem and each loading, may be interpreted in terms of the method of inner and outer matched asymptotic expansion (see for instance [1] and [5]). PE and Ve are then the expressions of the exact e > 0 problem in the outer and inner variables respectively, whereas PQ and Vo are the outer and inner limits respectively. We emphasize that convergence is proved for VE and Vo- The questions of the convergence of Pe and P 0 is open for / £ V'a. In any case, if convergence holds, by virtue of Theorem 1.1, its topology is certainly weaker than the energy one. In Section 2 we consider an example where the limit problem is elliptic (or microlocally elliptic along the singular curve, Remark 2.6). Problem PE (6) is variational, whereas P0 (9) with / $ V'a is not. Nevertheless, both problems enter in the framework of transposition solutions of Lions and Magenes [7]. The limit problem Vo describes the loss of regularity of the solution as e —> 0. It is an elliptic problem in the variable transversal to the singular curve, the tangential variable appearing merely as a parameter. Moreover, it deals with equivalence classes of functions, denned in some sense "up to the regular part", in other words, it describes the loss of regularity and nothing more. In Section 3 we consider an example concerning the equation (e2A2-d>)u*
= f,
(11)
which is elliptic for e > 0 and parabolic for e = 0. We study singularities along the characteristic a 2 = 0. We observe that the left hand side of (11) may be factorized so that the equation is equivalent to the system (-eA + d1)u = v, (-eA-d1)v = f,
, . [U)
so that it amounts to the reiterated solution of two transport-diffusion problems (see for instance [6] or [5]). We shall solve (11) taking advantage of the variational formulation of the problem Vs. In this case, the equation describing the asymptotic behaviour is a genuine partial differential equation in yx, yi (the tangential coordinante is no longer a parameter) accounting for the propagation of singularities along the characteristic. This point is not considered explicitely here, but may be handled as in [9]. Section 4 is devoted to some complements concerning the case when the singularity is not interior to the domain, but along a boundary with Neumann non-homogeneous boundary conditions. The case of equation (11) is then somewhat analogous to that of [8]. Notations are usual. In particular H3(Q), s € R are the Sobolev spaces (see definition in [7]) and Hg(Q.) are the (closed) subspaces of the functions for which all the traces which make sense vanish. The summation convention with respect to repeated indices is used.
464
2
Model problem for Pe and PQ elliptic of order 4 and 2 respectively
In this section we consider elliptic problems in the domain n = (0,7r) x (-1,+1)
(13)
of the variable x — (x\,x-2). The bilinear forms are a(u,w)=
I diudiwdx, n
(14)
b(u,w) = / dijudijwdx. n For e > 0 the energy space is V = # 0 2 (fi),
(15)
(16)
so that the classical formulation of problem Pe (6) is : ( - A + £2A2) ue = /
inH,
(17)
£
v? = dnu = 0 on 90,
(18)
where dn denotes the normal derivative on the boundary dQ. Obviously, using Poincare's inequality for w and its first order derivates in the bounded domain Q,, we have C\\w\\2v>a{w,w)
+ e2b(w,w)>ce?\\w\\l,
VweF,
(19)
so that the coerciveness constant tends to 0 as £ tends to 0. Clearly ||.|| a is a norm on V and the completion of V with respect to this norm is V* = H] (fi)
(20)
and the classical formulation of P0 (9) is -Au = f
in
u = 0 on dQ,.
ft,
(21) (22)
Let us consider the right hand side : f{xux2)
= 8'(x2)F{xl),
(23)
where 6 denotes the Dirac mass and F e I? (0,7r), so that (f,w) = -fF(x1)d2w(x1,0)dx1 o
(24)
465 and using the trace theorem \{f,w)\<\\F\\0\\d2w(.,0)\\o
*>3/2.
(25)
In addition, no analogous inequality holds for s < 3/2, so that f£H-2{Q) = V
• ftIT1(n)
= Va
(26)
and we are in the framework of Theorem 1.1 ii). Let us consider, for e > 0 the change : xi=yi
,
x2 = ey2
,
uE(x) =v%y).
(27)
Denoting Di = d/dyi
i = l,2,
(28)
the problem Pe (6) with (14), (15), (23) becomes the new one Ve : Find vE e Hi (BE), Be = (0, TT) X ( - 1 / e , 1/e) such that \fw 6 Hi (BE) : a0 (vc,w) + e2ax {vc, w) + eia2 (ve, w) = ( $ , w),
(29)
where OQ{V,W) ai(?;,w) a2(v,w)
=
/ D2vD2wdy + / D2vD2wdy, J Bc J Be = / {D1vD1w + 2D1D2vD1D2w)dy, J Be = / D\vD\wdy. J Bc
(30)
7T
{<S>,w) = -jF(y1)D2w(y1,0)dy1. o
(31)
R e m a r k 2.1 Obviously, vE and w in (29) may be extended with values 0 for | y2 | > 1/e, so i/iai i/iey are considered as functions on Bo = (0, IT) xH. As a consequence, integrals in (30) may be considered on B0. We are now constructing an energy space adapted to the "limit problem" of (29) as £ —* 0. We observe that the left hand side of (29) with e = 0 only involves the first and second order derivatives with respect to y2 so that it "ignores" additive functions of y\. The same holds for the right hand side. Let us denote by Hl(—l/e, l / e ) / R the space of the functions of class H2 vanishing for | y2 |> 1/e defined up to an additive constant (i.e., for each w S Hl(—l/e, 1/e) extended with value zero for | y2 |> 1/e, we consider the set of functions which differ from it by a constant, and we consider this set as an element or equivalence class). We then construct the space V = C | J L 2 ((0, n)m • Hi (-1/e, 1/e) / R ) ,
(32)
466 where U denotes the union and C the completion with respect to the norm (constructed from ao) :
\\v\\l = j[{D2vf+{Dlvf]dy.
(33)
Bo
It should be noticed that the completion implies loss of the property that the functions take the same value as y2 tends to +00 and —00 (see an example in (44)). We are now able to define the new limit problem Vo • Find v eV such that \/w e V : a0(v,w) = (*,w).
,„ ..
Obviously, by virtue of (30), (33) and Remark 2.1, the bilinear form at the left hand side of (34) is continuous and coercive oh V. Moreover : Lemma 2.2 The right hand side of (34) (see (31)) is a continuous functional on V. Proof. Immediate, as the functional is independent of additive functions of y-y. The continuity follows from the trace theorem for dw/dy2 in (for instance) /f 1 (—1,+1) and integration in j/i. • As a consequence, problem Vo has a unique solution in V (but as a function, it is only defined up to an additive function of y\). Remark 2.3 For each element v of HQ (Be), we may construct its extension with value zero to the strip BQ and consider it up to additive functions of y\. The equivalence class constructed in this way will be denoted by v, and is obviously an element ofV. We then have the convergence property : Theorem 2.4 Letv£ be the solution ofPE (29) andve the corresponding equivalence class defined according to Remark 2.3. Then : ve —» v
strongly in V,
(35)
where v is the solution ofVo (34). Proof. Taking in (29) w = ve and using Lemma 2.2, we obtain the a priori estimates :
WWv < C, ai(ve,ve)1/2
e l2
2
a2{v ,v f
(36) (37) (38)
Because of (36), there exists v* e V such that (at least for a subsequence) ve -> v* weakly in V.
(39)
467 Let us check that v* coincides with the solution v of (34). To this end, we fix w belonging to Hfi (Be) for some E\. Then, after extending it with value zero, it also belongs to H$ (Be) for e < £i and may be used as test-element in the corresponding Ve. Using (39) and the estimates (36)-(38) it follows from (29) that ao(v*,w) = (^,w),
(40)
but obviously, we may take in (40) w as well as w (see Remark 2.3) and according to (32) the test-functions so considered form a dense set in V, so that v* is the unique solution v of (34). In order to prove the strong convergence in (39) (which clearly holds for the whole sequence), we write the expression e V (ve, ve) + 6*0,2 {ve, vE) + a0 {vE - v, vE - v), £
(41)
e
which, using (29) with w = v , and w = v and (34) with w = v is equal to : <$,t/) + ( < M ) - 2 ( < M e > ,
(42)
which tends to zero by virtue of (39). In particular, the term do of (41) tends to zero, and the theorem follows. • Let us now solve the limit problem (34). Obviously, it is an ordinary differential equation in y2, with parameter y\. Denoting V = D^v and taking in (34) w obtained according to Remark 2.3 from w €T> (R), we get (-D2 + D32) V =
F(yi)6'(y2),
or (-l + Dl)V = F(yi)6(y2) + C,
(43)
2
where C is a constant. But V £ £ (R) is for j / 2 > 0 and y2 < 0 solution of the differential equation (43) with F = 0. It follows that C = 0 and V exponentially decreases at infinity. It then follows from (43) that V(y2) is even and more precisely : V(y1,ya)
=
- ^ e - M ,
so that v, defined up to an additive function of 3/1, may be represented by the odd function in 2/2 v (yuy2) = ±1-'f±
(1 - e-lwl)
for y2 < 0 (resp. y2 > 0).
(44)
Obviously, it is a piecewise smooth function tending to ±F{y{)/2 as y2 tends to ±oo having at y2 = 0 a jump of D\ equal to —F(yi). It should be considered up to an additive constant. Remark 2.5 Obviously Theorem 2.4 amounts to a property of convergence of functions °fyi> 2/2 v,p to an additive function ofyi or, equivalently, a convergence of the first order derivatives with respect to y2. On account of the change (27) and the form of the limit
468 function in (44) the functions ve(xi,x2/s) converge (always up to an additive function of Xi) to a step function of intensity —F(xi) at x2 = 0. On the other hand, the limit problem -Au = F(Xl)S'(x2),
(45)
enjoys classical elliptic regularity of order 2 inside the domain Q.. The right hand side of (45), with F e L2, belongs to HS(D,), s < —3/2, as follows immediately from trace theorems. Consequently, u locally belongs to Hr(Q), r < 1/2, which is consistent with a step function along x2 = 0. Moreover, an asymptotic expansion of singularities analogous to those used in [9] which are based on [3], shows that the leading order singularity of the solution u is effectively a step at x2 = 0 of intensity —F(xi). Theorem 2.4 may be seen as a description of the asymptotic formation of that step as e tends to zero. Remark 2.6 All the considerations of this section hold in the case where the laplacian A in (17) is replaced by d2/dx\. (46) In the case (46) the limit problem is no longer elliptic, but the differentiations are transversal to x2 = 0, so that it is microlocally elliptic of order two for £ = (0,1), i.e. along the considered singularity. Microlocal regularity theory then holds classically [2] and all this section, including Remark 2.5, holds.
3
Model problem for singularities along t h e characteristics of t h e limit problem, Pe elliptic of order 4, Po parabolic of order 2
In the same domain Q as in section 2, we now consider a(u, w) = I d\ud\wdx, Ja
(47)
b(u, UJ) = / dijudijwdx,
(48)
Jn f(x1,x2) with some F € L2(0,n).
= F{x1)ff(x3),
(49)
For e > 0 the energy space is V = H2(Q) so that Pe (6) is : {-d\ + e 2 A 2 ) u£ = F (Xl) 6' (x2)
in
ue = diM£ = 0 on dSL Clearly ||.||o is a norm on V, and the corresponding completion is K = L2((-1,+1),2 ; ^ ( 0 , 0 ,
fi,
(50) (51)
(52)
so that the limit problem is P0 : -dlue = F{xl)8'{x2)
inQ,
(53)
469 u = 0 o n i ! = ±1,
(54)
which is elliptic in X\ with parameter x2 or equivalently parabolic in (x\, x2) with double characteristics x2 = const. We note that the right hand side (49) is singular along the characteristic x2 = 0. Obviously, / in (49) is not in the dual V'a. Let us consider for e > 0, the change e = rf , xL = y! , x2 = ny2 , A = d/dy,, ue (x) = -v"
(y),
(55) (56)
which is suggested by the fact that the solutions of Po (which are not variational) are of the form S'(x2)V(xi). Making the change in PE for the unknown and the test functions, this problem is equivalent to V,, : Find v7* £ Hi (Bq), 5„ = (0,TT) X (—1/77,+1/TJ) such that : Vw € H§ (Br,) : a0 (v", w) + rj2^ (v", w) + rfa2 (v", w) = ($, 1
(57)
with a0(v,w) ai(v,w) a2(v,w)
=
/ DivDitudy + / D\vD\wdy, J Br, J Be = 2/ DiD2vDiD2wdy, J B, = / DivDiwdy
(58)
and the right hand side of (57) is again (31). As in Remark 2.3, we always consider functions of Hi (£?,,) extended with value zero to H2 (B0) but this time, we do not consider equivalence classes. We then construct the space
V = C [Jfl2(B,),
(59)
where C denotes the completion with the norm (see (58))
\Mv = ao(v,vf/2.
(60)
We then have : Lemma 3.1 In V holds the equivalence of norms : llt=(R V2i H o 1 (0,ir)» 1 )
+•
^((CTr^iH^R,,))'
(61)
470 Proof. The square of the norm of V is obviously less than a constant multiplied by the right hand side of (61). Let us prove the opposite. Using Poincare's inequality in Hi(0,7r) and integrating in y2, we see that the square of the norm in V is more than a constant multiplied by the first term of the right hand side of (61). It only remains to prove an analogous inequality for the second term in the right hand side of (61). But it follows from the previous considerations that
ll-llv>C||.|| L 2 ( R v 2 ; L 2 ( o,. ) v i )
(62)
and obviously \Mv>C\\Dlv\\LHRs2]L2(0ir)yiy
(63)
2
Moreover, in H (R) the norms IMI^(R)
and
2
IMI^(R) + ||£> HlL(R)
are equivalent (this is a classical result, which holds for bounded as well as for unbounded intervals, see [7], Theorem 2.3, p. 19). Then, adding (62) and (63) the conclusion follows.
• In addition, classical methods of translation and convolution with a smooth function ([7], p. 14) show that Lemma 3.2 We have the identity between spaces : V = L2 ( H ^ J H J I O , ^ ) f]L2
((0,7r) yi ;fl a (H, B )) .
(64)
Remark 3.3 We note that v e V implies that the traces of v on yi = 0 and yi = n vanish, but the trace of D\V does not vanish in general (compare with H^B^)). The process of completion to construct V implies a loss of principal boundary conditions at yi = 0 and y\ = IT. Lemma 3.4 The functional ($,w) defined in (31) (see also (50)) is continuous on V. Proof. Immediate, using Lemma 3.1 and the trace theorem in H2 (Rj,2).
•
As a consequence, the following limit problem VQ is a Lax-Milgram problem, enjoying existence and uniqueness : Find D S V such that V w e V : a 0 (f,w) = ( $ , w ) .
, . *• '
We then have the convergence theorem : Theorem 3.5 Let vn and v be the solutions of (57) and (65) respectively. Then : vn —> v
strongly in V-
(66)
471 The proof of this theorem is exactly analogous to that of Theorem 2.4 and even simpler, as equivalence classes are not involved. It will not be given here. Let us consider now the limit problem Vo (65). It is not hard to solve it by separation of variables, as it involves functions (and no longer equivalence classes, as in section 2). The equation associated with the variational formulation (65) is clearly (-D\ + B}) v = F(yi)6'(y2)
on B0 = (0, TT) X R.
(67)
We add the boundary conditions w(0,lft) = «(jr,ift) = 0, v (2/1,3/2) tends to 0 as \y2\ —> 00.
(68) (69)
Obviously, conditions (68) are contained in the definition of the energy space V (30). As for (69), they will be understood in the sense of the exponential decreasing for the Fourier components involved in the separation of variables process (see later on) ; they are consistent with the definition of V and allow to construct the unique solution of (65). Let An and wn{y1) be the eigenvalues and eigenvectors (normalized in L2(0,7r)) of —D\ with the boundary conditions w„(0) = wn(n) = 0.
(70)
Let OO
v = 5Z V n (&) Wn (y^'
( 71 )
n=l 00
F = Y, FnVn (Vl) ,
(72)
n=l
be the expansions of the unknown v and the datum F respectively. Equation (65) amounts to A»«» ( » ) + D\vn (jft) = FJ (a,). (73) The distribution 5' at the right hand side of (73) may be described in terms of the jumps. Denoting [a] = a(0+) — a(0—) :
K ] = 0, [D2vn] = 0, [Z?K] = F, lD32vn] = 0. The function vn (y2) are odd and satisfy : vn(0+) D22vn(0+)
= 0, = F/2
472 and we have explicitly ' un (2/2) is odd, < un (y2) = -^—
(e"*» - e ^ » )
for y2 > 0,
(74)
, where a± = 2 ^ / 2 A i / 4 ( _ 1 ± i ) We then have t h e explicit expression of the solution, which is exponentially decreasing for I2/2I - * 00. It should be mentioned that, as the exponents are not real, the graph of un (2/2) cuts t h e axis of abcissas infinitely many times. R e m a r k 3.6 The same problem may be considered with a singularity the right hand side, namely f{xux2) = F(x1)6(x2):
of lower order in
which does not belong to the dual of Va defined in (52). The developments to the previous ones, but (56) is replaced by uc(x)
are analogous
= -v^(y)
(75)
and the right hand side of (57) is presently TV
($,«,) = /
F{y1)w(y1,0)dyl.
0
Obviously, in the limit problem (67), 6' must be replaced by 6. Its explicit solution obtained as before and is now an even function in y2.
4
is
Layers along a boundary with a N e u m a n n boundary condition
In this section, we consider the previous problems in the domain n = (0,jr)x(0,l),
(76)
instead of (13). T h e boundary conditions will be of Neumann type on x2 = 0, whereas they remain unchanged on the rest of the boundary. We first consider (14), (15) on t h e space V = {v | v e H2(Q),
v = dnv = 0 on xx = 0, x1 = TT, X2 = l }
(77)
and the functional of the right hand side is still given by (24). This amounts to ( - A + e 2 A 2 ) u£ = 0
infi,
(78)
473 with the Neumann boundary conditions : (-<92 + 2e2<912d2 - e2<92) uc = 0 o n i 2 = 0, 2 2 e e
du
= -F
oni2=0
(79) (80)
and of course (18) on the rest of the boundary. After the change, the problem Ve is analogous to (29) but in the space {v | v G H2 (Be), v = dnv = 0 on yl = 0, y1 = TT, y2 = 1/e) B £ = (0,7r)x(0,l/e). Moreover, (32) is replaced by
V = C ULJ((M)B ;
ff2(0,l/e)/R),
(81)
e
where H2 (0,1/e) / R denotes the space of the functions of H2 (0,1/e) extended for y% > 1/e with a constant value equal to the trace at y2 = 1/e and considered up to an additive constant. Finally the solution of the limit problem is, instead of (44) : v(yi,y2) = F(yi)(l-e-y>).
(82)
The problem of section 3 admits an analogous treatment in the domain (76). In particular, Lemmas 3.1 and 3.2 hold, after replacing R ^ by the half-axis y2 > 0. The convergence Theorem 3.5 holds true, and the solution of the limit problem is merely twice the expression (74) for y2 > 0.
References [1] Eckhaus W., Asymptotic analysis of singular perturbations. North-Holland, Amsterdam, 1979. [2] Egorov Yu. V. and Shubin M.A., Linear partial differential equations. Foundations of classical theory. In : Encyclopaedia of Mathematical Sciences (Part. diff. Eq. I), Springer, 1992, 30, p. 345-375. [3] Gerard P., Solutions conormales analytiques d'equations hyperboliques non lineaires. Commun. Part. Diff. Eq., 13 (1988), 345-375. [4] Gerard P. and Sanchez-Palencia E., Sensitivity phenomena for certain thin elastic shells with edges. Math. Meth. Appl. Sci., 23 (2000), 379-399. [5] IPin A. M., Matching of asymptotic expansions of solutions of boundary value problems. Amer. Math. Soc, 1991.
474 [6] J.L. Lions, Perturbations singulieres dans les problemes aux limites et en controle optimal. Springer, Berlin, 1973. [7] Lions J.L. and Magenes E., Problemes aux limites non homogenes et applications. Vol. 1, Dunod, Paris, 1968. [8] Sanchez-Palencia E., On a singular perturbation going out of the energy space. J. Math. Pures Appl., 79 (2000), 591-602. [9] Sanchez-Palencia E., Propagation of singularities along a characteristic boundary for a model problem of shell theory and relation with the boundary layer. Compt. Rend. Ac. Sci., ser. lib, 329 (2001), 249-254. [10] Vishik M.I. and Lusternik L., Regular degenerescence and boundary layer for linear differential equations with small parameter. Usp. Mat. Nauk, 12(5) (1957), 1-122.
Regularity and uniqueness results for a phase change problem in binary alloys* Jean-Frangois Scheid1, Giulio Schimperna2 1
'
Institut de Mathematiques Elie Cartan, Universite de Nancy 1 F-54506 Vandoeuvre-les-Nancy Cedex, Prance 2 Dipartimento di Matematica, Universita di Pavia Via Ferrata, 1, 1-27100 Pavia, Italy Email : [email protected] ; [email protected]
Abstract An isothermal model describing the separation of the components of a binary metallic alloy is considered. A phase transition process is also assumed to occur in the solder; hence, the state of the material is described by two order parameters, i.e., the concentration c of the first component and the phase Held ip. Existence of a solution to the related initial and boundary value problem has been proved in a former paper, where, anyway, uniqueness was obtained only in a very special case. Here some further regularity and uniqueness results are shown in a more general setting using an a priori estimates - compactness argument. A key point of the proofs is the analysis of the fine continuity properties of the inverse map of the solution-dependent elliptic operator characterizing one of the equations of the system.
1
Introduction and mathematical preliminaries
In this paper, we aim at presenting some regularity and uniqueness results for t h e system dttp-Atp dtc — div (ji (tp, c) Vw) w
= F1{ip) + c + F2{^), = 0, e -Ac + (3 (c) + 7 (c) + g (tp),
(1)
describing the diffusive separation of components in a binary metallic alloy possibly undergoing a phase transition phenomenon. In t h e above relations (1), t h e unknown c represents t h e relative concentration of either of the components, while tp is the phase parameter, with
475
476 paper [4]. Finally, /i is the mobility coefficient, possibly depending on both the unknowns, but assumed to be nondegenerate, and (3 is a maximal monotone graph guaranteeing the "physical" constraint 0 < c < 1. Let us point out that the system (1) has to be complemented with the Cauchy conditions for ip and c and with homogeneous Neumann boundary conditions for
V^.= {CeV:Cn
= 0},
Let also 0 < a < fi0 be assigned constants and let fi 6 Lip ioc (R 2 ), with a< fi
V0:=VnVi.
a.e. in K2.
(3)
(4)
If v, z : Q —> R are measurable functions, then we can define the operators B(VlZ) : V -> V,
(BMu,
y) := / p. (v, z) VM • Vydx, for u, y € V, J n
(5)
477 B:V^V,
{Bu,y) := / Vu • Vycte, for u,yeV. (6) J n Clearly, B and f$(„,z) map V onto VJJ and their restrictions to VQ turn out to be isomorphisms of V0 onto VQ. Then, we can denote by Af the inverse of B and by Af(v,z) the inverse of -B(„jZ). Just by applying the definition (5), one can readily check that for any u € V and ( e VJ there holds = (B(v,z)NMt,u)
(BMu,N(?,z)0
= (C,u> .
(7)
Let now p £ [l,oo], u, Z be measurable functions on Q, and u, £ be measurable functions of time with values in V and VQ, respectively. Then, for a.e. t € (0,T), we can put (BMu)(t) := B(v(t)iZ(t)){u{t)) and (A/"(„iZ)C)(£) := A/(„(i),*(t))(C(<))- This provides a natural extension of the above operators to a time-dependent setting. Prom this point on, k will stand for a positive constant (possibly not always the same) depending only on a, fiQ, O. A positive constant depending on one, or more additional parameters (say 6), will be noted as kg, instead. We have (see [4, Sec. 2] or [1, Sec. 2] for the proof, which is essentially based on the Poincare-Wirtinger inequality): Lemma 1.1 For all ( £ V0' and all measurable functions v, z : Q —> M, we have ||JW||V ^
k
IKHv
md
(8)
Moreover, ifv, z are measurable functions of Q into K, p £ [l,oo], then BM
: 77(0, T; V) - V^T'X)
and MM
: i7(0,T; V0') - L"(0,T; V0),
introduced as noted above, are well defined. In addition, the S(^,z) are (surjective) isomorphisms of LP(0,T;V0) onto 1^(0,T; VQ). Finally, we have \\B(v'z)\\c(LP(0,T;V),LP(0,T;V^)
—k
and
I I ^ W ) \\c(LP(p,T;V£),W(0,T;Vo)) —
fc
>
^>
i.e., their norms do not depend on (v,z). A consequence of (8) and of the compact embedding V CC H is that, if 6 > 0, C, £ Vo, C — B(„]Zju for a suitable u £ V, then Kl2 < «|VC|2 + fcs Ut,
< <5|VC|2 +
fc,|V«|2.
(10)
Now, let us analyze some deeper continuity properties of the operator Af(v,z)- We report the proofs, which follow the lines of [1, pp. 493-494], for the sake of completeness: Lemma 1.2 Let v, z £ H2(fl), f 6 ii" n VJ,', and u £ V be such that £ := B(V
(11)
478 Proof.
Since £ £ H, our hypotheses on v, z, (J, guarantee that u is in H2(£l) and satisfies —div(/i(v, z)Vu) = £, in H.
We deduce that : —n(v, z)Au = (Vfi(v, z)) • Vu + ( in H and consequently |-Aw| 2 dz
/
(\Vv\2 +
\Vz\2)\Vu\2dx.
Now, using elementary interpolation, Sobolev's embedding theorem, and Young's inequality, we obtain 2
VI
IMI tf («)
VI
k(\u\2 + \-Au\2) k\C\2 + k(\\vfH^) +
\\Z\\H'(CI)J
|V«|
\H\H2{n)
< k\C\2 + k(\HiHHn) + IMIjfCn)) |V«|2 + \ ||«||^ ( n ) Lemma 1.3 Lei u, w, z, ( be functions from Q into R, wifft £ = B(V^)U, a.e. in (0,T). Then, the following formulas hold whenever they make sense : i
((BMu)t,u)
i
/»
= -—{BMu,u)+
n(v,z)t
({B(VlZ)u),ut) = -—(B(ViZ)u,u)-
|Vu| 2 dx,
fj,(v,z)t\Vu\2dx.
(12) (13)
Proof. Let us assume that all the functions are sufficiently regular to give sense to the integrations by parts below. Then we have ((•B(„,z)u)t, u)
= — {BMu,
u) - (BMu, 2
=
/
dt{ii{v,z)\Vu\ )dx-\
=
2/
dtu) n(v,z)dt\Vu\2dx
[
<9t(M^)|Vu| 2 )
•/It
^,z)t|Vu|2dz,
J it
so that (12) is proved in the regular case and in general we can conclude using a density argument. The proof of (13) is analogous. •
2
Main results
Let us give the main assumptions on the data of the problem. Let K > 0 and \Fi\ • l-Pal. M , Ifll, \F{\, \K\, W\, \g'\ < K a.e. in R,
>Po e H, co e y, /3 C 1 x R is a maximal monotone graph such that 0 £ /? (0), c n £ intD (/?), where ca = (ca)\n •
^ '
479 Here t/> : K—• [0, +00] is a convex and lower semicontinuous function such that 0 = dip and i> (0) = 0 and the domain of the graph (3 is required to fulfill £(/3):={reR:/?(r)^0}c[O,l].
(15)
Then, the following existence-regularity result [4, Thms. 3.1, 3.3, 3.4] holds : Theorem 2.1 Let assumptions (4), (14)-(15) hold. Then there exists a quadruple Up, c, w, f) such that c G H1 (0, T; V) n L°° (0, T; V) l~l L2 (0, T; if2 (fi)), «J6L2(0,T;F), £eL2(0,T;H).
(16)
The quadruple (tp,c,w,£) satisfies dtip + B
= = = € = =
Fi (p) + cF2 {ip) 0 Be + £ + 7 (c) + 5 (>) /3(c) 30(.), c(.,0) = co(.) cn
in V, a.e. in (0,T), in V, a.e. in (0,T). inV, a.e. in (0,T). o.e. in Q, o.e. infi, /oraW*e[0,T].
(17)
If in addition ¥>o e V,
(18)
i/ien > satisfies
C° ([0,T]; V) n L 2 (0,T; tf 2 (H))
(19)
5„v?0 = 0 a.e. onT,
(20)
and i/
tp e w1'™ (0, T; H) n ff1 (0, T; V) n L°° (0, T; ff2 («)) f l l 2 ( 0 , T; # 3 (fi)) . Finally, just under the assumptions (4), (14)> if M (fic) (ip,c,w,^) is unique.
(21)
*s constant, then the solution
Under more restrictive hypotheses on the data, we are able to prove additional regularity properties of the solution(s). We will possibly assume fi = \i Up) only depends on
new '™, ||MlU-
, , {
'
Theorem 2.2 Let (4), (14)-(15), (20) and (22) hold for a couple of initial data ipoti, Co,i and ?o,2) co,2 and assume that (cb,i)n = (co,2)„ • (23)
480 Moreover, suppose that
dm, \M\LH0,T;H)
%eL4(0,T;ff),
+ \\dt
for a given R > 0. TTien i/ (<^I,CI,ID 1 ,£I) and (
7n particular, the solution (
ft.
(25)
Then, there exists TQ € (0,T] such that the solution (ip,c,w,£) given by Theorem 2.1 fulfills the additional regularity c € H1 (0, T0; V) n i°° (0, T0; ff2 (ft)),
» e r (o, r0; y) n L2 (O, ro; # 2 (ft)).
,
. (M)
Moreover, if (22)\ holds, we can choose TQ = T. Remark 2.5 Thanks to the first of (26)\, under suitable hypotheses on <po (e.g. ipo € H3 (ft) and dnifo = 0, on T), further parabolic regularity could be deduced also for (p by differentiating (17)\ and multiplying the result by du
(27)
for a given R > 0. Then we have IIVl
_
f2\\L^(0,To;V)nL^(0,T0]H^(n))
+ ll C l ~~ C2|lz,»«(0,To;V')nL2(0,To;^)
(28)
< kR (||ipo,i - Vo,21|v + l|co,i - co,2|lv) • In particular, the solution (p,c,w,£)
given by Theorem 2.1 is unique on [0,To].
Remark 2.7 As before, (27) are not satisfied in the given regularity setting. However, if (20) and (25) are assumed, then (27) do hold at least up to a suitably small final time T0.
481
3
Proofs of the Theorems
Proof of Theorem 2.2. Set (ip,c,w,£) := {
(29)
Doing the same with (17)2,3, we have, respectively dtc +Bviw + (BV1-BV2)w2 = 0, w = Be + £ + 7 (ci) - 7 (c2) + g (fi) - g (p 2 ) •
. .
Then, we test (29) by if + Bip and integrate over (0, t) for t < T. By (14) lj2 , (15) and Young's inequality, we easily infer \ llv Wily + II V ^ l i 2 ( Q t ) + ^ l|B>ll£»W,) < 2 IIVo.1 " Vo,a||v + k \\
Vudxdtdx + \\Vc"2LHQt)
(32)
Let us deal with the first two terms I\ and h on the left hand side of the preceding inequality. For the first one, by (5), (8) and (22)2, we have \ (
h
>
n{^{t))\Vu{t)\2dx-l-[
fi&0tl)\Vu(0)\2dx
+1 [ [ v'(
(33) /
\dm\\Vu\2dxdt.
Then, using (11) and a Galiardo-Nirenberg inequality [5, p. 125], we infer |2 / / \dtipiWVu\Uxdt J oJ a
{\\u\\%\n) IVu| 1/2 + IVu| 2 ) dt
< kj
\dm\
< kj
\dtVl\ \Vu\2dt + kj^\dm\
< f \dm\\Vu\2dt Jo
+ k\\c\\2+k
\Vu\x'2 (|c| 3/2 + ||Hl5p ( n) |Vn| 3 / 2 ) dt f (Ift^il 4 + | S V l | 4 + l^l 4 ) |V«| 2 dt. Jo
(34)
482 The second term in (32) gives, thanks to the continuous embedding H2 (Q) C L°° (CI)
\h\ < k f
IML~(n) IVtual | V«| dt < | J* (|^| 2 + \B
\Vw2\2 \Vu\2 dt.
(35) Now, take the sum of (31) and (32) and note that (10) can be applied with c in place of (, so that the L2-norms of c on the right hand sides of (31) and (32) are controlled by taking an opportune 6. Taking advantage of (33)-(35) and recalling assumptions (24), we conclude applying Gronwall's Lemma in the form of, e.g. [3, Lemme A.4, p. 156], to the function t i-> |<^ (t)||J. + |Vu(i)| 2 . • Proof of T h e o r e m 2.4. We derive new a priori bounds for the solutions of (17). We point out that, at this level, some passages are formal. Anyway, they could be made rigorous by performing the estimates at an approximating level (cf. e.g. the Faedo-Galerkin argument of [4, Sec. 3]). Then, differentiate (17)3 with respect to time. This gives dtw = dtBc + dt£, + 7' (c) dtc + g' (
(36)
Now test (17)2 by dtw and (36) by dtc and integrate over (0,i), t < T. Noting that two terms cancel and using (6), (14)i]2, (22)2 and the monotonicity of /3, we get \\dtVc\\l2Kt)
^\Vw(t)\2<^\Vw(0)\2
+
+k Jo J
((\dM + \8tc\) \Vw\2 + k \\dt
Of course, the main trouble is given by the integral term J on the right hand side. Using Sobolev's embeddings, (11) and the Gagliardo-Nirenberg inequalities, we infer
(\\dM\v + \\dtc\\v)\\Vw\\2L12/Hn)dt
J < kj
< k•Jf (\\9M\v (\\dM\v + \\dtc\\ (\Vwf22 ||H|$ \\w\\U( 2n ) + |VH \Vw\22)) •dt c\\vv)) (\Vwf (38)
< k f (\\dM\y + \\dtc\\v) J 0
x ( | V H 3 / 2 \dtc\1/2 + \Vw\2 ( l + \\
where the integrals Jj are defined in this way : Ji
:= k [ \\dM\v\VMi/2\dtc\1/2dt,
J3 := kj
J2 := k f
\\dtc\\v\Vw\V2 \dtc\1/2 dt,
||3^|| v |V W | 2 (l + | M I ^ ( n ) ) ^ J* •= * / ||^|lv|Vw| 2 ||c| H (fl)
J5 := kJj\dtc\\v\Vw\2(l
2
+ MH2m)dt,
J6 := kj
||a ( c|| v |VH
C
"*>
IIH 2 (SJ) ^*-
483 Now we estimate the Jj's. The term J 3 does not give troubles. Instead, Jj, J 2 , J4, J5, owing to the third of (21), are bounded as follows:
k [ \Vw\2\\dtlp\\y3dt + k f \dtc\2dt, J Q
h
Jo -9
Ji < kj h
J 0
< 6 [ \\dtcfvdt + k6 f
J 0°
\Vwfdt,
\Vw\2\\c\\*H2{Q)dt + kJ
t
\Vw\2\\dM\fdt,
(39)
< S f \\dtcfvdt + ks [ \Vw\4dt. Jo Jo
The estimation of the latter term J 6 , instead, is more complicated and we proceed similarly as in [1, p. 497]. More precisely, we collect in the next computation also the first term resulting from J 4 on the right hand side of (39)3. By elliptic regularity :
Je + k [
\Vw\2\\c\\lHHn)dt
J 0t t < S I \\dtc\\2v dt + ks f (|VH 4 (|c|2 + |Bc| 2 ) + \Vw\2 (|c|4 + |Bc| 4 )) dt. Jo Jo
(40)
Then, we multiply (17)3 by Be. Owing to the monotonicity of /3, to (14)^2, and to the second of (16)2, we easily infer, a.e. in (0,T) \Bc\2 < k (1 + | V « f + |Vc| 2 ) < k (1 + |Vu;|2) •
(41)
Then Je + k J
\Vw\2\\c\\4H2(n)dt<6
J
\\dtc\\l dt + ks j
{\Vw\2 + \Vw\6)dt.
(42)
Finally, we have to bound the norm of w (0) in (37). Writing (17)3 for t = 0 (this could be made rigorous working on the discrete scheme) and recalling (25), we see that \Vw (0)| 2 < k ( l + | h | | ^ ( n ) + Uofv + \\
(43)
Now, let us take (38)-(43) into account and go back to (37). If (22)x holds, the situation is simple since J 2 = J4 = J5 = Je = 0. Actually, we can apply (10) with dtc, w, tp
in place of
£, u, v and z,
(44)
respectively. Thus choosing 8 suitably small and recalling (21), we see that Gronwall's lemma applies to the function t i-> \Vw(t)\2. This yields Vui € L°°(0,T;H) and also to the first of (26)i for T0 = T. The second of (26)i is a consequence of (41). Now the first of (26)2 would easily follow once we show that w 6 L°° (0,T0; H). Due to Neumann conditions, this is not obvious and has to be proved by suitably adapting the argument of e.g. [2, Subsec. 5.3]. At this point, also the second of (26)2 is proved since, looking
484 back at (17)2, we can exploit (11) with the choices (44) and take advantage of the third of (21) and of the second of (26)i. Instead, if (22) does not hold, we deduce from (37)-(43) that
liavc|kwo + |v™(t)|2 < k + kj* (/|VH 2 + |v
rfi,
where / := (l + ||%>||y 3 ) e 1/(0, T), for some p > 1. Then, using a generalized Gronwall's lemma (e.g. a slightly modified version of [6, Thm. 7.1] works for our case), we obtain the same relations as before, holding now up to a final time To 6 (0, T]. B Proof of Theorem 2.6. It is very similar to the proof of Theorem 2.2. Then, we just give the highlights. Proceeding as in that proof (but choosing now t 6 (0,T0j), we see that the last term in (33) now depends also on dtCi. However, it can still be estimated as in (34) by taking the contribution of dtc\ into account. Thus, the last term in (34) will now depend on all the norms in (27)2 but the latter. The bound (35) needs to be modified more carefully. Actually, we now have / k
J
/
(/•'(VijCi) - M ( V 2 , C 2 ) ) VM;2 • V« dxdt
\M\L°°(n)\V™2\\Vu\dt + kJ ||c|| Le(f!) l|Vw2||L3(fj) |Vu| dt.
Thus, the first term can now be bounded as before, while the second one is less than ^ IICIIL2(O T-V) + ^« Jo II Vw2||L3/m I V«| dt, so that, on account of the last of (27)2, a further application of Gronwall's Lemma [3, Lemma A.4, p. 156] permits us to conclude. •
References [1] J. W. Barrett and J. F. Blowey, Finite element approximation of the Cahn-Hilliard equation with concentration dependent mobility. Math. Comp., 68 (1999), 487-517. [2] E. Bonetti, W. Dreyer, and G. Schimperna, Global solution to a viscous Cahn-Hilliard equation for tin-lead alloys with mechanical stresses. Submitted. [3] H. Brezis, Operateurs maximaux monotones et semi-groupes de contractions dans les espaces de Hilbert. North-Holland Math. Studies, 5, North-Holland, Amsterdam, 1973. [4] D. Kessler, J.-F. Scheid, G. Schimperna and U. Stefanelli. Study of a system for the isothermal separation of components in a binary alloy with change of phase. Paper in preparation. [5] L. Nirenberg, On elliptic partial differential equations. Ann. Scuola Norm. Sup. Pisa (3), 13 (1959), 115-162. [6] M. Sassetti and A. Tarsia, On a nonlinear vibrating-string equation. Ann. Mat. Pura Appl. (4), 161 (1992), 1-42.
Bifurcation in population dynamics Kenichiro Umezu Faculty of Engineering, Maebashi Institute of Technology 460 Kamisadori, Maebashi 371-0816, Japan Email : [email protected] Abstract This paper is devoted to the study of a local bifurcation problem for a nonlinear elliptic boundary value problem arising in population dynamics, having nonlinear boundary conditions. Necessary and sufficient conditions for the existence of bifurcation points are studied. Ecological interpretations of our results are also given.
1
Introduction
Let D be a bounded domain of B.N, N > 2, with smooth boundary dD. In this paper we consider the following nonlinear elliptic boundary value problem: - A M = \(m(x) — au)u,
in D,
du — = b(x)g(u),
on dD,
(1)
where A is a positive parameter, m e Ce(D) is a real-valued Holder continuous function with exponent 0 < 6 < 1 on the closure D of D, and satisfies : m(xo) > 0, at some point XQ € D, a is a given positive constant, b 6 C1+e(dD), b > 0 and b ^ 0 on dD, g is a real-valued sufficiently smooth function near t = 0 which satisfies g(0) = 0, and n is the unit exterior normal to dD. It is well-known (cf. [3, 4]) that the equation : —AM = \(m(x) — au)u, in D, describes the steady state for the population density of some species which diffuses at the rate 1/A, where m(x) denotes its growth or decay rates at x G D, meaning that the regions {x € D : m(x) > 0} and {x e D : m(x) < 0} are respectively favorable and unfavorable for the species, and a represents the crowding effect! The nonlinear boundary condition du/drx = b(x)g{u) on dD means that the rate of the flux of population into the region D is governed in a nonlinear way by b(x)g(u) on the boundary dD. From this viewpoint, the existence of positive solutions is of ecological interest. Here a solution u 6 C2(D) of (1) is called positive if u > 0 on D. We assume here for g that g(t) is non-negative for all t > 0 close to the origin.
(2)
Ecologically speaking, assumption (2) means that only the influx of population can occur at the boundary, if its density is sufficiently small there.
485
486 In the Neumann case, that is g = 0, Hess proved in [6] the existence, uniqueness and asymptotic stability of positive solutions and also determined the qualitative behavior of the unique positive solution when parameter A runs, especially when A J. 0. By fJ.i{\), we denote the unique principal eigenvalue of the eigenvalue problem —A
— = 0,
on dD,
(3)
where principal eigenvalues mean eigenvalues with corresponding positive eigenfunctions. Brown and Lin proved in [2] that there exists a unique and positive Ai(m) such that /u1(Ai(m)) = 0, provided fDmdx < 0. Our consideration starts with the following result. Theorem 1.1 (Hess [6]) Let g = 0. Assume a unique positive solution u\ € C2+e(D) of (1), totically stable, and the trivial solution u = 0 have fD m(x)dx _ u\2 e a\D\
first that JDmdx > 0. Then there exists for every A > 0, which is globally asympis unstable for all X > 0. Moreover, we —>0,
asXIO,
C + (D)
where \D\ is the volume of D. Assume now that J„ mdx < 0. Then there exists a positive solution ux € C2+e(D) of (1) if and only if A > \\{m), u\ being unique, globally asymptotically stable and such that I I U A | | C 2 + 8 ( D ) — * °>
as
A
i
M m )-
Moreover, the trivial solution u = 0 is unstable for any A > Ai(m) and globally asymptotically stable for each 0 < A < Ai(m). Some ecological interpretation of Theorem 1.1 can be given concerning the qualitative behavior of the globally asymptotically stable, non-negative solution U\ when A J. 0. If fD mdx < 0, then Ux is identically equal to zero for 0 < A < Ai(m). This would say that the species becomes extinct if he diffuses at the so large rate. If fD mdx = 0, then U\ tends to zero as A J. 0, which means that the steady state for the population becomes small gradually and extinct asymptotically as the diffusion rate goes to infiriity. In the case JD mdx > 0, Ux keeps positive even when A J. 0. This implies that the species is persistent even if the diffusion rate goes to infinity. To sum up, if the food environment becomes rich, then the more the species diffuses, the more we find out clearly that it is favorable for him. In the case of the nonlinear boundary condition, the uniqueness of positive solutions does not necessarily hold (see Amann [1, Theorem 2.6] and Pao [7, Theorem 4.6.3] for criterions for the uniqueness). This suggests that it is quite difficult to obtain always a globally asymptotically stable solution for (1). For this reason we restrict our attention to the problem how the stable steady state behaves as A J. 0 for the corresponding initialboundary value problem — v(0,x) dv —
=
—At; + m(x)v — av2 in (0, oo) x D,
= uo(z)>0
inD,
= b(x)g(v)
on (0, oo) x dD,
(4)
487 with sufficiently small initial data u0 ^ 0. For this problem, the following bifurcation and stability problems arise naturally. Our main interest of this paper is to study the problem of bifurcation to the right at the origin for (1) of positive solutions, meaning the problem whether there exist sequences {A^} and {WA^} of positive solutions of (1) for A = \j, such that A.,- J. 0 and HMAJICCD) -» 0 as j -» oo, or not, and additionally to investigate the stability of the bifurcation positive solutions. Our results under the nonlinear boundary condition are compared with Theorem 1.1 in the framework of the local bifurcation problem. In the work [9], we study the behavior of the minimal positive solution of (1) when A | 0, under some growth condition on g, mainly for the case JD mdx > 0 by the method of super- and sub-solutions, in which there is the corresponding result for the case JD mdx = 0 that is rather weak. In the present work, we focus on the case fD mdx = 0 and on the existence of the local bifurcation branch, which represents the minimal positive solution, within the scope mentioned above. In the present work, our approach which relies on bifurcation theory, is different from that used in [9].
2
Main results
Before stating our main results, we remark that the condition g'(0) = 0 is necessary for the origin (X,u) = (0,0) to be a bifurcation point to the right for (1). Indeed, if Uj is a positive solution of (1) for A = Xj such that A., J. 0 and H^j Hcrrn) ~~* 0> then we can verify that the function Vj := Uj/llujllpfm tends to some v0 e C2(D) which satisfies -Av0 = 0,
in D,
- ^ = b(x)g'(0)v0,
on dD.
Now we can formulate our main results. First we have the following result for the case
JD mdx > 0. Theorem 2.1 Suppose that JDmdx > 0. // condition (2) is fulfilled, then the origin (A,u) = (0,0) is not a bifurcation point to the right for (1). Next we have the following result for the case JD mdx < 0. Theorem 2.2 Suppose that JD mdx < 0. / / there exists an integer k > 2 such that g'(0) = ---=gik-1)(0)=0
and
ffw(0)>0,
then the origin (A, u) = (0,0) is a bifurcation point to the right for (1). More precisely, the set of all non-trivial solutions u of (1) in A > 0 and ||W|| C( B) both small enough consists exactly of a unique smooth branch of positive solutions emanating from the origin. Remark 2.3 In the case JD mdx < 0 the origin (A, u) = (0,0) is not a bifurcation point to the right for (1) provided g(t) = 0 for all t > 0 close to the origin, see Theorem 1.1. This indicates that Theorem 2.2 would be optimal under the additional assumption that g is analytic at t = 0.
488 Finally we state our results for the case JD mdx = 0. Let 70 € C2+e(D) be the unique solution of the Neumann boundary value problem —A70 = m(x),
in D,
-^— = 0,
on dD,
on
/ y0dx = 0. JD
Theorem 2.4 Suppose that JD mdx = 0. / / «/(0)= fl "(0) = 0,
Lb{X)da>V0'-=29'"(*)5D\^0?dx>
(5) (6)
where da is the surface element of dD. Then the origin (A,u) = (0,0) is not a bifurcation point to the right for (1). More precisely, there is no non-trivial solution u of (1) for any A > 0 and |Mlc(5) both sufficiently small. On the other hand, we have the following. Theorem 2.5 Suppose that fD mdx = 0, and suppose condition (5) holds. If b(x)da < VQ,
(7)
JdL then the origin (A,u) = (0,0) is a bifurcation point to the right for (1). More precisely, for any A > 0 small, there exists a minimal positive solution ux € C2+e(D) of (1), which satisfies that ||"A|IC 2 + S (O) —• 0 as A J. 0. Additionally if g is analytic att = 0, then the set of all non-trivial solutions u of (1) in A > 0 and ||u||c(i5) both sufficiently small, consists exactly of two smooth branches of positive solutions emanating from the origin. The following two results are complementary to Theorems 2.4 and 2.5. Theorem 2.6 Suppose that JDmdx = 0, and that g'(0) = 0. Additionally if g"(0) > 0, then the origin (\,u) = (0,0) is not a bifurcation point to the right for (1). Theorem 2.7 Suppose that fDmdx = 0, and that condition (2) holds. Ifg'{0) = g"{0) = g'"(0) = 0, then the origin (A,u) = (0,0) is a bifurcation point to the right for (1). More precisely, problem (1) has a minimal positive solution ux G C2+e(D) for any A > 0 small such that ||uA|lc2+e(5) -^ 0 as A | 0.
3
Proofs of our main results
In this section we give a sketch of the proofs of our main results. Theorem 2.1 is proved having a priori lower bounds by a positive constant for positive solutions of (1), which is a direct consequence of [9, Theorem 3]. For the case fD mdx < 0, we can apply the wellknown local bifurcation theory for simple eigenvalues due to Crandall and Rabinowitz [5, Theorem 1.7] and then Theorem 2.2 follows.
489 Now we prove Theorems 2.4 and 2.5. It seems to be difficult to apply to the case JDmdx = 0 the local bifurcation theory due to [5, Theorem 1.7], where the so-called transversarity condition given as JD mdx =/=• 0 breaks down. To overcome the difficulty, we use the Lyapunov and Schmidt type of procedure in order to reduce (1) to a bifurcation equation in R 2 as follows. Assume that g(0) = g'(0) = 0. We see that C2+e(D) and Ce(D) are decomposed uniquely by R respectively as C2+e(D) = R ffi X2,
. with X2 = lue
C(D)
with Y2 = Iv G C 2 + e (S) : J
= R © Y2,
C2+e(D) : I
udx = o | ,
vdx = o\ .
As well, the projector Q of C6(D) onto Y2 is introduced as Qv = v--^-
veCe(D).
fvdx, D
The projector Q and Green's identity : [ -Audx JD
= - J -£-do, JdD ^n
u e C2(D),
are combined to prove that if u € C2+6{D) satisfies (1) then -Au2 + j--7 b(x)g{a + u2)da = \Qf(x,a l-^l J BD —^- = b(x)g(a + u2), ondD, $(A,a,u 2 ) := A /
f(x,a + u2)dx +
J D
+ u2),
in D,
b(x)g(a + u2)da = 0, J 3D
where u = a®u2eH®X2 and f(x, u) = (m(x) — au)u. Next we define the nonlinear mapping T : R x R x X2 —> Z\, where Z1 = /(>, V) e C(D) x C1+e{dD) : f
i/ida = o | ,
as T(A, a, u2) = ( - A u 2 + Tjj;
Hx)9(a + u2)dx - XQf (x, a + u2) , - ^ - 6(s)s(a + u2]
490
It is clear that : T(0,0,0) = (0,0). The Prechet derivative TU2(X,a,u2) : X2 -> Zx with respect to u2, is given as TU2(A, a,u2)v = I - A v + —- / b(x)g'(a + u2)vda - \Qfu (x, a + u2) v , V 1-^1 J OD b(x)g'{a + u2)v It follows that : TU2(0,0,0)v = (-Av,dv/dn), and thus that : TU2(0,0,0)v : X2 -> Zx is bijective. Therefore the implicit function theorem shows that : T(\,a,u2)
= (0,0)<*u2 = u2{\,a),
(8)
near (A, a, u2) = (0,0). To sum up, problem (1) near the origin (A, u) — (0,0) is equivalent to the equation : #(A,a)
:=
$(A,a,u 2 (A,a))
=
A/
/ (x,a + «2 (\,a))dx
+ /
b(x)g (a + u2 (A, a)) da = 0,
near the origin (A,u) = (0,0). The reduction is thus complete. Under the assumption that ^(0) = (0) = g"{0) = 0, all the partial derivatives cP<&/d\i~kdak vanish at the origin up to 2nd order, which means that the so-called Morse lemma is not applicable. To overcome the difficulty, we impose the additional restriction that g is analytic at t = 0 in order to verify the assertions of Theorems 2.4 and 2.5, first. The analyticity of g implies that of u2, via (8) with the aid of [10, Corollary 4.23], and consequently so is <E>(A, a). Some direct computations give that $(A, a) = ^(3AA2 - 3BAa + Co? + Vi(>,«)), 6 near (A, a) = (0,0). Here A, B and C are positive constants given by
a3* (0,0) dadX2
=• 2 /
m{x)-y0(x )dx,
J D
a3* B
= - ^
( 0
' ° )
=
C = 0(0,0)
2a
^>
= g"'(0)J
b(x)da
and ipi is a higher order term of the form oo
A;
ipx{\, a) = 5 Z H k=3
j=0
c
kjXak-j,
ckJ € R.
491 The requirement for Theorems 2.4 and 2.5 is 3A\2
- 3BXa + Ca2 + V>i(A, a) = 0,
near (A, a) = (0,0). Here we find that condition (6) <=> 9B2 - Y1AC < 0, condition (7) <S=> 9B 2 - Y1AC > 0. This implies that conditions (6) and (7) determine the sign of the discriminant for the quadratic function 3A\2 — 3B\a + Ca2. Prom this observation, the conclusions in Theorems 2.4 and 2.5 follow under the analyticity of g. The extension to the case g without the analyticity assumption is due to a comparison argument using the super- and sub-solution method as follows. Senn proved in [8] that if JD mdx = 0, then the principal eigenvalue Hi{\) of (3) is negative for any A > 0. Denote by ipi(X) the principal positive eigenfunction of (3). We can easily see under (2) that for every A > 0 a function E
t<™«-/ to K.)*> 55 $g p= . Let gi (t) = Sit3/6. Then there exists ri > 0 such that b(x)gi(t)
te[0,n}.
(9)
One sees that gi fulfills (5) and (6) and is analytic at t = 0, so that there is no positive solution u of (1) with g = gx for any A > 0 and ||M||C2+«(C) both small enough in the previous argument. By the well-known bootstrap argument, the same situation holds for A > 0 and u G C(D) both small enough. Now we assume to the contrary that the origin (A,u) = (0,0) is a bifurcation point to the right for the original problem (1) and, without loss of generality, there may exist positive solutions u^ of (1) such that ||WA||C(D) —» 0 as A J. 0, so that it follows that IIUA|IC(D) < TI for 0 < A < A* with some A* > 0. This implies that ux is a super-solution of (1) with g = gi for any 0 < A < A*, by virtue of (9). We recall the existence of such sub-solutions of (1) with g = g\ which are sufficiently small. Then the super- and subsolution method allows us to obtain that for any A > 0 small the existence of a positive solution v\ of (1) with g = g\ such that 0 < v\ < u\ on D. This proves that ||«A||C(5) ~~y 0 as A J. 0, which is a contradiction. The proof of Theorem 2.4 is complete. • Now we end the proof of Theorem 2.5 in an analogous way. Indeed, it suffices to note that for any g satisfying (5) and (7) without analyticity, we can take a constant &z > 0 such that r 3n2IDI2 b{x)da< * ' . 5"'(0)<52and /
492 Put gi{t) = 5 2 t 3 /6 and the rest of the proof is similar to the proof of Theorem 2.4.
D Theorems 2.6 and 2.7 can be also proved in the same way, using super- and subsolutions. Finally we refer to [9, Lemma 4.3] for the existence of the minimal positive solution in Theorems 2.5 and 2.7. •
4
Interpretations
This section is devoted to some ecological interpretation of our main results. Due to [9, Lemma 4.3], the minimal positive solution of (1) obtained in Theorems 2.5 and 2.7 is leftside asymptotically stable for the initial-boundary value problem (4) with u0 ^ 0. Here we refer to Pao [7, Chapter 5] for the definition of the left-side asymptotical stability. The critical value v0 given by Theorems 2.4 and 2.5 is of ecological interest in the following sense. We recall ([9, Theorem 2]) that the trivial solution u = 0 of (1) is unstable for all A > 0 under the condition J_ mdx = 0. Hereby Theorem 2.4 suggests that, in the case b(x)da > u0, JdL JdD the positive minimal stable steady state for the population obeying (4) may be persistent as the diffusion rate 1/A increases without limitation, even if the initial data u0 is small enough. Meanwhile, Theorem 2.5 suggests that it must be extinct asymptotically as the diffusion rate goes to infinity, provided b(x)da < u0. JdD
To be specific, for a constant c > 0, we put m{x) = cfh(x) where m 6 Ce(D), JDrhdx = 0 and Hwll^m = 1. Define a constant i>§ through the right-hand side of (6) with m = rh and we can verify that if c> then Theorem 2.4 applies, and meanwhile that Theorem 2.5 applies if 0
References [1] H. Amann, Nonlinear elliptic equations with nonlinear boundary conditions. In : New Developments in differential equations (W. Eckhaus ed.), Math. Studies, Vol. 21, North-Holland, Amsterdam, 1976, pp. 43-63.
493 [2] K. J. Brown and S. S. Lin, On the existence of positive eigenfunctions for an eigenvalue problem with indefinite weight function. J. Math. Anal. AppL, 75 (1980), 112-120. [3] R. S. Cantrell and C. Cosner, Diffusive logistic equations with indefinite weights: population models in disrupted environments. Proc. Roy. Soc. Edinburgh, 112A (1989), 293-318. [4] R. S. Cantrell and C. Cosner, Diffusive logistic equations with indefinite weights: population models in disrupted environments II. SIAM J. Math. Anal., 22 (1991), 1043-1064. [5] M. G. Crandall and P. H. Rabinowitz, Bifurcation from simple eigenvalues. J. Funct. Anal., 8 (1971), 321-340. [6] P. Hess, Periodic-parabolic boundary value problems and positivity. Pitman Research Notes in Math. Series, vol. 247, Longman Scientific & Technical, Harlow, 1991. [7] C. V. Pao, Nonlinear parabolic and elliptic equations. Plenum, New York, 1992. [8] S. Senn, On a nonlinear elliptic eigenvalue problem with Neumann boundary conditions, with an application to population. Comm. Partial Diff. Eq., 8 (1983), 1199— 1228. [9] K. Umezu, Behavior and stability of positive solutions of nonlinear elliptic boundary value problems arising in population dynamics. Nonlinear Analysis, TMA, to appear. [10] E. Zeidler, Nonlinear functional analysis and its applications I: Fixed-point theorems. Springer-Verlag, Heidelberg Tokyo, 1993.
Elliptic and Parabolic
SdtftJ-
ISM
Rolduc and Gaeta 2001
World Scientific www. worldscientific.com 5004 he