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!
'k depend on \x...in only as composite functions of \\ in through (7)2" • Obviously, one has to impose that in these composite functions, the contribute of velocity drops out. After that, eqs. (4) become pi,...i„
dh>
dX
L.J,
=
=
axl..jrdK...ln
h.,.tn
e\l..Jr
_ d(4>'k + h'vk)d\l„Jr
Fil...ink
dh'
—
_ d(
k + h'vk)
,,,.,„
which can be compared with eq. (7)i, (29) and giving the consequences " 9X1
%
'
d\'
l\ ... lr
[ )
i l\...lr
i.e., the equation (4) still hold also for the non-convective quantities! The principle of material frame indifference has simply given the further requirement that the composite functions h
' = ^...^(V^X)]
h'vk + 4>k = ^ [ A L o V ^ - i n V ) ]
have zero derivative with respect to V{, i.e., dh
Z^ r=0
Q\I
dX QVI
3I---]T
l^, r=0
Z_, ^•• l " n=r
d X ^ dvl
= v
!
395 JV-l
r=0
N
JV-1
n=r-\-\
and similarly
r—0
= h'5k + £
0= ^
(r + l)mk^^A^...^,
(12)
where we have used eqs. (9), (7) 2 , (28) and the fact that, when r=N then we must have n=N and X£;;£ = ^ x • • • < ^ } . The relations (11), (12) can be written also as
EV+1)037^^^.^=0 r=o N x
~
^
(13)
n—i-r ah'
( r + 1)
X
l..ir* + h'6!=0
9V
r=0
^4)
fej'i—jr
where we have used eqs. (9) and its consequence d6'k ^J—
=^
dti ~
^
r = 0,...,iV-l.
(15)
We want indicate another point which needs a judicious interpretation: In eq. (9)i with r = 1 and (9)2 with r = 0 one has
°=3AJ ;
(16
°=^
>
because m; = 0. We have Proposition 2: "Eqs. (16) don't mean that h' doesn't depend on \j and 4>'k on A7." To obtain this result, we have to remind that (4)i are to be interpreted as condition from which A J ,• can be abteined as functions of Fu'"ln: the result has to be inserted in (4)2 with n = N to obtain the costitutive function F i l - i " f e . If we use (7)i this relation became 8
F)h'
^:::i;(v)m-- = _ —
;
x'^V)™*"-*- = ^—. 71
(17)
The first of these gives A^...^ as function of m- --^ and v, while the second one gives the costitutive function m-7'1 ••••?'"*. The hypothesis that h' and ip'k — h'vk depend on Aj,...i„ only as composite functions through (7) has transformed (17)! in (9)i and (17)2 in (9)2 with r = n. Therefore, (9)i has to be interpreted as conditions from which A J, , can be obtained in terms of mPl—^\ the result can be substituted in
396
(9)2 with r = N in order to obtain the constitutive function m?'l—irk. The conditions (6) transform into (15). Therefore (9)i isn't a condition restricting h! but only a condition from which to obtain \\ i in terms of m i i . . . j r . it s value for r = 1 doesn't imply that h' doesn't depend on \{\ For the sake of greater security, let us confirm this with what occurs in the Kinetic case (see 3 ) , where we have .' = f F(X' + A?u* + • • • + Afu* 1 -*^' 1 • • • uiN)du fc
= J F(X' +\!ui + --- +
\Iiuil-iNuil---uiN)uku (18)
In this case, it is easy to see that eqs. (15) are verified and also (13) and (14), which are equivalent to
The eq. (9)i with r = 1 and (9) 2 with r = 0 become mjl = J F'(- • • )u*du and this doesn't mean that f F(- • • )uld\x doesn't depend on \{, as in our case! It is easy to check also Proposition 1 in the kinetic case ; in fact, from eqs. (7) 2 and (27) we have
N
N
,
v
n u(il
J2 £ ^i-*- ( ) n=0 r=0
N
•••^ " ^ •••"
in)
= £ *«!...<» ( u i i + y i i ) • • • («
n=0
^ '
from which
r h'=
N il i F(J2K...i nc ---c ")dc 71=0
y,'fc = 4>'k + tivk = /" F ( ^ Ai^.^c' 1 • • • c'")cfcc J
n=0
(21) with cl — v? + vl. The exploration of conditions (9), (13)-(15) will be the object of a future
397
work. The case N = 3 has been considered in ref. 4 . We conclude reporting, for the case of N odd, the particular solution h' = 0, oo
^
w'k — Y ^ — C f j ( Q i a 2 • ..nanrk')XI V
— /^i r=0
\
r
r y
y
-X1
A
A
ax...aN
aNr_N+i...aWr
(22)
of these conditions, with cr constants. This solution cannot be found with the kinetic approach (18), even if can be expressed in the kinetic forms as h' = jf(\l..iNui>---ui»)e-u2du,
(«,)] f(v)f(w).
(15)
2lr
Choosing (p(v) = v2 shows that the average kinetic energy of granular temperature T(t) = (v2) keeps decreasing at a rate proportional to the inelasticity 7p
=1-
f
f- (| S in 9\2+2*> + | cos 6\2+2P)
.
(16)
Thus, if the initial density has finite temperature, the solution to the inelastic Kac equation does not reach the Maxwellian u>(v) = (27r)~1/'2exp[—v2/2], but is approaching a Dirac delta function 5(v) for large times. 4. Steady states of infinite energy The Kac model (10) can be studied in weak form — hp(v)f(v)dv= dt
/
J
—- {
1. Remark 2. We notice that v = 0 => L = 0. Further if V> ^i T, v of T, by requiring the invariance of (17) and (18) with respect to the generator Y* = Y + T = 2 m ' > m" >inf.0, E' = E(4> < m') D E(
(£L-""MI-"")
(u
>
then (locally) L = 0&v
= 0.
(12)
Remark 3.Let us notice that: i) our results will be based on the (integral) energy estimates (29) and (40) obtained considering perturbations v such that u = U + v is a classical solution ii) (29) and (40) - and hence our results - continue to hold along the perturbations v such that u = U + v is a weak solution obtained - as it generally happens {Cfr. Sect. 3 of [11] and [12]} - as limit of sequences of classical solutions.
424
3. Two weighted energy relations Let w : x G Cl —> w(x) G R+
(13)
be a scalar weight function such that 'w£C2(£l),
|Aw|
\Vw\
(14) lim w = 0, I, |x|-»oo
a being a positive constant. We denote by I(n,w), functional space such that, Vi > 0
n G N, the weighted
( w{v2 + L 2 ("+!) + (VL n + 1 ) 2 ) G L(Q), w G I(n,w)
lim w[Z, 2n+1 |VZ,| + L2(™+1) + |L„AL|] = 0
(15)
Remark 4. We notice that - in view of (14) - the functions v G I(n, w) are allowed to grow (suitably) when |x| —> oo. In particular, if w decays exponentially to zero when |x| —> oo, then the functions v G I(n,w) are allowed to grow as \x\q, Vq > 0. Precisely, let \F{u)\ + \F'{u)\ + \F"(u)\ < ci\u\i\
WuGR (16)
|u| + |Vu| + |Au| < c 2 |x| 92 with Ci,qi = positive constants(i = 1,2). Then it turns out that, for each n, exist two positive constants cn and qn such that |L2n+lVL|
+
L 2(n+1) +
| £ u A £ | < ^|x|ff»
(17)
Theorem 1 - Let v G I(n, w) be a smooth solution of (5). Then setting ' Gn(U,v) = [VL2n+1(U,v)dv; Jo
Vn(U,v) = [ Jn
w(x)Gn(U,v)dy (18)
2
dx
the following weighted energy energi relations hold 1
K
f r2(n+i;^Aw dx 2(n + l ) i n """^
2n + l 2
[ w (VL( n+1 >) 2
(n+1) Jn
dx (19)
= - /' (Vti; • WL)Ltdx -
f
w^-(AL) dx.
425
Proof. Along the solutions of (5) it turns out that Vn=
[ {V • [wL2n+lWL]
- V(wL2n+1)
• VL} dx
£=
f wVL • VUdx. = f [V • (LtwVL) - L*V • (wVL)]dx. Jo. Jn But {v = 0} =» {L = 0} => {L t = 0}, hence in view of (5) and (15), (19) easily follow. 4. Energy inequalities Let us begin by recalling two lemmas which proof can be found in [5], [6] Lemma 1 - Let F'[u(x,t)]
> m,
Vt e [0,T]
(20)
with m and T positive constants. Then, Vt £ [0,T], it turns out that Gn < vL2n+1;
L 2( " +1 > > mGn; Gn > - ^ U
2
("+D
(21)
Lemma 2 - Let F'[u(x,t)}
<mu
Vt € [0,T]
(22)
with mi and T positive constants. Then, Vt E [0, T], it turns out that £2(»+i) <2(n + l)miGn.
(23)
Theorem 2 - Let (22) hold and let w = g = e~ar, r = \fx2 +y2 + z2;
a = const. > 0
(24)
(x, y, z) denoting rectangular Cartesian coordinates, when Q, is the exterior of Q.Q and w = g = e-a(z+p\
p = s]x2 + y2,
a = const.>0
(25)
when Q = H. Then it turns out that v£l{n,g);
Gn{U,v0)eL{n)
(26)
imply, Vt € [0, T] En(t) + f dr f (VL n + 1 ) 2 dx < oo Jo Ja
(27)
426
with
En(t)= f Gn(v,U)dx.
(28)
JQ
Further the following inequality holds E
^) + T^TTTI I
dT
I (Vi" +1 ) 2 ^x < En(0)
(n + ly J0
(29)
JQ
with En(0)=
f Gn{U,v0)dx. (30) Jn Proof. Let Q. be the exterior of the bounded domain £IQ. In view of (23)-(24), we obtain Ag < a2g;
[ L2(-n+l) Agdx < 2c?mx(n + l)Vn(t)
Jn
and (19)!, implies, Vi G [0,T]
Vn{t) < ahniVn{t) - - ^ ± 1 / g(VLn+1)2dx. {n + iy j n Integrating we obtain Vn(t)+{?n
+ ll [ e ^ ^ d r
(n+iy
J0
f g(VLn+1)2dx<ea2m^Vn(U,vQ)
Jn
(31)
Letting a —> 0, the right hand side - in view of (26)2 - converges to En(to), and the monotone convergence theorem implies (29). The previous proof continues to hold also when £1 = H. In fact (25) implies Ag < 2a2g and one arrives to (31) with 2a 2 at the place of a2. Theorem 3- Let (20) hold, v G 1(0, g) and l(VL)2}t=0
G L(Q).
(32)
Then [(VL) 2 ] G L(Q),
Vt > 0
(33)
and E*{t)=
f (VL)2dx
(34)
Jn is a nonincreasing function satisfying the inequality
E*(t) + m [ dr f (ALfdQ < E*(0). Jo Jn
(35)
427
Proof. In view of < - f \Vw\L2tdx + \ [ \Vw\(VL)2dx, 2 JQ 2 7n
- / LtVwVLdx Jo.
(36)
being (both in the cases (24)-(25)) \Vw\ < 2aw
(37)
from (19)2 it follows that £
ma) [ w{AL)2dx,
(38)
and the equality sign holds in the exterior domains. Integrating and letting a —> 0, (35) immediately follows. In particular it turns out that E*{t) < E*(0).
(39)
On integrating (38) on [to,*], with *o > 0, in view of (33) we can let a —> 0 and obtain, Vt > to E*(t)<E*(t0)
Vt>t0
(40)
i.e. E*(t) is a non increasing function. Remark 5. We notice that allowing m = 0 in the assumption (20), from (38) with m — 0, one obtains £ < a£ =*• £ < £(Q)eat. Letting a —> 0, it follows that E*(0) < 00 implies that E*(t) < E*(0), Vt > 0. Therefore - since analogously it follows that E*{t2) < E*(ti), Vt2 > t\ - it turns out that E* (t) is a nonincreasing function also allowing m = 0 in (20). 5. Stability Criteria At each time t, for any real valued function / defined on Q, x R+, we denote by I|/(:r,t)||p and ||/(x,*),D*|| p , Vp > 1, the LP(ft)-norm and the LP(D*)norm, D* being a measurable subdomain of Q. The Lebesgue measure of D* will be denoted by fi(D*). Lemma 3 - Let f(x,t) G Lp(fl), p > 1, Vt > 0. On denoting by x ^ i ( s , l/( ,*)|) the largest subdomain ofCl on each point of which, at time t, | / ( x , t)| is bigger than s(t) > 0 and by fx{e, | / ( x , t)|) the Lebesgue measure of£lx(e, | / ( x , t ) | ) , the following inequality holds ii(\\f(x,t)\\f",\f(x,t)\)<\\f(x,t)\\f1
,
Vt>0.
(41)
428
The proof of the lemma 3 can be found in [10]. We notice that on setting D(t) = Cl1 ( | | / ( x , i ) | | ! * M / ( x , * ) | ) , (41) implies /x[D(t)]<||/(x)t)|||?T (42) ||/(x,t),n/I>(t)||00<||/(x>t)|||
?T
.
On defining - as it is quite natural- U weakly pointwise attractive with respect to f when, Ve > 0, there exist a i and a measurable subdomain Do(t) such that t>t=>
(\\v(X,t),n/D0(t)\\oo<e I { n[D0(t)] < e
(43)
then ifU is attractive with respect to ||/(x,£)|| p , on choosing Do(t) — D(t) it follows that lim /i[D 0 (t)] = lim ||/(x,t),fi/A>(t)||oo = 0, t—>oo
(44)
t—>oo
and U is weakly pointwise attractive with respect to f. We denote by I* the set of functions growing polynomially as |x| —> oo and collect here the stability criteria implied by theorems 1-3 and lemma 3. Theorem 4 - Let U be a steady solution of (2). If (22) holds and a positive n exists such that v(x,t)€l*,
Vt>0 (45)
Gn(U,v0) e L(Q), then U is stable according to En{t) < En(0),
Vt > 0.
(46)
Further, if also (20) holds, then U is stable with respect to the L2<Jl+1^(Q) — norm according to i ) < E2(n + l ) £ . j/{n+l)2 ( «^^"i^ ^)v
+
dQ
( 47 )
Proof. In view of (45), (29) holds which implies (46). On the other hand (20) implies (21)3, hence (47) immediately follows. Remark 6. As noticed in [2] in the case of the half space, (45)2 can work as a selection criterion in the set of the steady states. In fact if
429 U(^ U) is another steady state then Gn(U,Vo) € L(Q) does not imply Gn(U,v0)GL(n). Theorem 5 - Let U be a steady solution of (2). If (20) holds and (v{x,t)el*, Vt>0 { l[(VL)2]4=0eL(n),
(48)
then U is stable according to f(VL)2dQ< [ {VLfdSl . in Lin J t=o Further U is also stable according to
(49)
[ wv2dx< ^E*(0) (50) in ™2 for any weight function 0 < w < 1 and any positive constant 7 such that the weighted Poincare inequality holds [6] f wL2dx < 7 / (VL)2dx. (51) in in In the case n = 2, if also the assumptions of theorem A holds, then U is
-*. - , ™^,. a. ™
/ ii. + (vim. J U . * in
/ [L2 + (VL)2}dQ < ^ ( / [L2 + {VL)2]dn\ in '1 I i n J t=o
(52)
«re£/i li = i n f ( — ,1), I2 = sup( — ,1). 2TO m Proof. In view of theorem 3, (49) holds . On the other hand (51) and (21) imply (50), while (21), (23), (29) and (49) yield (52). Theorem 6 - Let U be a steady solution and let the assumptions of theorems 2 (with n = 0) and 3 hold. Then U is attractive according to lim / [wL2 + (WL)2}dx < (1 + 7)£o lim ~*°° i n *^°° t and weak pointwise attractive according to
(53)
t
1/3
lim A(||VL|| 2 / 3 ,|VL|) < lim ||VL|| 2 / 3 < lim t—»oo
t—»oo
t—»oo \
(^
= 0
t
(54)
lim m^2\\l'\ _ t—>oo
|v^L|) < lim ( ^ ) t—>oo \
=0 I
/
430
Proof. For n — 0, (29) implies
/ Jo
E*(t)dr < E0(x)
(55)
with E0(x) = / G0(v0,U)dx. But in view of theorem 3, E* is is a nonmin creasing function, hence (55) implies implies tE*(t) < E0(x) i.e.
lim / (VLfdQ t—too JQ
< E0(x) lim - = 0. t—>00 t
and (41), (51) imply (53)-(54). Theorem 7 - Let U be a steady solution of (2) and let the assumptions of theorems 2-3 hold Vn > 0. Then U is asymptotically stable with respect to the pointwise norm (a.e.). Proof. See [10]. Remark 7. Assumption (20) and (22) have been used in a crucial way (with the exception of the case (49)) in order to establish the stability criteria in theorems 4-1. However conditions less stringent than (20) and (22) are considered in theorem 8, which proof can be obtained following the guidelines of theorems 4 of [2] and [8] and can be found in [10]. Theorem 8 -Let U be a steady solution of (2) such that inf U(x) > a > - o o ; supC/(x) < b < oo; m* > F'(U) > 2m,
(56)
m* and m being positive constants. If conditions (45) hold Vn e TV, then U is asymptotically stable in the pointwise norm (a.e.). Remark 8. Let us notice that, since F'(u) € C(R), (56) are implied by ai = inf U(x) > —oo;
b\ = supC/(x) < oo (57)
m 2 > F'[U(x)} > 2m 3 ,
Vxefi
m2 and ma being positive constants. On the contrary, appears critical the case F'(ai) = 0. In fact in this case mz = 0 and (56)$ does not hold.
431
6. L o n g t i m e b e h a v i o u r of t h e solutions of t h e p o r o u s m e d i u m a n d horizontal filtration e q u a t i o n s As is well know [9], the equation governing the diffusion of a gas of density u in a porous medium is given by (1) with F(u) = up,
(p > 1).
(58)
In particular, when (58) holds with p = 6, (2) governs the horizontal filtration of a moisture into a dry-soil. In view of (58), (6) becomes AUP = 0 x e O (59) Up = up
xeffi
where u\ is a positive constant. In particular, let us consider the steady state U = ui
(60)
In view of remark 5, it follows that U — u\ is stable in the / (VL) 2 dx Jn measure with respect to the perturbations growing polynomially at large spatial distances such that / (VL)2cZx Jo. Ua.
< oo, t=0
and, in view of (51), is also stable with respect to the measure /
[wL2 + (VL) 2 ] dx.
JQ Q
Passing to the asymptotic pointwise stability, we notice that setting Ul u 3 P P-1 w>-i a= y , ,b= -ui; m = -ap \ mi = pbp from p > 1 and F' = puP-1 > 0; F" = p(p - l)up-2
> 0; Vw S [a, b]
it turns out that 1 3 -m < U = in <-ui; 2m
432 filtration equations is globally pointwise asymptotically stable (a.e.) with respect to the perturbations satisfying (45) Vn > 0. We end by considering t h e special case fio = S{0,ro). Then the nontrivial steady states are given by U
r0
r
«i
(62)
with A G [—ro,oo[ , in order t o guarantee U > 0, Vx G S(0,ro). 5 G]0,1[, it follows t h a t :
-ro<-(l-S)=X<0^ v
y
(miU = 6pUl n { Tr I sup U = Ui
\ = -r0=>{
Letting
(inf£/ = 0 Q „ I S U P U — ui
inf U = Ui sup U = 1 H 1 u\ n \ r0J However, since m > 0 guarantees t h a t E*(t) is a nonincreasing function (cfr. Remark 5), it follows t h a t t h e steady state U = ui (— j
is stable
with respect this measure at least when p = 1 + 2q, with q = c o n s t . > 0 and the perturbations verifying one of t h e point i)-ii) of Remark 3. Acknowledgments This work has been performed under t h e auspices of the G.N.F.M. of I.N.D.A.M. and M.I.U.R.(P.R.I.N-): "Nonlinear mathematical problems of waves propagation and stability in models of continuous media". References 1. I. Torcicollo, M. Vitiello: On the nonlinear diffusion in the exterior of a sphere. Rendiconto Accademia Scienze Fisiche Matematiche Napoli LXVIII, 2001, pp. 139-146. 2. S. Rionero, F. Perrini: On the nonlinear diffusion in the half space with application to the porous media. Rendiconti Accademia Scienze Fisiche Matematiche Napoli LXIX, 2002, pp.71-81. 3. I. Torcicollo, M. Vitiello: A Note on the nonlinear pointwise stability in the exterior of a sphere, (to appear) 4. S.Rionero: On the longtime behaviour of the solutions of nonlinear parabolic equations in unbounded domains. Proceeding Wascom 2001, World Scientific, 2002, pp. 447-457.
433 5. S. Rionero: Asymptotic and other properties of some non linear diffusion models. Continuum mechanics and applications in geophysics and environment, Springer (Physics and Astronomy), eds. B. Straughan, R. Greve, H. Ehrentrant, Y. Wang, 2001. pp. 56-78. 6. J.N.Flavin, S.Rionero: Asymptotic and other properties of nonlinear diffusion model. Journal of Mathematical Analysis and Applications, 228, 119-140, (1998). 7. J.N.Flavin, S.Rionero: Qualitative Estimates for Partial Differential Equation. An Introduction. CRC Press, Boca Raton, FL, (1995). 8. J.N.Flavin, S. Rionero: Stability properties for nonlinear diffusion in porous and other media. Journal of Mathematical Analysis and Applications, (in press) 9. A.B. Tayler: Mathematical models in applied Mechanics. Clarendon Press Oxford (1986) 10. S. Rionero: Asymptotic properties of solutions to nonlinear possibly degenerated parabolic equations in unbounded domains, to appear on Mathematics and Mechanics of Solids (2003). 11. J.L. Vazquez: An Introduction to the Mathematical Theory of the Porous Medium equation., Shape Optimization and Free Boundaries, M.C. Delfour ed., Mathematical and physical Sciences, Series C, Kluwer Academic Publisher, Dordrecht, Boston and Leiden, (to be published) 12. D.G. Aronson, P. Benilan: Regularite des solutions de l'equation des milieux poreux dans Rn. Comptes Rendus Academy Sciences Paris, Serie A, t. 288, p. 103 (1979).
O N G R O U P ANALYSIS OF A CLASS OF E N E R G Y - T R A N S P O R T MODELS OF S E M I C O N D U C T O R S IN T H E TWO D I M E N S I O N A L STATIONARY CASE
V. R O M A N O , A. V A L E N T I Dipartimento
di Matematica e Informatica, Universita di Catania, viale A. Doria, 6, 95125 Catania, Italy E-mail: [email protected], valentiQdmi.unict.it
A group classification via equivalence transformations of a class of energy-transport models of semiconductors in the two dimensional stationary case is presented.
1. Introduction Continuum models for the description of charge carrier transport in semiconductors have attracted in the last years the attention of applied mathematicians and engineers on account of their applications in the design of electron devices. The energy transport models for semiconductors (hereafter ET models) are macroscopic models that take into account also the thermal effects related to the electron flow through the crystal at variance of the popular drift-diffusion 1 , 2 , s models that are based on the assumption of isothermal motion. The evolution equations are given by the balance equations for density and energy of the charge carriers, coupled to the Poisson equation for the electric potential 4'5>6-7. In these models there is the presence of some arbitrary functions as the mobilities, whose expression is based on fitting of experimental data or Monte Carlo simulations. In ref. 8 a symmetry analysis were performed for the one dimensional nonstationary case giving classes of functional forms of mobilities, energy relaxation time, doping profile and examples of exact solutions. However in the most applications one looks for problems in the two dimensional stationary case. In several models of mathematical physics, the differential equations governing the evolution phenomena contain arbitrary functions (constitutive function) which introduce non-linearities in the determining system when one looks for a symmetry analysis. So that, often it is impossible to
434
435
obtain the complete solution of the determining system. In these cases an effective method based on equivalence transformations can be utilized 9 . In the present paper we perform a group classification of the ET models in the two dimensional stationary case by using equivalence transformations. We get functional forms of the constitutive functions for mobilities, energy relaxation time and doping profile. The classical models of Chen et el. 4 and Lyumkis et al. 5 enter as particular cases in our analysis. 2. The mathematical model The unipolar ET models for charge carriers in semiconductors are given by the balance equations for density and energy density of electrons, coupled to the Poisson equation for the electric potential On — +div3 = 0, d(nW) v ; +divSJ Vcj> = nCw, A2A> = n - c(x).
(1) (2) (3)
n is the electron density, J is the electron momentum density, W the electron energy, S is the energy flux density, nCw is the energy production, A2 the dielectric constant,
w T
-l - ^-^iw'
J = _ V (n^Tn) S = -V
(M(2)T2U)
+ fi^nVcj), + /x(2)TnV
<4» (5) (6)
where, T is the electron temperature, scaled according to T —> ~ , Tj, = 1 is the scaled crystal temperature (taken as constant), Tw is the scaled energy relaxation time depending on T. / i ^ are the electron mobilities that depend on T as well.
436
Taking the system equations (l)-(3) and the constitutive relations (4)(6) into account, if we denote with x and y the independent variables, in the two dimensional stationary case the general energy transport model is given by the following class C of PDE's: +f/i(1)Tn)
(fi^Tn) \
/ xx
\
(^T2n) \
- L^Tn^)
' yy
V
+ (^T2n) / xx
+^Tn4>l
V
V
-U^Tn^) / yy
- $y U»Tn)
- U^Tn^) / x
= 0,
- U^Tn^y)
V
/ a;
+ ^Truft
(7)
/ y V
- <j>x U»Tn) / y
V.
/ x
- f n ^ - ^ 2 = 0,
(8)
>?{4>xx + 4>yy) -Tl + C = 0.
(9)
3. Equivalence transformations As it is well known, an equivalence transformation is an invertible transformation of independent and dependent variables which changes a system into a system belonging into the same class. Roughly speaking, the system of equations (7)-(9) is mapped into a system of the same form but with different ^
x
\ /J,^2\ TW and c.
With the aim of obtaining an equivalence classification u with respect to fj.^ and JJP^, we work in the extended space (x,y,n,T,
= (T\v)y
= {Tw)n
= (TW)> = 0,
Cn = CT = C
(10)
(11)
which characterize the functional dependence of the functions TW and c. In order to make use of the Lie infinitesimal criterion 9 , let us introduce the one-parameter group of equivalence transformations in the extended space (x,y,n,T,
+ 0(e2),
y = y + ee(x,y,n,T,
+ 0(e2),
n = n + er)1(x,y, n, T,
+ 0(e2),
4> = ct> + erf(x,y,n,T,
+ 0{E2), +0(e2), + 0(e2),
437 where £ is the group parameter. The associated equivalence algebra £ is the set of vector fields Y of the form .i9
Y = e-sdx +
d dytir+v dn
o d
a
d
3TW
,d_ 8c
(12)
The determining system of (7)-(9) and (10)-(11), arising from the invariance conditions, after some lengthy calculations leads to the relations Z1 = a\x - a2y + a 0 , 2
7] = (2ai 4- a3)T,
£2 = a 2 x + a\y + b0,
+2
2
r-n
(14)
* + M (1 V (2) ] "1 - {(2ox + a 3 )T 2 [(/x (1) )'(/x (2) )' [(3 f l l + 2a 3 )/i (1) (M (2) )' - (ax + a 3 )(A* (1) )y 2)
TL
+a3
v2 = a3c,
ry = (2ai + a3)4> + d0,
+2( a i + a s J ^ V ^ V 1 2 ' - (4oi + 3a3) T2
(13)
3
l W > ) ' - iV^V -M(1)(M(2))"
n1 = a3n,
( M (1) )V 2) -a 3
r
T-T,
•M
T2
M ( 1 ) (M ( a ) )'
T-TL
V a) }
TW
(15)
= 0,
(2a, + a3) | V 1 ) ^ 1 ) ) " - T [(/x*1))']* + M(1 V 1 } )'} = 0,
(16)
(20l + a3) {TOxW) V 2 ) ) " - 2 V 1 V 1 W 2 ) ) ' + 2(/x<1>) V 2 ) ) ' +T (M(1))' M( 2 ) _ / U ( i ) ( / i d ) ) y 2 )
(17)
where Oo, &0i <^o> ^l) 0,2, 03 are constants and primes mean differentiation with respect to T. From conditions (13)-(17) we get the following equivalence classification of the class C of PDE's (7)-(9): Case (I) fiW and fi^> arbitrary Z1 = aix - a2y + a0,
£2 = a2x + axy + b0,
771 = - 2 a m ,
V2 = 0,
^1 = 2a\Tw,
v2 = —2a\c
rf = d0,
(18) (19) (20)
and the generator Y of the equivalence algebra assumes the form: ., , . d , , , d Yi = (axx - a2y + a0) — + (a2a; + a ^ + 60) — 9 , d „ d -2a\n— + do^-r + 2aiTWon 0(p OTW
d 2aic—; llC d-c'
n
(21)
438
Case (II) /id) = /#> Tm,
^
i1 = aix-a2y 1
r] = a3n, v\ =
Tm + rf$ Tm~\
=^
+ a0,
£2 = a2x + axy + b0,
2
T} = (2ai + a3)T,
(2ai + o 3 )
a 3 + - 2 ax
(T - TL)
(22)
if = (2ai + a 3 ) <j> + d0, (23)
— 2mai — (1 + m)a3f w ,
(24)
v2 = a3c
(25)
and the generator Y of the equivalence algebra assumes the form: 1^ / Y11 = {aix-a2y
\ ^ / , , d d + a0)-^- + {a2x + axy + o0)—- + a 3 n—ax ay an
+(2ai + a 3 ) T — + [(2ax + a 3 )0 + a3} — (2ai + a 3 ) -
(T-TL)
a-iC
d_ dc
— 2moi — (1 + m)a 3Tw
d
(26)
w
where fi^ , /J,0 , /UQ0 and m are constitutive constants. If we set m = — 1, one recovers the expressions for J and S of the model of Chen et al. while if we set m — — | , one recovers the expressions for J and S of the model of Lyumkis et al. 4. Symmetry classification via equivalence transformations The most general symmetry algebra of the system (7)-(9), for arbitrary functions fi^, ^2\ TW and c, is said the Principal Lie Algebra and denoted by L-p. A symmetry classification and the corresponding extensions of Lp admitted by the system equations under consideration, in many applications of group analysis, can be obtained via equivalence transformations through the following proposition 11 ' 12 ' 13 : Proposition: The operator X, projection of the operator Y in the space x, y, n, T, and
*d
= t-+t
d
A
d
9
d
^ + v ^i + rgf + v-^,
is a symmetry operator of the system (7)-(9) if and only if the functions TW{T) and c(x,y) are invariant with respect to the operator Y. In order to find L-p and to give an application of the previous proposition, we consider the equivalence operator Yj of the Case (I) and by applying the invariant test YI(rw(T))
= YI(c(x,y))
= 0,
439
it follows oi rw = 0,
(27)
(oi x - a2 y + a0) cx + (a2 x + ax y + b0) cy - 2 ax c = 0.
(28)
Therefore taking into account that (i^ and JJP^1 are arbitrary, for T\y and c arbitrary, we get ao = bo = a,\ = «2 = 0 and the symmetry operator spanning L-p is
x -
d
(29)
A symmetry classification and the corresponding extensions of L-p are obtained by solving the previous equations (27) and (28) with respect to T\y{T) and c(x,y). The results are reported in table 1. A symmetry classification and the corresponding L-p extensions can be obtained, in the same way, in the Case (II). Table 1. A symmetry classification via equivalence transformations in the Case (I). The constitutive function c(z) depends on the similarity variable z while a and b are arbitrary constants. Case
Extensions of
Forms of TW(T)
and
c(x,y)
TW arbitrary
(/)a
c = c(z),
T\Y arbitrary
(I)b
c = c(z),
c = c(z),
& - a &
x
* = -b£ + &
z = x + by
Tw arbitrary
V)c
*2 =
z = ax + y
X2 = -(y + b)£ + (x + a)£
z = (x + a)2 + (y + b)2
Acknowledgements The author V.R. acknowledges the financial support by M.I.U.R. (COFIN 2002 Problemi Matematici delle teorie cinetiche) and TMR project Hyperbolic and kinetic equations: asymptotics, numerics and applications grant HPRN-CT-2002-00289. The author A.V. acknowledges the financial support by INdAM through the project Modellistica Numerica per il Calcolo Scientifico e Applicazioni Avanzate.
440
References 1. S. Selberherr, Analysis and simulation of semiconductor devices, Wien - New York, Springer-Verlag (1984). 2. W. Hansen, The drift-diffusion equation and its applications in MOSFET modeling, Wien, Springer-Verlag (1991). 3. P. Markowich, C. A. Ringhofer and C. Schmeiser, Semiconductor equations, Wien, Springer-Verlag (1990). 4. D. Chen, E. C. Kan, U. Ravaioli, C-W. Shu, R. Dutton, IEEE on Electron Device Letters 13, 26 (1992). 5. E. Lyumkis, B. Polsky, A. Shir and P. Visocky, Compel 1 1 , 311 (1992). 6. N. Ben Abdallah and P. Degong, J. Math. Phys. 37, 205 (1996). 7. A. Jiingel, Quasi-hydrodynamic semiconductor equations, Basel, Birkhauser (2001). 8. V. Romano and A. Valenti, J. Phys. A 35, 1751 (2002). 9. L. V. Ovsiannikov, Group Analysis of Differential Equations, New York, Academic Press (1982). 10. N. H. Ibragimov, CRC Hanbook of Lie Group Analysis of Differential Equations, Boca Raton, FL, CRC Press (1994). 11. M. Torrisi, R. Tracina and A. Valenti, J. Math. Phys 37, 4758 (1996). 12. N.H. Ibragimov and M. Torrisi, J. Math. Phy. 33, 3931 (1992). 13. M. Torrisi and R. Tracina, Int. J. Non-Linear Mech. 3 3 , 437 (1998).
SOME R E C E N T MATHEMATICAL RESULTS IN M I X T U R E S THEORY OF EULER FLUIDS
TOMMASO RUGGERI Department (CIRAM)
of Mathematics and Research Center of Applied Mathematics University of Bologna, Via Saragozza 8, 40123 Bologna, Italy E-mail: [email protected]
D e d i c a t e d t o S. Rionero in t h e occasion of his 7 0 t h b i r t h d a y
The entropy principle plays an important role on hyperbolic systems of balance laws: symmetrization, principal subsystems and nesting theories, equilibrium manifold. After a brief survey on these questions we present some recent results concerning the local and global well-posedness of the Cauchy problem for smooth solutions with particular attention to the genuine coupling Kawashima condition. These results are applied to t h e case of a binary mixture of ideal euler gas and we prove that the K-condition is satisfied only in presence of chemical reaction with consequence that there may exist global smooth solutions for small initial d a t a and the solution converge to a constat equilibrium state. Viceversa, if the mixture is not chemically reacting, the problem of global existence remains an open question.
1. Balance Laws Systems, Entropy and Generators Let us consider a general hyperbolic system of N balance laws: 8aFa(u)
= F(u)
(1)
where the densities F°, the fluxes F* and the productions F are SJ^-column vectors depending on the space variables x%, (i = 1,2,3) and the time t = x°, (a = 0,1,2,3; da = d/dxa) through the field u = \i{xa) G K N . Now we suppose, following Friedrichs & Lax x, that the system (1) satisfies an entropy principle, i.e. there exists an entropy density — h°(u), an entropy flux — hl(u), such that every solutions of (1) satisfies also a new balance law (entropy law): daha = £ < 0.
(2)
with a non negative entropy production —£(u). The compatibility between
441
442
(1) and (2) implies the existence of a main field u' such that daha-Y,
= u'-(0aFa-F).
1,;2
: (3)
As a consequence of the above identity, we have dha = u' • dFa,
£ = u' • F < 0.
(4)
Starting from an example of Godunov 3 , Boillat 4 (see also in a covariant formulation Ruggeri and Strumia 2 ) was able to introduce four potentials h'a: h'a = u' • F a - ha,
(5)
such that from (4)i 8h'a It follows that, upon selecting the main field as the field variables, the original system (1) can be written with Hessian matrices in the symmetric form dh'a\ d2h'a provided that h° is a convex function of u = F° (or equivalently the Legendre transform h! is a convex function of the dual field u' ). 2. Principal Subsystems We split the main field u' £ 5RW into two parts u' = (v',w'), v' € 3? M , w' G » W " M , (0 < M < N) and the system (7) with F = (f,g), reads:
Given some assigned value w^,(xQ) of w ' (in particular constant), we call principal subsystem of (7) the system 5 : a
\
9V
)=f(v'w*)'
(10)
In other words a principal subsystem (there are 2N — 2 of such subsystems) coincide with the first block of the system (8), (9) imposing w' = w'„.
443
Principal subsystems have two important properties: they admit also a convex sub-entropy law and the spectrum of the characteristic velocities is contained in the one of the full system (sub-characteristic conditions) 5 . 3. Equilibrium subsystem A particular case of (8), (9) is given when the first M equations are conservation laws, i.e. f = 0. In this case it is possible to define the equilibrium state as usual in thermodynamics: Definition 3.1. An equilibrium state is a state for which the entropy production —E|B vanishes and hence attains its minimum value. It is possible to prove the following theorem 5 ' 6 : Theorem 3.1. In an equilibrium state, under the assumption of dissipative productions i.e. if
2 1 dw'
\ aW'
is negative definite,
(11)
E
the production vanishes and the main field components vanish except for the first M ones. Thus g|E=0,
w'\E = 0.
(12)
Therefore in the main field components the equilibrium manifold is linear w' = 0 and this confirms once again the importance of the main field. In the case of one dimensional space the system (8), (9) assume the form:
v t + (fcC0x = o (13) .wt + ( 0 * = - G ( v ' , w ' ) W where v = h'v,, w = h'w, and G is a definite positive (N — M) x (N — M) matrix. 4. Qualitative Analysis In this section we discuss the importance of the entropy principle on the Cauchy problem.
444
4.1. Local well
posedness
In the general theory of hyperbolic conservation laws and hyperbolicparabolic conservation laws, the existence of a strictly convex entropy function is a basic condition for the well-posedness. In fact if the flux's F* and the production F are smooth enough, in a suitable convex open set D £ Rn, it is well known that system (1) has a unique local (in time) smooth solution for smooth initial data 1 ' 7 , 8 . However, in the general case, and even for arbitrarily small and smooth initial data, there is no global continuation for these smooth solutions, which may develop singularities, shocks or blowup, in finite time, see for instance 9 ' 1 0 . On the other hand, in many physical examples, thanks to the interplay between the source term and the hyperbolicity there exist global smooth solutions for a suitable set of initial data. This is the case for example of the isentropic Euler system with damping. Roughly speaking, for such a system the relaxation term induces a dissipative effect. This effect then competes with the hyperbolicity. If the dissipation is sufficiently strong to dominate the hyperbolicity, the system is dissipative, and we aspect that the classical solution exist for all time and converges to a constant state. Otherwise, if the dissipation and the hyperbolicity are equally important, we expect that only part of the perturbation diffuses. In the latter case the system is called of composite type by Zeng n . 4.2. The Kawashima
condition
In general, there are several ways to identify whether a hyperbolic system with relaxation is dissipative or of composite type. One way is completely parallel to the case of the hyperbolic-parabolic system, which was discussed first by Kawashima 7 and for this reason, is now called the Kawashima condition 12 or genuine coupling 13 : In the equilibrium manifold any characteristic eigenvector is not in the null space of V F . It is possible to verify that the Kawashima condition is equivalent in our notation to the following requirement (see 1 2 ): For every A € R and every X e i? m \{0}, the vector I in the null space of — AA0° + Ag1, where A 'o
.
d2*' , 9u'<9u''
A
'i_
**' du'du'
I e Rn is not
445
and the index 0 denotes the equilibrium state. If we denote d the right eigenvectors of the symmetric system (-AA^ + A ^ d ^ O , the K-condition is satisfied if every eigenvectors
<14>
**{*)• We observe that d = A °d. 4.3. Global Existence
and stability
of constant
state
For dissipative one dimensional systems (13) satisfying the K-condition it is possible to prove the following global existence theorem due to Hanouzet and Natalini 12 : Theorem 4.1. Assume that the system (13) is strictly dissipative and the K-condition is satisfied. Then there exists 5 > 0, such that, i/||u'(a;,0)|| 2 < 6, there is a unique global smooth solution, which verifies u €C°([0,oo);F 2 (i?)nC 1 ([0,oo);-H' 1 (i?)). Moreover Ruggeri and Serre stable:
13
have proved that the constant states are
Theorem 4.2. Under natural hypotheses of strongly convex entropy, strict dissipativeness, genuine coupling and "zero mass" initial for the perturbation of the equilibrium variables the constant solution stabilizes \\u(t)\\2 =
0(t-V2).
In 12 the authors report several examples of dissipative systems satisfying the K-condition: the p-system with damping, the Suliciu model for the isothermal viscoelasticity, the Kerr-Debye model in non linear electromagnetism and the Jin-Xin relaxation model. 4.4. A counterexample K-condition
of global existence
without
Zeng n considered a toy model for a vibrational non equilibrium gas in Lagrangian variables, proving that also if the system is of composite time
446
the global existence holds. Therefore the K-condition is only a sufficient condition for the global existence of smooth solutions. An intriguing open problem is if there exists a weaker K-condition that is also necessary to ensure global solutions. And if exists such condition the question is its physical meaning in order to consider it as a possible new principle of Extended Thermodynamics adding to the convexity of entropy 14 . Therefore in literature we have systems satisfying the K-condition s and other no. We can see in the following that the mixture theory have the interesting property that the K-condition is satisfied if we have chemical reaction while without chemical reaction the system is of composite type and the global existence of smooth solutions remains an open problem.
5. Mixtures of Fluids Mixtures of fluids exhibit a huge amount of diverse phenomena. The first rational treatise of mathematical model of homogeneous mixture of fluids was given by Truesdell in the context of Rational Thermodynamics 15 . The compatibility of the model with the second principle of thermodynamics was discussed by Miiller 16 and the mixture theory belongs to the Rational Extended Thermodynamics theory 14 . The balance equations of mass, momentum and energy of the mixture constituents read as follows: df>a
dt dt
+div(pava)
= Ta,
+ div (p a v a (g)v a - t n ) = m a ,
d{\pavl+pa£a) dt
+
d l V
J/1 \\2
o
\ \ nPaVa + Pa
(a = 1,2, ...n),
£
, a
(15)
, Va - t a V a + q a V = ea,
where pa, v a , ea, t a , q a are the mass density, velocity, internal energy, stress tensor and heat flux respectively of the a-component of the mixture. These equations have the same form as the balance equations for a single body, except for the non-zero right hand sides which represent the production of masses, momenta and energies. These productions are due to interaction between the different constituents. Since the total mass, momentum and energy of the mixture have to be conserved, we must impose following
447
restrictions: n
n
n
Y^ ra = 0,
^ m0 = 0,
^ ea = 0.
a=l
a=l
a=l
If we sum equations (15) over all constituents and introduce the following quantities: n
the density
n
p = ^2paj
the velocity
v = ^ J —v°,
a=l
the diffusion velocity
(16)
a=l P
u a = v a — v,
(17)
n
the stress tensor
t = V ] (t a — p0u0
(18)
a=l n
the intrinsic energy density
pej = 2_]paea,
(19)
o=l
the internal energy density
1 " pe = pej + — Y J Pauai 2
the intr. flux of int. energy
q7 = ^
(20)
a=i
{q a + pa£aua
- t a u a } , (21)
a=l
the flux of internal energy
1 " q = q j + - Y J pau^ua, 2 a
(22)
=l
we obtain the following equations for the total mixture: balance equation of mass dp dt+div(pv) = 0,
(23)
momentum - ^ + div (p v <8> v - t) = 0,
(24)
and energy d (\pv2 + pe) + div \[ -pv2 + pe v - t v + q } = 0. dt ' "" \\2
{&
(25)
As we consider a single absolute temperature T, the aim of extended thermodynamics of fluid mixtures is determination of the An + 1 fields: mass densities velocities temperature
pa va T
(a = 1,2,... n).
448
In order to solve the problem we need an appropriate number of equations. They are consisted by balance equations of mass (15)i and momentum (15)2 for each constituent and conservation of energy of the total mixture (25). 5.1. Balance
Equations
for Binary
Mixture
of Euler
Fluids
Let us consider a binary mixture of Euler fluids, i.e. fluids that are neither viscous nor heat-conducting: qa=0,
ta = -paI,
(a = 1,2).
Instead of the mass and momentum balance laws for the second component, we use the equivalent equations of total conservation for mass and momentum. Therefore, associated with the 9 unknown fields (pi,P2,vi, v 2 , T ) , or equivalently (p,pltv,vi,T), we have the 9 balance equations: ^+div(pv)=0
-Qj- + d i v ( p 1 v 1 ) = T - | ^ + div (pv®v-t)=0
(26)
—hr- + div (/»ivi®vi + pil) = m d(±pv2 + pe) __V2T dt r_j_
,. +
f/1 V2
I1
d i y
\
2
2
+
f ) £
\
v
_
t v
+
q
with
q = 53 Y* (£a + 2U»J + M U a ' 2
t = -pl-^2paua®ua,
(27)
ao = l 2
P = 5>ao=l
5.2. Mixture
Considered 17
as a Single Heat Conducting
Fluid
Ruggeri observed that it is possible to write the velocities of the two constituents in terms of the mass center velocity and the flux of internal
449 energy: a Vi=vH
a q,
v2 = v
Pi
where 1
_ (-
. . (28)
q P2
, V\ , 1 . 2 \
(
, P2 . 1
2
Introducing the concentration variable c = p^/p, equations (26)2 and (26)4 can be written in terms of p, c, v and J = aq, where J denotes diffusion flux vector. Thus, the system (26) becomes: ^+div(pv)=0 - ^ p + div (pcv+J) = T
^
+
div(pv^v
~^~
l
+
- J _
y j 0
j ) = o
(30)
- + div i pcv
ot
9
+ P
m
O 2 + ^ + div{fV + rp £ + P V + f - ^ T + ^ ) J H 0 . r ^
iV2
'
'V
\pc(l-c)
a
Hence we conclude that in an extended thermodynamic model with 9 fields the binary mixture can be considered as a single heat-conducting fluid with a variable concentration. The evolution equation (30)4 is a natural extension of the Cattaneo equation for the heat flux. 6. Entropy Principle and Thermodynamical Restrictions The compatibility with the entropy principle is expressed in the form: ^
+ div {pSv + * } > 0
which yields several restrictions on the constitutive equations. In particular: *
=
Y ~ f (Pl/ilUl
+
^2M2 U 2)
450
where \ia = ea + pa/pa ~ TSa is the chemical potential of constituent a = 1,2. We can see that entropy flux *& is not equal to q / / T by definition (Clausius-Duhem), but also contains an additional term which seems due to the compatibility between balance laws and the entropy inequality. This fact is in full agreement with the basic ideas of extended thermodynamics in which the entropy flux is not a priori assigned 14 . Thanks to the entropy principle and to the convexity of the entropy density h = — pS the original system can be put in the form of a symmetric hyperbolic one if we chose the main field components as variables 17 : u' = (A",A c ,A v ,A J ,A £ ) AP
C =
^-M _ J2 T 2 ( - l + c) 2 cT p* J •V
+( 1 - c ) A C =
A T
+
Tp
V2
2T
(-l + 2c)J2 2(1 - c ) 2 c 2 7 V
(l-c)Tp
+
J-v (l-c)cTp
T
J
A'=
(1 - c) cTp 1
A£=
rr,
T
with
A = /x1-/x2;
p, = c/i! + (1 - c)/i 2
Moreover, taking into account the Galilei invariance entropy inequality we obtain 14 : m = r v — 6J; 6.1. Equilibrium
r = —aAc;
18
and the residual
a, b > 0.
Manifold
The equilibrium manifold depends on the presence of chemical reaction. In the first case the equilibrium is characterized by the condition (see (12)2): A J = Ac = 0
<^>
m =
T
= 0
<=>
J = 0, p1 = p2-
451 and the equilibrium subsystem becomes the Euler system for a single fluid. While, without chemical reaction: AJ = 0
<£>
m = 0
<S=>
J = 0.
and the equilibrium subsystem becomes the Euler system for a single fluid plus the equation for the concentration: dt that implies the Constance of c along the path of the particle. 7. Constitutive Assumptions In the following sections we shall analyze binary mixtures of Euler fluids which obey the thermal equation of state of classical ideal gas: pa = !^-paT;
(a = 1,2),
where ma is atomic mass of constituent a and kB is the Boltzmann constant. Corresponding internal energy densities have the form: ea=
/ "
(a = 1,2),
lV
where 7 a is ratio of specific heats of constituent a. Therefore, the total pressure p = pi + P2 and the intrinsic internal energy density £j = ^-£i + ^-£2 could be written in a form similar to that of the single fluid case 19 : kB „ V = m(c) — r r P ^ ;'
kBT £/ = m(c) (7(c) — 1)'
provided we define average atomic mass and average ratio of specific heats in the following manner: 1
c
m(c)
mi
1 7(c) - 1 7.1. Kawashima
Condition
1-c
+ c
7712
m(c)
7i - 1 mi for the
1 — c m(c) 72 - *
m
2
mixture
In the following text the basic results about Kawashima condition are presented for the case m 1 ss m-i = m but the results remain the same also in the general case. In this case the characteristic eigenvalues A and the right eigenvectors d were evaluated by Ruggeri and Simic 19 for the one
452
dimensional system, choosing the field u = (p,c,v,J,T). equilibrium state UQ = (po,co,0,Q, To) we have: id)
(3)
(5)
_
\K0> = Cs
k >£' = 0,
-Cs,
(31)
_ A(4) — Cg 0
- —Cg, Ai 0W —
(32)
/
_£Q_
To(7o-l)
Jo
0 £s
dj
In particular in
r0(70-i)
0
V i
/
o
0
0 0 0
do1
U
!
Cs To ( 7 o - l )
0
V /
1
/
o \ i Po Ce
1
Po°e 0
0
/
\
\
En
/ T0(7o-l) *
0
1
V °7
V o / where k = ks/m, sound wave).
Cs = y/j0 kT0 (sound velocity), Cg — \fkT0
(adiabatic
du' Evaluating the matrix A' 0 = ——, we obtain the right eigenvectors of the OTl
symmetric system d' = A °d: ( <
=
% 7?(7o-l)
0
V
( -W \
\ d' 5 i
a
0
Tj Ivy.2 \ To ( 7 2_ -1 l\ ) \
fcr
d' 3 a0 —
To
0 0
j£
Tg(y0-i)
453 -c0Ce \
(CQC9\
-Ce d
0
-co
V 0 /
an —
V
-co 1 0 /
with r = i / ( 7 l - 1 ) - 1 / ( 7 2 - 1 ) . Looking the eigenvectors and (14), we can deduce immediately that in the case of chemical reactions the system satisfies the K-condition. In fact every eigenvectors does not have zero simultaneously in the second and forth position. While in the case without chemical reaction the system is of composite type because the eigenvectors corresponding to the sound velocities and the contact wave d 0 , d 0 , d 0 have a null forth component. Therefore for the previous general theorems, we conclude that mixtures of Euler fluids with chemical reaction having truly dissipative character have global smooth solutions that converge to an equilibrium state of the Euler single fluid provided the initial data are sufficiently smooth. While, without chemical reaction the global existence remains an open problem. Acknowledgments This paper was supported by fondi MIUR Progetto di interesse Nazionale Problemi Matematici Non Lineari di Propagazione e Stabilita nei Modelli del Continuo Coordinatore T. Ruggeri, by the GNFM-INDAM, and by Istituto Nazionale di Fisica Nucleare (INFN). References 1. K.O. Friedrichs, P.D. Lax, Systems of conservation equations with a convex extension. Proc. Nat. Acad. Sci. USA, 68 1686 (1971). 2. T. Ruggeri, A. Strumia, Main field and convex covariant density for quasi-linear hyperbolic systems. Relativistic fluid dynamics, Ann. Inst. H. PoincarS, 34 A 65 (1981). 3. S. K. Godunov, An interesting class of quasilinear systems. Sov. Math., 2 947 (1961). 4. G. Boillat, Sur l'Existence et la Recherche d'Equations de Conservation Supplementaires pour les Systemes Hyperboliques. C.R. Acad. Sc. Paris, 278 A 909 (1974). Non Linear Fields and Waves. In CIME Course, Recent Mathematical Methods in Nonlinear Wave Propagation, Lecture Notes in Mathematics 1640, 103-152 T. Ruggeri Ed. Springer-Verlag (1995). 5. G. Boillat, T. Ruggeri, Hyperbolic Principal Subsystems: Entropy Convexity and Sub characteristic Conditions. Arch. Rat. Mech. Anal. 137 305 (1997).
454 6. G. Boillat, T. Ruggeri, On the shock structure problem for hyperbolic system of balance laws and convex entropy, Continuum Mech. Thermodyn. 10, 285 (1998). 7. S. Kawashima, Large-time behavior of solutions to hyperbolic-parabolic systems of conservation laws and applications. Proc. Roy. Soc. Edimburgh, 106A 169 (1987). 8. A. E. Fischer, J.E. Marsden, The Einstein evolution equations as a firstorder quasi-linear symmetric hyperbolic system. Commun. Math. Phys. 28, 1 (1972). 9. A. Majda, Compressible fluid flow and systems of conservation laws in several space variables, Springer Verlag, NewYork, 1984. 10. C. Dafermos, Hyperbolic Conservation Laws in Continuum Physics. Springer Verlag, Berlin (2000). 11. Y. Zeng, Gas dynamics in thermal nonequilibrium and general hyperbolic systems with relaxation, Arch. Ration. Mech. Anal. 150 no. 3, 225 (1999). 12. B. Hanouzet, R. Natalini, Global existence of smooth solutions for partially dissipative hyperbolic systems with a convex entropy. Arch. Rat. Mech. Anal. 169 89 (2003). 13. T. Ruggeri, D. Serre, Stability of constant equilibrium state for dissipative balance laws system with a convex entropy. Quarterly of Applied Math. (2003). 14. I. Miiller, T. Ruggeri, Rational Extended Thermodynamics, 2nd ed., Springer Tracts in Natural Philosophy 3 7 , Springer-Verlag, New York (1998). 15. C. Truesdell, Sulle basi delta termomeccanica, Rend. Accad. Naz. Lincei 8, 158 (1957). 16. I. Miiller, A new approach to thermodynamics of simple mixtures, Zeitschrift fur Naturforschung 28a, 1801 (1973). 17. T. Ruggeri, The binary mixtures of Euler fluids: A unified theory of second sound phenomena in Continuum Mechanics and Applications in Geophysics and the Environment, Eds. B. Straughan, R. Greve, H. Ehrentraut and Y. Wang, p. 79, Springer-Verlag, Berlin (2001). 18. T. Ruggeri, Galilean invariance and entropy principle for systems of balance laws. Contin. Mech. Thermodyn. 1, No.l, 3 (1989). 19. T. Ruggeri, S. Simic, Non linear Wave Propagation in Binary Mixtures of Euler Fluids. Contin. Mech. Thermodyn. 15, No.7, (2003).
N O N LINEAR WAVE P R O P A G A T I O N IN B I N A R Y M I X T U R E S OF EULER FLUIDS*
T. RUGGERI Department of Mathematics and Research Center of Applied Mathematics (C.I.R.A.M.) University of Bologna, Via Saragozza 8, 40123 Bologna, Raly E-mail: [email protected] S. S I M I C Department of Mechanics, Faculty of Technical Sciences University of Novi Sad, Trg Dositeja Obradovica 6, 21000 Novi Sad, E-mail: [email protected]
Serbia
In this work we analyze propagation of non linear waves in mixtures of ideal Euler fluids. If the difference between molecular masses is negligible, we can separate the properties resembling the single fluid case from the ones peculiar to mixtures. We also showed that diffusive /c-shock is locally exceptional.
1. Introduction It is well-known that extended thermodynamic theory could be well applied to the study of the mixtures of Euler fluids, i.e. fluids which are neither viscous, nor heat-conducting (see Miiller and Ruggeri 1 ). Recently Ruggeri 2 observed that a binary mixture of Euler fluids could be seen as a single one heat-conducting fluid. Purpose of the present account is to give a brief survey of non linear wave propagation in such a medium. Namely, we shall see what are the main consequences of specific structure of the mathematical model on characteristic speeds and non linear waves, i.e. acceleration and shock waves. *This work was supported in part (T.R.) by fondi MIUR Progetto di interesse Nazionale Problemi Matematici Non Lineari di Propagazione e Stabilita nei Modelli del Continuo Coordinatore T. Ruggeri, by the GNFM-INdAM, and by Istituto Nazionale di Fisica Nucleare (INFN), and (S.S.) by the Ministry of Science, Technology and Development, Republic of Serbia (Project No. 1874).
455
456
2. Mathematical Model and Basic Assumptions Governing equations for the binary mixture of Euler fluid could be written in the form of a single fluid if the field variables describing behaviour of the mixture as a whole (density p, velocity v and temperature T) are extended with variables describing behaviour of one constituent (concentration variable c and diffusion flux vector J ) . Here we shall analyze one-dimensional case without chemical reactions, thus reducing the balance laws to the form: dtp + dx(pv) = 0; dt(pc) + dx(pcv + J) = 0; dt{pv) + dx (pv2+p+— dt{pcv + J)+dx
r)=0; pc(l - c)
(1)
(pcv2 + 2vJ + — + v j =
-P(T)J;
Governing equations consist of balance laws of mass, momentum and energy of the mixture (Eqs. (l)i, (1)3 and (1)5) and balance laws of mass and momentum of one constituent (Eqs. (1)2 and (1)4). In (1) we used p = Pi + V2 for the total pressure and pe = p\E\ + p^£2 for the total internal energy of the mixture, v = p\ for the partial pressure of one constituent while: 1
(„
+
, Pi
,
1
N
+_ 2 \ f
(,.
i = ^ £ i" M
, P2 , +1 £2+
o
S ^>
(2)
is the difference of enthalpies and m\ = —(3(T)J represents the momentum exchange between constituents. We shall assume that constituents obey thermal and caloric equation of state of ideal gas pa = (kB/ma)paT, ea = Pa/(pa(la-!•)), a = 1,2, fc# = 1.38 • 1 0 - 2 3 J/K - Boltzmann constant, ma molecuar masses of constituents. These constitutive assumptions, which are in accordance with entropy principle, give to the system (1) the structure of a quasi-linear hyperbolic system of balance laws. Following the same idea which led to specific structure of governing equations, we can derive unifying constitutive equations for the mixture p = p(p,c,T), e = i(c,T), provided an average atomic mass m = m(c) and average ratio of specific heats 7 = 7(c) are introduced. General form of these functions could be found in the recent paper of Ruggeri and Simic 3 .
457
Here we shall restrict our attention to the special case characterized by the equal masses assumption: m = mi « 1712 = const. It was shown1 that this assumption leads to decoupling of the equations for linear wave propagation in binary mixtures. At the same time unifying constitutive equations are reduced to the following form: p = kpT; v = p1 = cp;
kT = ——-; 7(c) - 1
£l
1 —— 7(c) - 1
=
(3)
c 1— c H ; 7i-l 72-1
, ks k = — , m
where ej is intrinsic internal energy and we shall assume 71 < 72. 3. Characteristic Speeds and Acceleration Waves Although equal masses assumption could be criticized as too restrictive, and seems to be too close to the single fluid model, it gives a possibility for the qualitative analysis of the problem. Namely, if the system (1) is transformed into normal form dtu + A(u)<9xu = f(u) for field variables u = (p, c, v, J, T)T, characteristic equation det(A — AI) = 0 is of the fifth degree with respect to characteristic speed: fcTp2 ( 7 ( c ) - l ) { ( l - 2 c ) J 3 - ( l - 5 c ( l - c ) ) -2c(l-c)(l-2c)p2U2J
-c2(l-c)2p3U(U2-kT)}
2
2 2
+ (J -2(l-c)pUJ 2
x (J -2cPUJ
pUJ2 (4)
2
+(l-c) p (U -kT)) 2 2
2
+c p {U -kT))
(PU + rJ(rf{c)-l))
= 0,
where U = v - A and T = 1/(71 - 1) - 1/(72 - 1). In the sequel we shall analyze the wave propagation into a region where the mixture is at rest and without diffusion (vo = 0, Jo = 0): Uo = ( P O ) C O , 0 , 0 , T O ) T . This assumption simplifies the structure of Eq. (4) and leads to a conclusion that characteristic speeds in unperturbed state Uo could be splitted in two groups. First group has the same form as in the case of single Euler fluid:
A^ = -V^kTo,
A 0 3 ) =0,
A05) = V7o~fcTb,
(5)
but carries the information about the mixture through the average ratio of specific heats 70 = 7(co). Second group, peculiar to mixture, corresponds to the propagation of the so-called second sound:
A02) - -y/kU,
A04) = Jk%,
(6)
458
here related to changes in concentration and diffusion flux. Propagation of the highest speed acceleration waves, which propagate along characteristic 4>(x,t) = x — Cot = 0, Co = ^"fokTo, is based upon amplitude equation of Bernoulli type. As shown by Boillat 4 (see also Ruggeri 5 ) it governs the behaviour of weak discontinuities for all hyperbolic systems of balance laws. It can be showed that critical times for the formation of shocks in horizontal and vertical direction, respectively: tcrt{co)
=
2C0 G0(7o + 1 ) '
tcrt{co)
2C0 M Ii #7o
57o
!>}
G0(7o + :
(7)
have the same form as in the single fluid case5 when expressed in terms of initial acceleration jump Go = [«t](0) = [u^,](/>t(0). Moreover, they are bounded by the corresponding values of critical times of the constituents:
4?t = *crt(0) < tcrt{c0) < tcrt{l) = t ( i )
(8)
4. Shock Waves Analysis of shock waves in binary mixture of Euler fluids will be based upon the solution of Rankine-Hugoniot equations which govern the jump of field variables across the wave front:
[H = 0; [pcu — J] = 0 ; pu
+p-
J2 pc{\ - c)
pcu — 2Ju H
J2
0;
(9)
hv - 0 ; J2
-pu
ps I u + pu + pc(\ — c)
1
J
a
= 0,
where u = s — v, s - speed of shock. Our attention will be restricted to fc-shocks - weak shock waves which bifurcate from trivial solution of (9) where the speed of shock corresponds to the characteristic speed. Along with the search for nontrivial solutions of Rankine-Hugoniot equations, a question of shock admissibility will be raised. Apart from the classical case of genuine nonlinearity (VUA • d ^ 0 for all u), where Lax condition Ao < s < A and entropy growth criterion rj = —s[h°] + [h] > 0 could be well applied, and linearly degenerate case (VUA • d = 0 for all u), where
459 s = Ao = A and rj = 0, we shall also encounter the case of local exceptionality. In this case condition of genuine nonlinearity is violated on some manifold in state space (so-called critical manifold). Consequently, Lax condition has to be substituted by more general Liu condition s(uo, p.) < s(uo, n) for all no < p. < H, and entropy growth criterion has to adjoined with the principle of superposition of shocks (see Liu and Ruggeri 6 ). Therefore, in the sequel we shall employ the following expression for the entropy growth function across the shock 77 = [puS] — [^]: s
= ^ f T T l n ( ^ ) ) - ' t ( < : l l " : + ( 1 -< : ) 1 °( 1 - c )>^
*= u { rln G)- ln (^)}-
<10>
(»)
We can distinguish following three cases. 4.1.
Sonic
Shock
This case is characterized by the absence of jumps of concentration variable c = Co and diffusion flux J = Jo = 0 and the following nontrivial solution of Rankine-Hugoniot equations: 7r=^- = (l+/ig)M02-/i§;
{1
(1
-rH^ "^ -^ e = ^ = ^ = ™ = ]L{(i where Mo =
UQ/CQ
+
;
(12)
^)M02-»l}{i-ri(i-MZ)},
is sonic Mach number and /i 2 (c) =
7
r : ~ j , /z0 = n(co)-
It bifurcates from the characeristic speed A0 - speed of sound, and obeys the very same properties of shock admissibility as in the single fluid case: it is genuinely nonlinear and admissible for Mo > 1. 4.2.
Diffusive
Shock
In this case nontrivial solution of the system (9) can be expressed in terms of concentration c as shock parameter. It is goverened by the solution of biquadratic equation: a0(c)ujQ + ai(c)wo + a2(c) = 0,
(13)
460
where
LOQ
=
UO/CQJJ, CQD
= VkTo, is diffusive Mach number, and reads: 1/2
U)0
a\{c) 2a 0 (c)
\f g x (c) \ VV 2 a o(c)y
a 2 (c) oo(c)
1 ±l+^o 2 (c), ( ) ' w o( c )
( 14 )
™=r = i - ^ p ! a c
P(CY\
0 = -
=7TW=
M , ,.2.
—r—
1
—
^
r a{c) V a(c)J J = /90 COD W 0 (C) (C - c 0 ),
2f \
'
o^(c)
where the explicit form of coefficients ao(c), ai(c), 02(c), a(c) and /3(c) will be omitted. It can be shown that solution (12) bifurcates from the second sound eigenspeed A0 . This case is particularly interesting since it obeys the property of local exceptionality. Namely, by a straightforward computation one can prove: V u A< 4 >.d( 4 )
= u0
xl
-2co ~^.y/k%
(15)
Co (1 — Co J
so that critical manifold has structure uo = (po,co,vo,0,To) for Co = 0.5 and arbitrary values of other field variables. This result strongly influences admissibility of diffusive shock in such a way that: (i) if Co < 0.5 shocks are admissible for Co < c < c; (ii) if Co > 0.5 shocks are admissible for c < c < Co; (iii) if Co = 0.5 there are no admissible shocks. Numerical computation of Liu condition and entropy growth across the shock confirms this assertion (see Fig. 1). In fact, we compared dimensionless speed of diffusive shock LD and diffusive Mach number LJQ. These results are in accordance with discussion of second sound phenomena in rigid heat conductor given by Ruggeri, Muracchini and Seccia7. 4.3. Characteristic
Shock
The final case is obtained for s = A'3) = A0 = 0 , and corresponds to the characteristic shock in the single fluid case. From the set of the RankineHugoniot equations (9) it is easy to obtain the following result nontrivial
461
solution: v = VQ = 0;
c = CQ;
P = Po;
J = Jo — 0;
p — arbitrary.
!
ID/
^
0.31
0.32
0.33
\
J
^
_, entropy growth region J \
/ 0.30
Lax region
(16)
0.29
0.30
0.31
0.32
0.33
c
6
xW
f
0.49
0.50
0.51
\r\D
0.49
0.50
0.51
xlO. 1.015
5.0 N^>
1.010 0.0
1.005 1.000 0.995
_ Lax region entropy growth region ^ 0.67
0.68
0.69
c
0.70
Lax region
J
^
-5.0 / L enfropy growth region J
0.67
0.68
0.69
0.70
c
Figure 1. Lax condition and entropy growth in diffusive shock for CQ = 0.3, Co = 0.5, co = 0.7 and 71 = 1.35, 72 = 1.40
Acknowledgments The present paper was developed during the stay of Srboljub Simic as a visiting professor Junior in CIRAM of the University of Bologna with a fellowship of the Italian GNFM-INdAM.
462
References 1. I. Miiller and T. Ruggeri, Rational Extended Thermodynamics, 2nd ed., Springer Tracts in Natural Philosophy 37, Springer-Verlag, New York (1998). 2. T. Ruggeri, The binary mixtures of Euler fluids: A unified theory of second sound phenomena in Continuum Mechanics and Applications in Geophysics and the Environment, Eds. B. Straughan, R. Greve, H. Ehrentraut and Y. Wang, p. 79, Springer-Verlag, Berlin (2001). 3. T. Ruggeri, S. Simic, Non linear Wave Propagation in Binary Mixtures of Euler Fluids, to be published in Continuum Mechanics and Thermodynamics. 4. G. Boillat, La Propagation des Ondes, Gauthier-Villars, Paris (1965). 5. T. Ruggeri Stability and Discontinuity Waves for Symmetric Hyperbolic Systems in Nonlinear Wave Motion, Ed. A. Jeffrey, pp. 148-161, Longman (1989). 6. T.-P. Liu, T. Ruggeri, Acta Math. Appl. Sinica 19, 1-12 (2003). 7. T. Ruggeri, A. Muracchini, L. Seccia, Nuovo Cimento D 16, 15 (1994).
H Y P E R B O L I C I T Y REGION FOR F E R M I A N D B O S E GASES
T. RUGGERI Department of Mathematics and Research Center of Applied Mathematics (CIRAM) University of Bologna, Via Saragozza 8, 40123 Bologna, Italy E-mail: [email protected]
M.TROVATO Department of Mathematics, University of Catania, viale A. Doria 6 - 95125 Catania, Italy E-mail: [email protected]
Using Extended Thermodynamics in the 13-moments approach, we write the hydrodynamic balance system for Fermi and Bose gases. We evaluated numerically the characteristic polynomial and the hyperbolicity region. Particular attention is devoted to the completely degenerate case when Fermi gas reaches the 0 °K and when the Bose gas is close to the transition temperature Tc.
1. The moment's method in Extended Thermodynamics The Entropy Principle (EP) plays a fundamental role in Extended Thermodynamics (ET) 1. In fact it provides a powerful constraint in order to select the physical constitutive equations in the case of classical solutions and it becomes a fruitful selection rule for admissible weak solutions 2 . Furthermore, if the principle is combined with the stability requirement of the concavity of the entropy density, it permits to rewrite the field equations in the form of a symmetric hyperbolic system through the introduction of the privileged main field components 3 ' 4 . In the case of a moment system associated to the Boltzmann Transport Equation (BTE) the principle permits the closure of the system and the corresponding distribution function coincides with the one obtained by the Maximum Entropy Principle (MEP) 5 ' 6 . The Entropy Principle is exploiting in the full non linear case without any assumptions about the non equilibrium processes. If we consider processes
463
464
not far from an equilibrium state, an approximate distribution function is usually derived through a formal expansion in the neighborhood of the local equilibrium and so we obtain Extended Thermodynamics theories of M moments and degree a (ET^ theories). The non-linear closure suffers by some analytical problems that were firstly discovered by JUNK and coworkers 7 in particular concerning the domain V of invertibility between the field and the the main field (Lagrange multipliers) and the integrability of the moments (see also 8 ) . Instead, in ET^ we do not have this kind of problems. In fact, thanks to the equilibrium distribution function that dominates any polynomial, all the expressions for the moments are integrable. In the present paper only the ET^ theory in the case of degenerate Fermi and Bose gas is considered. The system of balance equations for the first 13 moments FA = {F, Fi, Fij, Fin} can be compactly written as * **£ + **» =SA + PA , with A = l,. ..13 (1) at axk where FAk = {Fk,Fik,Fijk,Fiiik}, SA = { 0 , ^ , 2 ^ / ^ , 3 i f y / j ) } , U, and PA = {0,0,0, P(ij), PHI} indicate, respectively, the fluxes, the external field productions, the specific external forces and the collisional productions. The structure of this system of equations shows that there are some unknown constitutive functions HA — {F(ijk^,Fuik,P^, Pm} that must be determined in terms of the variables FA while the production terms P, Pi, Pu are null for the conservation of total particle number, momentum and energy respectively. Introducing the mean velocity Vi of gas it is possible, by Galilean invariance, to decompose both the fields {FA} and the functions {HA} in convective and non-convective parts x . Thus we obtain the new set of variables mA = {p, Vi, p, rn^ij), qi} and the new set of constitutive functions GA = {m(ijk), mink, R(ij), Rui}, being p the mass density, p the pressure, m^ij) the stress deviator, q^ the heat flux and RA the new production terms which will depend on the specific nature of the collisions. By following the MEP, or analogously the EP, we calculate the nonequilibrium distribution function and then all unknown constitutive functions can be determined. According to this approach, we have for the variables of local equilibrium {p, p} TT
p{r,t) = —l£(a),
2 hr, T 5 / 2 p(r,i) = - ^ - r / 4 ± ( a ) ,
(2)
465
with 7 — [y/ir/'lmfis + 1)] (2irh2/mkB)2 and sh the particle spin. In the framework of -ET/3, we obtain for the fluxes the closure relations m
= 0,
f 14 J± fcB \ It kB mu
being, in general, i ^ ( a ) the Fermi and the Bose integral functions r+oo ^n(")=
/
n 7
v
(3)
oVTT^2
w " ' J0 exp(a + x ) ± 1 where a = —/x/(/csT) with /i the chemical potential and T the temperature.
2. Analysis of the hyperbolicity zone In the one-dimensional case, we consider as vector of the independent variables UB = {p, v, T, a, q) with v = vi, q = q\, a = — m < n > . Being FA = FA{UB) and FAX = FAX(UB) the system of balance equation (3) can be written in the form Ac^f+AAc^~-=SA
+ PA,
with
A = l , . . . 5 (4)
where A°AC = dFA/dUc, AAC = dFA1/dUc. The system (4) is said to be hyperbolic if: i) the determinant of .4° is not zero, ii) all the roots A^' (with i = 1, 2 , . . . ) of the characteristic polynomial JAB = (AAC)~1ACB are real, iii) if A^ has multiplicity rW there exist in correspondence rW linearly independent eigenvectors. It is convenient to define a suitable speed c c=
10 It (<*) kB, 9 if {a) m
lV2
(5)
and to introduce the dimensionless quantities A = (A — v)/c, a — a/(c2 p),q = q/(c3 p). In this way the characteristic polynomial of JAB assumes the form A 5 +34 A 4 + 5 3 A 3 +32 A 2 + 5 ! A + g o = 0 , where the coefficients cji depend on the dimensionless variables For Fermi gas and Bose gas with T > Tc we obtain: dct (A0 ) det {AAB)-
6 kB o2 [5 / 4 ± ( a ) &mP \31±{a)
3/2±(Q) I ± 0 3 7 ± ( a ) p ° -
(6) {a,a,q}.
(7) W
466
"^ 7±
uii
+
15 7±
25/±
9I
I
25 - 4 :
+3
2
21 5
+
It
I±
6211
It
5 7±
bit
tf £
3 7±
5 7?
0
4
i
o J
3
J
2?-
90 = 0
Z
o
2
(8) 96 25
,52 1
15 :
_
CT+
0
We note that, as well know 1 ' 1 0 , the hyperbolicity is violate in the cases: i) a Completely Degenerate Fermi gas (when T —> 0°K) being 7+(a) = (—a) 2 / ( n +1) and consequently detA0 = 0; ii) a Bose gas in equilibrium condition ( a = q = 0) with T —> Tc+ (where for a —> 0 7^ (a) —> oo, and consequently also g~i = g\ = 0) and we have the eigenvalue A — 0 with multiplicity r = 3 and only two independent eigenvectors 10 . For Strongly degenerate Bose gas with T < Tc we obtain: In this case a = 0 and we have a corresponding phase transition for the system that will be considered as a mixture of two thermodynamic phases, the normal phase and the condensate phase (Bose-Einstein condensation). In this case we have 7~(0) = oo for n < 1, 7~ n > 1, while for {p,p} we obtain
(o) = r m e m / 2 for
'^-IS"-^
P=
VBT^\
(9)
where VB — ( 2 ^ B ^ ) / ( 3 7 m ) and <£> G [0,1] is the fraction of condensed matter. Analogously if we calculate the detA0 and the coefficients gi of the characteristic polynomial (6) we have det(.4%)=-10^V^(l-^), 9s = ~
189 1*1* 125 (/4")2
+
9 5
+
94 = 50 = 0 ,
14 72~ Je" _ 25 ? ^ F
h =
9i = •v.
-^Q
(10)
126 4"4" 125 (74")2 l - y ,
We note that the hyperbolicity is lost both for y> —> 1 (detA0 = 0 for T —» 0 ° K) and in the equilibrium state. In fact, also in this case (T < Tc), in equilibrium (q =a = 0) we obtain the root A = 0 with multiplicity r = 3 and we have only two independent eigenvectors 10 . In non-equilibrium state there are no vanish values of q and/or a for which we obtain five real and
467
distinct roots of the characteristic polynomial. In the figures [1-3] we report the hyperbolicity zone obtained through a numerical computation of the roots of the characteristic polynomial (6), by fixing the values of a and representing in plane {a, q} the regions with zero (white zone), two (light gray zone), or four (dark gray zone) complex conjugate roots. The figure [1] report the region of hyperbolicity, calculated only for a £ [—16,4] because in a different interval the numerical results remain unaltered. We note that the system is always hyperbolic in the neighbourhood of equilibrium, and that the hyperbolicity region increases if the a parameter increases reaching its greatest extension in the case of classic approximation (i.e. large a). In the case of Bose gas, with T > Tc, we represent, in the figure [2], the hy(1*orcl) a = - 1 6
-T
1
i'
(1 a ord.) a = - 7
,„
1
1
1
1.0 |
«
1
•
1
•
•
1
•"
1
•
• 1 . ( 1 s t ord) a = 4
1
*-
'
'
'
•
-0.5
0.0
0.5 •
ic
o.o ie
o.o •
-0.5 • —. 1 . ( 1 s t ord) n = 0
1
.
1
.1.0 I
0.5 I-
10
1C
0.0
0.0 •
-0.5 -
J -1.0
-0.5
0.0 q
Figure 1.
0.5
1.0
-1.0 I -1.0
'
0.5
1.0
q
Hyperbolicity zone of a degenerate Fermi gas for differents a
perbolicity region for a £ [0.01,4] while for larger a values the hyperbolicity zone remain unaltered. We observe a large variation of hyperbolicity for very small a (strongly degenerate condition) and that, on the contrary of the Fermi gas, for increasing a the region of hyperbolicity decreases reaching its smaller extension in the case of classic approximation. In Figure [3] we report the region of hyperbolicity for T
468
and for (p —> 1 (e.g. T —> 0 °K) the hyperbolicity of system is lost. The details of the present work can be found in a paper in preparation 11 . (r'ord.) a = 0.05
ie
o.o •
ie
o.o
IO
O.o •
10
0.0
-1.0
-0.5
0.0
0.5
1.0
-1.0
-0.5
q
Figure 2.
0.5
q
Hyperbolicity zone for a degenerate Bose gas with X > Tc. (1 s t ord.)
1.0
0.0
(r'ord.) cp = 0.4,T
0.5
to
0.0
IO
0.0 -
-0.5
(1 s1 ord.) (p = 0.95, r < T 0
1.0
d1" ord.)
1.0
cp = 0.999, T < T0
: 0.5
0.5
IO
0.0
IO
-0.5
-0.5
-1.0
-0.5
0.0
q Figure 3.
0.0
0.5
1.0
-
•
-1.0
-0.5
0.0
0.5
1.0
q
Hyperbolicity zone for a strongly degenerate Bose gas with T
469 Acknowledgments This paper was supported by fondi M I U R Progetto di interesse Nazionale Problemi Matematici Non Lineari di Propagazione e Stabilitd nei Modelli del Continuo. Coordinatore T . Ruggeri, by the GNFM-INDAM, and in part (M.T.) by a local contract of University of Catania.
References 1. I. Muller and T. Ruggeri, Rational Extended Thermodynamics, Springer Tracts in Natural Philosophy, 37, Springer-Verlag, New York (1998). 2. C. Dafermos, Hyperbolic Conservation Laws in Continuum Physics, SpringerVerlag, Berlin (2001). 3. G. Boillat, C.R. Acad. Sc. Paris, 278 A, 909 (1974). Non Linear Hyperbolic Fields and Waves in CIME Course Recent Mathematical Methods in Nonlinear Wave Propagation. T. Ruggeri Ed., Lecture Notes in Mathematics n. 1640, Springer-Verlag, Berlin, 1 (1996). 4. T. Ruggeri and A. Strumia, Ann. Inst. H. Poincare, 34 A, 65 (1981). 5. G. Boillat and T. Ruggeri, Continuum Mech. Thermodyn., 9, 205 (1997). 6. W. Dreyer, J. Phys. A: Math. Gen., 20, 6505 (1987). 7. M. Junk, J. Stat. Phys., 9 3 , 1143 (1998); W. Dreyer, M. Junk and M. Kunik, On the approximation of kinetic equations by moment systems, WIASPreprint No 592, Berlin (2000). 8. T. Ruggeri, WASCOM 99, 10th Conference on Waves and Stability in Continuous Media (Vulcano), 434, World Sci. Publishing, River Edge, NJ, (2001). 9. F. Brini, Continuum Mech. Thermodyn., 13, 1 (2001). 10. T.Ruggeri, L.Seccia, Meccanica, 24, 127 (1989). 11. T. Ruggeri and M. Trovato Hyperbolicity in Extended Thermodynamics of Fermi and Bose gases, in preparation (2003).
A CELLULAR N E U R A L N E T W O R K A P P L I E D TO T H E EQUATIONS OF MATHEMATICAL PHYSICS
FRANCESCO RUNDO* STMicroelectronics,
Consumer and Microcontroller Catania Italy, E-mail: [email protected]
Group,
In recent years the artificial neural networks have improved the method for solving complex problem in many different areas such as pattern recognitions, image processing, function approximation etc.. In this paper a particular kind of neural network called Cellular Neural Network will be used as a new method for solving partial differential equations. The Cellular Neural Networks (CNNs) are a powerful tools based on some aspects of neurobiology and adapted to integrated circuits or computer simulation. The CNN offer a new computational model which has important potential applications in such areas as image processing, signal processing, and finally in partial differential equations (PDEs) solving. As example of previous assertion, a classic P D E will be considered in order to show how the CNN can be applied successfully. The P D E considered is The Heat Conduction Equation.
1. The Cellular Neural Network: A n introduction 1.1. A Brief introduction
to the Cellular
Neural
Networks
The cellular neural network is an analog, nonlinear, real time parallel processing array network, which is a new model of neural networks. Since L.O. Chua and L.Yang first proposed the cellular neural network (CNN) [1] in 1988, the re-search on CNN has developed rapidly, and various applications^] of CNN have also been developed. There are many sorts of applications: image processing, handwriting recognition, moving direction detection, velocity estimation etc. An important application of the CNN is the Real Time computation which can be applied efficiently for solving the Partial Differential Equations without the classic tedious digital simulation used in this case. In this paper, an important kind of PDE is considered: the Heat*Work partially supported by University of Cagliari
470
471
Conduction Equation. The CNN is implemented by using integrated operational amplifier circuits[l][2] but a computer simulation can be performed[l]. CNN's hardware performance and its simulation results are satisfactory. 1.2. The Theory of the
CNNs
A basic processing unit in CNN is a cell, and each cell is made of the linear resistance, the linear capacitance, the independent current source and the nonlinear VCCS[1]. The cell in CNN is only connected to its neighboring cells. They have direct effects on each other, but non-neighboring cells only have an indirect influence [2]. In CNN, the r-neighborhood of a cell C(i, j) in a M x N CNN is defined by: Nr(i,j)
= {C{k, I) : max \k - i\, \l - j \ < r}
(1)
where r is a positive integer number and i € [1,M], j G [1,-W]. The equivalent circuit of a CNN cell C(i,j) is described in the original paper written by L. Chua and L. Yang in which the cirucits contains discrete components such as resistors, capacitors, operational amplifiers[1][4] etc... The state equation of the CNN cell can be descibed as follow: C
~^L--^+
x
+
E
Mi,3\k,l)ykl
c(k,i)eNr(i,j)
J2
B(i,j;k,l)ukl + I
(2)
c(k,i)eNAi,j) In which the x^ represent the state of the cell C(i,j), the yki represent the output signal of the single CNN cell and the Uki and the I represent the input signal the bias signal. The yui signal is a PWL function described by means of the following equation: Vij = 2^Xkl
+ 1
1.3. The CNNs used for solving
I ~ \Xkl ~~ 1 D
(3)
PDEs
It is well known that the heat-conduction equation is the following: [ ) dt 9*2 First select two mesh constants h and k, with the stipulation that m = l/h is an integer. The grid points for this situation are (xi,tj), for i = 0,1,..., m
472
and for j = 0,1, ...,n. We obtain the difference method by using Taylor's series to form the difference quotient: dU(xi,tj)
=
U(xj,tj)
-U(xi,tj^i)
at
d2U^Xj^j)
k
2'
k
ot2
u
where
_ U(xi+1,tj
dx2
- 2U(xj,tj)
+
U(xj-i,tj
+ *(&,*i) -r»W3
h2
~
(6)
in which the error function 6(t;i, tj) is equal to:
where & e [XJ_I,:EJ]. The partial-differential equation implies that at the interior grid point for each i = 1,2,..., m and j = 1,2,..., n we have:
~di
°
dt2
=
(8)
°
Using the backward-difference method and using the difference quotients Eq.(5) can rewritten as follow: Wj,j
- Wj,j-1 _
W 2
i+l,j -2Wj,j +Wj-ij
° '
k
^
_
-
U
W
where Wij approximates U(xi,tj). The local truncation error for this difference equation is: iJ
~
kdU2(Xi,^) 2 9*2
2 4
2h
a 12
d U(^,t3) 9a .4
UUj
Solving Eq(9) with A = k£: (1 + 2A)tyiij = Aiu i+ i,j + Awi_ij + Wij-i
(11)
for each i = 1,2, ...,m — 1 , and j = 1,2, ...,n — 1. Using the knowledge that for Wifl = f(xi) each i = 1,2,..., m — 1, and wmj — WQJ = 0 for each j = 1,2, ...,n — 1. The difference equation Eq.(ll) is similar to the state voltage of a cell Eq.(2). Their activity performance depend on the mutual effect of partial space points. They are both difference equations consisting of neighboring variable in dynamic procedure [3] [4]. Their difference is that Eq.(2) is a nonlinear difference equation, but Eq.(ll) is linear. If making Eq.(2) equivalent to Eq.(ll), we only restrict the input voltage of linear VCCS[1] to: \uk,i\ < 1, i.e. it works within linear region. So Eq.(ll) can
473
be implemented by the 3x3-neighborhhood CNN. At this point let us consider the heat-conduction equation: ^d2U{xutj) dt2
dU(xj,tj) dt
{
}
in which x G [0,1] and t > 0. If n=5 and m=4, step sizes h = l / 5 = 0 . 2 and k=0.125, together with Eq.(9): 3.125w i+li j + 3.125wi_i,j + Wj,j_i ~ 7^5
Wi j
'
{
'
which can be solved by 3x3 Cellular Neural Network implemented through the discrete components [1] [4]. 2. The C N N simulation: Results. The following table shows the results obtained comparing the CNNs method with the classic simulation method based on digital computations (only few cells have been considered): Table 1.
The results obtained using the CNNs approach CNNvalue
Error
i
h
U
1
1
0.125
0.2
0.258
0.266
3.2
1
2
0.125
0.4
0.412
0.431
4.4
2
1
0.25
0.2
0.118
0.1213
2.7
2
2
0.25
0.4
0.202
0.1960
3.0
2
3
0.25
0.6
0.204
0.1960
4.0
3
2
0.375
0.4
0.092
0.0890
3.3
3
3
0.375
0.6
0.089
0.0890
0
3
4
0.375
0.8
0.052
0.0540
3.7
3
Simulation
As it is possible to note looking at Table 1 (in percentage) the results obtained by means of CNN are satisfactory taking into account that the computation performed by CNN is a Real Time computation as consequence of analog processing of the data. Both the heat-conduction equation and the other PDEs, can be transformed into a set of differential equations at first, then can be implemented by using the hardware (IC OP-AMP or ASIC chip). Experimental results show that the hardware performance of CNN is in agreement with the computer simulation.
474
Acknowledgments T h e author t h a n k s Mr. Sebastiano Pennisi for his continuous encouragement.
References 1. L.O.Chua and L.Yang, Cellular Neural Networks: Theory 35, IEEE Trans, on Circuits and Systems, 1988, 35(10):1257. 2. T.Roska and L.O.Chua, The CNN Universal Machine: an Analogic Array 40,IEEE Trans, on Circuits and Systems, 1993, (40): 163 3. Richard L.Burden, Numerical Analysis 35, PWS-kent Publishing Compang, 1985. 4. L.O.Chua and T. Roska, The CNN Paradigm 40, IEEE Trans, on Circuits and Systems, 1993, VOL(40).
F R O M K I N E T I C SYSTEMS TO D I F F U S I O N EQUATIONS*
F. SALVARANI Dipartimento
di Matematica, Universitd degli Studi di Pavia, E-mail: [email protected]
ITALY
J. L. VAZQUEZ Departamento
de Matemdticas, Universidad Autonoma de Madrid, SPAIN E-mail: juanluis.vazquezQuam.es
We study the limiting behavior of the Cauchy problem for a class of Carlemanlike models in the diffusive scaling with data in L 1 . We show that, in the limit, the solution of such models converges towards the solution of a nonlinear diffusion equation with initial values determined by the data of the hyperbolic system.
1. Introduction We consider the Cauchy problem for the following system of balance equations: ( dus
ldue
1
dvs — K at
1 dv£ 1 , -^— = -rk{u £ ,v e ,x)[u £ -ve) e ox el
(i) xEK,t>0,
with initial conditions UQ{X) and vo(x), where ue — ue(x,t), ve = ve(x,t) and k(u6,ve,x) is a nonnegative function, called the interaction rate (or rate coefficient). Such a system can be interpreted as describing a gas composed of two kinds of particles moving parallel to the z-axis with constant and equal (in modulus) speeds, one in the positive a;-direction with density u£(x,t), the other in the negative ^-direction with density ve(x,t). 'Work partially supported by the EU financed network no. HPRN-CT-2002-00282.
475
476
The quantity e > 0 is a scaling parameter. The scaling leading to (1) is particularly interesting because the limit e —> 0 + (called hydrodynamical limit) leads, at least formally, to diffusive type equations which can be viewed as the Navier-Stokes equations of the fictitious gas. Hence the name of diffusive scaling. Note that the speed 1/e —> oo. The process is as follows: we introduce two macroscopic variables, the mass density p£ and the flux j £ , defined by p£ — (u£ +v£), j£ = (ue — v£)/e respectively. In the typical case when the rate (which characterizes the interactions between gas particles) has the form ka(u£,v£,x) = (u£ + v£)a, most considered in the literature (we note that, when a = 1, we recover the Carleman model), System (1) is equivalent to the following macroscopic equations for the mass density and the flux dt
dx (2)
6
dt
+
dx ~
zpe3e
posed in (x,f)£Mx (0, T) with initial data for density and flux, p£(x, 0) = uo(x) + vo(x) and je(x,0) — (uo(x) — v0(x))/e. If we now are allowed to disregard the term e2dj£/dt in the limit e —>• 0, we formally obtain the following nonlinear heat equation for the limit density p = l i m ^ o Ps-
dp=ld_(±dp\ dt 2dx\padxJ'
KJ
with initial conditions p(x, 0) = u0 + v0- This is called the diffusive limit. Much more general forms of the rate function k can be admitted in the study of this limit process (a set of convenient assumptions on k will be stated and discussed below). For such general k the factor pa in System (2) must be replaced by k{{pe + ejE)/2, (pe - ej£)/2, x). Assuming that in the limit ej£ -> 0, the denominator pa in the limit equation (3) becomes k(p/2,p/2,x). An interesting mathematical problem is posed, i.e., justifying this limit process for different choices of a (or, more generally, the function k) and under suitable assumptions on the data. It has been the object of a number of papers (a list is provided in another paper of the same authors 5 ), where information is given on the convergence of p£ — u£ + v£ as e —> 0 + to a function p(x,t), solution of the nonlinear diffusive process, particularly in the model cases ka(u£, ve,x) = (u£ + v£)a.
477
Our main interest is to develop a theory that applies to all nonnegative data in the typical functional space L1 (K), and to include interaction rates k of type ka(ue,vs,x) = (u£+v£)a for different values of the real parameter a. In this note we will simply provide the statements of the main theorems, as well as a little discussion on them. The detailled proof of the theorems of this article, as well as the extension to bounded domains fl with Neumann (specular) boundary conditions and to other classes of initial data, are available in the previously quoted paper of the authors 5 . 2. Preliminaries In what follows, we always assume that the interaction rate k satisfies some basic properties that we list next. Definition 2.1. We say that k(u£,ve,x) if:
is an admissible interaction rate
R l . k is a measurable real function defined for u£,v£ > 0 and x £ l ; R2. for every A > 0 there exists M = M(A) > 0 such that 1/M < k(u£,v£,x) < M for all 1/A < ue,v£ < A, and a ; £ l . These minimal conditions are used in the literature 2 . They will be always assumed throughout the work and are satisfied by the typical rate coefficients k = (u£ + v£)a for all a. They have to be strengthened in some developments as follows. Definition 2.2. We say that fcis a regular interaction rate if it is admissible and: R3. k(ue,ve,x) is continuous as a function of u£ and v£ for a.e. x. Moreover, the function k(uE,ve,x)(ve — ue) is uniformly Lipschitz continuous as a function of u£ and vs for bounded values of these arguments; R4. for every A > 0 there exist M = Af (A) and N = N(\) > 0 such that JV < k(u£,ve,x) < M for all x G K and 0 < u£,vs < A. It is easy to check that, for the choice k = ka = (u£ + v£)a, Conditions R l and R2 hold, R3 is satisfied for a > 0 and R4 is only satisfied for a = 0. All are satisfied when ka is replaced by k(u£,v£,x) = ka(u£ + 6, ve + 5,x), S>0. Another property that plays a role is dissipativity.
478
Definition 2.3. We say that an admissible rate k is dissipative if for every a a i, 2,b2,b2 G IS and a.e. i £ l (&i (61 - Oi) - k2 (62 - 02)) (sign [ai - a2] - sign [bi - 62]) < 0, with fci = k(ai,bi,x),
k2 =
fc2(a2,62,:r).
The rate functions fc(we,u£, 2:) = ( u £ + u £ ) a with |a| < 1 enjoy a stronger version of dissipativity, called T-dissipativity. Definition 2.4. We say that an admissible rate k is T-dissipative if, for every ai,a2,b2,b2 € 1R and a.e. 1 £ I , we have (k(a1,bi,x)(b1-a1)-k(a2,b2,x)(b2-a2))(
sign + [a1-a2\-
sign + [&i-&2]) < 0
We give now the definition of weak solution of our system. Definition 2.5. For given initial conditions uo(X), VQ(X) € L1(R)DL°°(R), we define a weak solution of System (1) as a pair of functions (ue,vE) G C([0,T] : L p (K)) n L°°(E x (0,T)), 1 < p < 00, such that the equation is satisfied in the sense of distributions and the initial data are recovered in the sense of traces as t -> 0. Lions and Toscani2 prove, in a slightly more general form, the following theorem: Theorem 2.1. Let 0 < u0(x),v0(x) £ L 1 (K)DL°° (E) and k be admissible. Then the initial value problem for System (1) admits a unique global weak solution ue{x,t),ve(x,t) £ i ° ° ( l x (0,T)) n C([0,oo);L p (E)). 3. Limiting behavior with a regular interaction rate After these considerations we will solve System (1), with u£(x,0) = uo(x) and ve(x,0) = vo(x), in the class of bounded solutions and initial data if the system has a regular interaction rate in the sense of Definition 2.2. This implies that there are suitable a priori estimates on the solutions u£, ve, needed to perform the passage to the limit e —> 0. Let us write K(p,x) = k(p/2,p/2,x). We denote by S = SR,T the bounded strips (—R,R) x (0,T) in space time. We prove 5 the following result: Theorem 3 . 1 . Let (ue,ve) a sequence of solutions for the initial value problem of System (1) with initial values ue(x,0) = uoe(x), vs(x,0) =
479
v0e(x), u06{x)1voe(x) G L°°(E) such that 0 < u0£(x), v0e(x) < M and UQE -*• UQ, voe —»• «o in the sense of weak convergence in Ll0C(M). If k is a regular interaction rate, then there exists a non-negative function p{x,t) 6 L°°(S) fl C(S) such that ps = u£ + vE converges strongly to p{x,t) in L2(S), for every bounded strip S, and the limit density p(x,t) is the unique bounded weak solution of the Cauchy problem for the nonlinear diffusion equation
I=i H i ) •
«>
with diffusivity D(p) = l/(2K(p)), and taking the initial data po{x) = UQ{X) + vo(x) in the sense of traces. Moreover, j e is uniformly bounded in L2(S). It follows that ue(x,t) —> p(x,t)/2, ve(x,t) —>• p(x,t)/2, and eje{x,t) -» 0 a.e. and strongly in L2(S) for every S. The proof of the previous theorem is based on an entropy estimate, as well as the div-curl lemma of compensated compactness theory. Finally, after having identified the limit of the kinetic system, we have proved that the limit equation takes on the initial condition when t -> 0 in the weak sense (trace sense). 4. Initial data in the class L\ for ka, \a\ < 1 The main analytical problem of these models lies in the fact that k often vanishes or becomes infinite, as in ka for a ^ 0. The approach of the previous section justifies the diffusive limit even for these rates if the initial data satisfy the conditions 0 < S
480
ka = (ue + ve + 25)a has all the properties required to develop a good theory, both in L1 and in L°°. These properties are listed in Definitions 2.1 and following. For general data in L1 (M) we still have to pass to the limit 6 —> 0 + to recover the solutions of the original problem. In the range \a\ < 1, the functional setting corresponds to T-dissipative operators. This is a typical property of nonlinear diffusion equations of the types dealt with here, and it implies that the equations generate semigroups of contractions in the corresponding functional space, here Z/^j.(ffi): Proposition 4.1. Let / i = (u\(x,t),vi(x,t)) and f2 — (u2(x,t),V2(x,t)) two solution of System (1) when \a\ < 1, with initial data of class L1 /o,i = (uifl(x),vift(x)) and / 0 ,2 = (u2,o(x),v2fi(x)) respectively. Then ll/i ~ /2II1 < ||/o,i
—
/o,2||i-
Moreover, if uifi(x) < v,2,o(x) and ^1,0(2:) < «2,o(a;) almost everywhere, then the solutions are such that ui(x,t) < U2(x,t) and v\{x,t) < V2{x,t) almost everywhere, for all t £ M + . Applying this concept with data in that space, and in particular the property of monotonicity of the solution with respect to the initial conditions, we prove the convergence result to the diffusive limit eliminating all additional conditions on the data of previous works. Definition 4.1. A miJd solution of System (1) is a pair of functions (ue,ve) 6 C([0,T] : L1^)), that can be obtained as limit in C([0,T] : L 1 (E)) of a sequence (un,vn) of weak solutions of the system with data (uon,von) 6 L\(R) (1 L°°(R). The following theorem holds: Theorem 4.1. Let k = ka with \a\ < 1, and let (ue,ve) a sequence of mild solutions for the initial value problem of System (1), with nonnegative initial values uoe(x,0) € i 1 (lR), uoe(a;,0) 6 L x (E) converging to some UQ, v0 € L 1 ^ ) as e ->0. Then there exists a positive and smooth function p such that p£ (x, t) converges to p(x,t) in C([0,T];L 1 (R)) for all T > 0. When — 1 < a < 1, the limit density p(x,t) is the unique weak solution of the Cauchy problem for the nonlinear heat equation (3) with initial data p0(x) = uo(x) + vo(x) G LX(K). If a = 1, we obtain the unique maximal solution of (3), characterized by the property of conservation of mass. We
481 also have ue,ve p < oo.
—» p/2 in LP(S)
on compact subsets S o / Q = l x
(0,oo),
We recall t h a t existence and uniqueness of nonnegative solutions of the Cauchy problem for t h e target equation is guaranteed for a < 1, but it does not hold for a > 1 if t h e d a t a are integrable. Indeed, for 1 < a < 2 the problem with (nontrivial) integrable initial d a t a pQ G L 1 (IR), po > 0, admits infinitely m a n y s m o o t h solutions p € C ( [ 0 , T ] ; L 1 ( K ) ) n C o o ( E x [O.T)) 1 - 3 ' 4 . O u r limit selects t h e maximal one, which exists for all times t > 0 and is characterized by t h e property of mass conservation: / p(x,t)dx JR
= / po(x)dx
Vt > 0.
JR
Finally, let us mention t h a t the subject contains a number of open problems with m a t h e m a t i c a l interest. Thus, the convergence t o t h e diffusive limit for exponents 1 < a < 2 is still poorly understood.
Acknowledgments T h e authors t h a n k G. Toscani for having pointed out t o us the interest of the problem, and for useful information and discussions. F.S. acknowledges moreover the hospitality of the Universidad Autonoma de Madrid, where this paper has been written. References 1. J. R. Esteban, A. Rodriguez, J. L. Vazquez, A nonlinear heat equation with singular diffusivity, Comm. Partial Differential Equations 13, 985 (1988). 2. P. L. Lions, G. Toscani, Diffusive limits for finite velocities Boltzmann kinetic models, Rev. Mat. Iberoamericana 13, 473 (1997). 3. A. Rodriguez, J. L. Vazquez, A well posed problem in singular Fickian diffusion, Arch. Rat. Mech. Anal. 110, 141 (1990). 4. A. Rodriguez, J. L. Vazquez, Nonuniqueness of solutions of nonlinear heat equations of fast diffusion type, Ann. Inst. H. Poincare, Anal, non Lineaire 12, 173 (1995). 5. F. Salvarani, J. L. Vazquez, The diffusive limit for Carleman-type kinetic models, submitted (2003).
A S Y M P T O T I C M E T H O D S IN O P T I O N PRICING*
MARCO SAMMARTINO Dipartimento
di Matematica ed Applicazioni Via Archirafi 34, 9012S Palermo, Italy E-mail: marco6math.unipa.it
We consider the problem of the pricing a European option with transaction costs and non-constant volatility. Using a utility maximization procedure we obtain a fully non-linear Hamilton-Jacobi-Bellman equation. We solve this equation in the asymptotic regime of fast-mean reverting volatility and low transaction costs.
1. Introduction A call option on a stock is a financial derivative that gives the right (but not the obligation) to buy at a specified time in the future and at specified price the underlying asset. The option is therefore characterized by the strike price K and by the expiry time T. A put option instead gives the right to sell at a specified strike price and at a specified time T the underlying asset. Is we denote by St the price in the market of the stock at time t, the value of a call option C(t, St) at the expiry time is therefore given by: C(T,
ST)
= max (ST - K, 0) .
(1)
The fundamental problem of the option pricing theory is to find the value of the option C(t, St) before the expiry time. In particular it is important to determine the right value of the option at the time t — 0 given the stock price So. In fact C(0, So) is the "right" price at which the option should be sold in the market. In 1973 Black and Scholes proposed a pricing formula that is the cornerstone of modern option pricing theory. The assumptions which the theory of Black and Scholes (BS) is based on, are the following: ""This work is supported by the INDAM-GNFM grant "insiemi assorbenti, attrattori e varieta inerziali nella fluidodinamica esterna ed applicazioni alia geofisica"
482
483
(1) The price of the underlying stock St follows a geometric Brownian motion: dSt = St (adt + adWt) .
(2)
In the above formula a is the drift rate of the stock, and is related to the expected return of the stock; a is the volatility of the stock, and gives a measure of how risky the stock is; Wt is a Brownian motion. Both a and a are constant during the life of the option. (2) There exists a riskless asset (e.g. a treasury bond) whose rate of return (interest rate) r, is constant during the life of the option. (3) There is no friction in the market, i.e. the investor does not pay taxes or transaction costs, nor bid-ask spread is present. (4) There is no arbitrage opportunity. This means that any riskless investment must have a rate of return lower than r. The idea of BS was to construct a portfolio made up of a combination of options and stocks, in such a way that risk is hedged away: In the BS portfolio one puts —1 call option and A stocks. The value of the portfolio il(i, St) is therefore given by: II(i, St) = —C(t, St) + StA. Using Ito formula one can compute the infinitesimal variation of the portfolio with time: dU = (-dtC
- aSdsC
- ]-(r2S2dssC
+ ASa\
dt+{-aSdsC
+ AaS) dWt
If one takes A = dsC one can hedge away the stochastic term in the above stochastic differential equation and get a portfolio whose value increases deterministically with time. Using the no-arbitrage hypothesis one must have that the rate of return of the above portfolio must be r, i.e.: dll = (-dtC
- ]-a2S2dssC\
dt = rUdt .
The above relation finally leads to the celebrated Black and Scholes PDE: dtC + ]-a2S2d2sC
+ rSdsC
- rC = 0 .
(3)
The above equation is a backward parabolic PDE. The final condition is given by 1. With a change of variable one can recast the above equation as the heat equation and get an explicit formula for the "right" (in the sense that avoids arbitrage opportunity) price of a call option. For more details and for an alternative approach to option pricing based on the risk-neutral probability measure see e.g. Refs. 1,2,1 .
484
The treatment we sketched above is valid for European style options, i.e. options that can be exercised only at the expiry time. American style options, i.e. options that can be exercised at any time during the life of the option, need a more complicated treatment, because the holder is faced with the problem of exercising the option at the optimal time. This leads to a free boundary parabolic problem and no exact pricing formula can be written. In the case of American option one has to resort to numerical techniques. In the market there are several type of options with payoffs more complicated than the European and American options; these financial instrument are sometimes called exotic options. Since the BS pricing formula was derived, the option market has experienced an exponential growth in the volume of the money traded. Options are used both by speculators for their leverage effect, and by hedgers to reduce or balance risk in a portfolio. Options are the basic tool for financial engineering or risk management. Despite of the importance of the BS pricing formula it is however clear that the assumptions which it is based on are too crude to be considered realistic. The two more serious weakness in the BS model are clearly the fact that it ignores transactions costs and the fact that volatility is considered constant during the life of the option. Regarding transaction costs one has in fact to observe that the hedging strategy used to derive the pricing formula requires to keep in the portfolio A = dsC stocks. This means that the strategy requires to continuously trade the stock to perfectly hedge the risk. When transaction costs are present this strategy can be very expensive and not optimal. Regarding volatility, it is a well known phenomenon the so called volatility smile. In fact, when the BS pricing formula is used to infer volatility from the observed market prices of the options, the plots of strike prices versus volatility exhibits smiles that are incompatible with the BS assumptions. In this paper we shall consider the problem of pricing an option taking into account the presence of transaction costs and modelling the volatility as an Ornstein-Uhlenbeck stochastic process. Stochastic volatility is a widely used tool for modelling uncertainty in financial markets. One can find more on this in Ref.4. An asymptotic analysis of a stochastic volatility model is in Ref.5. To price the option we shall follow a utility maximization procedure, first introduced in option pricing theory in Ref.6. Therefore we will have to solve a stochastic optimal control problem, the control being the amount of stocks held in the portfolio. The resulting Hamilton-Jacobi-Bellman equation will be analyzed in the asymptotic regime of fast mean-reverting
485
stochastic volatility and (very) low transaction costs. A singular perturbation problem arises, that will be tackled with the classical method of the rescaled coordinate. The rescaled coordinate is needed to resolve the no-transaction region, i.e. the region where it is optimal to tolerate some risk and do not rebalance the portfolio. This kind of asymptotic methods, typical in the portfolio theory with transaction costs, where first introduced in option pricing in a series of papers by Whalley and Wilmott, see e.g. Refs. 7,8 . We shall therefore write the price of the option in the form of an asymptotic series, the leading term being the BS price with averaged volatility. The average is taken with respect to the Ornstein-Uhlenbeck invariant measure. 2. Stochastic volatility and transaction costs We want to price a European call option with strike price K and expiry time T, written on a stock. We suppose that the stock follows the geometric Brownian motion (2). Moreover we shall suppose that the volatility at depends on a zt where zt follows a Ornstein-Uhlenbeck process: °t = f(zt)
(4)
dzt = | (m - zt) dt + /3dXt •
(5)
In the above expression £ is the rate of mean reversion, m is the long term mean, /? is the volatility of the volatility. For a study on the reliability of the above model in describing the actual behavior of the volatility see e.g. Ref.9. The Brownian motion Xt is usually correlated (negatively) with Wt, the process driving the underlying stock, i.e. one has:
Xt=pdWt
+
y/l-p2dZt,
where p is the rate -• f correlation, and Zt is an independent Brownian motion. When buying or selling stocks the investor must face transaction costs. We shall suppose proportional transaction costs with rate A and fj, for buying and selling respectively. We shall therefore consider an investor that invests on a single stock St and in a riskless account Bt- He keeps in his portfolio yt stocks, selling or buying stocks at the rate dMt or dLt. The portfolio is self financing in the sense that the investor to buy the stocks withdraws the money from his account; similarly when selling the stocks he puts the money in the account.
486
One can summarize the situation just described in the following system of stochastic differential equations: dBt = rBtdt - (1 + \)StdLt
+ (1 - n)StdMt
(6)
dyt = dLt - dMt
(7)
dSt = St(adt + f(zt)dWt)
(8)
dzt = £(m - zt)dt + P (pdWt + y/l - pPdZt) .
(9)
We shall suppose to deal with absolutely continuous strategies, so that: Lt =
lsds Mt = / msds . Jo Jo This introduces relevant simplifications in the process (6)-(9). 3. Utility maximization When transaction costs are absent one can perfectly hedge away the risk. On the other hand when one has to consider transaction costs, perfect replication is not optimal, and some risk must be tolerated. Therefore it is important to introduce a measure of the risk aversion of the investor giving a measure of how he perceives his wealth. We introduce the utility function U{x) of the investor. In what follows we shall consider the exponential utility function: U(x) = 1 — exp (—7x) , where 7 is the investor's risk aversion. To price the option the main idea, first introduced in Ref.6, is to consider two investors: the investor 1 enters into the market trading only stocks and bonds; investor w instead, enters into the market writing a European call option (i.e. subscribing an obligation), and trading with stocks and bonds to hedge this option. A market strategy IT of the investors is expressed by the path y™. We call the set of admissible strategies (e.g. those strategies that do not allow the investor to borrow money above a certain limit) T. The final value of a portfolio of the writer of a European option with strike price K, after following the strategy TT is:
$w(T,BZ,yZ,ST,zT) = BZ + I{ST
,
(10)
where IA is the indicator function of A. On the other hand the final value of the portfolio 1 which does not include the option is simply:
= B$ + y$ST.
(11)
487
We can now define the following value functions: Vj{B) = supE(U($j(T,BT,yT,ST,zT)))
(12)
for j = l,w. Notice how this value functions depend on the initial endowment B. Following 6 we now define: Bj =mi{B:Vj{B)
> 0}
The fair price of the option C to avoid arbitrage, i.e. the amount of money that the writer has to receive to accept the obligation implicit in writing the option, will therefore be: C = BW-BX.
(13)
We are therefore dealing with a stochastic control problem, where the control function is the strategy yt. The Hamilton-J acobi-Bellman equation relative to the controlled stochastic process (6)-(9) with performance functional Vj is: max 0
( (dyVj - (1 + X)SdBVj) I - (dyVj - (1 - n)SdBVj)
m+
l
dtVj + rBdsVj
+ aSdsVj + £(m - z)dzV3 +
i[/(z)]2529l^- +
l
-pdlVj+{3fSpdSzV3}
= 0
One can prove that the state space (y, S, z) is divided in three regions, the buy, sell and no-transaction region: see figure 1, where we have not drawn the z variable. The buy and sell region do not intersect because it is not optimal to buy and sell at the same time. Moreover is one allows large transactions, and continuous trading, i.e. the possibility of taking immediately action as soon as one exits the no-transaction region, one can easily see that: (1) in the buy region: (dyVj - (1 + X)SdBVj) = 0 (2) in the sell region: (dyVj - (1 - n)SdBVj) = 0 (3) in the no-transaction region: dtVj + rBdeVj + aSdsVj
+ £(m - z)dzVj +
i [ / ( z ) ] 2 S 2 c ^ . + \p2dlV3
+ PfSpdstV,
=0
(14)
Now in the no-transaction region we change variable passing Vj —> Wj
Vj = l-eXp(--t{B
+ Wjj) ,
488
no-transactipm
Figure 1. Buy, sell and no-transaction regions. When transaction costs are small to resolve the no-transaction region one introduces the rescaled coordinate Y. The curve y* expresses the classical Black and Scholes hedging strategy.
where S is the discount factor: 5 = exp [-r{T - t)] . With the above expression for Vj the price of the option C is now given by: C = Wl-Ww.
(15)
4. The asymptotic procedure We now suppose small transaction costs and fast mean reverting volatility. Moreover we will assume that the transaction costs are much smaller than the rate of mean reversion. X = /j, = e
H
P
V2v V~e
Buying and selling costs are assumed to be the same for simplicity. We believe that our asymptotic assumptions are consistent with a situation where a large investor, facing very small transaction costs, is involved. In fact, in the empirical study 9 it is found that e ~ .005. For large investor, typically A < .01%. In the limit £ —> 0 one would have deterministic volatility and no transaction costs. The investor would therefore be able to hedge away risk adopting the BS strategy, i.e. staying on the curve y — y*. For small e one has a small no-transaction region. To resolve this strip one introduces the rescaled coordinate Y defined as y = y* + eaY. The scaling parameter a will be found to be a — 1/3 imposing the matching conditions between the no-transaction region and the sell and buy regions. The parameter a
489 gives the width of the no-transaction region. We denote the Y-coordinates of the upper and lower boundaries of the no-transaction region as Y+ and Y~ respectively. Imposing the above scaling, equation (14), written in terms of Wj, is: dtWj - rWj + aSdsWj
1 2 dl . W
\lf(z)}2S2[d2sWj-l(dsWj)2
+
+
1
+ -(m-
z)dzWj
_
±{dzWif
£
0(16)
-=vy/2fSp
We look for a solution of the above equation in the form: 13
WNT
= S(y* + e^Y)
+ U0(S,t, z) + £ ^ ^ ( S , t , z ) + U14(S,t,z,Y)
+ ...
i=2
(17) The reasons for the above expansion are the following: the zeroth and first order terms contain Sy* and SY to allow the matching between the "inner" solution (17) and the solutions in the buy and sell regions. Moreover in the equation (16) is present the scaling £ 1//2 , while the width of the notransaction region is e 1 ' 3 . Therefore the natural expansion parameter for W is e1'6. 4.1. The boundaries
of the no-transaction
region
We now define the linear operators £, and the non linear operator He'£0U = (m - z)Uz + u2Uzz dU
= -vV2p(a~^Uz
C2U = Ut + -fiS'Uss-rU McU -
+
(18) vV2fSPU.Sz
(19)
1 7 (a — r) 2 + rSUs + -j/2
(20)
-v2lb{Uzf
(21)
The 0(e~1) equation can therefore be written as: £0Uo + McU0 = 0 .
(22)
The above equation can be considered an ODE in z for UQ. v2U0zz + (m - z)U0z - v2\
{U0z)
0
(23)
490
In Ref.5 it is proved that a solution of the equation CQU = 0 that does not grow with an unreasonable exponentially quadratic rate when z —> oo, is a U which does not depend on z. The same situation occurs here. In fact the non linear term would give positive contribution to the exponential growth. One must therefore have: dzU0 = 0 . With a similar reasoning one gets that dzUi = 0 for i = 1 , . . . , 5. Proceeding further in the expansion one gets the expression for the leading order hedging strategy y*: (a — r)S V- = -Uos + * ^ ,
(24)
and the following equation for U6 and UQ\ C0U6 + dtU0 + ^f2S2d2sU0
+ rSdsUo - r « 7 0 + | j ~ P ~
= 0
(25)
The above equation, considered as an ODE for U&, is of the form: CQU
= 4>.
(26)
In Ref.5 it is shown that the solvability condition for the equation (26) is: <>}= 0 ,
(27)
where the average (•) is taken with respect to the Ornstein-Uhlenbeck process invariant measure:
(
(28)
Therefore the solvability condition for (25) is: dtUo + hs2d2sU0
- rU0 + rSdsUo - - *
( a
~
r ) 2
l ,
(29)
where a is the effective constant volatility: a
2
I t1
2>>
and r is defined as:
-
=
(-
Equation (29), supplemented with the final conditions appropriate for the investor with and without option liability, gives the leading order solution.
491
Using the expression (24) for y*, the 0(e1^2)
equation can be written
as: £2U3 + C^e
+ £0U9 = 0
(30)
Notice that in the above equation appears UQ which, until now we have derived only up to the function U§(S,t), to be determined to a higher order asymptotics. However, in (30) U% is hit by the operator C\, which cancels U0{S,t). Therefore one can consider (30) as a Poisson problem for Ug, whose solvability condition is: (C2) U3 = - (dUe)
(31)
The above equation is a Black and Scholes equation for U3 with 0 final condition and with a source term. Imposing the appropriate final condition one gets the 0{e1/2) correction to the Black and Scholes value. In Fig. 2 we report an example of the correction where we have used the Scott model (f(z) = ez) for the stochastic volatility. One can see that the order of the magnitude of the correction is much higher for deep out-of-the-money options. We can now pass to the 0(e2/3) equation. One gets: £2U4+£0U10
+ v2{y*z)2UUYY
- \lf2S2Y2
=0.
(32)
I 0
From this equation, and imposing the appropriate matching conditions, one can recover an expression for the boundaries of the no-transaction region. In Fig. 3 it is reported the shape of the no-transaction region for different values of the volatility. Going further in the asymptotic analysis one can get the following pricing formula, corrected up to order O(e): C = C0 + V^C3 + eCe .
(33)
The explicit form for the terms in (33) will appear elsewhere11. 5. Conclusions In this paper we have considered the problem of the pricing a European option when transaction costs are taken into account and when the volatility is not constant. The problem of hedging an option with transaction costs and non constant volatility has been recently considered in Refs. 12,13
492
-0.2,
Figure 2. X = 100, T = 3 months, long term volatility = 30%, v = 1, p (i = 10%, r = 4%, £ = 1/200.
10
12
14
6
8
10
12
14
8
10
12
14
Figure 3. T h e boundaries of t h e no transaction zone for different values of t h e volatility a = exp(z). In all of the above figures K = 10, m = log(0.2), 7 = 1, T = 1, t = . 1 , e = 1/200.
where the volatility was modelled as a GARCH process, and through numerical simulation of the pricing process on a discrete lattice of possible scenarios. In this paper we have adopted a different perspective. We have assumed that the volatility follows a continuous time stochastic process (an
493 Ornstein-Uhlenbeck process) a n d have followed a stochastic optimization procedure on the value function of two investors: t h e first with option liability, the second without the option liability. T h e price of t h e option has been defined t o be t h e difference of t h e initial endowments necessary to the two investors to enter into t h e market. T h e resulting complicated Hamilton-Jacobi-Bellman equation has been analyzed in t h e asymptotic regime of fast mean-reverting volatility and low proportional transaction costs. T h e main results of this analysis are t h e pricing formula (33) and t h e explicit derivation of t h e optimal hedging strategy as shown in Fig.3 (in the case of t h e Scott model for the stochastic volatility).
Acknowledgements P a r t of this work has been done while t h e author was visiting t h e M a t h . Dept of UCLA. He gratefully acknowledges the warm and friendly hospitality he received.
References 1. J.C.Hull, Options, Futures and Other Derivatives Prentice-Hall, (2002). 2. P. Wilmott, S. Howison and J. Dewynne, The Mathematics of Financial Derivatives Cambridge University Press (1995). 3. A.Etheridge, A Course in Financial Calculus Cambridge University Press (2002). 4. J.-P. Fouque, G. Papanicolaou and R. Sircar, Derivatives in Financial Markets with Stochastic Volatility, Cambridge University Press, (2001). 5. J.-P. Fouque, G. Papanicolaou and R. Sircar, K. Solna, SI AM J. Appl. Math., 63, 1648, (2003). 6. M.A.H. Davis, V.G. Panas and T. Zariphopoulou, SIAM J. Control and Optimization 3 1 , 470 (1993). 7. A. E. Whalley and P. Wilmott, Math. Finance 7, 307 (1997). 8. A. E. Whalley and P. Wilmott, Euro. J. of Applied Mathematics 10, 117 (1999). 9. J.-P. Fouque, G. Papanicolaou, R. Sircar, K. Solna, Mean reversion of S&P 500 volatility, preprint, (1999). 10. M. Jonsson and R. Sircar, Math. Finance, 12, 375 (2002). 11. R. Caflisch and M. Sammartino, Option pricing with transaction costs and stochastic volatility, in preparation, (2003). 12. J. Gondzio, R. Kouwenberg and T. Vorst, Journal of Economic Dynamics & Control, 27, 1045 (2003). 13. Ritchken, P., Trevor, R., Journal of Finance, 54, 377 (1999).
A P P R O X I M A T E INERTIAL M A N I F O L D S FOR T H E R M O D I F F U S I O N EQUATIONS *
M. S A M M A R T I N O A N D V . S C I A C C A Dipartimento
E-mail:
di Matematica, Universita di Palermo via Archirafi 34, 90123 Palermo, ITALY [email protected], [email protected]
In this paper, we consider the two dimensional equations of thermohydraulics, i.e. the coupled system of equations of fluid and temperature in the Boussinesq approximation. We construct a family of approximate Inertial Manifolds whose order decreases exponentially fast with respect to the dimension of the manifold. We give the explicit expression of the order of the constructed manifolds.
1. Introduction The concept of inertial manifold was introduced in 1988 by Foias, Sell and Temam 9 in the context of the theory of dissipative dynamical systems. It is an important development in the study of system with a complicated attractor since it reduces an infinite dimensional problem to a finite dimensional one without introducing any error. We consider an abstract evolution equation of the form flu
— +Au = f(u)
(1.1)
•u(O) = w 0 ,
in a separable Hilbert space H, which defines a continuous semigroup {S(t)}t>o. A is a positive operator with discrete spectrum. In what follows, we shall assume that A is a linear continuous map from D(Ae) x R into H, 0 < 9 < 1, satisfying suitable boundedness and Lipschitz properties. Furthermore it is assumed that {efc} is the orthonormal basis in H consisting of the eigenfunctions of the operator A, and 0 < \\ < A2 < . . . , lim/.-.co Ak = 00 the respective eigenvalues. We suppose that A defines a strongly continuous linear semigroup {e~At}t>o- Consider the projector P of H into the subspace generated by the first n eigenfunction of "This work has been supported by the INDAM-GNFM under the grant "Insiemi assorbenti, attrattori e varieta inerziali nella fluidodinamica esterna ed applicazioni alia geofisica".
494
495
A, and define Q — I — P. We shall use the notation Pu = y, Qu — z. We recall that an Inertial Manifold for (1.1) is a finite dimensional Lipschitz manifold M, whit the following properties: (i) M is positively invariant for the semigroup (i.e. S(t)M. C M, Wt > 0); (ii) M. attracts all the orbits of (1.1) at an exponential rate. We start to explain the construction 17 ' 6 ' 4 of the Inertial Manifold M. = {y, $(y)} as the fixed point of a suitable map. Assuming that $ is known and its graph is invariant for {S(t)}t>o, then for an orbit u(t) lying on M (if i*o £ -M), we have z(t) = *(y(t)), (1.2) and the equation (1.1) can be written as d(y + $(y)) A(y + $(y)) = f(y + $(y)), dt If we use the projections P and1 Q, Q, we we have have &y_
+ Ay = Pf(y
+ $())
t > 0.
(1.3)
(1.4)
dt ^Ml+A$(y) = Qf(y + $(y)). (1.5) Equations (1.4) is a system of finite dimensional differential equations, the Inertial System, which determines univocally y(t) = y(t;y0,&) for every t g M . To compute the function $, we use (1.5), obtaining *(l/0)= f
eAsQ(f(y(s)
+ t>(y(s)))ds.
(1.6)
J — oo
Equation (1.6) suggests to construct the function $ as the fixed point of the contractive map <j> —> T4> (the Lyapunov-Perron operator) defined by (1.6) and where <j> is an element of the complete metric space T^ = {cj>:PH-> QH, Lip0 < l, I^U < b},
(1.7)
of the bounded Lipschitz function from PH into QH and y is the solution of the Inertial System. We recall that this proof is based on a spectral gap condition, i.e. the difference A n + i — A„ should grow sufficiently fast as n —> co. Many physically interesting equations do not satisfy this condition (such as the Navier Stokes equations and the thermodiffusion equations) and it is an open problem to prove the existence of Inertial Manifolds for these systems. For this reason, it was introduced the concept of Approximate Inertial Manifolds 4 . The idea is to approximate T with an explicit Euler method. Choosing a positive integer N and a time step r > 0, then we define recursively a family of yk, with k > 1, by yk+1 yk
~ +Ayk
= Pf(yk+({>(yk)),
(1.8)
496
where yk is the approximation of y(—kr). Now we construct the function yT(s) = yk for — {k + l ) r < s < —kr and yT(s) = yN for s < —NT.; then the approximation J-^ of T is N-l
??
-(>l)-1e-^TQ((»N+^N))). (1.9) To construct the family of Approximate Inertial Manifolds, we consider a sequence (TN)N(-N , N £ N, and we define the manifolds M.N as the graph of the function $JV, constructed iteratively by $N+i=FTNN(<S>N), (1.10) for N > 0 in the complete metric space Tiyh = {
(2.1)
d9 — + KA26
+ B2{V,6)
= R2V.
(2.2)
in the domain Q, =]0, l[x]0,1[. We consider the space H = Hi x H2, where Hi = {v £ L2(Q)2,V • v = 0,u 2 U 2 =o = v2\x2=i,vi\Xl=0 — ^ l U ^ i } , and H2 = L 2 (fi). We use the same notation (•, •), | • | for the scalar product and the norm in Hi,H2,H. We also consider V = V\ x V2, where V2 = H1^)
497 vanishing at X2 = 0 and %2 = 1 and Vi = {v G V22, d\w = 0}, and we also denote by ((•,•)) and || • || the canonical scalar product and norm in V1; V2 and V. Moreover, we consider also the space D(A) = D(Ai)xD(Az), where D(Ai) = {v e Vi n # 2 ( f i ) 2 , J ^ k = o - ^ 7 l x 1 = i } , and let a be the linear operator from D(A) into iJ and from V into V defined by (Aui,U2) = " ( ( « I . V 2 ) ) + K((0I,02)),
VU
* = {vi.fc} G D(.4),
i = 1,2.
We denote
by A„ the eigenvalues of A\ and by A„ the eigenvalues of A2- As usual 6i(y,z,u>) = E,lj=1 JnVi^Wjdx, and b2(y,4>,ip) = £ 2 = i Jnyi§^-ipdx. They define a trilinear continuous operator on V or even on iJ 1 (fi) 2 x H1(Q,). Finally, the continuous operator {—e29,v2} = {RiS,R2v}, with e 2 the vertical direction. The existence of absorbing set 17 in H and V and the existence of the global compact attractor A was proved in 7 . It is possible to consider other boundary condition to the Benard equations in the x\ direction; they all lead to the same results, the only differences being the functional setting. 3. C o n s t r u c t i o n of A p p r o x i m a t e Inertial Manifolds We start the construction 14 of the approximate inertial manifolds for the thermodiffusion equations (2.1,2.2). In the following, we consider , for M and N e N fixed, the projector Pi = P\M on the space spanned by the first eigenfunctions of A\, and Q\ = QiM = I — P\M. Analogously, we denote with P2 = PIN the projector on the space spanned by the first eigenfunctions of A2, and Q2 = Q2N = I — P2N- It is clear that P = Pi x P2 and Q = Q\ x Q2 are orthogonal projectors in H, V and D(AS); moreover these projectors commute with A. An Inertial Manifold for Benard problem is: M = {{y + $ i ( y ) , 7 + $2(7)) : * i * $2 : A # i x P2H2 -» Q ^ i x Q2H2, Lip($i x $ 2 ) < 1 , |$i|oo < b i and |$ 2 |oo < b 2 } and the associated discretized inertial system is: Vk+l
JT
Vk
lk+1
_~
lk
+ "AlVk
= - P i B i ( » f e 4- MVk),Vk
+ MVk))
(3-1)
+i?i(7fc+4>2(7fc)) + KA2ik = -P 2 B 2 (y f c + 0i(yfc),7fc + Mlk))
(3-2)
+R2(yk + Mvk)) where yk and 7^ approximates y(—kr) and -y(—kr), and r is the fixed time-discretization mesh. We now write the discretized Lyapunov-Perron operator: ^ i ( W o ) = Mi)_1(/-e-,'yllT)x
498 n-l X
e-kvA^Ql
£
( - Br{yk + MVk), Vk + MVk))
+ #i(7fc + Mlk))
fc=0 fc=0
+(^ J 4 1 )- 1 e-"^^g 1 ( - B!(yn + 4>i(yn),yn + MvJ)
+ Rift* + Mln)
^2^2(70) = ( « A 2 ) - 1 ( / - e - ^ T ) x nn — - l1 X
^
e - f c ^ T Q 2 ( - B 2 (y fe + MVk)nk
+ Mlk)) + #2(3/* + MvS)
fc=0
+(KA,)-1e-nKA^Q2(
- B2(yn
+ 4>i{yn),ln +
defined in the Banach Space FiM x TiM = {fa x <j>2 : PXHX x P2H2 -> Q ^ x Q 2 tf 2 , Lip(0i x ^2) < 2, |
V 2
+ A^+1) e-"A«^*
< K ((^)"1/2 +A]&)
e
(3.3)
- « W
(3.4)
l^'e-^^QiU^,^) < ^ A ^ e - " ^ ^ ' , 1
l^2" e-'
eAat
(3.5)
Q2|c(H2lWil) < KK^lle-^"^
(3.6)
IJ + i r ^ i l ^ v j ) < (1 + VT\M),
(3.7)
l-f|£(P1fl'1,p1v1) < AM , | / + KTA2\C{V2)
(3.8) 2
< (1 + K T A N ) ,
|/|£(P 2 H 2 ,P 2 y 2) < A # .
(3.9)
Proposition 3.1. Assume that (3.3)-(3.9) hold. There exists a constant C such that if one takes r small enough to have (n + l)r< .
l / 2
°
l / 2
M +AN
A
(3-10)
then there exists I and 61, b2 such that the operator J-\n x T2Tn maps Fifa x J-i,b2 into itself. The proof14 is constructive and gives the explicit expression of 61, b2 and I. Condition (3.10) expresses the time up to which one can solve backward (1.4,1.5). this time is short due to the backward instability of the dissipative equations (1.4,1.5). At this point, we analyze the semidistance dtfaXH 2 (A,M„) := s u p ( V ; ( 9 ) 6 ^ i n i { W t 0 e M n (\\v-w\\Hl + ||0-£||ff 2 ), which is the width of a neighborhood of Mn that contains the attractor.
499 T h e o r e m 3 . 1 . Choosing a sequence ofrn and supposing that \ is any fixed positive constant such that\ < (n+l)Tn < 1/2C 1 / 2 , then the Approximate A
Inertial
Manifolds
(1-10),
satisfy
A4n
= g r a p h ! $ i i T l x $ 2 , n ) ; n G N, constructed
dHlxHAA,Mn) < ^ f ^ ^
+W ^ x W - 1 / 2
"I
for n sufficiently
+AN
M
X
-
+
using
(3ill)
1 / 2 + A N + I
A
'
large.
T h e previous theorem provides a finite-dimensional smooth manifold Mn, with dimension N x M , which, for suitable n, approximates the attractor A at an exponential order. T h e interested reader can be found the complete proof in 1 4 . A c k n o w l e d g m e n t s : We sincerely t h a n k prof. G. Mulone for introducing us into the problem and for many useful discussions. References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17.
S.Agmon, Lectures on Elliptic boundary value problems. H.Brezis, T.Gallouet, Nonlinear Anal. TMA, 4, 677 (1980). Dan Henry, Lee. Notes in Math. , 840 Springer-Verlag. A.Debussche, T.Dobois, PhysicaDJ2, 372 (1994). A.Debussche, T.Dubois, R.Temam, Theoret. Comp. Fluid Dynamics, 7, 279 (1995). A.Debussche, R.Temam, J.Math.Pures Appl. , 73, 489 (1994). C.Foias, O.Manley, R.Temam, Nonlinear Anal, 11, 939 (1987). C.Foias, O.Manley, R.Temam, R.Rosa, Navier-stokes equations and turbolence, (2001), Cambridge University Press. C.Foias, R.Sell, R.Temam, J. Diff. Eq., 73, 309 (1988). B. Garcia-Arhilla, J Novo, E.S. Titi, Mathematics of Computation, Vol. 68 (227), 893 (1999). M.D. Graham, R H . Steen, E.S. Titi, J. Nonlinear Sci., 3, 153 (1993). M.Jolly, L.G.Margolin, E.S.Titi, Theoret. Comp. Fluid Dyn., 7, 243 (1995). G. Metivier, J. Math. Pures Appl., 57, 133 (1978). M. Sammartino, V. Sciacca, Preprint 197 Dip. Math, of Palermo (2003) R. Rosa, Discrete Contin. Dynam. System, 1, 421 (1995). G.R. Sell, Y. You, Dynamics of evolutionary equations, A M S 143, Springer Verlag. R. Temam, Infinite dimensional dynamical systems in mechanics and phisics, second editions, A M S 68, Springer Verlag.
L I N E A R STABILITY OF SOME E X A C T SOLUTIONS TO IDEAL M A G N E T O - G A S - D Y N A M I C S EQUATIONS
MARIA PAOLA SPECIALE, F R A N C E S C O OLIVERI* Department of Mathematics, University of Messina Contrada Papardo, Salita Sperone 31, 98166 Sant'Agata, Messina, E-mail: [email protected]; oliveriQmat520.unime.it
Italy
Some of the exact solutions obtained in Ref. l for the equations of Ideal MagnetoGas-Dynamics are here considered in connection to initial and boundary value problems of physical interest, and their linear stability is investigated.
1. Introduction We consider the equations governing the flow of an inviscid, thermally nonconducting fluid of infinite electrical conductivity in the presence of a magnetic field and subject to no extraneous force
g+V.(pv) = 0, P (-^
+ (v • V)v ) + V p + / i H x ( V x H ) = 0 ,
V - H = 0,
(1)
— - V x (v x H) = 0, ds „ - + v . V S = 0, where p is the mass density, p the pressure, s (a function of) the entropy, p. the constant magnetic permeability, v and H the velocity and magnetic fields respectively. Moreover, the system (1) is closed with the equation of a perfect gas P = p1/V-s, *Work supported by funds of Progetto Intergruppo 2003 dell'I.N.d.A.M. numerica per il calcolo scientifico e applicazioni avanzate".
500
(2) "Modellistica
501 where V is the adiabatic index. In Ref. 1 various classes of exact solutions (containing also arbitrary functions) to the system (1) have been determined through the use of Lie point symmetries (see Refs. 2>3-4'5'6) and some finite transformations known in literature as Substitution Principles. In what follows we examine some of these solutions (that generalize well known ones) by considering some initial and boundary value problems of physical interest; further, we face the problem of investigating their linear stability. In Ref. 7 it has been proved that the steady ideal magneto-gas-dynamics equations with a separable equation of state are invariant with respect to the family of transformations MxjrV
p*=p,
H*=H,
s* = [m(x)] 2 s.
m(x) being an arbitrary scalar function of x satisfying the conditions
E
dm
^-^
TT
dm
i.e., the function m(x) must be constant along each individual gas streamline and along each individual magnetic line. In Ref. 8 the Smith's result has been extended to unsteady equations provided that the total magnetic pressure steady whereas in Refs. *, in the case of a planar motion with a transverse magnetic field (and taking the adiabatic index T = 2), it has been proved that from a steady solution a new unsteady solution can be generated. 2. 2 D steady solution In Ref. 1 it has been found the following 2D steady solution (describing an incompressible flow) vi = -v0{x2
- k2)ra~l,
H1 = -{x2-k2)r-1X(r), ,w ^
fci)ra-1,
v2 = t>o(zi -
H2 = (Xl - ^ r ^ r ) ,
{V' + VXJr^X + x'))
(3)
_2a+1
where r = \J{x\ — fcj) + (x2 — k2) , whereas a, k\, k2 and VQ are arbitrary constants, and x{r) a n d ty(r) arbitrary functions of r. By using the Smith's result 7 , after solving the constraint , . dm , , . dm „ , .
- ^ - ^ ^ -
+( , , - ^ ^ = 0 ,
(4)
502
whereupon it follows m = M(r), M(r) being an arbitrary function of r, we get the new solution vi = -(x2-k2)$(r),
v2 =
1
H1 = -(x2-k2)r' X(r), lT,/
(xi-ki)$(r),
H2 = (an - k^r^x^r),
(5)
l
x
^' + PX{r~ X + x')
where $(r) = ra~1M(r). Solution (5), written in cylindrical coordinates, reads: vr=vz=0, P=*W.
vg = $(r), s
He=r-1X(r),
Hr = Hz=0,
^' + px{r'lx = ^^TJf
* ' + M x ( r " 1 X + x') P= ^2 "•
+ x')
(6)
If we consider the flow between two rigid coaxial cylinders under the influence of a circular magnetic field, we may specify the arbitrary functions assigning the field variables at the boundary r = R\ and r = R2, R\ and R2 being the radii of inner and outer cylinders, respectively. The previous solution contains a well-known solution described in Ref. 9 : vr = vz=0,
v0=r$(r),
p = p0 = const.,
Hr=Hz
= 0,
He =
—((p + ^ H • H)) = r $ 2 -
x(r),
(7) AW-1*2,
that is recovered from solution (6) when
tt(r) = p0f ($2r - A«--y )dr - | X 2 + k0,
(8)
ko being a constant. Chandrasekhar's solution (7) is stable 9 if the Rayleigh's discriminant ^(r) = ^ ( r
2
* )
2
- ^ )
a
> 0 ,
(9)
that implies some restrictions on the functions $(r) and x(r) (involved in the espressions of velocity and magnetic field). Now, let us examine the linear stability of solution (6), in which, though the flow is incompressible, p is not constant; by substituting in the governing equations the perturbations = e^Xz+k^ur(r), ve = e^Xz+k^u9{r) + r$(r), vz = i( Xz+k li Xz+kt Hr = e - ^Hr{r), He = e ">H0{r) + x ( r ) , Hz = P = e'V'+Wftr) + ^ (*' + px (r-'x + x')) , Vr
e^Xz+kt)uz(r), e^Xz+kt">Hz{r),
503
where A and k are constant, and solving the linearized system, after some algebra what we recover is fc2
I T (P(U'r + r~lUr^) ~ pX2Ur ) =
-^E(r)ur,
(10)
that has to be solved along with the boundary conditions ur = 0,
at
r — R\
and
r = R2.
By multiplying (10) by rur and integrating by parts over the domain, we arrive at the condition \*f%E(r)v?rrdr
HI k2= /^^ r p (;;: ~:Z~' > K + r - 1 u r ) 2 + A2«2)d:
w
whereupon the Rayleigh's discriminant (that must be positive in order k be real) is given by
E(r) = ~ ( p ( r a * ) a ) - ^
f
= 1 A ( ^ + ,xx>) + 4M(^)2 > 0;
therefore, the linear stability requires some functional restrictions on the functions \I>(r) (involved in the espression of p) and x(r), but does not impose restrictions on $(r) (involved in the expression of the velocity) as in the case considered by Chandrasekhar. 3 . 3D s t e a d y solution The 2D steady solution found in the previous section has a natural extension in the 3D case fc3))i?a_1,
vi = {~VQ{X2 - k2) + w0(x3 v2 = (v0{xi - ki) + z0(x3 -
fc3))i?a_1, k2))Ra~l,
v3 = -(w0(x1
- ki) + z0(x2 -
Hi = (-v0{x2
- k2) + w0(x3 - fc3))iT ^ ( f l ) ,
H2 = (v0(x! - fcj) + 2:0(2:3 H3 = ~(w0(xi P
= *(R),
- ki) + z0(x2 -
(12)
-1
fc3))fl xCR), k2))R~1x(R),
s= ^7f (*' + nxiR^x + x') R~2a+l,
where a, fci, k2, fc3, VQ, WO and ZQ are arbitrary constants while x(R) $(R) are arbitrary functions of
an
d
R = {(v% + u;g)(n - fci)2 + {vl + zl){x2 - k2f + {vol + 4)(*3 - k3)2 +2w0z0(xi
- h)(x2
- k2) + 2v0(x3 - k3)(z0(xi
- fci) - w0(x2 -
k2)))1/2.
504
By solving the constraint dm
dm
VITT-+
dm
v2—+
OXx
v3—
0X2
OX3
= 0,
13
whereupon m = M(R, Q), with M(R, Q) arbitrary function of R and Q = z0{xi - ki) - w0(x2 - k2) - v0(x3 - k3),
(14)
we get, through the use of Smith's Substitution Principle, a new steady solution that we can write in the compact form v = <jj x (x - k)$(i?, w • x), P
*W>
s
H = OJ x (x -
k)R-lx(R),
#$2#i/r a 1
where $(i?, w • x) = R ~ M(R, k are
OJ • x), whereas the constant vectors u> and
u; = [z0,-w0,-vo]T
k=[k1:k2,k3]T.
(15)
In particular, for WQ = ZQ = 0, we have, in cylindrical coordinates vr=vz=0,
* = *(r)>
vg = $(r,z),
Hr = Hz=0,
s
=r^l,i/r^'+^(r~lX
He=r~1x(r),
+ x'))-
(16)
Nevertheless, this solution does not possess cylindrical symmetry. 4. Unsteady planar motion with H transverse By considering a planar motion with a transverse magnetic field and taking r = 2, the following unsteady solution has been recovered1 _ (2at + b)xi - 2rr2$(r?) Vl
2(at2 + bt + c)
~
y
*(r?) (at +bt + c)2' 2
_ 2XI$(T?) + (2at + b)x2 V2
'
h =
~~
X(V) at +bt + c 2
2{at2 + bt + c) s =
'
8(*' + nxx') 1 4$2 + 62-4aC%/^'
where a, b, c, k\ and k2 are constant, whereas $(77), x(v) arbitrary functions of
an
(
d *(7?) a r e
^ a t 2 + bt + c This solution, written in cylindrical coordinates, reads -
(2at + b)r - 2(at2 + bt + cY #(»/) (at + bt + c)2' 2
'
(18)
<&{r})r v"e - 2(at2 + bt + c)' h =
x(v) at + bt + c' 2
s =
8(*'_+juxxO_J_ 4$2 + b2 - 4ac ^ '
l
}
505
The arbitrary functions occurring in the solution (17) can be determined by assigning appropriate initial and boundary data. At a first look, this solution can explode in a finite time due to the presence of the i-polynomial in the denominators. However, we want to notice that blow up of the solution may be avoided for instance by choosing a, b and c such that b2 - 4ac < 0
(20)
so that at2 + bt + c =£ 0 for all t > 0. If we choose a = b = 0 and c = 1 then solution (17) becomes steady, say vr = 0 , *,< \ p = *(r),
v $ = $(r)r, j, { \ h = X(r),
1
s=Wr
W + PXX') -^ .
(21)
Let us now consider an incompressible flow between two rigid coaxial cylinders under the influence of an external applied axial magnetic field (see Ref. 9 ) . If we choose: *(r) = Po [r^2dr-'~h2
+ k0,
k0 being an arbitrary constant, solution (21) reduces to the well-known solution reported in Ref. 9 vr — Vz=Q,
ve = r$(r),
P = Po = const.,
Hr = He = 0,
HZ = h0,
p = / $2rdr — — h%,
(22)
that is linearly stable 9 when S%{i& - E{r))v*rdr phi > —s , /gV(«+r-iur)2 + AV)* where the Rayleigh's discriminant is
(23)
LA
E{r) = ~(r2*)*(24) r 3 dr Now, let us study the linear stability of the solution (21) when x — ^0i whereupon it results P = 7&>
(25)
with the same boundary conditions considered by Chandrasekhar (perturbation of radial velocity vanishing at the boundaries).
506 By perturbing t h e field variables, substituting t h e m in the governing equations, and solving the linearized system, we arrive a t t h e condition
I
r{pk2 - ph2X2)2{(u'r
+ r-lurf
+ X2u2r)dr =
Ri
i-R2
R2 2
X2 ['
2
(pk2 ~ fxh20X2)E(r)u2rrdr
iii,
+ 4/jhlX4 f ''r$2u2dr, J Ri
(26)
where, since p is not constant, t h e Rayleigh's discriminant writes E(r) = - ^ ( p ( r 2 $ ) 2 ) = * " + - * '
(27)
Since t h e imaginary part of (26) is zero, we can solve it with respect t o k2 and finally obtain t h a t k2 is real if
>
jZ^-Ejr)^ IRI ( « + r-1^)2
+ X2u2)rdr
Therefore, also in this case t h e requirement of linear stability of our solution implies a functional restriction on 'l'(r) (the function entering in the espression of p).
References 1. F. Oliveri, M. P. Speciale, Exact solutions to the ideal magneto-gas-dynamic equations through lie group analysis and substitution principles, preprint (2003). 2. L. V. Ovsiannikov, Group analysis of differential equations, Academic Press, New York (1982). 3. N. H. Ibragimov, Transformation groups applied to mathematical physics, D. Reidel Publishing Company, Dordrecht (1985). 4. P. J. Olver, Applications of Lie groups to differential equations, Springer, New York (1986). 5. G. W. Bluman, S. Kumei, Symmetries and differential equations, Springer, New York (1989). 6. G. Baumann, Symmetry analysis of differential equations with Mathematica, Springer-Verlag, New York (2000). 7. P. Smith, J. Math. Mech., 1, 505 (1963). 8. G. Power, C. Rogers, Appl. Sci. Res., 2 1 , 176 (1969). 9. S. Chandrasekhar, Hydrodynamic and Hydromagnetic stability, Oxford University Press, London (1961).
N O N - B O U S S I N E S Q C O N V E C T I O N IN P O R O U S M E D I A
B. STRAUGHAN Department of Mathematical Sciences, University of Durham, DH1 3LE, U.K. E-mail: [email protected] A model is given of a general density law which permits oscillatory convection when an internal heat source is also present in a porous medium. Linear and nonlinear analyses are discussed.
1. Introduction In this article we consider equations for convection in a porous layer when the density in the buoyancy force may be a nonlinear function of temperature, T. We also allow an internal heat source/sink Q which may be a function of the vertical coordinate z, i.e. Q — Q(z). Both of these effects may be adopted to be applicable as models for penetrative convection. The topic of penetrative convection is explained in detail in chapter 17, in the book by Straughan 38 where various models for the phenomenon are exposited. It is noteworthy that the subject of penetrative convection has been attracting the attention of many writers recently, cf. Carr 2 ' 3 ' 4 ' 5 ' 6 , Chasnov & Tse 7 , Hill 16 ' 15 , Krishnamurti 19 , Larson 20 ' 21 , Mahidjiba 22 , Normand & Azouni 24 , Payne & Straughan 25 , Straughan 36,37 , Straughan & Walker 39 , Tse & Chasnov 40 , Zhang & Schubert 42 ' 43 . We here discuss general models for thermal convection in a porous medium, and we analyse the techniques of linearised instability theory together with nonlinear energy stability theory in this context. The energy method for deriving nonlinear stability is explained in depth in the book by Straughan 38 . It is also pertinent to mention at this juncture that Rionero 27 ' 28,29 ' 30 has made fundamental contributions to the energy stability technique in a hydrodynamic setting and has especially contributed by establishing existence of a maximising solution in the energy stability variational maximisation problem. His later work on weighted energy methods for the Navier-Stokes equations on an unbounded domain, Rionero 31 , Rionero & Galdi 32 ' 33 , has led to much use of weighted energy methods
507
508
in several other areas of mathematics, cf. the accounts in the books by Flavin & Rionero 8 and Ames & Straughan 1 . The recent contributions of Rionero in deriving natural energy functionals, especially for obtaining unconditional nonlinear stability in a variety of partial differential equation contexts, cf. Flavin and Rionero 9 ' 10 ' 11 ' 12 ' 13 , Mulone & Rionero 23 , and Rionero & Maiellaro 34 , are noteworthy. 2. General equations If we adopt the Brinkman-Forchheimer model for thermal convection in a porous medium, cf. Qin & Kaloni 26 , then assuming a general function of temperature p{T) in the buoyancy force, we may write the relevant equations as Avitt + —Vi + \F\v\vi
= -p,i - ghp(T) +
XBAvi,
vt,i = 0, Tt+viTii
(!) = K&T + Q(z).
In these equations Vi, p are the velocity and pressure fields, A,fi,K,AF,g,^B,K are the inertia coefficient, dynamic viscosity, permeability, Forchheimer coefficient, gravity, Brinkman coefficient, and thermal diffusivity, and k = (0,0,1). Standard indicial notation is employed and A is the Laplacian operator. Equations (1) hold in the domain {z 6 (0, d)} x M2 x {t > 0}. On the boundaries z = 0,z = d we assume the temperatures are prescribed as constants T = Ti,T = Tu, and the velocity field is zero there. A basic steady state (Vi, T,p) may be found by looking for a solution of form Vi = 0, T = T{z), (with p subsequently determined by (l)i), to find
f=-~fZ
f Q(0dtds + TL + (Tu^TL)z+^[
f Qd&s. (2)
Then, the gradient of temperature in the steady state assumes the form f'(z)
= - - f Q(s)ds + (TU~TL) K Jo ^ d
+ - I ['Qd£ds. ) K JQ J0
(3)
3. Perturbation equations To study the linear instability / nonlinear stability of the solution (vi,T,p) we introduce perturbations (ui,8:ir), and non-dimensionalize. If we let Ra = R2 be a Rayleigh number, A, Ai, A2 be non-dimensional forms for
509 A, XF,XB, and let T(z) be a non-dimensional form for T(z), then the nondimensional perturbation equations have form Auiit + ui + X1\u\ui = -Trti-R[p(T
+ 9) -p(f)]ki
+
\2Aui,
Ui,i = 0,
0tt+Ui0ti
(4)
=
A9-RT'(z)w,
where w = u3, and (4) hold on R2 x (0,1) x {t > 0}. The boundary conditions on the perturbation equations become m = 0,6 = 0,
2 = 0,1,
(5)
and (ui, 6, n) satisfy a plane tiling periodicity, the period cell being denoted byV. To investigate linear instability we linearize (4) and introduce a time dependence like Ui = e
+ A2AUJ,
Ui,i = 0,
(6)
a6 = A6-
RT'(z)w.
Rionero & Straughan 35 study a system of form (6) when A = 0, A2 = 0, and so study Ui = — 7T,, +
H(z)Rki9,
Ui,i = 0,
oO = A6 +
(7)
RN(z)w.
More generally, from (6) we may study a system of form (Aa + l)ut = —Kti + H(z)Rki9 + X2Aui: Ui,i = 0,
ad = A9 +
(8)
RN{z)w.
A nonlinear energy analysis of (4) might begin by multiplying (4)i by Ui and integrating over V, then multiplying (4)3 by 6 and integrating over V, to find the energy identities ^ I H
2
= - | | u | | 2 - X1\\u\\l - A 2 ||Vu|| 2 - R(p(f
£1-\\9tf dt
= -\\V9f
- R{T{z)w,9).
+ 9)- p(f),w),
(9)
(10)
510 An argument like t h a t of P a y n e & S t r a u g h a n 2 5 shows t h a t in general, L2 theory, i.e. (9)+C(10) for a coupling parameter Q > 0, does not lead to unconditional nonlinear stability. A suitable functional for unconditional stability may need other t e r m s in a Lyapunov function, such as other Lp norms. 4. F u r t h e r r e s u l t s Straughan 3 8 has analysed (4) in detail when A2 = 0, Q = constant and p = po(l — [T — 4] 2 ). Interesting resonance-like effects are predicted from a linearized instability analysis. A nonlinear energy stability analysis employing the appropriate froms of (9) and (10) together with an energy identity for the L 3 norm of 9 is shown in S t r a u g h a n 3 8 to lead t o interesting unconditional nonlinear energy stability thresholds which are close t o the linear instability threshold and so are very useful from a practical viewpoint. A richer layer like structure involving t h e potential for resonance between multiple layers may be found employing equations (4). We should point out t h a t N o r m a n d & Azouni 2 4 have derived very interesting results for t h e Navier-Stokes equations when p has a quadratic t e m p e r a t u r e dependence and a constant heat source/sink is additionally employed. T h e results of N o r m a n d & Azouni 2 4 are also explained in the book by S t r a u g h a n 3 8 . Acknowledgments It is a pleasure t o t h a n k t h e conference organisers Professor S. Rionero, Professor T. Ruggeri, Professor S. Pennisi and Professor F . Borghero for their kindness. References 1. K.A. Ames and B. Straughan, Non-standard and Improperly Posed Problems, (Academic Press, San Diego, 1997). 2. M. Carr, Convection in porous media flows, (Ph.D. Thesis, University of Durham, 2003). 3. M. Carr, Math. Models Meth. Appl. Sci. 132072003. 4. M. Carr, Continuum Mech. Thermodyn.15452003. 5. M. Carr and S. de Putter, Continuum Mech. Thermodyn.15332003. 6. M. Carr and B. Straughan, Advances in Water Resources262632003. 7. J.R. Chasnov and K.L. Tse, Fluid Dyn. Res.283972001. 8. J.N. Flavin and S. Rionero, Qualitative Estimates for Partial Differential Equations, (CRC Press, Boca Raton, 1995).
511 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42. 43.
J.N. Flavin and S. Rionero, Rend. Matem. Acad. Lmce«992991997. J.N. Flavin and S. Rionero, J. Math. Anal.Appl.2281191998. J.N. Flavin and S. Rionero, Q. J. Mech. Appl. Math.524411999. J.N. Flavin and S. Rionero, Continuum Mech. Thermodyn.111731999. J.N. Flavin and S. Rionero, J. Math. Anal.Appl.2812212003. G.P.Galdi and S. Rionero, Weighted energy methods in fluid dynamics and elasticity, (Springer, Heidelberg, 1985). A.A. Hill, Penetrative convection in porous media flows, (Ph.D. Thesis, University of Durham, to appear). A.A. Hill, Continuum Mech. Thermodyn.152752003. I. Herron, SIAM J. Appl. Math.6113622000. I. Herron, Int. J. Engng. 5cz.392012001. R. Krishnamurti, Dynamics of Atmospheres and Oceons273671997. V.E. Larson, Q. J. Royal Meteorological Soc.1261452000. V.E. Larson, Dynamics of Atmospheres and Oceans34452001. A. Mahidjiba, L. Robillard and P. Vasseur, Int. J. Heat Mass Trans/er463232003. G. Mulone and S. Rionero, Arch. Rational Mech. .AnaJ.1661972003. C. Normand and A. Azouni, Phys. Fluids ,442431992. L.E. Payne and B. Straughan, Stud. Appl. Matfi.105592000. Y. Qin and P.N. Kaloni, Quart. Appl. Mai/i.56711998. S. Rionero, Ann. Matem. Pura ,4ppZ.76751967. S. Rionero, Ricerche Matem.162501967. S. Rionero, Ann. Matem. Pura Appl.783391968. S. Rionero, Ricerche Matem.17641968. S. Rionero, On the use of weighted norms in stabilty questions on exterior domains, (Conf. Sem. Matem. Univ. Bari, volume 157, 1979). S. Rionero and G.P. Galdi, Arch. Rational Mech. AnaZ.622951976. S. Rionero and G.P. Galdi, Arch. Rational Mech. AnaZ.69371979. S. Rionero and M. Maiellaro, Rend. Ace. Sc. Fis. Mat. JVapo/i623151995. S. Rionero and B. Straughan, Int. J. Engng. Scz.284971990. B. Straughan, Dynamics of Atmospheres and Oceans353512002. B. Straughan, Resonant porous penetrative convection, Manuscript (2003). B. Straughan, The Energy Method, Stability, and Nonlinear Convection, 2nd edition, (Springer, New York, 2003). B. Straughan and D.W. Walker, Proc. Roy. Soc. London A452971996. K.L.Tse and J.R. Chasnov, J. Computational Phy sicsl424891998. G. Veronis, Astrophys. J.1376411963. K.K. Zhang and G. Schubert, 5cience29O19442000. K.K. Zhang and G. Schubert, Astrophys. J.5724612002.
A N E W C O N T I N U U M MODEL OF SOLIDS I N C O R P O R A T I N G MICROSCOPIC T H E R M A L V I B R A T I O N A N D ITS A P P L I C A T I O N TO WAVE P R O P A G A T I O N P H E N O M E N A
M. S U G I Y A M A Graduate
School of Engineering, Nagoya Institute of Nagoya 466-8555, Japan E-mail: [email protected]
Technology,
A new continuum model of solids incorporating microscopic thermal vibration explicitly, which is valid over a wide temperature range including the melting temperature, is explained. Mechanical and thermal properties of solids can be studied in a unified and consistent way by the model. As an application of the model, linear wave propagation phenomena in solids at finite temperatures are analyzed and discussed. Propagation speeds and amplitude ratios of both longitudinal and transverse waves are derived from the model as the functions of the temperature. Furthermore, the local equilibrium assumption, that has usually been adopted in thermodynamics of irreversible processes, is reexamined in the course of the analysis.
1. Introduction Comprehensive study on the dynamical properties of solids over a wide temperature range including the melting temperature as a limiting case is important from both practical and theoretical points of view. For an example of the practical solid materials are nowadays often utilized in many engineering systems at high temperatures even near the melting point, engineers are inevitably faced with the problem to estimate the detailed dynamical behavior of solids in such situations. The other remarkable example is that of shock wave phenomena in solids, in which local melting may occur. Temperature dependence of the Rankine-Hugoniot relation for shock waves in solids, for example, is not fully understood at present. As far as the author knows, however, there are few models of solids that can describe appropriately the dynamical properties at a high temperature in a consistent way. In the present paper, therefore, we firstly explain and discuss a new
512
513
continuum model of solids that is suitable for the above-mentioned purpose, that is, a continuum model of solids incorporating explicitly microscopic thermal vibration of constituent atoms as a field variable 1 . On the basis of the model, mechanical and thermal properties of solids can be studied in a unified and consistent way over a wide temperature range including the melting point. Instability phenomena accompanied by the melting can also be analyzed by using the model. Secondly, as a typical application of the model, linear harmonic waves, that is, the fundamental dynamical modes in solids at finite temperatures are studied in detail 2 . The propagation speeds of longitudinal and transverse waves, and the amplitude ratios of the waves are explicitly shown as the functions of the temperature. In the analysis, the importance of the thermal vibration of constituent atoms is emphasized. One of the findings of interest is the singular temperature-dependence of the quantities at the melting point. Thirdly, the so-called local equilibrium assumption, which has usually been adopted in thermodynamics of irreversible processes, is reexamined on the basis of the model through studying the linear wave propagation phenomena. Here the assumption says that thermodynamic properties of a subsystem, which is sufficiently large microscopically but sufficiently small macroscopically, can be described by the same relationships as those of a globally equilibrium system. We will study later the validity range of the assumption. In this context, extended thermodynamic approach 3 is explored and discussed briefly within the framework of the model. In the last section, it is also pointed out, especially, that the study of nonlinear waves such as acceleration waves and shock waves is one of the next important studies.
2. A new continuum model of crystalline solids A new three-dimensional continuum model with the characteristic features mentioned above was recently proposed on the basis of a nonequilibrium statistical-mechanical model of crystal lattices by using the continuum approximation 1 . The statistical-mechanical model 4 is the dynamical version of the self-consistent Einstein model 5 ' 6 . We call hereafter the continuum theory as the macroscopic theory, whereas the statistical-mechanical theory as the microscopic theory. In the present section, we summarize briefly the set of field equations expressing the continuum model. In this paper, for simplicity, we confine our study within solids with special sym-
514
metry, that is, isotropic solids. The field variables in the set of the field equations that describe the model, or basic equations of the model in brief, are the deformation gradient F(X,t), the dimensionless velocity q(X,t), and the thermal dimensionless quantities g(X,t) and r(X,t). Here X is the position of a material point in the reference configuration, and t is the time. Hereafter, the thermal equilibrium state at absolute temperature T with no external force and no translational motion is adopted as the reference configuration, and a single Cartesian coordinate system for both reference and current configurations is adopted. Then, the above-mentioned field variables are defined in a component form as follows: {wi(X,t))
=
^qi(X,t),
{(Wi(X,t)
- ((Wi(X:t)))(Wj(X,t)
-
(Wj(X,t))))
2
= a- [A<%+ 5 i ,-(JM)] ,
((m(X,t)
- ((^(X,0))(^(X,t) = =
D_ ~M
+ r(X,t)
(1)
(wjiXj))))
Sij
D where ( } stands for a statistical average with respect to the nonequilibrium distribution function, w(X,t) is the displacement of a constituent atom from its thermal equilibrium point corresponding to a material point X at a time t, £(X,t) = dC,(X,t)/dt for a generic quantity C, M mass of an atom, D depth of the atomic pair potential, a~1 microscopic characteristic length, fee Boltzmann constant and 5^ Kronecker's delta. The quantity A is the reduced mean square displacement due to the thermal vibration in the reference equilibrium state: XSij = a2(wi(X,t)wj(X,t))equiVlbrium
•
(2)
Here we have taken the symmetry of solids into consideration. This quantity A can be determined as the function of the temperature T. For its details see the reference 4 . It should be remarked that, in the third relation in the definitions (1), we have adopted the local equilibrium assumption. Then the quantity r has been introduced as a scalar quantity. A possible generalization of the assumption will be proposed in the last section. The quantities g and r indicate the amplitude increment of the thermal vibration of constituent atoms and the temperature increment from the reference equilibrium state, respectively. Therefore, owing to the simultaneous
515
use of the quantities, we can obtain the detailed information of the thermal vibration in various dynamical situations by analyzing the new continuum model. In terms of the dimensionless quantities defined above, the basic equations expressing the continuum model are given as follows1: (Hereafter the summation convention is adopted.) pdet[Fik]
= p0 ,
l ft ~ qi = a
da dF~k
da d~F~ ik
d
1
0Xk
jk — "i
da dFijk d
ft - l .
da Qi
dXk .
(\5im + gh
(3)
da dgr,
d~F~k kBT 2D
1+
D k^TJ
6a
where p and po are mass densities in the current and reference configurations, respectively, and the quantity po is the known function of the temperature 1 . The quantity a is the dimensionless potential energy density, and it is the function of the quantities F and g. Explicit functional form of a can be derived from the microscopic theory : . The quantity Q. is the characteristic microscopic frequency defined by ft
(4)
and e is the dimensionless energy density defined by 1 e
=2
q
+
(kBT 2 \ D
(5)
The basic equations (3)i_4 mean the conservation laws of mass, momentum, angular momentum and energy, respectively, while the relation (3) 5 means the equation of state. To sum up, the equations (3) are the basic equations that describe finite deformation continuum model incorporating explicitly microscopic thermal vibration of constituent atoms. In the first step of our approach, let us now analyze linear wave propagation phenomena in solids at finite temperatures by using the model in the next section.
516
3. Linear wave propagation In this section, numerical results of linear harmonic wave propagation in aluminum(Al) obtained from the linearized basic equations are shown as a typical example and their physical implications are discussed. Results for the other several fee metals are omitted here for simplicity as these are qualitatively similar to that of Al. The linearized basic equations derived from the equations (3) are expressed as follows: £l-lfi
= qt ,
to-% = 4a-2(2/?6 + / ? 7 ) 7 J ^ - + 4a-2/?7, 'dXidXk
i o — 1 a V9pp
^
dXkdXk
, o —1/3 ®9ik
dXi
dXk (6)
2^ + Pi9Pp = 0 , 2a
0 ^ — + a
/35A {—
+ —
j
+2f32\5ijgPP + (j8i + 2/?3A)5ij - - r % = 0 where /?'s are the expansion coefficients in the dimensionless potential energy density a:
a (F, g)=/30 + fch + ft J? + /33I2 + fahh +/35/7 + /? 6 /| + (37h
(7)
with the invariants defined by
h = 9ik9ki , I\ = Bss — 3 ,
(8)
h = {Bst - 6st) {Bst ^7 = 9st {Bts -5ts)
Sat),
•
Here B is the left Cauchy-Green tensor defined by (9)
517
The temperature dependence of the expansion coefficients ft's in the expansion (7) can be estimated if the atomic pair potential VE(X) with x being a distance between two atoms is fixed. For their details, see the reference 1. We hereafter restrict ourselves within studying the harmonic plane waves in the long wavelength limit for simplicity. We can therefore compare macroscopic and microscopic results with each other on the common ground as will be discussed in the next subsections. Furthermore, for the sake of concreteness, we adopt the following Morse function as the interatomic pair potential 7 ' 8 : vE{x) = D[exp{-2a(x
- R0)} - 2exp{-a(a; - i?0)}] ,
(10)
where D , a and i?o are the material constants. 3.1. Propagation waves
speeds of the longitudinal
and
transverse
The temperature dependence of the propagation speeds of the longitudinal and transverse waves in Al is shown in Fig.l. The solid lines are those predicted by the microscopic theory, while the broken lines by the macroscopic theory. For details of the method for obtaining the results from microscopic theory, see the reference 2. Two results derived from the microscopic and macroscopic theories are qualitatively consistent to each other. The other points to be noticed from Fig.l are summarized as follows: (1) Although the present model is quite simple, the results agree fairly well with the available experiental data. The difference between them are within about 10 percent. There are, however, no exerimental data at high temperatures near the melting point as far as the author knows. It is hoped strongly that the present theoretical prediction near the melting point will be checked by some experiments in near future. (2) The propagation speeds of longitudinal and transverse waves are both decreasing functions of the temperature T. And as the temperature T tends to the melting temperature TM from lower temperature side, there are singularities such that dVL dVT
HT - * " 0 0 ' IT - > - ° ° -
(11)
The values of 14 and VT themselves at TM are, however, finite. (3) As is expected, the propagation speeds of the harmonic waves derived from the macroscopic theory are exactly same as those of acceleration waves obtained recently by using the theory of singular surfaces9.
518
D
,
0
I
0.1
,
I
0.2
,
£
,_
0.3 kBT D
Figure 1. Temperature dependence of the propagation speeds of the longitudinal waves VL a n d the transverse waves Vx in Al. T M is the melting temperature. Solid lines: Microscopic theory. Broken lines: Macroscopic theory.
3.2. Amplitude
ratios
Hereafter, we assume, without loss of generality, that the harmonic plane waves propagate in the xi-direction. Furthermore we assume that the polarization of the transverse waves is in the a^-direction. Figures 2 and 3 show the temperature dependence of the amplitude ratios of r to Q for the longitudinal and transverse waves in Al, respectively. Here T and Q are the complex amplitudes of a wave corresponding to the quantities g and q, respectively. As before, the solid lines are those predicted by the microscopic theory, while the broken lines by the macroscopic theory. In the case of longitudinal waves shown in Fig.2, only the diagonal
519
0.05
Figure 2. Temperature dependence of the amplitude ratio of T to Q for the longitudinal waves in Al. Only the nonzero components of r are shown. T\j is t h e melting temperature. Solid lines: Microscopic theory. Broken lines: Macroscopic theory.
components of the amplitude T have nonzero values. And the component parallel to the propagation direction F n is more enhanced by a wave than the components perpendicular to the direction 1^2 and ^ 3 . We also notice that there is a similar singularity at the melting temperature TM to that found above. In the case of transverse waves shown in Fig.3, on the other hand, the nonzero components of the amplitude T are only off-diagonal ones. From these results, we can obtain the information about the change of thermal vibration of constituent atoms induced by a wave. Such a change is an observable quantity measured by, for example, X-ray diffraction experiment. Then we can check the appropriateness of the model by comparing the theoretical prediction with experimental data about the thermal vibration.
520 I
0.07
rn
0.06 -
1
1
..
,
\
Qi
J
•
Al "
r2i Qi
0.05
1 1 /
0.04
/ / /
0.03 0.02 /
0.01
^ y
//
y
/
0.2
/
-
/
. •
/ *s
y 2^--—
0.1
/
^y
*-* ^~-^^ ^ ^ --"_—-~ -"
0
.
/ //
%^M
-
D // // / //
0.3 k^r_
Figure 3. Temperature dependence of the amplitude ratio of T to Q for the transverse waves in Al. Only the nonzero components of V are shown. T M is the melting temperature. Solid line: Microscopic theory. Broken line: Macroscopic theory.
Figure 4 shows the temperature dependence of the amplitude ratio of R to Q for the longitudinal waves, where R is the complex amplitude of a wave correspnding to the quantity r. The quantity Rij is the complex amplitude correspnding to the quantity r^ that will be introduced in the next section. From Fig.4, we can estimate the temperature change induced by a longitudinal wave. As can be seen, when the value of Qi is fixed, the temperature change becomes large as the temperatur T approaches the melting temperature TM- In the case of transverse waves, it is easy to show that the temperature change induced by a wave vanishes. This is true, of course, only within linear approximation adopted here. The temperature change is intimately related to the volume change induced by a wave, and it is
521 !
Al _
0.2
R <2i
0.1
tr (Rt) 3fii
^ ^ \
s^
0
i
0.1
>
0.2
:
D
•
'V.
0.3
D Figure 4. Temperature dependence of the amplitude ratios of R to Q and t r J t / 3 to Q for the longitudinal waves in Al. Tyi is the melting temperature. Solid line: Microscopic theory. Broken line: Macroscopic theory.
well-known that there occurs no volume change in a transverse wave within a linear theory. The study of nonlinear effect of the temperature change should be interesting. 4. Discussions From the analysis in the previous section, we found that the propagation speeds and the amplitude ratios of the harmonic waves derived from the macroscopic theory and the microscopic theory agree fairly well with each other at least qualitatively, while quantitatively there are some discrepancies between them. The discrepancies become more evident as the temperature T approaches the melting temperature Ty[. In order to understand the origin of the discrepancies, we now try to
522
adopt the following equations instead of the third and the fourth equatons in the linearized basic equations (6), respectively. ^fpp + fcgpp = 0 ,
+2p2X5ijgpp + (ft + 2(33\)gij
(12)
- -rij
=0.
We, however, retain the other two equations in eqs.(6). These four equations constitute the second new system of macroscopic basic equations. The above two equations (12) and (13) have been introduced by taking into account the fact that the quantity r in the right-hand side of the third equation in the definitions (1) is, in general, not a scalar. Therefore we have adopted a tensor r^ instead of the scalor r. Physically, this means that we have gone beyond the local equilibrium assumption. We can check numerically that the propagation speeds of the longitudinal and transverse waves and the amplitude ratios of the waves obtained by using the second new system of macroscopic basic equations are exactly the same as those derived from the microscopic theory in the long wavelength limit. From Fig.4, we notice that, if we may assume that the concept of temperature is still valid in this case, the quantity (tri?)/3 becomes a good index for the temperature increment. In conclusion, if we want to obtain quantitatively better results especially at high temperatures near the melting point, we must abandon the local equilibrium assumption. It must be emphasized that this is true even in the long wavelength limit. Here, as we have adopted one of the basic idea of extended thermodynamics, systematic investigation of nonequilibrium phenomena in solids by using the present model on the line of extended thermodynamics is highly expected. The other subject of interest to be studied next is that of nonlinear waves such as acceleration waves and shock waves. Shock wave phenomena in solids at finite temperatures are attracting much attention recently 10 . The study of acceleration waves can also be regarded as the study of the formation process of shock waves, and its details can be seen in the reference9. We have studied only isotropic solids in this paper. We are now planning to analyze wave propagation phenomena in anisotropic, or crystalline solids at finite temperatures by the model adopted above.
523 In the present paper, the continuum model has been proposed by taking the continuum limit in t h e statistical-mechanical theory. It is, however, important to construct a similar continuum model in a phenomenological, or thermodynamic way. Because of the reason mentioned above, extended thermodynamic approach seems t o be the most promising. T h e study must be also useful for making clear the mathematical structure of t h e present continuum model.
Acknowledgments T h e present work was m a d e in collaboration with G. Valenti, C. Curro, K. Goto and K. Takada, t o whom the author would like t o express his sincere thanks. T h e author is also grateful t o K. Kawai and H. Suzumura for their kind help in the preparation of t h e figures. This work was supported by the Grant-in-Aid from J a p a n Society of Promotion of Science (No. 15560042).
References 1. M. Sugiyama, J. Phys. Soc. Jpn. 72, 1989 (2003). 2. M. Sugiyama, K. Goto, K. Takada G. Valenti and C. Curro, J. Phys. Soc. Jpn. 72 (2003) (to be published). 3. I. Miiller and T. Ruggeri, Extended Thermodynamics (Springer, New York, 1993), Rational Extended Thermodynamics (Springer, New York, 1998). 4. M. Sugiyama and K. Goto, J. Phys. Soc. Jpn. 72, 545 (2003). 5. T. Matsubara and K. Kamiya, Prog. Theor. Phys. 58, 767 (1977). 6. T. Hama and T. Matsubara, Prog. Theor Phys. 59, 1407 (1978) 7. I. M. Torrens, Interatomic Potentials (Academic Press, New York, 1972). 8. L. A. Girifalco and V. G. Weizer, Phys. Rev. 114, 687 (1959). 9. G. Valenti, C. Curro and M. Sugiyama, Continuum Mech. Thermodyn. 15 (2003). 10. Y. Horie, L. Davison and N. N. Thadhani eds., High-Pressure Shock Compression of Solids VI (Springer, New York, 2002).
O N PARTIALLY I N V A R I A N T SOLUTIONS OF T H E NAVIER-STOKES EQUATIONS
K. T H A I L E R T * A N D S.V. M E L E S H K O
E-mail:
Suranaree University of Technology School of Mathematics Nakhon Ratchasima, 30000, Thailand. [email protected], [email protected]
There is interest in obtaining exact solutions of the Navier-Stokes equations. One of methods for constructing exact solutions is group analysis : . By this method two classes of solutions can be found. One class is a class of invariant solutions. Another class is a class of partially invariant solutions. Study of partially invariant solutions is more difficult, since analysis of compatibility for them is more complicated. The main problem in obtaining partially invariant solutions consists of studying compatibility of a reduced system, obtained after substituting a representation of a partially invariant solution into the initial system of equations. Since the property of the group to be admitted is not involved in constructing a representation of a solution, this fact gave an idea for constructing partially invariant solutions for groups, which are not admitted by the initial system of differential equations. In the manuscript examples of such solutions for the Navier-Stokes equations are presented.
1. Introduction Mathematical modelling is a basis for analyzing physical phenomena. Almost all fundamental equations of mathematical physics are nonlinear, and in general, are very difficult to solve explicitly. Group analysis is a method for constructing exact solutions of differential equations. This method uses symmetry properties for constructing exact solutions. There are two types of solutions, which can be obtained by group analysis. The first class is a class of invariant solutions. The manuscript is devoted to the second class: partially invariant solutions. Constructing of partially invariant solutions consists of some steps: choosing a subgroup of the admitted group, finding a representation of solution, substituting the representation into the *Work is supported by Ministry of University Affairs of Thailand.
524
525
studied system of equations and the study of compatibility of the obtained (reduced) system of equations. It should be noted that the notion of compatibility plays the key role in constructing partially invariant solutions. At the same time in constructing a representation of partially invariant solution the property of the group to be admitted is not involved. These facts allow assuming that one can construct partially invariant solution with respect to a Lie group, which is not necessary admitted. In the next sections examples of such partially invariant solutions are presented for the Navier-Stokes equations. Examples of such solutions are presented in the manuscript. The proposed research will deal with partially invariant solutions of the Navier-Stokes equations with the defect 6 = 1 and the rank a = 1. Subgroups for studying are taken from the part of optimal system of subalgebras considered for the gas dynamics equations 2 . 2. Partially invariant solutions The notion of a partially invariant solution was introduced by L.V.Ovsiannikov 3 . This notion generalizes the notion of an invariant solution. The generalization extends an area of applications of group analysis for constructing exact solutions of partial differential equations. The algorithm of finding partially invariant solutions consists of the following steps. Let U be a Lie algebra with the basis X\,..., Xr. The universal invariant J consists of s = m + n — r* functionally independent invariants J = ( J 1 (x,u), J2(x, u),..., Jm+n~r-
(x,u)),
where r* is the total rank of the matrix composed by the coefficients of the generators Xi, (i = l , 2 , . . . , r ) . If the rank of the Jacobi matrix Q(jl
jm+n-r,\
' ,' d(u1,...,um)
r
is equal to a, then one can choose the first a < m inH
H,
*_
dU1,..., Jq) variants J 1 ,..., J9 such that the rank of the Jacobi matrix ——'-^is d[ui,...,um)
equal to q. A partially invariant solution is characterized by the two integer numbers: the rank a = 6 + n — r , > 0 and the defect 6 > 0. These solutions are also called H(a, (S)-solutions. The rank a and the defect 6 must satisfy the inequalities p < a < n, max{r, — n,m — q, 0} < 6 < min{r, — 1 , m — 1}, where p is a maximum number of invariants only dependent on the independent variables. Note that for invariant solutions the defect 6 = 0 and q = m.
526
For constructing a representation oiH(a, <S)-solution one needs to choose I = m — S invariants and separate the universal invariant in two parts: 7 = (J 1 ,.... J1), 1 = (Jl+1,Jl+2,
..., J m + n - r * ) .
The number I satisfies the inequality I < I < q <m. The representation of the H (
(1)
Equations (1) form an invariant part of the representation a solution. The next assumption about a partially invariant solution is: equations (1) can be solved with respect to I dependent functions, for example: ui =
(i = l,...,Z).
(2)
It is very important to note that the functions Wl, (i = l,...,l) are involved in the expressions for the functions
In this case the partially invariant solution is called regular otherwise it is irregular 4 . The number a — pis called a measure of irregularity. The process of studying compatibility consists of reducing the overdetermined system of partial differential equations to involutive system. During this process different subclasses of H(a, S) partially invariant solutions can be obtained. Some of these subclasses can be Hi{ai,5\)-solutions with the
527
subalgebra Hi C H. In this case o\ > a, Si < S x. The study of compatibility of partially invariant solutions with the same rank ci = a but with the fewer defect di < S is simpler than the study of compatibility for H(a, <5)-solution. In many applications there is a reduction of H(a,S)solution to H'(a,0). In this case the H(a, <5)-solution is called reducible to an invariant solution. The problem of .reduction to invariant solution is important since invariant solutions are studied first. 3. The unsteady Navier-Stokes equations Unsteady motion of incompressible viscous fluid is governed by the NavierStokes equations u t + u Vu = - V p + Au,
V • u = 0,
(3)
where u = (u\, 1*2,1/3) = (u, v, w) is the velocity field, p is the fluid pressure, V is the gradient operator in the three-dimensional space x = (a;i, #2, £3) = (x, y, z) and A is the Laplacian. A group classification of the Navier-Stokes equations in the three-dimensional case a was done in 5 . The Lie group admitted by the Navier-Stokes equations is infinite. Its Lie algebra can be presented in the form of the direct sum L°° © L5, where the infinitedimensional ideal L°° is generated by the operators1" x^
= 4>i{t)dXi + 4>'i(t)dUi -
with arbitrary functions >i(t), (i = 1,2,3) and ip(t). The subalgebra L5 has the following basis: Y = 2tdt + XidXi - UidUi - 2pdp, Z0 = dt, Zik = XidXk - xkdXi + UidUk - ukdui, (i < k < 3). The Galilean algebra L 1 0 is contained in L°° @ L5. Several articles 8,9,10,11,12,13,14,15 a r e devoted to invariant solutions of the Navier-Stokes equations 0 . While partially invariant solutions of the Navier-Stokes equations have been less studied d , there has been substantial progress in studying such classes of solutions of inviscid gas dynamics equations 1,2,4,19,20,21,22,23,24
a
A classification of the two-dimensional Navier-Stokes equations was studied in 6 . T h e r e is still no complete classification of the subalgebras of the Lie algebra L°° © L 5 . Classification of infinite-dimensional subalgebras of this algebra was studied in 7 . c Short reviews devoted to invariant solutions of the Navier-Stokes equations can be found
b
j n 8,10,16,17,18 d
Firstly the approach of partially invariant solutions to the Navier-Stokes equations was applied in 8 .
528
4. Some particular solutions of the Navier-Stokes equations In this section partially invariant solutions of the Navier-Stokes equations are studied. In the manuscript we study the following subgroups {Xi,X±,X\Q,X\\},
{Xi,Xt,X6,aX5
{X2,X3,Xe,/3X4
+ In},
+ aX5 + I n } .
Here X\ = dx, X2 = dy, X3 = dz, X\ = tdx + du, X$ = tdy + dv, Xe - tdz + dw, X10 = dt, Xu = tdt + xdx + ydv + zdz. Note that the generator Xn is not admitted by the Navier-Stokes equations. The groups are taken from the optimal system constructed for the gas dynamics equations 2 5 . Partially invariant solutions of the gas dynamics equations for these groups were considered in 2 . The Navier-Stokes equations are used in the component form: Ut + UUX + VUy + WUZ = ~PX + UXX + Uyy + UZZ, Vt + UVX + VVy + WVZ = -Py Wt + UWX + VWy + WWZ = ~px
+ VXX + Vyy + VZZ ,
(5)
+ WXX + Wyy + WZZ,
(6)
UX + Vy + Wz = 0.
The dependent variables u,v,w x,y,z and time t.
(4)
(7)
and p are functions of the space variables
4.1. Partially invariant solutions {-X"i, X4, X10, X n }
with respect
We start from the simplest case {Xi, X±, X\o, Xn}. sentation of the partially invariant solution is
to In this case, a repre-
v = V(s), w = W(s), p = P(s),
(8)
where s = z/y. For the function u = u(t,x,y,z) there is no restrictions. Substituting the representation of partially invariant solution (8) into the Navier-Stokes equations (4)-(7), we obtain Ut + UUX + VUy + WUZ
{{W - sV)V
- (UXX + Uyy + UZZ) = 0,
(9)
- sP')y - ((s 2 + 1)V" + 2sV) = 0,
(10)
529 {(W - sV)W
+ P')y - ((s 2 + 1)W" + 2sW") = 0,
(11)
yux - (sV + W) = 0.
(12)
Since V and W only depend on s, equations (10) and (11) can be split with respect to y: (W - sV)V
- sP' = 0, {W ~ sV)W
+ P' = 0,
(13)
(s 2 + 1)V" + 2sV = 0, (s 2 4- l)W" + 2sW = 0.
(14)
Solving equations (14), we have V = Ciarctan(s)
+ C2, W = C^arctan(s) + C4.
Multiplying the first equation by s and combining it with the second equation of (13), we obtain (W - sV)(V
+ sW) = 0.
Let W - sV = 0, then Cx = C2 = C3 - CA = 0. This means that V = 0,W = 0. If V + sW = 0, then V = C2,W = d. In this case P = C5. Note that the Galilei transformation applied to V and W, also change s. Substituting V and W in equation (12), we have ux = 0. It means that u depend on t,y,z or u = U(t7s,y). Equation (9) becomes {{Uyy-C2Uy-Ut)y-2sUsy)y+{s2
+ l)Uss + {{C2y+2)s-Ciy)Us
= 0. (15)
Thus, there is a solution of the Navier-Stokes equations of the type u = U(t,s,y),
v = C2, w = C4, p = C 5 ,
where the function U(t,s,y) satisfies equation (15). This solution is a partially invariant solution with respect to the group correspondent to {X 1 ,X 4 ,Xi 0 ,-X'ii}. 4.2. Partially invariant solutions {X1,X4,X6,aX5 + Xu}
with respect
to
A representation of the partially invariant solution with respect to {Xi,X4,Xe,aX5 + Xu} is v = V{s) + y/t, w = W(s) + z/t, p = P(s), where s — y/t — a In t. The function u — u(t, x, y, z).
(16)
530
Substituting the representation of partially invariant solution into the Navier-Stokes equations (4)-(7), we obtain ut + uux + {V + -)uy + (W + -)uz - (uxx + uyy + uzz) = 0,
y ' ( - - -a)-lL W\--t
+
~a)-Z1
(y + y.)(y> + i) + p> _ — = o, (V+ ±)W + (W + ^ ) - ^ = 0,
+
tux + V + 2 = 0. Excluding y — ts + at\nt
(17)
(18) (19) (20)
in equation (18) and (19), we have
t(P' - aV + W + V) - V" = 0, t{VW - aW + W)- W" = 0. Because V and W only depend on s, we can split the last two equations with respect to the variable t, V" = 0, W" = 0, P' - aV + VV + V = 0, VW -aW' + W = 0.
(21)
Thus, V = ClS + C2, W = Czs + C4, P = (aCl - dC* ~ C2)s -
Cl(Cl
+
1)S2
+ g„
and the last equation of (21) becomes sC3(Ci + 1) - aC3 + C2C3 + Ci = 0.
(22)
After substituting V into equation (20), we get (Ci+2) ux = -±-±-—'-
or u=-
(Ci+2)x ., TTU j — ^ - + U(t,s,z).
where z = | . Equation (17) can be written as t2Ut + t[(ClS + C2- a)Us + (C3s + C4)Uiz - (Ci + 2)U] -U„-U-z-z + (Ci+2){C1+3)x = 0,
{
'
531
Splitting the equations (22) with respect to s and equation (23) with respect to a;, we obtain C73(Ci + l) = 0, CA + (C2 - a)C3 = 0, ( C 1 + 2 ) ( C 1 + 3 ) = 0, t2Ut+t[(C1s+C2-a)Us
(24)
+ (C3s+C4)U„-(C1+2)U]-Uss-U,i
= 0. (25)
If C3 = 0, then d = 0, and hence, W = 0. In this case equation (25) is t 2 [/ t + t[(ClS + C2- a)Us - (Ci + 2)[/] - [/„ - Ua = 0. Let d = - 2 , then V = - 2 s + C2, W = 0, P = (-2a + C2)s -s2 + C5 and the function u = U(t, s, z) satisfies the equation t2Ut + t[(-2s + C2- a)Us] - Uss - Un = 0.
(26)
If d = - 3 , then V = - 3 s + C2, W = 0, P= ( - 3 a + 2C 2 )s - 3s 2 + C 5 and u = -j + U(t, s, z) where t2Ut + t[(-3s + C2~ a)Us + U}- Uss - Ua = 0.
(27)
In the case C3 ^ 0 one has C\ = — 1, which contradicts to equation (24). In this case there exist partially invariant solutions of the Navier-Stokes equations which is partially invariant with respect to not admitted Lie group corresponding to the Lie algebra {X\, X4, X6, aX5 + I n } . One of the solutions is u = U(t,s,z),
v=-2s
where the function U(t,s,z) Another solution is u=--
+ U(t,s,z),
+ C2, W = 0, P = (-2a + C2)s - s2 + C5 satisfies equation (26).
v = -3s + C2, W = 0, P = ( - 3 a + C2)s - 3s 2 + C5
and the function U(t,s,z)
satisfies equation (27).
4.3. Partially invariant solutions with respect { X 2 , X3, X6, PX4 + aX5 + X11}
to
A representation of the partially invariant solution with respect to {X2, X3,X0, PX4 + aX5 + Xn} is u = U(s) + x/t, v = V(s) +alnt,
p = P(s),
(28)
532
where s = x/t — /?lni, and for the function w — w(t,x,y,z) there is no restrictions. Substituting the representation of partially invariant solution into the Navier-Stokes equations (4)-(7), we obtain t{P' - 0U' + UU' + U)- U" = 0, t{-pV
(29)
+ UV + a)- V" = 0,
(30)
X
wt + {U + -)wx + (V + a In t)wv + wwz - (wxx + wyy + wzz) = 0, (31) twz+U'
+ 1 = 0.
(32)
Because U and V only depend on s, then we can split the equations (29) and (30) with respect to the variable t, U" = 0,
V" = 0,
P' - (JU' + UU' + U = 0, UV -pV
(33)
+ a = 0.
(34)
Thus U = C\s + C2 and V = C^s + C4. After substituting them into equations (32) and (33), and solving them, we have
P=^C1-C1C2-C2)s-Cl{C\+1)s\c5,W
= -{-^p^
+ W(t,s,y),
where y = | . Equation (31) and (34) become t2Wt + t[(Cis + C2- P)WS + (C3s + C4 + a\nt~WSS - Wyy + (Cl + 2)(Cl + \)Z = 0,
y)Wy - {Cx + 1)W] (35)
SC1C3 + {C2C3 + o - PC3) = 0. Splitting the last equation with respect to s and equation (35) with respect to z, we obtain CXC3 = 0, a + (C2- p)C3 = 0, ( C i + 2 ) ( C i + l) = 0, t2Wt + t[(ClS + C2- P)WS + (C3s + C4+a\nt-{Ci + \)W]-W,a-Wyy = Q.
(36) y)Wy [
'
533
Since C\ = 0 contradicts to equation (36), then C3 = 0. In this case a = 0 and equation (37) becomes t2Wt + t[{ClS + C2~ P)WS + (C4 - V)Wy - (Cl + 1)W] - WSS - Wyy = 0. Let d = - 2 , then U =-2s + C2, V = C4, P= (-2/3 + C2)s - s2 + C5 and the function w — W(t, s, y) satisfies the equation t2Wt + t[(-2s + C2- P)WS + (C 4 - y)Wy + W}- Wss - WiS = 0. If Ci = - 1 , then U - -s + C2, V = C 4 , P = -/3s + C5 and the function w = 3 | + W(t,s,y) satisfies the equation t2Wt + t[(S + C2~ P)WS + {Ci - y)Wy] - WSS - Wyy = 0. Thus, as in the previous case there are solutions of the Navier-Stokes equations, which are partially invariant with respect to not admitted Lie algebra {X2,X3,X6,/?X4 + Xn}5. Conclusion The algorithm of obtaining partially invariant solutions was applied to the Navier-Stokes equations. Examples given in the manuscript showed that this algorithm can be applied to groups, which are not admitted. These possibilities extend an area of using group analysis for constructing exact solutions. References 1. L. V. Ovsiannikov, Group analysis of differential equations, Moscow: Nauka (1978). (English translation, Ames, W.F., Ed., published by Academic Press, New York (1982).) 2. L. V. Ovsiannikov and A. P. Chupakhin, J. Appl. Math. Mech. 60(6), 990999 (1996). 3. L. V. Ovsiannikov, In: Proceedings of 11-th Int. Congr. Appl. Mech. (1964). 4. L. V. Ovsiannikov, Dokl. RAS 343(2), 156-159 (1995). 5. V. O. Bytev, Chislennye metody mehaniki sploshnoi sredy (Novosibirsk) 3(3), 13-17 (1972). 6. V. V. Pukhnachov, J. Appl. Mech. Techn. Phys., (1960). 7. S. V. Khabirov, Partially invariant solutions of equations of hydrodynamics, In: Exact solutions of differential equations and their assymptotics. 8. V.V. Pukhnachov, Free boundary problems of the Navier-Stokes equations, Doctoral thesis (1974). 9. B. J. Cantwell, /. Fluid Mech. 85, 257-271 (1978). 10. B. J. Cantwell, Introduction to symmetry analysis, Camridge: Camridge University Press, (2002).
534 11. S. P. Lloyd, Acta Math. 38, 85-98 (1981). 12. R. E. Boisvert, W. F. Ames, and U. N. Srivastava, J. Eng. Math. 17, 203-221 (1983). 13. A. Grauel, and W.H. Steeb, Int. J. Theor. Phys. 24, 255-265 (1985). 14. N. H. Ibragimov and G. Unal, Bulletin of the Technical University of Istanbul 47(1-2), 203-207 (1994). 15. R. O. Popovych, Nonl. Math. Phys. 2(3-4), 301-311 (1995). 16. W. I. Fushchich and R. O. Popovych, Nonl. Math. Phys. 1(1), 75-113 (1994). 17. W. I. Fushchich a n d R . O. Popovych, Nonl. Math. Phys. 1(2), 158-188 (1994). 18. D. K. Ludlow, P. A. Clarkson, and A. P. Bassom, Studies in Appl. Math. 103, 183-240 (1999). 19. A. F. Sidorov, V. P. Shapeev, and N. N. Yanenko, The method of differential constraints and its applications in gas dynamics, Novosibirsk: Nauka (1984). 20. S. V. Meleshko, Classification of the solutions with degenerate hodograph of the gas dynamics and plasticity equations, Doctoral thesis (1991). 21. S. V. Meleshko, Differential Equations 30(10), 1690-1693 (1994). 22. L. V. Ovsiannikov, Differential Equations 30(10), 1792-1799 (1994). 23. A. P. Chupakhin, Dokl. RAS 352(5), 624-626 (1997). 24. A. M. Grundland and L. Lalague, J. Phys. A: Math. Gen. 29, 1723-1739 (1996). 25. L. V. Ovsiannikov, J. Appl. Math. Mech. 58, (1994).
COUPLED-MODE VERSUS NONLINEAR SCHRODINGER EQUATIONS FOR E L E C T R O M A G N E T I C WAVE P R O P A G A T I O N IN C O N T I N U O U S MEDIA
R. T O N E L L I , G. C A P P E L L I N I , F . M E L O N I INFM-Physics
Department, E-mail:
University of Cagliari - 09042 [email protected]
Italy.
S. T R I L L O INFM-Faculty
of Engineering,
University
of Ferrara - 44100
Italy.
We investigate Electromagnetic-Wave propagation in continuous media in which the presence of nonlinearities gives rise to parametric mixing. The case of a laser pulse injected in an optical fiber has been analyzed. This propagation is described by means of the Nonlinear Schrodinger Equation (NLS). On one side, we integrate numerically the NLS with different injection conditions describing phase and amplitude modulation as well as wave mixing phenomena. On the other side we illustrate a simplified model of coupled-mode equations giving an Hamiltonian description of the problem which enable us to obtain analytical solutions. We compare and discuss the results obtained with the two different approaches relatively to the same input conditions.
1. Introduction The problem of wave propagation in continuous media is one of the most investigated in both Physics and Mathematics. One of the most challenging case is the one in which the presence of nonlinearities in the response function makes the problem insolvable analytically. Very often the nonlinearities create instability with consequent complicate behaviors, exponential growths, bifurcations and so on. Here we would like to address as particular case the electromagnetic waves propagation in optical fibers in which linear and nonlinear components of the susceptibility are relevant 1 2 . Spatial instability effects appear with unstable eigensolutions, bifurcations and exchange of energy between different waves. Coupled mode equations are illustrate to give good agreement with the exact numerical solutions in many cases. The coupled-mode equations are thus useful to exploit many features of the system dynamics. In the case of light propagation in optical
535
536
fibers the process is called four wave mixing, given that four photons are involved in the interaction. When the two initial frequencies are degenerate, the parametric generation of two side-band waves, with frequencies up and down shifted with respect to the initial one, called Stokes and Anti-Stokes components, can occur. On the other hand, the same process can be seen more in general as the exponential growth of a perturbation, corresponding to the sidebands waves, propagating in nonlinear dispersive media. In this context the process is named Modulational Instability (MI) 4 . In nonlinear optics MI may occur when propagation is governed by the Nonlinear Schrodinger Equation (NLS) (in the anomalous group-velocity dispersion regime) 3 : .du 8-Z
1
+
ld2u 2W
.. + M
ll2 U
„ = °-
,„, (1)
Here we focus our attention on the case of the propagation of a modulated wave in a single mode optical fiber. This process is named parametric Three Wave Mixing (TWM). The carrier wave, at frequency LOO, mixes with two sidebands at frequencies UIQ ± 0, exchanging energy. Under these conditions a set of coupled-mode equations may be a good substitute of the NLS equation. In the present work we limit our investigation to the case of collinearly polarized waves. The coupled equations for the TWM, considering the three waves interacting through the intensity-dependent refractive index can be written as 2 3 : .dEp dz .dEx dz .dE2 dz
R[\\E0\\2 + 2(\\E1\\2 + ||£ 2 || 2 )]£o +
2RE*ElE2exp{iAkz),
RlWEiW2 + 2(\\E2\\2 + ||#>|| 2 )]£i + RE;E0E0exp(-iAkz), R[\\E2\\2 + 2(\\E1\\2 + ||£o|| 2 )]£ 2 +
(2)
RElE0E0exp(-iAkz),
where Ak = ki+k2 — 2fco is the low power propagation constant mismatch and the modes are normalized so that ||-E||2 gives the power in watts. Hence R = 27rn2/\oA, where n2 is the nonlinear refractive index at wo and A is the effective core area. The Eqs. (2), using conservation of energy, can be written in dimensionless variables, and give rise to a three-degree-offreedom Hamiltonian system 2 . In the MI case the coupled-mode scheme gives fair results with respect to the NLS ones while in the opposite case (two injected strong sidebands) the truncated model does not reproduce the correct wave propagation.
537 1.2 1 0.8 in
.1 °- 6 _3 O
£ u
0.4
O)
ili
0.2 0 -0.2 - 6
-
4
-
2
0 Mismatch k'
2
4
6
Figure 1. Nonlinear eigensolutions for a = 0: normalized pump power r/e for the stable (solid lines) and unstable (dashed lines) nonlinear eigensolutions versus normalized mismatch k'.
2. Hamiltonian Formulation Starting from Eqs. (2) we introduce 2 the amplitudes and phases of the waves, A 0 ,i, 2 (z) = ||.Eo,i,2(z)|| and 4>o,i,2(z)> where Ej = Aj-exp(i^j), j = 0,1,2. We use normalized dimensionless variables using the conservation of the total power P 0 = A2, + A2 + A 2 , introducing the normalized pump power r)(z) = [AQ]/PO, and the normalized sidebands amplitudes c*i,2(z) = Aj ) 2(z)/P 0 ' so that 7] + a\ + a\ — 1. Besides the total power the system admits also two other invariants: a = a\ — a\ 3 H = 47]aia2cos((f)) - (k' - l)t] - -rf
(3) (4)
where k' = Ak/RPo is the normalized mismatch. Given the three invariants, one can express the two normalized amplitudes a\^ as a function of the pump power rj and the last invariant results in an Hamiltonian formulation of the problem with 77 and <j> as conjugate variables: H = H(r,, >) = 2r?[(l - r?)2 - a2\l'2cos{c}>) - (k1 - 1)V - ^r,2
(5)
with the associated Hamiltonian equations 2 . We analyse the system us-
538
-1
o
1 - 1
o
1 - 1
0
1
r\ cos 0 Figure 2. Phase portraits for different values of normalized mismatch after the CoupledMode Hamiltonian model.
ing this Hamiltonian expression in the case of equal sidebands amplitude, so that a = 0, for different initial conditions and different values for the normalized mismatch k'. In this case the power exchanged with the pump is equally split between the two sidebands. There are fixed points for the Hamiltonian (5), corresponding to different eigensolutions that remains constant along the fiber like r](^ = 0) = 0 and rj((, = 0) = 1 or others r?(£ = 0) = rje. These and their stability properties depend on the value of k', as Fig. 1 clearly shows. The topology of the phase space depends on the existence of unstable eigensolutions, as illustrated in Fig. 2, where we report the trajectories in the phase plane (r^cos^sin^) for different normalized mismatch values.
3. Numerical Integration of NLS Next we numerically solved the NLS equation for the evolution of an initially weekly modulated wave, that provides modulational instability: u(£ = 0,t) = 1 + ee i * o / 2 cos(m)
(6)
539
Whenever e
77 c o s
$
Figure 3. Phase-space portrait for k' = -2 after projection from the exact solution of the NLS equation.
to exchange power between these three modes in a periodic fashion. We compare the phase portrait obtained by numerical integration of the NLS equation (Fig. 3) to the ones obtained by the model. Figures 2 and 3 show an excellent qualitative agreement between the two cases for the value k' = -2. The evolution of the solution both in time and in the Fourier space shows that only the first two modes, detuned at ft, exchange power with the central pump while the amount of power exchanged with higher order modes is negligible, both in the cases of amplitude or phase modulation (4>(0) = 0 or <^>(0) = 7r). The situation changes when we investigate the case of strong initial sidebands and small pump. The NLS numerical solution shows that higher order modes are involved in the exchange of power and are not negligible anymore (Fig. 4), contrary to the predictions of the model. In fact four wave mixing prevails with the generation of sidebands at 3fi 5 . In this case the agreement between NLS and Coupled-Mode Equations model ceases since the model can not take into account the generation
540
Figure 4. Propagation in the Fourier Space of a frequency modulated wave in which the pump is negligible with respect to the injected sidebands (pump = 0.01 and e = 2). In the frequency domain units of Cl have been used.
of waves detuned at frequencies different t h a n tt.
References 1. J. A. Armstrong, N. Bloenbergen, J. Ducuing, and P. S. Pershan, Phys. Rev. 127, 1918 (1962). 2. G. Cappellini, and S. Trillo, J. Opt. Soc. Am. B, Vol. 8, No. 4 (1991). 3. S. Trillo, and S. Wabnitz, Opt. Lett. Vol. 16, 986 (1991). 4. K. Tai, A. Hasegawa, and A. Tomita, Phys. Rev. Lett. 56, 135 (1986). 5. S. Trillo, S. Wabnitz and T. A. B. Kennedy, Phys. Rev. A 50, 1732 (1994).
A N E W CLASS OF LINEARIZABLE WAVE EQUATIONS
M. TORRISI, R. TRACINA, A. VALENTI Dipartimento
di Matematica e Informatica, Universitd di Catania, viale A. Doria, 6, 95125 Catania, Italy E-mail: [email protected], [email protected], [email protected] By using the differential invariants, with respect to the equivalence transformation algebra of the class of wave equations utt — uxx — f(u,ut,ux), we characterize a subclass of linearizable equations.
1. Introduction The search for transformations mapping nonlinear differential equations to linear differential equations has attracted in the recent decades the interest of many researchers. A pioneering paper in this field is due to Kumei and Bluman 1 , which following a group analysis approach get necessary and sufficient conditions for the existence of a linearizing map. In this short paper we consider the class of semilinear differential equations utt-uXx=
f{u,uuux).
(1)
In ref. 2, starting from the equivalence algebra Ls obtained in ref. 3 by using the infinitesimal method 4 ' 5 ' 6 , the differential invariants have been determined. These results have been applied in ref. 7 and some subclasses of the equations (1) which can be linearized through an equivalence transformation have been characterized. Here, by considering the linear equation utt - uxx = k(ut - ux)
(k / 0)
(2)
belonging to the class (1), we search for subclasses of the equations (1) which can be mapped by an equivalence transformation of Ls in a linear equation of the form (2). The outline of the paper is the following. In Section 2 we recall in short the results obtained, in reff. 2 and 3.
541
542
In Section 3, after specializing the differential invariants of (1) for the equations (2), we characterize the subclass of (1) that can be mapped into a linear equation of the form (2). 2. Lie equivalence algebra and differential invariant In ref. 3, by using the Lie infinitesimal criterion 8 , we have found that the equivalence Lie algebra for the class of equations (1) is infinite-dimensional and is spanned by the following infinitesimal operators: Y0 = xdt + tdx-ux
dUt - ut dUx,
Yy = dt,
Y2 — dx,
Y3=tdt+xdx-2fdf-utdUt-uxdUx,
(3) (4)
2
2
Yv = ipdu + [p'f +
(5)
where (p' and ip" are the first and second derivatives with respect to u of an arbitrary function <~p(u). In the following with prime we denote the derivative of a function with respect to the only variable on which it depends. It has been showed, in ref. 2, that the equations (1), with respect to G&, at order zero do not have differential invariants but possess the invariant equation (u2 — u2.) = 0 which implies / = 0. At first order, instead, the following cases arise: • The equations (1) admit the following differential invariant A =
2 / - ( u t -ux){fUt 2f-(ut+ux)(fUt+fUx)
- fuJ
(6) { )
provided that (ut + ux)(fUt + fuJ - 2 / ^ 0. • The equations (1) admit the following system of two invariant equations: («t - « x ) ( / » , - / « . ) - 2 / = 0,
(7)
( u t + « x ) ( / U t + / O - 2 / = 0.
(8)
3. Linearizing equivalence transformations An equivalence transformation of (1) is an invertible transformation of independent and dependent variables, t = a(i,x,v),
x = /3(t,x,v),
u = ,y(t,x,v),
(9)
that changes the equation (1) into an equation of the same form, vn - vxx = f(v, Vi, vx).
(10)
543
The function / , in general, may be different from the original function / . After observing that for the equations vii-vxx
= k(vi-vx),
(11)
the differential invariant (6) is zero, we solve, with respect to / , 2f-(ut-ux)(fUt-fuJ=0
(12)
and get / = (ut -ux)g(u,ut
+ ux).
(13)
So, all equations belonging to the class (1) and admitting A = 0 must assume the form utt-uxx
= (ut-ux)g(u,ut
+ ux).
(14)
By putting a = ut—ux,
r — ut+ux
(15)
equation (14) reads utt-uxx
=crg(u,T).
(16)
Now we calculate the second order differential invariants of equations (16) in order to get further restrictions about the function g(u,r). We consider the following change of variables: t = t,
x — x,
cr = ut-ux,
u = u,
T = ut+ux,
(17) ag(u,T)=f
(18)
in order to get the infinitesimal equivalence generator of (16): T = ?% + ?ds + 7)du + Vda + TdT + vdg.
(19)
Taking into account the procedure showed in ref. 9, concerned with the change of variables, we link the old coordinates £*, £ 2 , 77, // of the generator Y = CiYi + YV (CJ, i = 0 , . . . , 3 arbitrary constants) to the new coordinates £X>
(20) 2
= ?dt + t, dx + r]du + C ^ U t + C dUl + \ids + + ?% + ^dx + fjd* + ?
544
From the invariance of (17), it follows that:
?=e,i2=e,
n=v,
(21)
while from invariance of (18) it follows:
S = C 1 -C 2 ,
T-C'+C2,
£s + ™ = M,
(22)
from where we get the new infinitesimal components £, T and v. After some calculations we are able to write the equivalence generator T as T = (c3t + c0x + c{]dt + (c0t + C3,x + c2)dx + ip{u)du +(co - c 3 +
- g = 0.
(24)
We remark that for the linear equations (11) is g = k then rgT — g ^ 0. Looking for second order differential invariants, after performing the invariant tests ~£{2)(J(t,x,u,T,g,gu,gT,guu,gUT,gTT))
=0,
(25)
we get that the general form of second order differential invariants is J = J(ri, r2) where the two independent variables ri and r2 are defined as rgu - r2gUT i = -/ To".
r
(9-rgTy
r
r2gTT 2 =
•
26)
g~rgT
For the linear equation (11) is r\ = 0 and r2 = 0, that is rgu - r2gUT = 0,
T2gTT = 0.
(27)
By solving the equations (27), with respect to g, we get 9 = a[rh(u) + l0] where h is an arbitrary function of u and ^o is an arbitrary constant. So the equations of the class (1) which can be transformed in the linear form (11) must be of the form: utt - Uxx = cr[rh(u) + l0}. Then we can state:
(28)
545
Theorem. An equation belonging to the class (1) can be transformed in the linear form (11) by an equivalence transformation of Gs if and only if the function f is given by f = (ut - ux)[(ut + ux)h(u) + l0}.
(29)
Proof. The condition (29) is necessary as the equations like (28) admit the same differential invariant of the target (11). In order to demonstrate that the condition is sufficient, we must show that it exists at least an equivalence transformation which maps in (11) an equation of the form: utt - Uxx — (ut - ux)[(ut + ux)h(u) + lo\.
(30)
By applying to equation (30) the equivalence transformation of Gg t — i,
x = x,
u = xjj{v),
(31)
where ip(v) is an arbitrary function of v(i, x) with ip' ^ 0, we get vu - va = (ut- - v&) (Vi + vx)(iP'h(iP)-^)
+ l0
(32)
Then the equation (32) assumes the form (11), with k = lo, when the following relation is satisfied:
| £ =/ity(»))- n
(33)
E x a m p l e : We consider the equation: utt - uxx = (ut - ux) ( —
+ k) .
(34)
In this case, (33) becomes ip'2
ip
(35)
and, by integrating, we get V' = c D e ClV ,
(36)
with Co and c\ arbitrary non zero constants. By applying the change of variables u = c0eCl,\ equation (34) is transformed in the form (11).
(37)
546 Acknowledgments This paper was supported by University of Catania, by I N d A M through the project Modellistica Numerica per il Calcolo Scientifico e Applicazioni Avanzate and by G.N.F.M. ( G r u p p o Nazionale per la Fisica M a t e m a t i c a ) .
References 1. S. Kumei, G. W. Bluman, SIAM J. Appl. Math., 42, 1157 (1982). 2. R. Tracina, Communications in Nonlinear Science and Numerical Simulation, 9, 127 (2004). 3. M. Torrisi, R. Tracina, A. Valenti, On equivalence transformations applied to a non-linear wave equation, Modern Group Analysis: Advanced Analytical and Computational Methods in Mathematical Physics, N. H. Ibragimov et al. (eds.), Kluwer Academic Publishers, pp. 367-375, (1993). 4. N. H. Ibragimov, Elementary Lie Group Analysis and Ordinary Differential Equations, John Wiley & Sons, Chichester (1999). 5. N. H. Ibragimov, Nonlinear Dynamics, 30, 155 (2002). 6. N. H. Ibragimov, M. Torrisi, A. Valenti, Communications in Nonlinear Science and Numerical Simulation, 9, 69 (2004). 7. M. Torrisi, R. Tracina, A. Valenti, Nonlinear Dynamics, (to appear). 8. L. V. Ovsiannikov, Group Analysis of Differential Equations, Academic Press, New York (1982). 9. M. Torrisi, R. Tracina, Int. J. Non-Linear Mechanics, 3 3 , 473 (1998).
WAVE F E A T U R E S FOR A N E W C O N T I N U U M M O D E L OF ISOTROPIC SOLIDS *
G . V A L E N T I A N D C. C U R R O Department of Mathematics, University of Messina Contrada Papardo, Salita Sperone 31, 98166 Messina, Italy M. S U G I Y A M A Graduate
School of Engineering, Nagoya Institute of Showa-ku, Nagoya 466-8555, Japan
Technology
Acceleration waves propagating in isotropic solids at finite temperatures are studied on the basis of a new continuum model. The propagation speeds and the differential equations governing the time-variation of the wave amplitudes are derived. The analytical results, valid in a wide temperature range including the melting point, are evaluated numerically for several materials and their physical implications are discussed.
1. Introduction Recently a new continuum model has been proposed in order to analyze three-dimensional crystalline solids1. This model, which was derived statistical-mechanically from a three-dimensional anharmonic crystal lattice 2 , takes into account explicitly microscopic thermal vibration of constituent atoms, which can be observed, for example, by X-ray diffraction experiments. By using the new continuum model, linear harmonic waves have already been analyzed and discussed3. The aim of the present paper is to analyze the propagation speeds and the amplitude equation for acceleration waves propagating in isotropic solids by using the new continuum model 1 . Temperature dependences of the propagation speeds for longitudinal and transverse waves in Ag, Al, Cu, Ni, and Pb are numerically evaluated. "This work is supported by Intergroups Project 2003 of INdAM-GNFM,GNCS (coordinator A. Quarteroni) and by PRA 2001 of University of Messina (coordinator D. Fusco).
547
548
Moreover, the temperature dependence of the critical time is shown. The numerical results are consistent with the experimental data, although the data available are those observed at temperatures which are relatively lower than the melting temperature Tu2. Field equations The field equations for 3D crystalline solids at finite temperature incorporating microscopic thermal vibration of costituent atoms 1 can be written in matrix form as follows: fi-:U
- A ' ( U ) U X i = 0,
(1)
where Q, is the characteristic microscopic frequency, a"1 is the microscopic characteristic length, U is the field vector given by: U=
q
Fi
F2
(2)
while: 0 ISn I6i2 ISis
V F l ( V F , <*) 0 0 0
L-£
VF2(VF,
v F 3 (V F ,cr)
*)
0 0 0 0
0 0 0 0
o
Vr(VFia) 0 0 0 0
(3)
where I is the identity matrix, Sij the Kronecker symbol and the following relations hold: d
f da
' jk
da (Pi) fe
d / fin \ ur \ ura / dgpq A=
(4) 3
d(T
*2
ijQpq Orki
dgpn <^ypq
9
9pq dr
The field variables are the dimensionless velocity q, the deformation gradient F and the deviation of the temperature r induced by a wave from the reference equilibrium state. F 1 ; F2 and F3 in (2) denote the column vectors of the tensor F . The symmetric tensor g, namely the deviation of the atomic thermal vibration, is related to F and r by the following equation of state: (AI+g)V g ( r=
kBT 2D
1 +
D
k^f1
(5)
549 where D is the depth of the atomic pair potential, fcp the Boltzmann constant, T the absolute temperature of the reference equilibrium state and a = o (F, g) the dimensionless potential energy. In what follows, we adopt for a the expansion a (F, g) = A) + Pih + foil + £3/2 + £4/1/4 + foh + foil + £ 7 / 5 + /3sIf + foh + Pioh + Pnh + foih +£13/1/2 + foihh + £15/1/7 + / M A + £17/4 +foshh + £19/4/7 + £ 2 o/i 2 / 4 + £21/11\
(6)
with respect to the basic invariants defined by I\ h h h
= 9ii, II = 9ik9ki, h = 9ikgkj9ji, h = BSs ~ 3, = (Bst - 5st) {Bst - 6st), h = {Bst - S3t) (BtP - 8tp) (Bps - Sps) = 9st (Bts — fits), Is = 9ps {Bst —fist)(BtP — fitp), = 9tp9ps {Bst — fist),
(7)
being B = F F the left Cauchy-Green tensor. The expansion coefficients £'s occurring in (6) are functions of T and they can be evaluated in terms of the pair potential. 3. Acceleration waves In order to investigate the propagation of acceleration waves in crystalline solids, we consider a moving surface £(£) (wave front) of cartesian equation
^ |Vx4>| '
_ n
V ( i>
x? |VX0| •
/Sx W
As known, the normal speed of propagation V is equal to a characteristic velocity evaluated in the unperturbed field Uo while the jump vector II is proportional to the right eigenvector d evaluated in Uo, namely II = TTd(Uo). The amplitude ir of the jump satisfies a Bernoulli-like equation which, if Uo is constant, reduces to 4 : ~
+ j f (log 6) 7T + | V0| (VA • d) Q 7T2 = 0,
(9)
where ^ is the time derivative along bicharacteristics and 9 = -4? with b the Gaussian curvature of the wave front S(t).
550
By introducing the dimensionless propagation speed V = ^ , the characteristic equation cfet(A*n; -f- VI) = 0 becomes: V7 (V6 - 1AV4 + IIAV2
- 777 A ) = 0 ,
(10)
being 7^, 77^, 777^ the principal invariants of the matrix A whose components are given by: 3 , da 2 "• dgpq
dgPq \ dr
L
d ( 9a_„ \~\
J*-ii — (l
d ( da„
\ da (dgpq „ \
d
4. $"
9pi\
^2 ^ aSp, c>r J
It is easy to see that the characteristic equation (10) gives rise to two possibilities, namely V = 0 or V ^ 0, therefore the components of the corresponding right eigenvector d = [Q $ i 3>2 * 3 R] satisfy the following conditions: i) if V = 0 (standing waves) Q = o, V F i [(VFfc^)nfe] # i + RVr [(VFka)nk}
= 0;
(11)
ii) if V ^ 0 {V 2 I - mVFi [(VFfca)nfc] + V r [(VFka)nk) *i
=
- ^ Q ,
Pn} Q = 0 , (12)
7?=ip„-Q.
Now we restrict our analysis to the case in which the unperturbed state is the reference equilibrium state Uo = [0 5n fe <5i3 0] , so that the relation (11)2 evaluated in UQ reads as * $iknk + 9kink
N
= —-rii
with N =
(a-b
L = 4/37
7? (3/34 + A>) + + 6\/32) (a - 6 + 6A/?2) 2A/?,•22 a —b
/? 4 (3/3 4 +2/? 5 )
8A2/32/3f ( a - 6 ) ( o - 6 + 6A/3a)
Therefore, as expected, for standing waves, 7? and two vectors 3>, remain arbitrary.
551
Furthermore, the relation (12)i, evaluated at U 0 , implies two different propagation modes. ii)i Transverse waves: the dimensionless characteristic speed and the corresponding right eigenvector are, respectively, 2A/3f
V2
^ Q VT
dx =
4/?7
- ^ Q VT
0
being Qn = Q • n = 0. Since transverse acceleration waves are double waves, they are exceptional in the sense of Lax-Boillat 6,8 . Therefore, the amplitude equation becomes linear and, despite of the nonlinearity of the system, shocks can never occur on the wave front 4 ' 9,10 . ,
r
- .
i
1
.
r
4
'
i
_ .,. . , ,
Ag
H
<^
i
1^
Ag
Ni
Ni "
S
Cu Pby
Y\\
Pb
Al /
"
Al
Cu
/
-
2
-
-
°- 2
kBT/D
0 4
0 2
kBT/D °- 4
Figure 1. Temperature dependence of the dimensionless propagation speeds for transverse waves (left) and for longitunal waves (right).
ii)2 Longitudinal waves: the dimensionless characteristic speed is V2 = 4 2 (Pe + Pi) -
, o k (3/34 + 2/?5) ( a - 6 ) ( a + 26)_ a + 26
(13)
with a = |/3i + 2A/?2 + 2A/?3, 6 = §/?i + 2A/?2. The right eigenvector corre-
552
sponding to Vjr, is dL =
nin
n2n
n3n
4A/3i (30 4 + 0s)
Vb
Vi
Vi
3V L (a + 26)
1 T
(14)
In this case the amplitude equation (9) specializes into: d7r d „ _ + _(lo
„. g 6
|
) T
AQ, , __T »=0,
(15)
where the coefficient A is defined by A = 12 (0 6 + 0 7 ) + 24 (fi10 + ft? + 0is) + £ [3 (0 4 + 05 ) + 12 (011 + 014 + 019 + 021)] +K [3 (304 + 0 5 ) + 12 ( 0 n + 0 1 9 ) + 36'(0i4 + 02i)] +
EK_ [ g ^
+
Wfe
+ 1 2 A (/?i2 + 2/?i5 +
^i6
+
g^^j
[I (ft + 05) + 6A (012 + 015 + 016 + 020)]
+'
(16)
+¥- [3 (304 + 0 5 ) + 6A (0i 2 + 30i5 + 30i6 + 902o)]
+
BK2
[9 (30 2 + 0 3 ) + 9A (90 8 + 09 + 3/?i3)]
+ M1K [1302 + f 03 + 3A (90 8 + 30 9 + 50 13 )] + TT [I (& + ft) + 3A (0 8 + 0 9 + 0 13 )] +^
[9 (302 + 0 3 ) + 9A (90s + 09 + 30 13 )] ,
being K=^it%?:l7b)b)) and £ , - ^ In the particular case of plane waves (# = 1), from (15), by integration, we obtain: TO (17) v?
TT0t
with 7To being the initial value of 7r at t = 0 . If the condition J4.7TO > 0 is satisfied, the critical time, that is the time in which weak discontinuities may evolve into a shock wave, is given by tr.
(18) A-KQQ.
In the spherical case, 9 (t) = - g ^ , with i?o being the initial radius and /to
R(t) = R0 + Vtt- If VL > 0 the wave is outgoing, while if VL < 0 it is
553
u
Al
^^N -
-10
:
|
•
-20
D i
i
0.1
0.2
.
>*
0.3 kBT/D
Figure 2. Temperature dependence of the quantity A/V£ equation for longitudinal wave.
occurring in the amplitude
incoming. Then, the integration of the amplitude equation (15) leads to (19)
TT(i)
R(t)\vl
-
aA%ir0\og(m)
For both incoming and outgoing acceleration waves, the critical time is given by aRo
exp [ — ^
j - 1
(20)
\aARoir0J if the condition ATTQ > 0 is satisfied. 4. Conclusions In Figs.l the temperature dependences of the dimensionless propagation speeds for transverse and longitudinal waves in several materials are shown. A direct inspection of these figures shows that as T tends to TM from lower temperature side, the propagation speeds decrease rapidly and singularities occur at TM- In both cases their values at TM are, however, finite.
554 As explained above, the critical time depends essentially on the factor A/V£ if t h e condition A-KQ > 0 is fulfilled. T h e t e m p e r a t u r e dependence of this factor is shown in Fig.2. From t h e figure we notice t h a t t h e factor is nearly constant in a relatively lower t e m p e r a t u r e region while as T tends t o Tyi this q u a n t i t y decreases rapidly a n d a singularity occurs b u t its value is again finite. Therefore, once ITQ is fixed, t h e critical times becomes smaller as T t e n d s to TMReferences 1. M. Sugiyama, J. Phys. Soc. Jpn. 72, 1989 (2003). 2. M. Sugiyama, K. Goto, J. Phys. Soc. Jpn. 72, 545 (2003). 3. M. Sugiyama, K. Goto, K. Takada, G. Valenti, C. Curro, to appear on J. Phys. Soc. Jpn. 72 (12), (2003). 4. G. Boillat, La propagation des ondes, Gauthier Villars, Paris, 1975. 5. T.Y. Thomas, J. Math. Mech. 6, 311 (1957). 6. G. Boillat, C.R. Acad. Sci. Paris 274 A , 1018 (1972). 7. G. Boillat, T. Rugged, Wave Motion 1 (2), 149 (1979). 8. P.D. Lax, Contribution to the theory of partial differential equations, Princeton University Press, 1954. 9. T. Ruggeri, in Nonlinear wave motioned. A. Jeffrey, Longman, 43, 148 (1989). 10. A. Jeffrey, Quasilinear Hyperbolic Systems and Waves, Pitman, London, 1976. 11. L.A. Girifalco, V.G. Weizer, Phys. Rev. 114, 687 (1959). 12. I.M. Torrens, Interatomic potentials, Chapt. IV, p.54, Academic Press, New York, 1972.
STATIONARY KINETIC EQUATIONS W I T H COLLISION T E R M S RELATIVELY B O U N D E D W I T H R E S P E C T TO T H E COLLISION F R E Q U E N C Y *
CORNELIS VAN DER MEE Dipartimento
di Matematica e Informatica Universita di Cagliari Viale Merello 92, 09123 Cagliari, Italy Email: cornelis6bugs.unica.it
In this article the existence and uniqueness theory of stationary kinetic equations in L 1 -spaces is developed for collision terms dominated in the norm by the collision frequency.
1. Introduction In this article we study boundary value problems of the type Oti
Ou
v • — + a(x, v) • — + h(x, v)u(x, v) — (Ju)(x, v) + f(x, v), u-(x,v)
= (Ku+)(x,v)
+g-(x,v),
(x, v) G S; (1) {x,v) G £ _ ; (2)
where the position x G Q (Q an open subset of M"), the velocity v G V (V a subset of R" equipped with a positive Borel measure JIQ such that all bounded Borel sets in K™ have finite /^o-measure), and S = Q, x V equipped with the product measure dn(x,v) = dxdfj,o(v). We assume that a(x,v) is real and continuous in (x, v) and Lipschitz continuous in v on the closure of S, introduce the vector field v
d
-u f
^
d
"Research supported by MIUR under COFIN grant No. 2002014121, and by INdAMGNFM.
555
556 and suppose that for any C1-function <> / of compact support in £
L
Xcj)d[iT = 0,
/£
meaning that X is divergence free. Then through every point of E there passes exactly one integral curve of X. The left endpoints form the incoming boundary E_ and the right endpoints the outgoing boundary £ + . We assume in addition that no maximal integral curve of X can have a left or right endpoint in <9£ where v = a(x, v) = 0. Next, we assume that (1) the function h(x,v) is nonnegative and locally /x-integrable, (2) the operator J is real and satisfies
||Ju||i<*IIMIi
(3)
for some 5 £ (0,1), and (3) the operator K has norm strictly less than 1. If J and K are positive operators (in lattice sense), then we shall allow K to have unit norm and 5 to equal 1. A comprehensive theory of the existence and uniqueness of the time dependent counterpart of Eqs. (l)-(2) has been developed by Beals and Protopopescu 3 (also Chapter XI of Greenberg et al. 9 ) to cover situations where the operator J is bounded. It has recently been extended by Van der Mee 15 to deal with situations where J is the sum of a bounded operator and one satisfying (3). We mention that important earlier work on the time dependent problem was done by Voigt 21 for the case where a = 0 and J = 0 and Ukai 20 for J = 0. In addition to these papers, the literature is littered with treatments of particular examples, but discussing them is beyond the scope of this article. Let us outline the basic method of Beals and Protopopescu, 3 Greenberg et al., 9 and Van der Mee. 15 Assuming a phase space £ equipped with a Borel measure /j, and a vector field independent of t and writing 9
,
s
d
X = v.-+a(x,v).-, the fact that the vector field is divergence free may be expressed through the Green's identity
/ X(j)dfi=
7s
/ JT,+
4>dfj,+ — /
(j>d^i.
JT,-
for 4> in a suitable test function space, where /J± are suitable measures on E±. After constructing the boundary measures / ^ and the test function space pertaining to the vector field Y ~--§j+X, Eqs. (l)-(2) with J = 0 and K = 0 reduce to ordinary first order differential equations along the integral
557
curves of Y which can be solved trivially. Two perturbation arguments then allow one to incorporate a bounded J and K with \\K\\ < 1 into the theory. If J and K are positive operators, a monotonicity argument allows one to extend the existence and uniqueness result to operators K of unit norm. When developing existence and uniqueness theory for the stationary kinetic boundary value problem (l)-(2), there are essentially two approaches. One approach, favored by the French school in kinetic theory, is to prove that the corresponding time evolution semigroup has a negative spectral bound and hence has A = 0 in its resolvent set. This immediately implies that Eq. (1) with the homogeneous boundary condition (2) (i.e., with g_ = 0) is uniquely solvable in the functional setting to which the spectral result pertains. The case (x) + f{x),
x G (0, r ) ,
(4)
on a finite interval or on the half-line under boundary conditions involving projected boundary data. Starting from the operator-theoretic formulation of the one-speed neutron transport equation with isotropic scattering by Hangelbroek and Lekkerkerker,10 one can in fact distinguish two major subapproaches. In the subapproach launched by Beals 1 the solutions tp(x) are sought in an extended Hilbert space for which two natural scalar products are proven to be equivalent and to yield existence and uniqueness as a corollary. This subapproach, originally developed for positive selfadjoint operators A, has been made to apply also to indefinite Sturm-Liouville boundary value problems 2 and bounded and accretive A.16 In the second subapproach, initiated by Van der Mee, 13 by assuming compactness of B — I — A one is able to (1) seek solutions within the given Hilbert space, and (2) convert Eq. (4) with boundary conditions into a vector-valued convolution equation of the form 1>{x)- f •H{x-y)B^{y)dy = uj{x), x e ( 0 , r ) , (5) Jo where Fredholm techniques can be applied to either the given boundary value problem or the convolution equation (5). An up-to-date account of the two subapproaches can be found in Chapters II-IX of Greenberg et al. 9
558
In problems where the natural functional space is L : (E;ti/i) and some equilibrium condition demands that J {hu - Ju} dfi = 0,
wei'lEid/jJnL^EjMja).
(6)
a theory in an L 1 -setting for J satisfying (3) comes to mind in a natural way. To mention a few applications with unbounded h > 0 and positive J satisfying ||Ju||i < <5||/iu||i for some 5 £ [0,1], just consider (1) neutron transport where the collision frequency dominates the collision kernel integrated over outgoing velocities if the medium is nonmultiplying, 4 (2) radiative transfer where the phase function integrated over postscattering directions is dominated by the extinction coefficient,6 19 (3) cell growth modeling, 18 14 (4) electron transport in weakly ionized gases, 8 (5) rarefied gas dynamics, 5 (6) electron-phonon interaction in semiconductors, 11 12 and (7) the linearized Boltzmann equation with infinite range forces.17 r In many (if not all) of these applications, the integrated (nonnegative) collision kernel is exactly equal to the collision frequency. In fact, Eq. (6) is the linear counterpart of the balance condition involved in the nonlinear Boltzmann equation. In the time dependent counterpart of Eqs. (l)-(2) the vector field to consider on the spatial-velocity-time phase space AT = £ X (0, T), namely Y = {d/dt) + X, has only integral curves on which the travel time does not exceed T and which have both a left and right endpoint. a This allows us to parametrize the points of E x (0, T) as (z, s), where z is a left endpoint of a maximal integral curve of Y and s € (0, £(z)) is the travel time parameter, £(z) standing for the total travel time along this curve. We may then write the initial-boundary value problem as an elementary initial value problem by combining the initial data go and the boundary data g_ into one initial data g~ = (go, g_) on the incoming boundary (AT)" and solve the resulting initial value problem in L1 (AT , dfix), where dfiT = dfj,dt is the product measure. This can be done explicitly if K = 0 and J = 0 and by contraction mapping and monotonicity arguments for more general K and J. The more extensive variety of integral curve parametrizations complicates the study of the stationary Eqs. (l)-(2) in comparison to their time dependent counterpart. Integral curves may or may not have a left and/or a right endpoint, may be closed loops and may allow an infinite travel time. Thus when parametrizing them using the travel time parameter s, the doa
O n e also assumes that the integral curves of Y do not run off to infinity in finite time.
559 main of parametrization is either a finite interval, a left half-line, a right half-line, the full real line, or a circle. In this article we shall limit ourselves to the case in which all of the integral curves have a left endpoint. We can then parametrize E as E = { ( z , s ) : z G E _ , s € (0,£(z))} and identify the measure /J, with the product measure d^-ds. We now briefly describe the organization of the paper. In Sec. 2 we discuss the Green's identity for the vector field X in the case in which all integral curves of X have a left endpoint. We also derive solutions in L 1 (E; hd/j,) if J = 0 and K = 0. In Sec. 3 we obtain solutions in L ^ E ; hdfi) for general J and K with 6, K G [0,1) and explore extension to the case 5 < 1 and K = 1. In Sec. 4 we explore the stationary problems for which X has only closed loop integral curves. 2. The Green's Identity Let us define L 1 , ' o c (E; dfj.) as the linear space of all /x-measurable functions u on E which are /x-integrable on every bounded /i-measurable subset of E on which £(z, s) = £(z) is bounded away from zero. Further, let &T be the test function space of all Borel functions u on E such that (i) u is continuously differentiable on each integral curve of X, (ii) u and Xu are bounded, and (iii) the support of u is bounded and the travel time along the integral curves meeting the support of u is bounded away from zero. Then if u,Xu £ L1(E; / G $ < Xu,4> > + < u,X(f> >= I
u+4>d/j,+ — /
u-4>d/j,-.
Then if {u, (X + h)u} C L1(E,dyLj), u has a unique trace u±. Moreover, if u_ £ L1(E_;
\u+\dfi++
/ h\u\dfi=
/
\u-\dfi-+
/ sgn(u)(X + h)ud/i.
(7)
Observing that u(z, s) = exp
/ h(z,a)da u_(z)+ exp — / Jo J Jo L Jo
h(z,a)da f{z,o)da, (8)
we now immediately have
560
Proposition 2.1. Given f € L 1 (S,d/i) andg^ e L 1 (S_,d/i_), the unique solution u = S{f, g_) of the boundary value problem Ou v- — +a(x,v)
On • — + h(x,v)u(x,v)
= f(x,v),
u„{x,v) =g_(x,v),
(x,v)e'E;
(9)
(x,v)eT,-;
(10)
satisfies | | H l i + l l « + l | i < H / | | i + ||5-||i,
(11)
where the equality sign holds if f > 0 and g_ > 0. We remark that the above solution u of Eqs. (9)-(10) belongs to L 1 (E;d|x) whenever h is essentially bounded away from zero (i.e., if h~~l € L°°(E;dA0). 3. Using the Method of Characteristics In this section we shall prove the unique solvability of Eqs. (l)-(2). Theorem 3.1. Given f £ Lx{Y,,dy?j and g^ 6 L 1 ( S _ , d ^ _ ) , the boundary value problem (l)-(2) has a unique solution u € L1(E;/irf/i) having trace u± 6 L1(T,±;dfi±), provided there exist 5,K£ [0,1) such that \\Ju\\i < 5\\hu\\i,
\\Ku+\\i <
K||U+||I.
Further, u is nonnegative if J, K, f and <7_ are nonnegative. u £ Lx(S;
Finally,
Proof. Suppose J = 0 and K e [0,1)- Then any solution of Eqs. (l)-(2) satisfies u = S(f, Ku+ + g~), where u+ = S(0,Ku+)+
+
S(f,g_)+.
Since ||5(0,.Jsru + ) + ||i < ||i^u + ||i < K | | U + | | I , a contraction mapping argument yields u+ £ L 1 (E + ;d/x + ) uniquely. Denoting the so-obtained solution by u = Z(f,g_), we have H W . S - ) l l i + \\Z(f,g-)+h
< ll/lli + \\KZ{f,9-)+
< H/lli + \\g_\\x + K\\Z(f,g.)+h
< ^—
+ S-||i
(H/lli + ||5-||i) •
Let us now consider Eqs. (l)-(2) for J and K with 5 + K < 1. Then any solution u satisfies u = Z ( J u + /,
561 Moreover, since
11^(^,0)11! < - L . \\Ju\u < -±- \\huw,, J. — K
1
K
a contraction mapping argument yields the existence of u if 5 + K < 1. Denoting the so-obtained solution by u = W ( / , g_), we get
WhWif^h
+ WWfrg^+hiZ
*IIHIi + ll/lli + lb-Ik 1
- K
||/||l + ||fl-lll 1-5-K
"
, V
}
^
Applying (7) to Eqs. (l)-(2) we find
\\hu\U + IKIU < H^+lli + ||5-||i + ||J«||i + ll/Hi,
(13)
where the equality sign occurs if J and K are positive operators, / > 0 and g_ > 0. Hence ( l - 5 ) | | H l i + ( l - « ) l l « + l l i < l l / l l i + llfl-lli,
(14)
which suggests that the restriction to 5, K £ [0,1) with S + K < 1 is not necessary. Let us now extend the above estimates for W{f,gJ) to the case where 5 < 1 and K < 1, without assuming that 8 + K < 1. Now choose KO,K\ £ [0,1) such that K = KO + K\ and S + Ko < 1, and let u = V(f,g_) denote the solution of Eqs. (l)-(2) with K replaced by (KO/K)K and J = J. Replacing K by (KO/K)K and observing that the latter boundary operator has norm Ko and that 5 + «o < 1, we obtain from Eq. (12) the bound
||/iK(/.5-)lli + \\V(f,g-)+h < l l { l | l ^ l | g " 1 ' 1 . 1 — O — /Co
Now observe that the solution of Eqs. (l)-(2) has the form « = V(f, {KI/K)KU+
+g-)
Since Eq. (14) (applied for {KQ/K)K the place of K) implies that V(0AKU+)+
1
<
= V(0, (Kl/K)Ku+)
+
V(f,g_).
instead of K, and hence with Ko taking
1 — K0
<
l|u+||i,
1 — Ko
and since ( K I / ( 1 — «o)) < 1, a contraction argument yields the existence of the solution u of Eqs. (l)-(2). As a result of Eq. (14), we now obtain the estimate \\hWU^)h
+ l|W(/,*-) + Hi <
ll/
"l1+l'g"1'1 + 1—0
ll/lll
+llg 1
1 —K
'"1,
(15)
562
valid under the hypothesis that S, K € [0,1).
•
Using the monotonicity argument of Beals and Protopopescu XI.5 of Greenberg et al., 9 we obtain
3
and Sec.
Theorem 3.2. The boundary value problem (l)-(2) has a unique solution u G L X (S; hdjj) for every f G L 1 (S; d/j.) and g_ G L 1 (E_; d/j,^), provided J and K are positive operators and ||J«||i<J||Hli,
pru+||i
for certain S G [0,1) and K G [0,1]. Finally, u G L 1 (S;d/i) whenever h is essentially bounded away from zero. In general, under the conditions of Theorem 3.2 the solution u of Eqs. (l)-(2) need not satisfy u± G L1 (£±; d^±) if 6 = 1. 4. Closed Loop Integral Curves To illustrate the pitfalls of having X with closed loop integral curves, we consider the vector field X
= y0-X-X0y-
^
2
on E = R equipped with the Lebesgue measure and the nonnegative measurable function h on R 2 . Defining (x,y) = ^Jx1 + y2 (cos s, sin s) and z(x,y) = (A/X 2 + y2,0), we see that any solution of Eq. (7) must satisfy (8), where /•27T
h(z,a)da
u(z,0) = exp /•27T
+ / JO
r
u(z,0)
nlix
exp — /
h(z,a)da
L Jo
f(z,a)da. J
allowing one to compute u(z:0) uniquely from / G L J (R 2 ), provided h(z,a) ^ O o n the integral curve passing through (z,0). Integrating along integral curves, one obtains the estimate
|IHIi
(17)
where equality holds whenever / > 0. Thus Eq. (7) has a unique solution u G Ll{R2,hdxdy) for every / G L1(R2;c!a:ciy), unless h vanishes a.e. on some annulus about the origin.
563 Instead of denning t h e vector field l o n R 2 , we can also define it on E = R \ {(x,0) : x > 0} endowed with Lebesgue measure. T h e n every integral curve is a circle a b o u t the origin with left and right endpoint on the "cut" E-t = {(ar,0) : x > 0 } , which is equipped with the measure dfi±(x) = xdx. Using the "periodic" b o u n d a r y operator K : L 1 ( E + ; d / u + ) —> L 1 ( E _ ; d / / _ ) which acts as t h e identity on Lx(M.+ ;xdx), we convert the problem described in the preceding p a r a g r a p h into a problem as t r e a t e d in Sec. 3, where g~ = 0 and J = 0. 2
References 1. 2. 3. 4.
R. Beals, J. Fund. Anal. 34, 1 (1979). R. Beals, J. Diff. Eqs. 56, 391 (1985). R. Beals and V. Protopopescu, J. Math. Anal. Appl. 121, 370 (1987). K.M. Case and P.F. Zweifel, Linear Transport Theory, Addison-Wesley, Reading, MA, 1967. 5. C. Cercignani, The Boltzmann Equation and its Applications, Appl. Math. Sciences 67, Springer, Berlin, 1988. 6. S. Chandrasekhar, Radiative Transfer, Oxford Univ. Press, London, 1950; also: Dover P u b l , New York, 1960. 7. F. Chvala, T. Gustafsson and R. Pettersson, SIAM J. Math. Anal. 24, 583 (1993). 8. G. Frosali, C.V.M. van der Mee and S.L. Paveri-Fontana, J. Math. Phys. 30, 1177 (1989). 9. W. Greenberg, C. van der Mee, and V. Protopopescu, Boundary Value Problems in Abstract Kinetic Theory, Birkhauser OT 23, Basel-Boston-Stuttgart, 1987. 10. R.J. Hangelbroek and C.G. Lekkerkerker, SIAM J. Math. Anal. 8, 458 (1977). 11. A. Majorana, Transport Theory Statist. Phys. 20, 261 (1991). 12. A. Majorana and C. Milazzo, J. Math. Anal. Appl. 259, 609 (2001). 13. C.V.M. van der Mee, Int. Eqs. Oper. Theor. 3, 529 (1980). 14. C.V.M. van der Mee, A transport equation modelling cell growth. In: P. Tautu (Ed.), Stochastic Modelling in Biology. Relevant Mathematical Concepts and Recent Applications, World Scientific, Singapore, 1990, pp. 381. 15. C.V.M. van der Mee, Transp. Theory Stat. Phys. 30, 63 (2001). 16. Cornelis van der Mee, Andre Ran, and Leiba Rodman, J. Fund. Anal. 174, 478 (2000). 17. R. Pettersson, J. Statist. Phys. 59, 403 (1990). 18. M. Rotenberg, J. Theor. Biol. 103, 181 (1983). 19. V.V. Sobolev, Light Scattering in Planetary Atmospheres, Pergamon Press, Oxford, 1975; also: Nauka, Moscow, 1972 [Russian]. 20. S. Ukai, Stud. Math. Appl. 18, 37 (1986). 21. J. Voigt, Functional-analytic treatment of the initial-boundary value problem for collisionless gases, Habilitationsschrift, Ludwig Maximilians Universitat Munchen, 1980.
ON RELATIVE A S Y M P T O T I C STABILITY W I T H TWO M E A S U R E S VIA LIMITING EQUATIONS GIUSEPPE ZAPPALA' Dipartimento di Matematica e Informatica dell'Universita' di Catania 6 Viale Doria Catania 95125 Italy E-mail: [email protected]
The aim of this paper is to study the relative asymptotic stability and to mix the method of Liapunov's functionsfamilies(Matrosov, Rouche, Salvadori) with the method of limiting equations (Sell, Artstein, Andreev).
1. Start points Two differential systems of Caratheodory's type x = f(t, x) teR+
y = g(t,y)
tel
= I x€Rn
y 6 R'
= tf x(t') = x'
y{t') = y'
R' x R' = R"
(1)
(2)
where: a.)x(t) = x(t,t',x') and y(t) = y{t,t',y*) 3 for t > 0. Two measures h and h' € C[(t,x,y) —> I] s.t.: inf./i' = 0 for every t £ I and also 3A > 0, 3m = m{u) € A'(Hahn) so that(s.t.) h' < A implies (—>•) h < m(h') < m(A) . The Cartan's Silov's limit theory and the Definition 1.1. The systems (1,2) are said to be relatively (h',h)- stable if (Vc > 0)(W G I)(35 > 0)(Va:/,j/' e R' • h'(t',x',y') < S) we obtain h[t, x(t), y(t)] < e V< >t' -(uniformly stable if S = S(e)). -attractive if (W e I){3z > 0)(Ve > 0)(Vx',t/ 6 R' : h'{t',x',y') < z)(3T > 0) s.t. h[t,x(t),y(t))] < e Vt > t' + T (uniformly attractive if T = T(c)). asymptotically stable = stable+attractive. uniformly asymptotically stable=uniformly stable+uniformly attractive We denote: AF = {S : (t,x,y) -)• 7,sup5 > 0}, C F = {F 6 X F n C } , CL = {V € C(f, x, y) —> R and x, y lipschitzian }.
564
565
2. S t a b i l i t y a n d L i a p u n o v ' s f u n c t i o n s T h e o r e m 2 . 1 . Suppose that for every s > 0 a function V = V(t,x,y) G CL a constant I > 0 exist s.t. : a) h(t, x,y) = s —> V(t, x, y) > I; b) h —)• 0 —> V —y 0; c) V(t,x,y) < 0 . Then the systems (1,2) are relatively (h', h)- stable. T h e o r e m 2.2. Suppose that two functions V = V(t,x,y), W = W(t,x,y) G CL a constant c > 0 exist s.t. : a) V > 0, W > —c; b)/j -s- 0 —> V, W ->• 0 c) V, W < 0, ; d) for every s > 0 two numbers r, b > 0 exist s i . ; h = s and V < r implies W > b. Then we obtain a parameter function family useful for the theorem2.1 T h e o r e m 2 . 3 . Suppose that for every s > 0 there exist a function V G CL and two constants I, L G R so that: a)V < 0, 0 < / < L; b) h = s —> V > I; h < s —> V < L; c) s ->• 0 implies I ->• 0, L ->• 0; d) i / 0 < s' < s, h{i,x,y) < s' implies V(t,x,y < V'(t,x,y). Then the systems (1.2) a r e relatively (h' ,h)-uniformly stable. T h e o r e m 2 . 4 . Suppose that the theorems 2.2, 2.3 /io/d a n d a/so a) Vs > 0 3u > 0 s.t.: h(t,x,y) < s we have V, W < u; b)0 < s' < s implies r'/(c + 6') > rj{c + b); c) s —>• 0 implies r, u »-> 0. T/«en 2/ie systems (1.2) a r e (h,h')-relatively uniformly stable. L e m m a 2.1. Suppose that four functions
Vi(u)Qc(z
- u)du;
u,z£RxR".
(3)
JRxR" We have a,- G C 0 0 ; 0 < a,- < 1; supp a,- C A , Q« = 1 for (*,a:,y) G D",| | g r a d t t j | | < A^(constant> 0). Let (t,x,y) G AZ put J = J(t,x,y) = E(i=i,2)(-l)i+1[aiT}(t,x,y) we
566
obtain J>0;
J=
Y,
{-l)i+1[<XiT + aiT]<3NS.
(4)
(• = 1,2)
li{t,x,y) G D"TinAZ (i = 1 or 2) we obtain a,- = 1, J = (-l)i+lT hence J < —S'. Since, for every r > 0 the previous results hold, we conclude. Lemma 2.2. Suppose that for the functions
3. Precompactness, limiting equations and asymptotic stability In this chapter we suppose that: a) 3Ai > 0 s.t. sup {||a;||, \\y\\ : h'(t,x,y) < Ai, for some t £ 1} < +oo; b) 3A' > m(A) s.t the set D = {(x,y) : h(t,x,y) < A' for some t G / } C (C) x (C) = (D); where (£>)=compact set C R" c) the functions f(t,x) : I x (C) -> R', g(t,y) : I x {C) -> R', F = —Y(t, x,y) : I x (D) —> R verify the Artstein's-Andreev's conditions for the precompactness (/4/,/2/), we write f,g,F& PR; d) fit, x):Ix{C)^ R', g'(t, y) : I x (C) -> i?', F'(i, z, y) : / x (D) ->• /? are the correspondent limiting functions (Sell, Artstein, Andreev) with respect to a divergent sequence tm —> +oo; Definition 3.1.Given Z — Z(t, x,y) : I x R" —>• R, t G / ana" c' G R, we denote Z^{t,c') the set {(x,y)} so that / 2 / there exist three sequences tm ->• +oo, xm ->• x, ym ->• y for which lim Z[tm + t, xm,ym] = c'. m—t+oo
Theorem 3.1.-Suppose that two functions V — V(t,x,y), W = W(t, x, y) G CL, a constant c > 0 ezisf so that: a) V > 0, W > —c, K < 0, for every r > 0 two numbers r, b > 0 ezz'sf s i . : h = s and V < r implies W > b b)the lemmas 2.1 and 2.2 Zio/d, ana" also h —> 0 <£> K, W, J —)• 0. Tften / o r every /z G]0, r/(c + 6)] and A G [A', A"] the functions family Z(t, x, y) = V + A*(W + AJ) is useful for theorem 2.1 . Theorem 3.2. The previous theorem holds, put Z = —F, suppose that for every triplet (f',g',F') limit of {f,g,F), the set (F' = Q)r\Z^{t,c') contain not solutions of x = f'(t, x), y = g'(t, y) for Vc' > 0. Then the systems (1,2) are relatively (h',h)~ asympotically stable.
567
Proof.- The systems (1,2) are relatively (h',h)-stable. Suppose that [Vr > 0] [3s > 0] [3f G R+, 3x',y' G Rn : h'{t',x',y') < r], [3tm -» +oo] so that put [t'm = tm + t'] we have h[t'm,x(t'm),y(t'm)] > s where x(t) = x{t,t',x'), y{t) = y{t,t',y'). Let z(t) = Z[t,x{t),y(t)], since z < 0 3 lim z(t) = c" > 0, when c" = 0 we have the proof. Otherwise, suppose t-*+oo
c" > 0 and consider the following sequences of translates fm(t,x) = f(t+t'm,x), gm{t,y) =g{t+t'm,y), Fm(t,x,y) = F(t+t'm,x,y), xm{t) = x{t + t'm,t',x'), ym(t) = y(t + t'm,t',y') where T > 0 is arbitrary, t G [OjT], x and y G (C), we have xm{t) — fm{t,zm{t)), ym{t)=gm(t,ym(t)), xm{0) = x{t'm), ym(0) = y(t'm). Since the functions of the sequences {xm(t)}, {ym{t)} are equibounded and equicontinuous we can select (Ascoli-Arzela') five converging subsequences /,- -)• /'(«, x), 9i -» $'(*, x), Fi -> f'(t, i , y), t G [0, T], x, y G (C) ar,-(t) = » *'(<), y,(i) = » ?/(*) Vt G [0, T], (T > 0)
( = > uniformity).
1
We have ar'(t) = f'(t,x'(t)), y ' (t) = g'{t,y'(t)) i.e. x'(i), ?/(*) are solutions of limiting equations and also lim Z[t + t\, Xi(t),yi(t)] = c". i—>+oo
Obviously [x'(i), y'(i)] G Z^(t,c") with c" > 0. From Z < -F < 0 we obtain: , z(t+ti)-z(t<) < - fi;+t F[u,x{v),y(u)]du= - ft Fi[u,Xi(v),yi(u)]du < 0, c" - c" < - / F'[«, z'(u), j/(u)]du < 0. Jo
(5)
Hence F'[t,x'(t), (*)] = 0 V* G [0,7], (T > 0 arbitrary) it is a contradiction. Remark 3.1. Observe that we can use different strategies. Theorem 3.3. Suppose that theorems 2.4 and 3.2 hold and also: a) h G PR, for every triplet (/', g', /«") h'mii o/ (/, 5, h) the set /ioo -1 (f,0) contains only the solutions x = <j>{t), y = ip(t) of the limiting systems x = f'(t,x), y = g'(t,x) so that h"(0, 0(0), if>(0)) = 0. Then the systems (1.2) are relatively (h',h)-uniformly syrnptoiically stable. Proof. The systems (1.2) are relatively (ft', ft)-uniformly stable and relatively (h', ft)-attractive. Assume that: Vs > 0 3/ > 0 3tm G / 3xm and ym G R' : h'(tm,xm,ym) < s 3t'm -» +00 put t"m = tm + t'm xm(t) = x(t,tm,xm>) ym{t) - ym(t,tm,ym) we have h[t"m,xm(t"m),ym(t"m)] > I. If we consider the sequence of translates fm(t,x)
= f(t+t"m,x), £mm\t)
gm{t,y) =
Xm(t
=g(t+t"m,y),
-f- I
m
) , t/mm(')
hm(t,x,y) =
=
J/m (* "t" ' m J
h(t+t"m,x,y)
568
where t G [0,T(> 0)], x and y G (C), (£,2/) G (-D), we obtain %mm\t)
—- J v ^ i ^ m m ^ J J j «Emm \v) — Xm\t ymm\y)
~- ymmy1
m)i
m) ) ymm [J) ~- 9m \t > Vram v,J)
^m (U, ^ m m i ^ J , Vrnm \y) ) -^_ *• •
Since the set (C) is bounded, since xmm(t), ymm(i) are equicontinuous and, from the precompacness, we can select five converging subsequences fi{t,x)
-> f'{t,x)V{t,x)
G [0,T]x(C); gi{t,y)-+g'(t,y)V(t,y)
/»,-(<,x,y) -> h"[t,x,y)V(t,x,y)
G [0,T]x(C)
G [0,T] x (£>)
*,-,•(*) = > 0 ( 0 , »»(*) = > V*W V* G [0,T]. Where = > denote the uniform convergence, obviously
<j>(t) = f(t,
ft"[o^(o),V(o)]>/.
lim h[t + f _,-, Zj(i), yj{t)] = 0 we deduce [«£(<), ^(i)] G /ioo _1 (t,0).
Then we have a contradiction. 4.
Applications to the analytical mechanics
From the well known Lagrange's formulas [in Matrosov (1962) T = T2] d 8T 8T . dU 8R ,. x -r.-^^ — = fi+9isQs + -j. -z-r ( i , s = 1,2, ...,n) (6) at aqi oqi aqi oqi when T = T2+T1+T0 = ^arsqrqs+arqr + ao and ars = asr(q,t) we obtain the n differential equations (A/i = §f:,A/t — ^f-; i,r,s= l,2...,n) . . . 1 . . . . d(T0 + U) OR ,„. aisqs +aisqs +a,i - -arsiiqrqs -asnqs = fi+giSqs -\ -^-r (7) 5 2 ' ' aqi oqi can to describe a motion of a mechanical system with n degrees of freedom. Lemma 4 , 1 . Suppose that ar = ar(q,t), ars = asr(q,t), fi = fi(q,q,t), 2R = brsqrq's, brs = bsr(q,t), U = U(q,t), grs = -gSr(q,t). Select V = T2 — To — U we deduce that the sign of V can be variable. Assumption 4.1. Suppose that exists a non empty set T s.t. (q,q,t) G T implies 0 < V < p(=const.) and {q,q,t) (£ T —> V < 0. Lemma 4.2. Put J(q,q,t) — [§j-qi]2 we have J = 2qi(aisqs + ai)[ST2 + 27i + ( — - + /,- + dqt
—)qi Oqt
+ gisqiqs].
(8)
Oqi
Assumption 4.2. Suppose that there exists a non empty set T' so that (q,q,t) ET' —)• J < - V and when (q, q, t) £ T' => J < 0.
569 L e m m a 4 . 3 . Suppose that V C V put Z = V + uJ (p > 0) we obtain Z — V + pJ < p — ^v', in the sequel we select p' : plv' > p therefore Z <0. L e m m a 4 . 4 . Suppose that : i) Ti = \arsqrqs > ctiqf, p'J — To — U > faqi2 where a,- = a,-(i) > 0, /?,• = /?,-(*) > 0, 2i) h(q,q,t) = atq? + faq2 and h' = £h {£> 1). Then we obtain Z = V + p'J > h and h < h'. L e m m a 4.5. Suppose that for t —> + o o we have: dJE M1
^ ,
an,
% * , b4j, ( / < - * * ) , ( 2 o j + 5 y ? i ) - > 0 a n d a/so
a>i{fi H— Q 0+ ' ) —¥ c*ij where ajjg,gj is a definite
quadratic
form.
Then
Z — 0 implies ctijqiqj = 0 O ; = 0 solutions of limit equations . L e m m a 4.6. / / T o ( 0 , i ) + U(0,t) = 0 for t > 0 then £ ( 0 , 0 , * ) = 0 hence not exists solution of limit equations £ Z^(t,c) fl ( F ' = 0) for c > 0, then the system (6) is relatively (h', h)-asymptotically stable. T h e o r e m 4 . 1 . Given a second system relatively (ft'j, hi)-asymptotically stable and suppose that (h + hi) < ra(/i' + h'j) when ft' + h[ < lambda. Then the two systems are relatively (h' + h[,h + h\)asymptotically stable.
References 1. A.S. Andreev ,0n the asymptotic stability and instability of zeroth solution of an non-autonomous system, P M M , 4 8 , M2 (1984) 2. A.S. Andreev ,Sulla stabilita asintotica ed instability, Rend. Sem. Mat. Univ. Padova, 75 (1986) 3. A.S. Andreev and G. Zappala, On stability for perturbed differential equations, Le Matematiche, L I (1996) 4. Z. Artstein, Topological dynamics of an ordinary differential equation, J. of Differ. E q u a t , 2 3 (1977) 5. Z. Artstein, The limiting equations of non autonomous ordinary differential equations, J. of Differ. E q u a t , 2 5 (1977) 6. Z. Artstein , Uniform asymptotic stability via the limiting equations, J. of Differ. Equat, 2 7 (1977) 7. V. Lakshmikanthan, S. Leela, Differential and integral inequalities theory and applications, Accademic Press 1969 8. V. Matrosov ,On the stability of motion, P M M , 2 6 (1962) 9. N. Rouche, P. Habets P and M. Laloy, Stability theory by Liapunov Direct Method, Springer Verlag N.Y. Berlin 1977 10. L. Salvadori Famiglie ad un parametro di funzioni di Liapunov nello studio delta stabilita, Symposia M a t h . Accademic Press. London, 1971 11. G. Sell Non autonomous differential equations and topological dinamies, I- II Trans. Amer. M a t h . Soc, 1 2 7 (1967) 12. G. E. Silov Analisi Matematica, Funzioni di una variabile, ed. MIR. Mosca 1978.
www.worldscientific.com 5464 he