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!
/v+(l -e)
/w+(l + e)
jfhjmvnbjhgj where £^ and £;}" are denoted by \fvz moreover (^ > 1 and £g" > 1. 4. For any £ G [1, ^TWv
2
it holds that
(£ 3 + ^ + 3 £ - l )
1 +
4(£+l)
= ^ V v ° and x/^ 7 = ^J-y/vo respectively,
2
( 1 - g ) tf + i ) ( i + g) ' ( 1 + e)
4
(l-e)
(Z+l) 4
(1 + e) '(1-e)
> U
5. For any £ 6 [1, ' f f l + i h , it holds that 1 , (£3+£2 + 3 £ - l ) ( 1 - g ) 2+ 4(£ + l ) 2 ' ( 1 + e) 6. Denote ^
= ( ^ - ) 3 , # 1 ( 6 , 6 , 1 7 1 , ^ ) = 6 - 171(6 + 6 ) + 1711726,
#2(fi,6,i7i> m^i,S2)
= £i€i(l + »7i) + £2*716(1 + 92), it holds that
jhjfhugju ^ i ( ^ , € o " , % ^ o " ) + ^ 2 ( ^ , ^ 0 " , % ,»7o",?,e) < ( l - » ? o " % ) m a x { ^ , ^ }
jfhugjjfnvh hfgyhcgfh lklfkgjuhjythyjhdfhgfjhf R e m a r k 2.1.8. When £ and e tend to zero, the inequalities B^,B^ and Be hold automatically. £1 holds as well if r l T ( x ) = 0 and r 2q: (x) = 0, and the
48
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
Bn and B$ hold under condition A. This means that the condition B is a direct generalization of condition A. Moreover, it is easy to claim that for any fixed number 6, 0 < 6 < -i 23-1
there exist positive constant e, i and SQ such that all of the above inequalities in condition B hold for this pair of (e, e) if max{\Ar% vv+}
,
and osci; T (x)| + | r l T ( x ) | + |r 2 : f (x)| < £o (for x < 0 and x > 0 respectively). By using the initial data on x > 0 and x < 0 we solve the corresponding initial value problem for (2.3). Furthermore, by the similar argument as used in Lemma 2.1.4, it is not difficult to prove T h e o r e m 2.1.9. Suppose ( n ^ x ) , r 2:p (x)) are C 1 functions with bounded Clnorm such that q* > 0, then there exists a globally defined classical solution (ri+(t1x),r2+(t,x)) € Cl and (ri_(/, x), r 2 _(<,x)) G C 1 in the region # + and R- respectively such that In R- : r7„ < r 2 < supr 2 _(x) x<0
v1 € [£~",?~],
( v " ? r i r + U2) € [
In R+ : r £ < r 2 < supr 2 + (x) x>0 +
v* €[£ ,g+],
(v~*ri x + i;2) G [£+,tf+],
( v " ^ 2 r + i;2) £ [q+,<7+].
Where the subscript with the solution (ri, r 2 ) is omitted, rj„ and r£, are defined by r j # = — ln(supv_(x)) and r £ = — ln(supv+(x)) respectively. x<0
x>0
As same as the Riemann problem, the discontinuous initial value problem (2.3) (2.4) admits a unique discontinuous solution at least on a local domain R(S) in a class of piecewise smooth functions and this solution contains only a backward shock x = Xi(t) and a forward shock x = x2(*) passing through the origin. Moreover, the solution on the left side of x = xi(t) and on the right side of x = x 2 (t) is furnished by (ri_(*,x),r 2 _(tf,x)) and (ri+(tf,x),r 2 +(tf,x)) respectively. And, one is required to solve the same free boundary problem (FBP) on the angular domain R in order to construct the globally defined discontinuous solution for (2.3) (2.4) which contains only two shocks.
Frictional damping: Globally defined weak solutions...
49
Lemma 2.1.10. Suppose that the classical solution of (2.3) (2.13) and (2.14) exists in R(T), then it holds along x = xi(t) that r = ^ , - r — ^ ) {r + J—(n_) ) + —[r J) jfhgyjhfugjjhjghjjgh V
3
2x
r
r
^^^T-v^L^
lx
9
2
x
1 ■ AA ri ^ hdgtfhy r
g
v,yt^+t;-v^+3v/tiCt;--t;Vt; Ay/vvZ(y/vI + >/v)2
,(ri.)g * v^^
_.
jfhyghu while along # = X2{t) that
nchbfgy hfgyhf i ^+ v / JM:+f+v^+3Vt^:D-t)-yA
,(r 2 + )„
__,
where v_ = v_(t,ari(0),
ri_ = ri_(t,a?i(t)),
r 2 - = r 2 -(t,xi(t)),
w+ = t ; + ( ^ 2 W ) >
ri+ = ri + (<,x 2 (t)),
r 2 + = r2+(t,x2(t)),
etc..
Theorem 2.1.11. Under the condition B, the perturbed free boundary prob lem admits a global classical solution (ri, r 2 ) G C1 in R with a?,-(J) G C 2 (i = 1, 2) on which (2.15) holds. Furthermore, this solution possesses the following prop erties. For any (t, x) G R, r2* < r2 < supr 2 _(z), x<0
min<
-(supr 2 _(z)) < n < r j ,
min v*, (v~*r2x + t;i)(a,xi(a)) >
x 2 (r;<,r) V "
<
~"
y/v < max{g-,g+},
x<0
^
t;~£r2a. + t;£ < m a x <
max «*, (v~*r2x + v 5 )(a,xi(a)) > ,
x2(r,t,x) k <*
I )
50
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
min<
min v*, (y~*r\x + v*)(/3, X2{f3)) > x1(r;f,x)
k 0
<
)
v~?rix + t/2 < max< max v* , {v~*r\x + v*)(f3, x2(f3)) } , X!(r;<,x)
k 0
J
where the notation of x,(r; t, x), (/?, £2(/?)) and (a, #i(a)) is defined in the same way as in Theorem 2.1.1, r2*_is defined as the r2-value at the interaction point of ri + r2 = 0 and v — (max{g+,g~}) 2 , namely, r2* = — 21n(max{g+, g~}), and r\ is defined as the revalue at the intersection point of r2 = sup r 2 _(#) and the x<0
curve rm - rx - h2(r2R - r2) given in (2.10)2, where r2R = supr 2+ (a?),ri jR = x>0
f2R + 41ng+, namely, r* = supr 2 +(z) + 41ng+ — h2(sup r2+(x) — supr 2 _(z)). a;>0
x>0
x<0
Moreover, both shocks diminish exponentially, namely, there exist positive con stants A\, A2, As and A* such that
hdgyfht along x =
x\(t)\
jdhfyhfgtrhyfhugj along x =
x2(t).
Proof. Let us denote the following statement by H'2. {H'2) Along x = xi(t), 'it holds that
o< ^ + \ A < VtF(i + s)
-1 ^Tp^G[A4yl3]) hdgtfhcbgfh along a; = x 2(f), it holds that 0< ^ 1
"(>?-V5)
+ V? < V^+(1 + e) d(y^-y^)
*
€[
2
'
lL
where A,-(i = 1,2,3,4) are positive constants. We can prove the Theorem 2.1.11 by following the same framework used in Theorem 2.1.1.
Frictional damping: Globally defined weak solutions...
51
First, we can show by the same argument as in Lemma 2.1.4 that under condition Bi the classical solution defined in R(T) satisfies the estimates cited in Theorem 2.1.11 if (Hi) and (H'2) hold in R(T). It is similar to Lemma 2.1.6 to prove that (Hi) and (H2) hold locally in t. Then, it can be proved that the solution constructed by extension reserves (Hi) and (H' 2 ) in each step under condition B. Namely, if (Hi) and (H' 2 ) hold in R(To) where the classical solution of the free boundary problem is defined, then the solution defined in R(To + S) still satisfies (Hi) and (H'2), provided S is small enough. The arguments for the difference from the unperturbed case can be found in [78]. The method used for 7 = 1 can be used to investigate the case when 1 < 7 < 3. However, the result will be not as good if 7 is not very close to one. /2(7 — 1)
474
3—Y
For instance, the condition infy a > 4/ — sup a (a = v 4 ;) would Rm y 7 + 1 R{^yyy 3~7 make a much more strict restriction if 7 is not nearly 1. We omit the detail of the discussion which follows the same lines of arguement used to establish the result for the case when 7 is nearly one. Instead, we introduce a complete investigation for 1 < 7 < 3 in [79] by different methods. 2.2
Nonlinear diffusive phenomena for entropy weak solution
Consider the problem (2.3) (2.6) where v* > 0 and p(v) = av"1 with a > 0 , l < 7 < 3 , which is the state function for poly tropic gas. We compare the solution of (2.3) (2.6) with those of (1.4), namely [vt = -aP{v)xx {P(v)x = —<*U
(2.27)
in this section to show that the system (2.27) still model the time-asymptotic behavior of (2.3) even if shock waves may develop in the solution of (2.3) with rough initial data. Without loss of generality, we only deal with the case in which the two states (t/_,i;_) and (w+,v+) (we use the notation u^,v^ instead of u T , v T for convenience) are connected by a forward shock curve in the phase plane and the entropy condition is satisfied as well, namely
i ti+ - ti. = -(«+ - v-)yjfo+1ZPil~\
(2.28)
I v+ > t/_. It is known from [78] and [79] that if the initial shock is suitably weak, then the problem (2.3) (2.6) with (2.28) has a unique entropy weak solution globally
52
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
in time which is piece wise smooth and contains a forward shock x = X2{t)(0 < t < +00), satisfying the entropy condition, namely, it holds that
hdgftbcvhfgyhfg V VR >
VLy
where (uR,vR) = (u, v)(x2{t) + 0,0, ("L, VL) = (u,v)(x 2 (*) - 0 , t ) . Moreover, it is proved in [78] and [79] that (i) the solution (iz, v) is smooth in the domain Qi and Q2 up to the boundary respectively, where fti =:{(*,<) : * < x 2 ( < ) , < > 0 } , fi2=:{(x,*)
:x>x2(*),*>0};
(ii) in the domain fi2, (tx, v) is defined as (tz,t;) = {ti+(x,*),v + (;M)} = { w + e ~S v +}> and in the domain i i , (u, v) is defined as
(2.30)
(ti,v) = {ir(x,*),v~(x,*)} = {ti_e~',i;_}, where ii =: {(x,<) : a: < xi(t),t satisfying
> 0} and xi(tf) is the backward characteristic
/ ^ = A(._) = - V f ( t > - ) - * \ l « i ( 0 ) = 0;
(2.31)
(iii) along the shock curve x = x 2 (f), it holds (t>+ - v-)e~Alt
< v+ - v < (v+ - v_)e" A 2 t ,
*> 0
(2.32)
where v = t;(x2(f) — 0,*)>^4» is a positive constant (i = 1,2); (iv) t>-< v(x,*) < v+, f o r x G f l , * > 0 . (2.33) On the other hand, it is known from section 1.1 that (2.27) has a unique similarity solution v*(rj),ri = .x , satisfying v*(±oo) = v±, and V*(T;) is strictly monotone for T) £ R. The main result contained in this section, established by Hsiao and Luo in [74], states that there exists a unique x0 G R such that if S = \u+—tz_ | + |v+— v_ | is suitably small, then sup |ti(x, *) - u*(x + x 0 , *)l + sup |v(x, t) - v*(x + x 0 ,01 *€*
= 0(l)(* + l ) - i ,
r€*
as*->+oo
(2.34)
Frictional damping: Globally defined weak solutions... where v*{x + x0,t) = v* (j**\)
,t**(s+ *o,<) = =£p(v*(x+
x0,i))x.
53 This
shows that nonlinear diffusive phenomena also occur for entropy weak solutions of (2.3). To prove the result in (2.34), we first deduce certain decay estimates along the shock curve x = £2(^)5 which plays a key role for establishing the main result. Consider the system (2.3) with a — 1 and a = 1 for convenience, namely
{
v t - ux — 0 u
t + p{v)x = - t i ,
p(v) = v
1
(2.35) (1 < 7 < 3).
It is easy to observe that the characteristic speed of (2.35) takes the form -y.fi
-y + 1
A = —^/*jv~ 2 and fi = -y/~jv~ 2 respectively, corresponding to the backward and forward family. Introduce Riemann invariants (
2y^7
ri=u-
Tftjv
izJL 2
J
{ ^ r2 = u + jtrjv
(2.36) 2 .
Thus
jdhfyh l/ = I vr^ r2 " ri ^J and the system (2.35) can be written as (2.7) which is equivalent to (2.35) wherever the solution is smooth. By the notation introduced in section 2.1, (2.7) becomes
(*=i
(ri+rs)
(**)
lr 2 ' = -5(^1 +r 2 ). It is known from [78] and [79] that it holds along the shock curve x = x2(t)(t > 0) that
2
Wiv-
+]J
v+_+
I
jdhfgynvbh \
2V/7(«+ - «)(t>-T - »;7)
J
54
Quasilinear Hyperbolic Systems and Dissipative Mechanisms r
I z^
—7 1
(
\
\
2yJj(v+-v)(v-~t-v-~<)
(2.39)
J
where r\Xi r2x and v are taking values at (#2(2) — 0, / ) , and it is also shown that \r2*(x2(i)-0,i)\
- 0,<)| < Ce~plt,
t > 0, for some ft > 0.
To get the decay estimates of r2x along x = x2(t), lemma. Lemma 2.2.1.
(2.40)
we need the following
If S = \u+ — u_ | + |v + — v_ | is suitably small, then \r2xx(x,t)\
for x < x2{t),t
> 0,
particularly
1^2^(^2(0- 0,01 <>
^>0.
Proo/. The existence of all of the derivatives r\xx, r2xxi rixxxi rixxU r2xxX) v2xxU etc., in the domain fii = { ( x , t ) : x < x2{i)yt > 0} can be obtained, based on (2.7), by using difference quotients and taking limits. Let z = fil/2r2x
+ |,
for ( * , < ) € Oi,
where 9=
474 o
v
3 ^ 4 •
3-7 With the help of (2.7), it is easy to find that zt + V 7 * «
2
**
= " / ( * - W (*-7TT$), for (*,*)€«!, where / = ^ _ ^9~l- (the notation 2? and / are corresponding to z and / in section 2.1 respectively)
Frictional damping: Globally defined weak solutions...
55
Differentiating the above equation with respect to x, we arrive, by virtue of (2.36), (2.37) and (2.38), at
hdgfvcgbfhncbjv = fi(x,t) - / 21 - j ^ ~ | j g | zx+f2(x,t),
for {x,t) € «i,
where
fi(x,t) = -~h [? - ^4}ff2 + 2 ^ T V 2 ]
(M)
'
and
hdgfgbcvgfhbchbfgjcnbfjjshfd •y+1
Multiply the above equation with v~ 4 then , we get
<—->' - - {/ [» - &3}»] + ^ » - ' - } where Let
(2.41)
ncbhvnbfh
By the definition of z, f and g, it is easy to see that
K*,t) = \ + ^~^L(v-1r2r)(x,t), (x,t) e III. It is known from [78] and [79] that \r2x\ is small if S is small. This implies that
&(*,*)> J,
forOM)€«i,
(2.42)
foi{x,t)€Qi.
(2.43)
provided S is suitably small. We also know that |/30M)|
For any fixed (x,t) € ^ 1 , let x 2 (r;x,t)(0 < r < t) denote the forward characteristic passing through (#, t) such that x2(t] t, x) = x and x2(0; x,t) = f3. Let -y.fi
*(r) = v"
4
zx(x2{r;xJt),r)
56
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
where (x,t) £ ^ i is fixed. We still use the same notations b and fy to denote the corresponding values taken along X2{r;x,t) respectively, namely, b(r) — 6(a?2(r;a?,^),r),
and
h{r) = / 3 ( x 2 ( r ; x ^ ) , r ) . Then, (2.41) implies that ^
= -6(r)*(r) + / 3 ( r ) ,
0 < r < t,
(2.44)
where it is known from (2.43) that l/3(r)|
0
Multiplying (2.44) with exp
/ &(£)d£ and integrating the resulting equaUo J tion over [0,<], it turns from (2.42)-(2.44) then \k(r)\
0
particularly,
\k(t)\ < a Due to the arbitrarity of (x,t) 6 fli, we obtain I t T ^ ^ * , * ) ! < C,
for (x,t) e fii.
Lemma 2.2.1 follows then by virtue of the fact: V- < v(x,t) < v+, the uniform boundedness of ux and vx in Qi (see [78] and [79]) and the definition of z. The decay estimate of r^x along the shock curve x = x^ii) is given in the next lemma. Lemma 2.2.2. If S is suitably small, then it holds \r2x{x2(t) - 0,t)| < Ce-^\
t > 0,
for some fa > 0.
Proof, Let [M(v+)]1/2-j^-(S'2r2*+'f)(x2(t)-0,t),t>0, judhfghnvbhgj jfhhbchbfgbvh
jh(x2(t),t)
=:
and Y(t)
=: ft (ara (*),*),
where fi(v+) is the value of fi(v) at v = v+.
Frictional damping: Globally defined weak solutions...
57
It is easy to verify that d
-T- = f [»M1,2J^
- hi - nt)] K ) 1 " ^
_ x^s
- nt)]
+ M M 0 - 0,<) U?$l - fi(v(x2(t) - 0,<))1 (2.45) where g and / are taken values at v = v(x2(t) — 0,0Define
k1(t)=:fJt(v+)^-^±--±g(x2(t)-0,t),
jdhfbgvjh hdgvcgdfgch MO =: ylx{x2{t) - 0,t) [ ^ £ 1 - f*(v(x2(t) - 0,*))] • We can show (see [74] ) that |Ari(0| < Ce ^3*, t > 0, for some /?3 > 0, &2(0 > di,t > 0, for a positive constant a\ > 0, and 1*3(01 < Ce~P4t,t > 0, In terms of M 0 > M 0 ^ -
an<
for some /?4 > 0.
^ M 0 > (2-45) can be written as
= -fh(t)Y(t)
+ fh(0*2(0
+ M0,
*> 0
(2.46)
where / M 0 > ^0 > 0 (2 > 0),
for some positive constant do > 0,
and I / M 0 M 0 + M 0 1 < Ce-p5t(t
> 0) for some positive constant #> > 0.
Thus, it follows |Y(0| < Ce~p6t(t > 0) for some positive constant 0e > 0.
(2.47)
It is clear from (2.32) that
| ^ K ) 1 / 2 ^ - §(M0 - <M)| < Ce-**(t > o) for some positive constant /?7 > 0.
jdhfg
58
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
Lemma 2.2.2 follows then from (2.47), (2.48) and the definition of Y. Differentiating (2.39) with respect to t along x = #2(^)5 a n d using (2.7), we can find a relation between r\xx and r2xx. This, combined with Lemma 2.2.1 and the uniform boundedness of r\x, r2xi r\t and r2t, yields that \risx(x2(t)-0,t)\
i>0.
It follows from (2.29), (2.30)i, (2.32) and (2.37)i that \u{x2{t) - <M) - u{x2(t) + 0,*)| < Ce'^\
t > 0,
and p 4 - ^ ( * 2 ( t ) " 0,t) < Ce-p'\
t >0
for some positive constant /?' > 0. This, with (2.7), (2.40) and Lemma 2.2.2 yeilds | r H ( z 2 ( 0 - ° > 0 l + l r 2 t ( M 0 - M ) l < Ce-&\ t > 0, for some positive f3[ > 0. Hence ( M + |ti*| + k l + K D M * ) - 0,t) < Ce-fi*M,t
> 0,
for some /?2 > 0.
Similarly, we can obtain that (|t*«| + Wtx\ + 111**1 + \vtt\ + K | + \vxx\){x2(t)
- 0, t) < C,t > 0.
Differentiating (2.41) with respect to #, and using (2.42) and other estab lished estimates, we can show, with a similar approach as used for Lemma 2.2.1, that \Sxxx(x,t)\ < C, for (x,t) e QiLet fe(*2(*)-(M)=:
(v-^z^faW-Ott)
and
^)=:fcNt)-0,f) where z is defined in the proof of Lemma 2.2.1. It is not difficult to get, from (2.41), that ^
=
-b(x2(t)
- 0,0 • Z(t) + f3(x2(t)
- 0,t)
+y2*(*2(<) ~ <M)[i2(*) - ti(v(x2{t) - 0,t))], where the functions 6 and / 3 are defined in the proof of Lemma 2.2.1. By a similar argument as used in Lemma 2.2.2, we obtain \Z(t)\
t>0
Frictional damping: Globally defined weak solutions...
59
for some positive constant /3'3. It turns then \r2xx(x2{t) ~ 0,t)\ < Ce~P'*\t > 0 for some positive constant f3'4 > 0. (2.49) Due to (2.49), the relation between r\xx and r2xx along x = X2(t), mentioned before, and other obtained estimates, we can show that Vixx{x2{t) - 0,t)\ < Ce~^ 5 ',t > 0,
for some positive constant /3'5 > 0.
Thus, it is possible to get the exponential decay rates for other second order partial derivatives of rx and r 2 along x = x 2 W, w ^ h the help of (2.7). On the other hand, it holds along the backward characteristic curve x = x\ (t) that
jfhbvghf v(xi(t),t)
= V-,
u(xi(t),t)
= w_e _ t .
Moreover, it is easy to see from (2.30) and the higher smoothness of (u,v) in the domain I\ that all of vti vx, vxx, vxxx and ect. equal to zero along x = xi(t), while all of utl ux,uxx, uxxx and etc. decay exponentially fast as t —> -foo along x = xi(t). We now collect all of the estimates obtained so far in the following lemma. Lemma 2.2.3. Along the shock curve x = X2(t), it holds \u(x2{t) - 0,^)1 + Hx2{t) + | t l « | + \Vx\ + M
- 0,t) - t/+| + (\ux\ + \ut\ + \uxx\ + \uxt\
+ \VXX\ + |Vart| + \vtt\)(x2{t)
< Ce~P *,t > 0,
~ 0,t)
for some positive constant /?* > 0.
(2.50) A similar estimate along x = xi(t) can be expressed if we use V- and xi(t) instead of v+ and x 2 (t) — 0 in (2.50) respectively. We compare the solution («, v) of (2.3) and (2.6) with the similarity solution v* of (2.27)i and establish the main theorem next. It is known that there exists a unique similarity solution V*(T;), TJ = , x , v r "T" i
satisfying v*(=foo) = vT, for the equation (2.27)i, namely v*(rj) satisfies
([-pK(9))u + i^; = o, \t;* (Too) = « T . Moreover, t^(7?) > 0 for r\ £ R if v_ < v+, which implies v_ < v*(r;) < v+.
60
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
The following estimates on v* (77) = v* I
,x
) are needed in our analysis
(see [78] and [79] for detail). L e m m a 2.2.4. It holds for the similarity solution v*(r)),r) =
>x
, that
1^(1)1+1^)1+Kwi+{pgj:;:j i\\\
/ OO
/
\vx(x,t)\2dx
■OO
[\v;{x,t)\2 + \v*tx{x,t)\2)dx < C(v+ - v.)2(t + 1)-*,
-OO OO
/
\vxt(x,t)\2dx
< C(v+ - v.)2(t + I ) " 5 / 2 ,
■OO
\v*xxt(x,t)\2dx
r
< C(v+ - v-)2(t + I ) " * ,
J — OO OO
/
\v*xtt(x,t)\2dx
■OO
where v*(x,i) = v* [
,g
].
It follows from (2.29), (2.30)i and (2.32) that it holds Dit < x2(t) < D2t,
t > 0,
(2.51)
for some positive constants Di and D2. Take any smooth function m 0 (x) with suppm0(z)c[-l,l],
(2.52)
and / -'-oo
m0{x)dx =
m0(x)dx=l.
(2.53)
J-l
Due to (2.51) and (2.31), it is possible to choose Tx > 0 such that i-hl]C[x 1(T1)1x2(T1)l hdgvcfgdh
bcghfb (2.54)
Frictional damping: Globally defined weak solutions...
61
Define
hdgbchf ■*—-(ftfOH'Kftft))].r+oo
We can show that F(x0) =: - /
*/(x0,*)cft is well defined for x0 G # .
«/Ti
(see [74] for detail) Let G(x0)=
r
T l
J*i(Ti)
L x ^ - v * [
(^^)\dx-F(x0). \ y/Ti ) \
We can verify, by using (2.29), (2.30),(2.32) and the facts of v*(±oo) = v±, v_ < v+ and ti*(±oo) = 0, that lim
G(xo) = +oo and
x0-+ — oo
lim
G(XQ)
= — oo.
ar0->+oo
Furthermore, a careful calculation shows that G(XQ) is a monotone funtion of xo if T\ is chosen large enough. Thus, there exists a unique x~o € R such that G(x0) = 0.
(2.55)
Moreover, we can show that XQ does not depend on T\. Let m{x,t) = m0(x)h(t)
for any {x,t) G ft3 =: {*i(*) < * < *2(<)>* > Ti}> (2.56)
where +oo
/
v{*o,Z)dt
(2.57)
and let w(x,t) = v(x,t) - v* ( ^ = = f ) " m(x,t) z{x,t) = u(x,t) - u* ( ^ ~ f )
" £
rnt(t,t)d$
62
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
where (x,t) £ Q3. Thus, (2.3) and (2.27)i imply that Wt = z
r
*
{ zt + \p(vm + m + w)- p(v*)]g + z- p(v*)xt = -[ht{t) + M * ) ] J
™o«K> (2.58)
in &3. Introduce
f
w(£, t)dt,
(z, t ) €3f .l
(2.59)
It follows then
+ w(xi{t),t)ii(t).
- w(xi(*),*)xi(*). It turns that
¥>« + \p{v* + T71 + P*) - ?(«*)]* + ^ ~ P(t>*)a*
= -(M<) + M<)) T ™o(0# + $(*) " *(*))• Denote the right-hand side by A(x,t), namely A(x,t) =: -(A t (t) + A«(*)) ^
m 0 (Ode + (0t(*) " *(<))■
In view of lemma2.2.3, lemma2.2.4, (2.52), (2.54) and the uniform boundedness of Xi(t) and Xi(t)(i = 1, 2), we can show that \A(x,t)\ + IM*,*)| + |A*0M)| < Ce-M,
/% > 0.
Denote(t). We claim that tl>(t) = 0 for
t>Ti.
(2.60)
In fact, it follows from (2.52)-(2.54) and (2.55) that 1>(Ti) = 0. Furthermore, due to (2.52)-(2.54), (2.56)-(2.59) and (2.30), we can show that m )
- 0 t > T
Frictional damping: Globally defined weak solutions...
63
This, with V(7i) = 0 together, implies (2.60). The main result in the present section is that T h e o r e m 2.2.5. For the solution (tx, v) of (2.3) and (2.6) with (2.28), it holds that sup{|ti(a?,t) - u*(x + x0,t)\ + \v[x,t) - v*{x + x0,i)\} **R
KCil + t)'1'2,
(2.61)
t>Tu
where v*(x + x0,t) = v* ( j = | ) ,«*(* + «o^) = -p(w*(aJ + xo,<))*. Proof. It follows from lemma 2.2.4, (2.51) and (2.30) that < C\v+ - v-\e~C{
sup \v*(x + x0,t)-v(x,t)\
>/^ r ° ) < C(l + *)~ 1 / 2 , t > T i .
a7>ar2(*)
Similarly, sup |u(x,/) — u*(# -h x"o,^)| x>x2{t)
+
sup {|v(a?,t) - v * ( z + x 0 ,*)| + |ti(a:,t) - ti*(a? + x 0 ,t)|} a?
< c ( i + 0" 1 / 2 ,
*>ri.
Thus, to prove (2.61) it is sufficient to show that sup {\u(x,t)-u*{x+x0,t)\
+ \v{x,t)-v%x+x0it)tt
< C ( l + * ) - 1 / 2 , < > Ti.
xi(t)<x<ar 2 (t)
By virtue of the fact \h(t)\ < Ce _ / ? 9 t ,
for some positive constant /? 9
Xl
X
> 0,
lot
and \v* ( v) + ° ) - v_ < ce~^ , for some positive /310 > 0, we only need
I
\ v * +1 /
to show that
sup
I
[\
t>Tu
x1(t)<x<x2(t)
which can be established by the energy method, similar to the one used in Chapter 1 (see [74] for detail).
64 2.3
Quasilinear Hyperbolic Systems and Dissipative Mechanisms G l o b a l l y defined a d m i s s i b l e BV s o l u t i o n s We consider the existence of admissible BV solutions of the Cauchy problem
,2 62 v
r Vt - ux = o \ ut + p(v)x = -au,a
> 0
u(x, 0) = u0(x), v(x, 0) = v0(x), - o c < x < oo,
^'
'
(2.63)
where p(v) is a given smooth and decreasing function: p'{v) < 0 for v G (—00,00) instead of v £ (0,oo), as assumed in previous sections. The rea son is that v may denote specific volume (v > 0 then ) and p is the pressure in fluids, as discussed before, but v also can denote deformation gradient — strain in solids and (— p) is the stress. For example, the system (2.62) governs, in Lagrangian coordinates, the motion of one-dimensional elastic continua in teracting with media exerting frictional forces. Such motion includes the flow of an elastic fluid through a porous medium or in a pipeline, and the oscillation of an elastic string immersed in a fluid. Since the solution of (2.62) (2.63) will eventually develop singularities when p(v) is nonlinear, unless \UQ\ and \v'0\ are suitably small (see Nishida [152]), it is necessary to consider weak solutions. For p with at most one inflection point, the method of compensated compactness yields solutions in 1^, with p = 00 in the strain hardening case (see Dipara [34]) and p = 2 in the strain softening case (see Lin [107]). So far, BV solutions have been constructed only by a handful of people. By exploiting the special geometric properties of the shock curves, Luskin and Temple (see [135]) established the result for p = v~l (isothermal flow of an ideal gas). Dafermos (see [26]) obtained BV solutions on (—00, 00) x [0, 00) with small oscillation about some fixed equilibrium state (0, v) for p allowed to be any smooth, decreasing function. We describe this result in [26] next. For convenience, we take a = 2 and replace v by the new variable v — v, which we denote v again, so that the solutions will be taking values in a small ball centered at the origin in the (u,v) plane. The initial data (1/0,^0) will be Ll(—00,00) functions with small total variation over (—00,00). It is easy to find that the system (2.62) can not be handed directly by the method of Dafermos and Hsiao in [28] since the so-called diagonally dominant condition (see Hsiao and Li [68], Dafermos and Hsiao [28]) does not hold (see Hsiao [6Q] and Dafermos [26]). To overcome this obstacle, we perform the change of variable u = u+u (2.64) as used by Feireisl in [43], where dxu = v,
dtu = u.
(2.65)
Frictional damping: Globally defined weak solutions...
65
Thus, (2.62) and (2.63) can be reduced to the system dtv — dxu + v = 0 dtu + dxp(v) +u = uj
(2.66)
with initial data fio = u0 + wo,
uo(x) = /
v0(y)dy.
(2.67)
«/ — oo
The damping has been redistributed now, and it can be easily verified that the new system satisfies the diagonally dominant condition. Therefore, we can employ the algorithm in [28] which combines fractional stepping with the ran dom choice scheme. However, the price to be paid is the appearance of the nonlocal term LJ on the right-hand side of (2.65)2- Due to (2.64) and (2.65), it holds u{x,t) = e-tu0(x)+
f e-^-T^u{x,T)dr. (2.68) Jo In constructing solutions of (2.66) and (2.67), LJ can be determined, as a given source term, through (2.68) by the past history of the solution. In or der to secure that the dissipative action of friction overpowers the potentially destabilizing influence of this source term, an a priori bound is needed on the total variation of w(-,t) over (—00,00), uniformly with respect to t on [0,oo). With the help of an entropy estimate, the required bound can be obtained and the following theorem is established. We refer to [26] for the detail. Theorem 2.3.1. Assume w0, ^0 are in BV D L1 and set M =
TVu0+TVvo, oo
/
(\u0(x) + v0(x)\)dx. -00
There are positive constants M, /, a, 6,/i 0 , ^0 such that when M < M,I < I, there exists an admissible BV solution (u, v) of (2.62) and (2.63) on (-00, 00) x [0,00) with u(-,t) and w(-, t) in BV H L1 for any t e [0, 00) and TVxu(>,t) + TV9v{'>*) < ae~"otM + / i 0 / , oo
(|ti(*,<)|+Kaj,t)|)d*<6/. /
-OO
Furthermore, as M and / shrink to zero, I/Q T 1-
66 Quasilinear Hyperbolic Systems and Dissipative Mechanisms In study of the large time behavior of admissible BV solutions of (2.62), (2.63), it is of interest to find out whether the solution are still governed asymp totically by the corresponding system (2.27) as was proved in section 2.2. A related discussion has been carried by Marcati [137], and Marcati and Milani [138].
Chapter 3 Relaxation Many physical situations demand the consideration of relaxation effect. For instance, in the kinetic theory the relaxation time is the mean free path and in viscoelasticity the strength of memory. In gas dynamics the relaxation phenomenon occurs when the gas is in thermo-non-equilibrium. It also occurs in chromatography, river flow, traffic flow and etc. (see Whitham [181]). Consider the hyperbolic system of conservation laws with relaxation
( ut+ f(u,v)x
I
_L /
=0
\
yvt+g(u,v)g=
v*M-v ^^
where the second equation contains a relaxation mechanism with V*(u) as the equilibrium value for v and r(u) the relaxation time. We are interested in the zero relaxation limit r(u) —> 0. In many physical situations certain stabil ity criterion holds and the solution would converge to that of the equilibrium hyperbolic conservation law
ut+f{u,V*{u))x
= 0.
The zero relaxation limit reduces the number of equations and is highly singular. We are also interested in the stability of elementary waves for the equilibrium hyperbolic conservation laws, related to the corresponding hyperbolic system with relaxation. We are concerned with the question of stability of elementary waves and singular limits of zero relaxation time in this chapter. The former are investi gated in Section 3.1 and Section 3.2, corresponding to one-space dimensional case and multi-space dimensional case respectively. The later is discussed in Section 3.3. 3.1
The stability of shock profiles for a rate-type viscoelastic system with relaxation Consider the following rate-type viscoelastic system 67
68 Quasilinear Hyperbolic Systems and Dissipative Mechanisms
{
vt - ux = 0 ut+Px =0
(3.1)
hfgtdgyfhcbvgfh
-(P-PT«W) {p + Ev)t = where v and (—p) denote strain and stress, u is related to the particle velocity, E is a positive constant, called the dynamic Young's modulus, and r > 0 is a relaxation time. The system was proposed by Suliciu in [170] to approximate the system
jfhbcvhfb and studied numerically by Faciu, Pitman and Ni in [41] and [160] respectively. The system (3.1) is a kind of regularization of (3.2), which is quasilinear hy perbolic system of conservation laws if PR(V) is nonlinear and decreasing, even mixed-type if PR(V) is not monotone. It is because that the semi-linear system (3.1) is always hyperbolic and standard existence and uniqueness results apply for any fixed r > 0, and that the system (3.1) approaches (3.2) in the limit r -> 0 formally. It is proved by Hsiao and Luo in [75] that for any given shock wave (5; v~, T/~;t; + ,w + ) which corresponds to the reduced system (3.2) and satisfies the so called generalized entropy condition, the system (3.1) admits a smooth travel ling wave solution (v1u,p)(x,t)
= (v,%p){0,
€= T
with
lim tJ(^) = v^, f-j-T 00
lim w(f) = uT, C-fr-T00
lim p(£) = PR(VT) = p* A -P
Z-*T°°
no matter whether PR(V) is monotone or not. We call (v, u, p)(£) a shock profile. When PR(V) is monotone decreasing in a neighborhood of (v~, v + ) (if v~ < + v ) or (v+,v~) (if v+ < v~) and the given shock wave is suitably weak, the nonlinear stability of the corresponding shock profile for the system (3.1) is obtained by the authers in [75] under the condition /
[vo{x)-v(^]dx
= 0,
J
[ t t O ( * ) - t t ( J ) ] < f e = 0.
(3.3)
Namely, the solution of (3.1) with initial data (vo(x)1u0(x)Jp0(x))J which is a small perturbation of the shock profile (v,w,p)(f) with t = 0 and satisfies the restriction (3.3), exists globally and converges, in the i^-norm, to this travelling wave solution as t -> +oo. This will be explained in the first part of this section.
Relaxation 69 In the second part, we will introduce the linear stability of shock profile, without the restriction (3.3), which is investigated by Luo and Serre in [132]. To show the stability results, we first make assumptions on the function PR{V)-
H) There exists an open set N C R1 such that PR{V) is differentiate up to the third order in AT, and it holds —E\ < P'R(V) < —Ei, for v £ AT, for some positive constants E\ and E2 with Ei < E(i = 1,2) and PR{V) > 0 for v 6 N. Thus, the system (3.2) is hyperbolic and genuinely nonlinear for v G N. A discontinuity (5; v" , u~; v + , u+)(v^ £ N) in the weak solution of (3.2) is called a shock wave if A. The Rankine-Hugoniot condition is satisfied, namely S(v+ — v~) = — (u+ — u~) S{u+ - u~) = pR{v+)
-
PR{V~),
B. The entropy condition holds, namely, for any v between v~ and v + ,
_Pn{v)-Pn{v+)<s2<_Pniv)-PR{v-) V — V^
PR{V)
-
PR{V+
V —V
tf
V — V
)
> S2 >
PR(V) - PR(V~)
as = S2
V — v~
R e m a r k . The above entropy condition is called the Generalized Entropy con dition in [75]. Compared to the Generalized shock E condition, introduced for mixed type system of conservation laws by Hsiao in [63] [64] [65] and Hsiao and de Mottoni in [72], the Generalized Entropy condition used here is much stronger. The later is possibly satisfied only when both (v~,u~) and (v+,u+) are located in the hyperbolic region of the system (3.2). But, it is possible for the former to be satisfied when one of the state (v~, u~) and (v + , u+) is located in the elliptic region. To be definite, we assume S = S2 > 0. Thus, it holds v+ > v~ due to the entropy condition and the assumption H. A shock profile is a smooth travelling wave solution for (3.1), i.e., a solution in the form (v,¥,!>)(£),£ = x ~r , satisfying U(=foo) = v T , w(^oo) = tz^jp^oo) = PR^) =: p*
(3.4)
and the following system in which we assume r = 1 for simplicity, that is (-Sv'-u1 = 0 ) -Su' + ? = 0 [s&+Ev)'=p-pR{v).
(3.5)
70
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
It is proved in [75] that the system (3.1) admits a smooth travelling wave solution under the following subcharacteristic condition 0 < S < y/E
(3.6)
((3.6) becomes —E < S < 0 if we discuss the case S < 0). Moreover, it holds v' > 0
(3.7)
1^1 < C(v+ - tT), \v!\ < C{v+ - v-), \pf\ < C(v+ - tT),
(3.8)
and the same estimates as in (3.8) are true for the second and third derivatives of v, u and p, respectively. To get the stability result for this shock profile, we reset the problem on the moving coordinate (£,*),£ = x — St. Let {V,TI,xmt)
= {v{x,t) - ? ( £ ) , u ( M ) -u(Z),p{x,t)-p{Z))
(3.9)
where (v, u,p)(x,t) is the solution of (3.1) with initial data (vo, uo,po)(x). We rewrite the system (3.1) in the form
{
Vt — SJ7^ — /J^ = 0 P t - ^ + ^ = 0 (X + EV)t - S(x + EV)t = -x + (PR(V + V) -
(3.10) pR(v)).
Introducing
HZ,t)= I
V{y,t)dy,n{Z,t)= [
J — oo
Ji(y,t)dy,x(t,t)=x(U),
(3.11)
J—oo
we seek the solution (^,//,x)(£>*) with the property
("M),/i(-,*))e# 3 ,
x(-,t)eH2
for the following system
{
Vt - SV£ - fi£ = 0
/it-5//c+x = 0 (X + Evi)t - S(x + Evdt = - X + PR{V + i/$) - p*(i7).
(3.12)
The result of nonlinear stability for the above shock profile is T h e o r e m 3.1.1. Suppose the function PR(V) satisfies the hypothesis H and the initial data (vo, v>o,po)(x) satisfy (v0 - v, wo - u,p0 - p) e # 2 ,
Relaxation furthermore, the functions vo(x) =
(VQ - v)(y)dy and /io(x) = J — oo
71 (u0 -
J — oo
u){y)dy a r e w e ^ defined for x E R, with (^0,^0) € L2. Then, there exist suitably small positive constants So and 770, such that if (v+ — v~) < So and \\(v0 - v, uo - u,po - P)\\H* + ||(^o, A*O)||L* < *7o then the initial value problem for (3.1) has a unique globally defined smooth solution (v,u,p) which satisfies eH2,
(v-v,u-u,p-p)(,t)
t>0
and asymptotically converges to this travelly wave solution in the H 1 -norm, namely \\(v,u,p)(x,t)
— (v,u,p)(x
— Stf)||jji -» 0 as t -> +00.
First, we investigate the Cauchy problem for (3.12) with the initial data (uo - v)(y)dy,no(£)
MO = / J — oo
= /
{u0-u)(y)dy
J — 00
XO(0 = ( P O - P ) ( 0 in the Banach space X(0,T) = {(1/,/i.x) : (v,ii) € C ° ( 0 , T ; H 3 ) , X 6 C 0 ( 0 , T ; F 2 ) } , the norm for which is defined as sup (||(^/i)(*)||Ir. + llx(Oll5) 1/2 .
N(v,v,x,T)=
0
In the following, we assume a priorily that (*/, /z,x) is the smooth solution of (3.12) and ( I / , J I , X ) € X ( 0 , T ) .
To prove our main result, we need the following a priori estimates. Lemma 3.1.2. Suppose the conditions in Theorem 3.1.1 are satisfied, then there exist suitably small constants 771 and <5i such that if |v + — v~ \ < Si and N{v, //, x, T) < 771 for some T > 0, then it holds
N{V,p,X,T)2+
I \\{vi,lH,x)i;,t)\?H** JO
a
,it,x,0)^K*N'{0),
„ „, (6.16)
72
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
for (v, fi,x) £ ^ ( O J ^ ) where K > 1 is a positive constant which does not depend on T. To prove this lemma, we establish the following Lemmas 3.1.3-3.1.6. By Sobolev embedding theorem, Hk+1 <-+Ck, k > 0. Thus, if N N(v,ii,x,T) < Si then Mc*, H e * and |x|ci < 0(l)SXj hereafter we use the symbol 0(1) to denote a generic constant which does not depend on t. Lemma 3.1.3. Suppose 7V(^,/i,x,T) < 771, \v+ — v~ \ < S\ for suitably small 7/1 and Si, and the conditions in Theorem 3.1.1 are satisfied, then we have,
IM<)II2 + IIMOII2 + \\x(t)\? +
beJO
+ O(l)\v+-v-\
PRM?(Z,
T)d£dr
J-oo
f ™(x + Ev()2dl;dT,
[ JO
(3 14)
'
J-00
0
jghbvjg
PR(V)=PR(V+V)-PR(V).
Proof. We will work with /7, V and \ since ^ = JI, v^ = v and \ — XDefine $(z,v,y) =PR{V + Z) -pR(v) + Ez-y. It is obvious that $(0,v~,0) = 0, - T T - ( 0 , V " , 0 ) =P'R(V~)
+ E>
0 (due to the hypothesis^).
Then, by using the existence theorem of implicit function, we can show that there exists a positive constant €0 > 0 such that a smooth function z = h(y,v) is determined from 3>(2,tT, y) = 0 uniquely if \z\ < £ 0 , |y| < £o, \v — v~\ < e0Namely, for any given (y,v) with \y\ < So and \v — v~\ < €0, there exists a unique root z with \z\ < So for the equation 9(z,v,y)
= 0,
(3.15)
expressed by z = h(y,v). That is Pfi(^+/i(2/,v))-pi2(i7) + JE,/i(2/,v)-?/ = 0,
for
| y | < e o , | v - v - | < e 0 . (3.16)
Due to Uf > 0, it is known that \v(£) — v~\ < \v+ — v~\ < Si which implies 1^(0 — v~\ < 6o,£ £ R, if Si is suitably small.
Relaxation
73
Let y = PR(p) + EV. Choose 771 and S\ suitably small so that \PR(V) + EV\ < So if \V\ < 7/1.
In view of the fact that z = V satisfies (3.15) with y = PR(V) + E17, and the fact that |F| < £0 if 771 is suitably small, it turns that V — h{y,v), namely V = h{PR{V) + ET7,v).
(3.17)
For any w with \w\ < £0, we define rW
= / Jo
\pR(v) - pR(v + /i(C, v))]dC
= /
-PR(h(C,v))dC
Jo It is not difficult to calculate with (3.10) and to show that
{ ^ + ^
+ ¥>(x+ £*,*)} +[x + $£(x +
= { f Stf + f x 2 - W
tt,v)](x-PRm
+ S
(3-18) where we have made 771 suitably small such that it holds \\ + Ev\ < €0 when |x| < m and \V\ < 771, and | ^ = -PR(h(x +EV,v)). Define A(A) = -PR(h(X(x - Pfl(F)) + Pfl(F) + £F), tJ)). It is clear that A(l) = ^(x
+ EV,v)
A(0) = -PR(h((PR(p)
(3.19)
+ EV),v)).
Furthermore, it follows, with the help of (3.17) that A(0) = -PR{V) =PR{V) -PR(V
(3.20)
+ V).
By the mean value theorem, A(l) - A(0) = A'(ft)
for some 0 < ft < 1
(3.21)
where A'(ft) can be written as
A'(ft) = -p'R[v + Mft(x " ftW + p*iv) + Ev>v)\ •||(^i(x -
PRW)
¥m(x-PR(17)).
+ PRP)
+ EV,V)-(X-
PR{P))
(3.22)
74
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
We estimate § p now. Differentiating (3.16) with respect to y, one obtains
fy{y^)=
(3 23)
E + Mv\h{yx,v))
-
where Vi = 0i{x-
PR{»))
+
PR&)
+ Eu.
(3.24)
By a similar argument as used to obtain (3.17), it is easy to see h{0,v) = Q.
(3.25)
Thus, due to the continuity of h(y,v), \imh(y,v)
= 0.
(3.26)
y-+0
Therefore, it can be guaranteed that v + h(yi,v) S N, by choosing 771 and <5i suitably small, and it follows then from the hypothesis H that -Ex
<-E2,Q<Ex,E2<E.
(3.27)
This, combined with (3.19), (3.20), (3.21), (3.22)—(3.24), yields [X+$%(X+EV,V)]\X-PR(V)) jgmnmbjszhnv
=
(x-PR(V))2(l+m)
= (X-P fl (F)) 2 g + p , f l ( 4 % i | - ) }
jfhbcghf
>-ErSTT2(x-P«W)2Next, we estimate
Ev,v).
Pn(h(<;,v)) =pR(v + h(<;,v))-PR(v) (3.29) =
P H ( ^ + 02MC, V))/I(C,
v),
for some
0 < 02 < 1.
Similar to (3.23), we can show that 1 —( - i = Vy't'' £ + p'fl(t7+%,?))•
Since
hfgbchnvbhgjnvbhfy
jfhbchbf
Relaxation
75
if 12/| is small, and it holds A(C, v) = h((,v) - h(0}v) =
d h
^
v
\
for
some 0 < 0 3 < 1,
(3,31)
we arrive at
jnvbhfjnvhkjf provided \r\ is suitably small. This implies that if |x| < tyi, |F| < i/i, and It;"1" — v~ | < S\ then it holds ViX
+ EV,v) >
E 2{E
2E2)(X
+ Evf
where 771 and 8\ are suitably small. On the other hand, it follows from (3.32) and the definition of (p that + E?)2.
\
S^{x
+ EV,v)vv
It is known from (3.8) that
fa I < 0(\)\v+
-v'\.
By the definition of
It is easy to show by differentiating (3.16) with respect to v, that dh(C,v) ^p'R(v)-p'R(v + h(C,v)) dv E + p'R{v + h(C,v)) Thus,
(3.33)
76
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
Due to (3.30) and (3.31), it holds
IMC,*0l
\p'R(v)-p'R(v + h(C,v))\ < 0(1)|MC,")I < 0(1)|C|, if |C| is small. Then, it follows from the above estimates that ^^(x
+ EV^v^KOil^-v-lix-^Eu)2
(3.34)
if rji and-<$i are choosen suitally small. Integrating (3.18) over (—oo, +oo) x [0,tf](0 < t < T) and using the estimate obtained, (3.14) is proved. For getting further estimates, we rewrite (3.12) as Li(i/, fi) = vt - Sv^ - //£ = 0 ^{^
l*) = Xt- 5xc + Enx + x + A(v, u ^ = 0
where X = Snz -fit, A(v,iss) = - f p'R(v + evz)d0. Jo Thus, it is known from H that 0<E2
<EX
(3.35)
+
if \u^\ and \v — v~ | are suitably small. L e m m a 3.J..4. Suppose N ( J / , / / , X , T ) < rji and \v+ - v" \ < Si, if 771 and Si are suitably small, then we have ft
\W(t)\\h +
IMOIISP
r+00
+ Wx{t)\? + / / Jo
(X2 + v\ + £ + Svttt2m
T)dtdr
J-00
(4+4+xD^rKrfr,
0 < * < T.
Jo J-oo
Proof. By a similar approach as used by Kawashima and Matsumura in [92] and Xin in [185] respectively, we discuss the equation I/£I(I/, JI)
- A-lnL2{v,ii)
= 0.
(3.36)
Relaxation 77 The left-hand side can be reduced to [\v* + Ji4" V ~ A-lnx + is(A-1)iu2}t +\S(A~1)^
+ EA~lne
- A-'x2 + (E- S2)(A~1)^ ■ K
(3.37)
+(n*+%-}/<)+{-}{ where {• • -}f denotes the term which disappears after integration with respect to £ E R. In order to dominate x, we make a calculation on the following equation,
the left of which can be reduced to {\EA-^\
+ \A-lx2
+ Wh + (A-1 + $S(A-l)s -
^(A-l)t)x2
-(1 + $SE(A-1)! + f (A"1),)/!* + S ^ " 1 W « + {• • •}{. Hence, the combination of (3.36)+ (3.37) xA with a positive constant A yields Ft + [EA-1 - A(l + $SE{A~l)e + %{A-l)t\n\ + {(A - DA'1 + §(A-i)tS - §(A-i)t} +(E - SPXA-1)^ ■ H + (A-^ix +\E(A-1)sw€
x2 +
iS(A~W jfhbcg
+ SK - $/,)
+{■■■}! = <)
where ^2+^A-1tt2-A-1nx+ls(A-1)(^+^XEA-1fil+\\EA-1x2+\^^-
F=
It can be calculated that
,i dA-x = A' J p'h(v + Ovefadff + -^- ■!/«.
(3-39)
2
In view of H and (3.35), it holds o
Pfi(^ + Ov()dQ > —7 > 0 for some positive constant E1
ao > 0. (3.40)
78
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
This, combined with (3.39) and the fact of Sv^ > 0, yields
Taking A = ^(1 + E/E\),
it is known, due to E > E\, that
On the other hand,
"-■ * f Therefore, it holds that X-1>0J(EA'1-X)
> ^ - A d = f 6 ! >0. Ei
In view of (3.39) and (3.40), it is possible to choose rji so small that if \fi^\ < 771 and \i/tf\ < 771, then (A - l)A~l + ^(A*1)^
> 62 > 0 for some constant 62,
{EA-1 - A(l + ^SEiA'1)^
+ f {A-1)^
and
> 63 > 0 for some constant 63.
Thus, we are able to show, by integrating (3.38) over (—00,-foo) x (0,t) and using lemma 3.1.2, that IK*)II2 + IIM<)II2 + *J f\\x(r)\\2dT Jo tt
+ b3 f IKWIfrfr Jo
r+00
khjunbh
rt
r + OO
-XEiA-^ntuddtdr.
-
Relaxation 79 It is known from integration by parts,
Hi7_r(^)"'""H <0(i)m
/
Sv^dZdr + oWm
«/0 J — oo r>t r + o o
+0(1)7,! /
<
/
/
342)
(i/| + 4)d^r
Jo J—oo
(i/| + A«|)dedr.
It can be claimed, due to Young's inequality and lemma 3.1.3 that |
rt
r+OO
/ \J0
=
/
(S2 - E^A-1)^
J-oo
\LLJ
+ /
I
■ fx^drl |
E)
S
- —■•»" •*.«*! (S2-E)^—.^mdidr\
/ rt
r + OO
- ^ H i-oo +0(l)m
(3.43) /•£ /» + 00
5F
2d
^ ^
r+o 1
( )i
t,+
;
-' "iy0 y
^
r f r
j j v^dZdr + 0(1)7,1 J f nldtdr.
With the help of integration by parts and the fact /A_U
^
dA-1
dA-1 ._
)' = - 9 ^ • ^ = -^f(S^ + /»«)'
one obtains l y j(A-1)t^(x
- SusWdrl
< 0(l)m J J(4 + ^ 2 f + x 2 + / i f K d r ,
|//<-)'^H//{(^ + ^)^ + -»^ <0(l)mJJ
Svifl2
+ 0(1)7,! y J ( 4 + / ^ + „|)dedr. (3.44)
80
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
Finally,
< J J XEiA-^Snt < 0(l)(m
- x)t*sdSdT\
(3.45)
+ \v+ - v- \)J j b \ + X2)dtdT.
The estimates (3.41)-(3.45) imply that ft
ft
f+oo
IH*)I| 2 + I H < ) I | 2 + / (llx(r)|| 2 + | | ^ ( r ) | | 2 ) ^ + / JO
JO
ft
J-oo
Svytdt-dr
r+oo
/ JO
+0(1)1,1 /
/
vl{Z,T)dZdT
J-OO
f + OO
/
( 4 + fa + X |)(e, T)dZdT, J-oo
(3.46) provided 771 and \v+ — v | are suitably small. By virtue of Lemma 3.1.3, it holds ft
r+oo
/ / {x-PR(vt)?dZdT J J ° -°° , + oo ft <0(l)|t»+-t>-| / / (x + EvtfdtdT JO
(3-47) +
O(l)N(0).
J-oo
Due to H, it is easy to obtain E'vl<(PR(V())><E>v* if 771 and Si are suitably small. It follows that < 2[(X - PR(^))2
Eiv\ < (PRW
+
2 X
].
Therefore, (3.47) and (3.46) imply ft
/ Jo
f+OO
/
ft
^didr
< O(l) /
J—oo
Jo
f + OO
x2dtdr,
/ J—oo
+
if \v — v~ | is small. This, combined with (3.46) and lemma 3.1.3 together, yields ft
IM*)II2P + \W)\\h
f + OO
+ \\x(t)\\2 + ft
< O(l)JV2(0) + 0(l)m
/ Jo
(x 2 + 4 + Sven* + 4)(t
r)didr
Jo J-oo f+OO
/ J-oo
(^ + ^
+ x\m,r)didr,
0
Relaxation 81 Thus, the Lemma 3.1.4 is proved. The estimates on higher order derivatives, cited in the following Lemma 3.1.5 and Lemma 3.1.6, can be obtained by a similar approach which we refer to [75] for detail. Lemma 3.1.5. If rji and S\ are suitably small, then it holds that
ll/^WII2 + ll"«(*)ll2 + \\xdt)\\2 + [\\\X( (r)\\2 + ||^«(r)|| 2 + ||*«(r)||2)dr Jo
< O{l)N2(0) 0
II/*«€(')II2 + II/*«€(<)II2 + I|X«(*)||2 + A l k a l i 2 + IK«(r)||2 + Ilx«(r)||2)dr < O(l)iV2(0). JO
Thus, lemma 3.1.2 can be proved by combining lemmas 3.1.3 through 3.1.6. Now, we prove theorem 3.1.1. By a standard method, the following local existence result can be obtained. Suppose ^,0)eUV(£,0)e#3
and
x(£,0)<=H2,
then there exists a To > 0 such that the problem of (3.12) with initial data //(£, 0) = po, *(*, 0) = vo, x(£, 0) = xo
(3.48)
possesses a unique smooth solution (/i,^, x) o n [Oi^o]- Moreover, x(;t)eH2.
(v(;t),p(;t))eH3,
(3.49)
In fact, we may first consider the system (3.10) with initial data
(*,ftx)fc,o) = KiMe,Xo)(0 2
(3-50)
which belong to H . This is a Cauchy problem for a hyperbolic system and the well-known local (in time) existence theory (see Majda [136]) gives that these is a positive constant To, depending only on the upper bound of ||07,7T,x)(0)||jj2, such that the above problem has a unique solution (J7,/7,X) £ C°(0,To,H2) fl C^OjTcff 1 ). Furthermore, sup ||(I7,%x){t)\\H* 0
82
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
for some positive constant c depending only on E\,E2 p»>)(i = l , 2 , 3 ) L _ With the (^,//,x) obtained, we define
\v£,t)=
and the bounds of
f v(i,t)d£ + f{t)
Jo
|M*.*) = J ffi,t)dt + g(t) U(£,*) = x°te,t). Letting i/(£,0) = i*(0 = /
o
F(£,0)#, we get /(0) = /
J — oo
V{£,0)dt =
J — oo
i/0(0). Similarly,flr(O)= ^o(O). Let t f(t) = MO) + / ( ^ ( 0 , r) + jT(0, r))dr, Jo 0defined above, is a smooth solution of (3.12) and (3.48). To see (*/,//,x) € X(0,To), we use the first two equations in (3.12). The first equation gives "(*>*) = MO + / Jo
[SV{Z,T)
+ ]i{t,T)]dT,
for
* G [0,T0].
This representation immediately yields v G C°(0,T, H2) and sup ||i/(., r ) | | „ 2 < ||i/ 0 || 2 + t sup (|5|||F(., r ) | | H a + 0
||A*(-,
r)||jy a ).
0
It follows from ^ = 77 G C°(0, T, if 2 ) that v G C°(0, T, # 3 ) . Similarly, we can show that // G C°(0, T, # 3 ) and SU
P IMr)ll#2 < IMI jj2-ft
0
sup (|S'|||i/(r)||^2 + ||x( r )||i/ 2 )0
By combining the above two estimates and (3.49), we get sup (||A«(r)||H. + IKr)||„3 + ||x(r)||„ 2 ) 0
< C{1 + tMfioWH* + =
C(l+t)N2{0),
\\VO\\H>
+ IIXolM
0 < t < To
(3.51)
Relaxation
83
where C > 0 is a constant, independing on t. Since the smooth solution (77, /J, x) of (3.10) and (3.50) is unique, and ( ^ , /i^, x), defined from the smooth solution (i/, fi, x) of (3.12) and (3.48), satisfy (3.10) and (3.50), the smooth solution of (3.12) and (3.48) is unique. The local existence and uniqueness is finished. We now turn to the proof of Theorem 3.1.1. In view of (3.51), we can choose To such that W2(^,X,*)<2CAT2(0) 0 < t < T o . (3.52) We take = JV2(0) = ,» = min j J L j , . ^ } , ?
N*(v,n,X,0)
where rji and K > 1 are the constants cited in Lemma 3.1.2. Thus, (3.52) implies that N2{v,ii,x,t)<\rii
0 < * < T0.
Next, we take So — \v+ — v~\ = S\ (S\ is the constant cited in Lemma 3.1.2). It follows from Lemma 3.1.2 N2(v,ti,x,t)
< K2N2(0)
< j^rft
= \n\
0
To,
particularly,
IKr0),/i(To)||ir. + ||x(r0)||if. < \r,l By using the local existence theory, there exists a constant t > 0 so that the solution of (3.12) and (3.48) exists on [T0,T0 + i\ and sup
(||(i/,Ai)Wlllr. + l | x W l l 2 r O < ^ -
To
Therefore it follows by using Lemma 3.1.2 that (||(i/,/i)(*)l&. + ||x(<)H2r») < ^ 2 ^ 2 ( ° ) <
sup To
h i *
In particular,
\\(v,it)(T0 + i)\\H> + \\x(n + mh < \il Repeating the above procedure, we obtain the global existence of smooth solu tions.
84 Quasilinear Hyperbolic Systems and Dissipative Mechanisms Finally, we establish the delay result. Lemma 3.1.2 and the above analysis show that r+oo
/
(IK^,A*€)(*)lllr» + llxWllir»)*<+«>-
Jo
Thus, it follows from (3.12) that r+oo
J
r+oo I i
\\^(t)fdt <+oo, Jo
5 IM0H
I
a
* <+oo.
This implies IkeWII2 -* 0 as t -► +oo. Similarly, we can show that
IkcOOII2-^0 ^
*-»+°o.
Thus, \\^\\m -► 0 as t -> +oo. The same result holds for ||/if||#i and ||x||/p. This finishes the proof. Next, we turn to the discussion on the stability of shock profile under a generic perturbation without the restriction of (3.3). The stability of elementary wave for a 2 x 2 relaxation model has been proved by Liu (See [122]) In his paper, the corresponding equilibrium equation is a scalar equation of conservation law, thus, a generic perturbation of a shock profile produces only a translation. However, for the system (3.1) the corresponding equilibrium system (3.2) is a 2 x 2 system. And a generic perturbation of a shock profile will produce not only a translation but also some new waves. It is known that the second order expansion of (3.1) is similar to the ChapmanEnskog expansion for the Boltzman equations, which is a 2 x 2 system with viscosity. Based on this fact, some new waves with diffusive properties are constructed by Luo and Serre in [132] where the new waves are composed y m( y 1 *—*■ ) ri, Vt + 1 \ vt + 1 / which is required to carry the "excess mass". The second part has the form of
of two parts. The first part has the form of i _L I m i (
/
— ) A which carries the net zero "mass" where Ai and r\
are the eigenvalue and right eigenvector of (3.2) respectively. To be difinite, we call the first part as diffusive wave and the second part as high order correction. The idea to construct the above waves is motivated by the following timeasymptotic expansion. Suppose that the shock wave belongs to the second family of (3.2), and (v, u,p) is the shock profile. We may perform the following
Relaxation
85
time-asymptotic expansion for the solution (v, u,p) of (3.1), a
(x - Ai(* + \)\
,
6
/g-Aift + lA ,
Q1
6i
/'x-Al(t-^l)^ , / x - A i ( t - f i)A ,
The expansion of p has the same form. There is an extensive literature on the stability of viscous waves (see chapter 5). When dealing with the stability of shock profiles for the system with relax ation, the main difficulty is that the dissipation of relaxation is weaker than the viscosity. For simplicity, we discuss the linear stability only. The idea developed in [132] is basic to the study of nonlinear stability although the expression of the corresponding diffusive wave would be more complicated and a lot of techniques on the study of the high order corrections would be involved. The linearized system of (3.1) at the shock profile (t7,tl,p) is the following system, vt — ux = 0 tlt + Par = 0 (3.53)
{
(p + Ev)
= - p + PR{V) + PR(V)(V -
We consider (3.53) with the t initial data o{x)1u(x10)
= uo(x),p(x,0)
v).
= po{x)
(3.54)
v(x,0) = v
where
I
r°°
I
I f°°
I
/ (VQ(X) — v(x))dx\ < +oo, / (u0(x) - u(:c))cto < +oo. J — oo I \Jweak — oo so that the vectors I (v+— v", tz+ — Assume that the shock wave is sufficiently v~)T and (1, \/—PR(V~))T be decomposed as (I (v0(x)-v(x))dx\ I / (u0(x)-u(x))dx
are linearly independent, then the initial data can , v
J
v +
_v-\ '
(
!
w
FRK
\
hfgbch
;/
where the translation #o of the shock profile and S are determined uniquely. We decompose the solution as follows v{x,t) = v{x + x0,t) + j
a
rn{x,t) + j-^ymi(z,*) + vi(x,i)
u(x,t) = u(x + x0,t) +
. h'
m(x,t) + jz£jrni(x,t)
+ tii(x,t)
p(x,t) = p(x + x0,t) +
,c
m(x,t) + j^jmi{x,t)
+ pi{x,i)
(3.56)
86
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
with the hope that the new waves (the second and third terms of (3.56)1,2) with the base state (v~,t/~) have some diffusive properties, would move along the 1-characteristic direction and carry the "excess mass" <J(1, y/—PR{V~)). Then, m(x,t) is required to be in the form of
m(z) z
(5)
' -—vm—
and CO
/
m(z)dz = 6.
(3.58)
■00
Similarly, we let mi(xyt) = mi(z). A natural choice of (a, 6, c) is a = \,b=^-p'R{v-)
= -\l{v-)
=: - A i , c = pfe(tT) = -A?.
(3.59)
Inserting (3.56) into (3.53) and taking ai = l,ci = 0,6i = — 9A1, we arrive at
{
fit ~ Mia: = jfl^Aj" 1 ^ +
l)~1(zm)xx 1
uu+Pix = ~5a*(/ + l)" (zm)xx (pi + ^ 1 ) , = - p i +i^(tOvi + (* + l)-1/2\p'R(v) 1
1
~p'R(v-)](m
+ a'A^m.)
+ J(t + l)" ^™)* + a * ^ ^ , ]
(3.60) where a* = E + p'R(v~) and we have used the choice on mi and m such that mi = -X[lzm
(3.61)
a*m' + zm = 0.
(3.62)
and As a stability criteria in [122], the following subcharacteristic condition should be satisfied, namely E
+ PR(V) > 0
for
V" < v < v + ,
(3.63)
espcially, a*>0. The combination of (3.62) and (3.58) gives
m{z) =
vhexp(-&)-
<3-64>
mfvnjgn
Relaxation
87
This, with (3.61) together, implies \^1a*m,(z).
7711(2) =
Thus, oo
m
/
1(z)dz
=0
(3.65)
-00
which means that the high order correction carries zero mass. Due to (3.55) and the facts that oo
/
and
-00 oo
/
(v(x + xo) — v(x))dx — xo(v+ — v~) (u(x + xo) — u(x))dx = x0(u+ — w"),
■00
one has oo
/»00
/ Hence, (3.60) 1 and (3.60)2 imply that oo
/
/»oo
vi(;r,*)cfo = 0 and
/
u^x^dx
= 0 for all t > 0.
(3.66)
■00 J — oo With the help of the explicit expression of diffusive wave and the high order correction, obtained above, and the following estimates on the shock profile Lemma 3.1.7, the stability Theorem 3.1.8 can be proved by a similar approach
as used in [75]. Lemma 3.1.7. The shock profile (t7, w,p)(£), discussed in this section, satisfies 0
u,Po-p)
e H1 (R)
88
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
and oo
/
(1 + x2)[vQ(x)
- v(x + x0)]2dx
< +00
(1 + x2)[u0(x)
- u(x + x0)]2dx
< +00.
-00
/oo -00
Then, it holds oo
/
___
[W(x, t) - W(x - St + xQ)]2dx = 0 -00
lim
sup \W{x, t) - W(x - St + x0) \ = 0
*-++<*>
xeRi
for the smooth solution W(x,t)
=: (v,u,p)(x,t)
of (3.53) (3.54) and the shock
profile W =: (t7,w,p),provided the shock is weak, i.e. e is suitably small. Proof. To prove this theorem, we reset the problem on the moving coordinate (£, t) with £ = x -f xo — St and introduce
v(t,t)= I J — oo
vtic.tw, u(t,t)= I MC,t)d
Due to (3.66), it holds that V(±oo,t) = 0 and U(±oo,t) - 0. Let P{£,t) = p(^, t). We get the system for (V, U, P). The a priori estimates on (V, U, P) can be obtained by a similar approach as used in lemma 3.1.2, and is sufficient for the proof of Theorem 3.1.8. We refer the reader to [132] for the detail. The phenomena of relaxation with the Broad well model, which is also a 3 x 3 semilinear hyperbolic system and has served to understand the transition from a microscopic to a macroscopic description of gases in the kinetic the ory, have been investigated a lot. The relaxation time is the mean free path. We will describe it more in section 3.3. The limit where the mean free path approaches zero is known as the fluid dynamic limit which is understood for smooth solutions of the limit "Euler equations" (see Inoue and Nishida [84], Caflish and Papanicolaou [11]). Regarding the case of solutions with shocks, we refer the reader to Kawashima and Matsumura [92], and Matsumura [140] for the studies on stability of traveling wave solutions or rarefaction wave solutions respectively. The nonlinear stability o f elementary waves for relaxation approximation in a 2 x 2 system was established by Liu [122]. The case of 2 x 2 system with relaxation are considered also by Chern [16], Mascia and Natalini [139] to study the L1 -nonlinear stability.
Relaxation 3.2
89
The stability of planar rarefaction waves and shock fronts for the relaxation approximations of conservation laws in several dimensions To approximate the system of conservation laws in several space variables m
i= l
the following relaxation system was proposed by Jin and Xin in [89] from the numerical point of view, m
ut + X^')*; =0,
« € R", vf € R"
«=l
vit + AiUXt = - ^
V i
- ^ \
i=l,2,-..,m,
where A{ = a,/, a,- > 0 is a positive constant, i = 1, 2, • • •, m, / is a n x n unit matrix. In this section, we consider the simplest case-the scalar equation of con servation laws in several space variables and the corresponding relaxation sys tem. We also investigate the nonlinear asymptotic stability of planar rarefaction waves and shock fronts. Let us concentrate on the scalar equation of conservation laws in 2-D case for simplicity. The case of arbitrary space dimensions can be treated using the same arguments. Consider ut + f{u)x + g (u)y = 0, ueR1. (3.67) The corresponding relaxation approximation is ut + vix + v2y = 0 Vu
+ aiUx
=
.(vi
-/(»))
(3.68)
We assume that the flux functions are smooth enough. And that the equation (3.67) is genuinely nonlinear in the x-direction, i.e., for a fixed constant a > 0, it holds /"(«) > a. (3.69) The following subcharacteristic condition (SC), proposed in [89], plays an im portant role in the stability analysis A
90
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
where M C
Rl.
The initial data for (3.68) are given by 0(x, y), vi(x, y, 0) = v10(x, y), v2{x, y, 0) = v2o{x, y) u(x, y, 0) = u
(3.70)
satisfying lim
\\u0(xr)-u:*\\LoofRi)
= 0
xiimoo l ^ 1 0 ^ ' ' ) " /( tt=F )IU 00 (« 1 ) = ° lim
H^O^JO-^^JIU
00
(3-71)
1
^ ) = °
a:—>:foo
where w~ < w + and w^ are two constants such that u+ £ M and w~ £ M . A planar rarefaction wave (in x-direction) w r ( x , t ) is a solution of the fol lowing initial value problem / ut + [f(u)]x = 0, \ i * ( * , 0 ) = ti5(*)
*£flV>0
,„ 9v ^ ' ^
where UQ(#) satisfies lim Un(x) — vF x-^oo
and -j-ur0(x) > 0, a.e. . (3.73) ax Since any rarefaction wave of (3.72)i with the same end states is time asymp totically equivalent to each other, we study the smooth rarefaction wave of (3.72) with the initial data UQ(X) such that UQ(X) = u~, when x < —Aro, and UQ(X) = u + , when x > ko for some positive constant ATQ, moreover, ur0(x) £ C7{Rl) with bounded norm. Let 7 measure the initial perturbation 7 = \\M') - UI(-)\\HA{B?) + ||[vio(-) ~ /K(-))Lr||i^(i*2) + ||No(')]ylltf 3 (* 2 )The following stability theorem is established by Luo in [130] T h e o r e m 3 . 2 . 1 . Suppose that ur(x,t) is a smooth planar rarefaction wave solution of (3.72) and subcharacteristic condition (SC) is satisfied. Then there exists a constant 70 such that if \u+ - u~\ + 7 < 70 then the problem (3.68)-(3.70) has a global smooth solution (u(x,y, v1(x,y,t)J v2{x,y,t)) satisfying lim ||ti(-,*) - ti r (-,*)|| L oo( H3 ) = 0 lim|M-,<)-/(wr(-,*))||Lco(.R3)=0 lun ||t»2(-, t) - g(ur(; t))\\L-(R>)
= 0.
t),
Relaxation
91
For convenience, we assume r = 1 in (3.68). The proof of this theorem is based on a stability result of rarefaction waves for the system of conservation laws with relaxation term in one space dimension, given by Liu in [122], and an Z,2-energy estimate. Motivated by the above stability result, one constructs a planar solution (U(x,tf), vi(x,t)) of the problem
{
ut + vix = 0 vu + axux = -(wi - f{u))
(3.74)
ti(a:,0) = ti5(x),t;i( a r,0) = /(ti5(x)) for which it holds, due to the stability results in [122], that lim ||ti(. > <)-« r (-^)llL«(iit) = 0 lim HM-,*) - /(u r (.,*))|| L oo (jR1) = 0.
(3.75)
We compare the solution («, v\, v2){x1y1t) of (3.68) (3.70) with the function (w(x,t), vi(x,t),g(u(x,t))) for the proof of theorem 3.2.1. Let U{x,y,t) = u(x,y,t)-u(x,t) V\(*, y, t) = vx(x, t/,*) - vi(x, t) V2(x,y,t)
= v 2 (x,y,^) -5f(w(x,^)).
It is not difficult to obtain, with the help of (3.68) and (3.74), that Utt + [f{u + U) - f{u)]x - ait/** +g(u + U)y - a2Uyy + C/t = 0
(3.76)
with initial data t/(x, t/, 0) G # 4 (M 2 )
and
Ut[x, y, 0) € # 3 (M 2 ).
(3.77)
To make use of L2-energy method to the above equation, one should well un derstand the properties of iJ, such as the expansive property and the estimates of C m -norm. The former can be carried out by some ideas with comparision principle due to Natalini, Hanouzet and Kruzkov ([150], [54], [100]), while the later can be acquired by characteristic method. We state those in lemma 3.2.2 and lemma 3.2.3 and refer the reader to [130] for the detail. L e m m a 3.2.2. Suppose that the subcharacteristic condition (SC) holds and \u+ — u~ | is sutiably small, then the following properties hold for the smooth solution of (3.74) ux{x,t) > 0, xG i ^ / ^ O |ut(x,t)| <
y/aiux(x1t).
92
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
Lemma 3.2.3. If S = |u + — u~ | is small enough, then 7
i
■
xen-,ten-r j=1 \~T1— \ oxJ
1 6 \ ^°( )
where O(l) denotes a generic positive constant, independent of t. We solve the Cauchy problem for (3.76) in the space 1(0,T) = {U(x,y,t)
e C ° ( 0 , T ; # 4 ) , Ut G L 2 (0,T;tf 3 ),
UX)Uy e L 2 ( 0 , T ; # 3 ) } . Suppose that the solution U(x, y,t) belongs to X(0,T)
for some T > 0 and set
N(t)= sup (||C/(.3C)||^(^) + II^(-,C)||^(^))C€[0,t]
By using the energy method, with the help of Lemma 3.2.2 and 3.2.3, we arrive at the following basic a priori estimate Lemma 3.2.4. There exist positive constants SQ and So such that if N(T) < £ 0j S = \u+ — u~ | < Joj then it holds
ll^(-.0ll 2 + ll^(-,<)ll2 + ll^(-,*)ll 2 + ll^(-,0ll 2 + / {|l^(->C)||2 + ||M-.C)||2 + l|Uv(-,C)|| 2 K+/ / Jo
JO J-ooJ
/ uxU2dxdydC — oo
< 0(1){\\U(; 0)|| 2 + \\Ut(; 0)|| 2 + \\UX(; 0)|| 2 + \\Uy(; 0)|| 2 }
where || • || denotes the norm in L2(R2). By making use of the above basic estimate, one can establish energy esti mates on higher order derivatives of U. Lemma 3.2.5. If N(T) and S are suitably small, then
\M'MH*(&) + \\Ut(',t)U*(R>)+ I \\(u*,Uy,Ut){-A)\\H*(R*)dt Jo
\M-MH*(R>) + \\Ut(-,mH>{&) + J 2
\\(u*,uy,Ut)(-X)\\hwdC
0
Relaxation
93
The global existence of the unique smooth solution for (3.76) (3.77) and its large time behavior can be obtained then by using (3.78). In fact, by combining the standard theory of existence and uniqueness of local (in time) solution with the time-uniform estimates (3.78), we can extend the local solution for (3.76) (3.77) globally by usual continuity process and show that the estimate (3.78) holds for any time t > 0. Thus, it turns that Lemma 3.2.7. There exists positive constant 70 such that if \u+—u~ |+AT(0) < 70, then the initial value problem (3.76) (3.77) has a unique globally defined solution U G X(0,oo) satisfying SUP(||^(-,t)||if4(H2)-t-||^(-,t)||//3(/l2))
- roc + / Wx,Uy,Ut){.MH>{R*)dt Jo
(3.79) <0{l)N\V).
Now, we turn to the proof of Theorem 3.2.1. Due to (3.75), it suffices to show that lim ||t/(-,*)|| L oo (il2) = 0,
t — r OO
lim ||Vi(.,f)|| L oo ( * a) = 0,
T ~mr OO
and Hm||^(-,*)IU~(fl=) = 0.
(3.80)
For any fixed (x,y,t) € R2 x R+, it holds u2(x,y,t)
< \M;y,t)\\hm
+ HtM-.y.*)llL»(K»)
which implies \\U(;i)\\l-w
<
SU
P W(;y,t)\\h(V)+
yeR1
sup1 yeR
\\Us{;y,t)\\bw
Furthermore, it follows, by using Cauchy inequality, that ™p \\U(-,yMh(W)
yeR1
<\\U(-,i)\\\\Uy(-,t)\\
sup \\Ux(;y,t)\\h{Ri) < IIM^IIII^yM)ll-
yeR1
On the other hand, it is known from (3.79) that
r{ii(^.^v)(-.*)ii2+|^ii(^.^)(-.*)ii2|}'ft<+00 which implies
iim||(tf„cgMH 2 = o. t—)>oo
94
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
Therefore, (3.80)i is proved. With the same approach as above, it is easy to show that g('MLoo(R2h\\Uy(',t)\\LoolR2))^0 as *->oo. (\\U Due to (3.68) and (3.74), it is not difficult to get V1(x,y,t)
= e-tV1{x,y,0)+
f
C -<'-«[/(ti
(3.81)
+ U) - f(u) - axUx]{x, yX)d(
Jo
V2(x1y,t) = e-tV2(x,y,0)+
f e~^%(u
+ U) - g(u) - a2Uy -g{u)t]
Jo
(*,y,CKThis, with (3.81) and the fact | | ^ ( - , 0 I | L ~ -» 0
as
t -+ oo
(see [122]) together, yields (3.80),-, i = 2,3. Next, we turn to the study on the stability of planar shock fronts for (3.68), under the assumptions (3.69) and (SC). Give two constant states u_ and t/+, satisfying Rankine-Hugoniout condition and Lax-shock condition; i.e., «(«+-«_) = / ( « + ) - / ( « _ ) ,
(3.82)
u+<«_.
(3.83)
We consider the solution of (3.68) of the form {«, vi, v2} = {U, Vi, V2}{x — si), for which it holds, -sU' + V{ = 0, (3.84) -sV{ + aiU' = f(U) - Vi,
(3.85)
-»Vi = g{U) - V2,
(3.86)
and tf(±oo) = u ± J Vi(±oo) = /(ti±), 7 2 (±oo) = g(u±).
(3.87)
We call (U, VUV2) the planar shock profile of (3.68). Consider the initial data y>(x, y, 0) = u0(x, y), vi(x, y, 0) = v10(x, y),v2(x, y, 0) = v20(x, y), satisfying
Jl&o HW°^'') "
U
O\\L»(RI)
= 0,
^iimoo l | i ; i o ( x ' '> ~ /( ll± )IU-(« 1 ) = °»
(3.88) (3-89)
Relaxation ^Urn^ ||v2o(a?, •) - ^ ( ^ ± ) | | L O C ( / 1 I ) = 0.
95 (3.90)
For simplicity, we assume r = 1 in (3.68). The following stability result is given by Luo and Xin in [133]. T h e o r e m 3 . 2 . 8 . Suppose the subcharacteristic condition (SC) is satisfied. If v =: \u+ — U-\ is small enough, for any p > 1 there exists a 70 such that E
E
H<3 / /
x2)p[da(u(x,y,0)-U(x))}2dxdy<1(h
(l +
M<3 / /
(l +
x2)p[da(vi(x,y,0)-V1(x))]2dxdy<1(h
S|a|<2 / /
(l +
x2y[da(v2(x,y,0)-V2(x))]2dxdy
then the problem (3.68) with the initial data (wo, ^io, ^20) has a unique global smooth solution ( t i ( x , y , t ) , v\(x,y,t), V2(x, y , t ) ) , satisfying sup
|t/(x,y,t) — U(x — st)\ -> 0
(ar,y)Gl2
sup
|vi(x,y,<) - V i ( x - stf)| - » 0
(r,y)€M2
sup
| v 2 ( x , y , t ) - V-^x - st)\ ->• 0
(x,y)GK 2 a s t —»■ 0 0 .
The proof of Theorem 3.2.8 is based on the L2 energy estimates. It is clear that we can not write the perturbation as u — U =Xi with E L2, unless a shift S(y,t) is introduced. Now oo
/
roo
(u(x,y,t)-U(x
+ S))dx = /
{u(x,y,t)-U(x))dx
J — OO
■ 00
so it is possible to choose S such that oo
{u(x, y, t)-U{x
/
+ S))dx = 0
-OO
for all y, t. The programme is to decompose the solution as u(x9 y, t) = U{x + 5{y, t)) +
+
6-{u+-u-),
96
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
In the case of g — 0 and 5 = 0, the estimates are much simpler. For the general case, we need a preliminary normalization. Choose f = t,x' = x — st,y' = y -h k(x — st) — at and let f\{u) = f(u) — su, g2(u)
= gi{u) - au, where Ar = -fffi(0]],
9\{u)
= 9(u) + Ar/i(u),
g
"u I « " " • * n ^ n e s e n e w coordinates, (3.68) becomes (hereafter, we still use (t,x,y) to denote (tf,xf,y') for the simplicity of notation) ut + vlx + t>2y = 0 i>it + ^iU;r - 2s£i a: 4- &iUt/ - hhy V2t + hv>x — hhx
- sv2x + A2uy
(3.91)
— sv2y = / l ( ^ ) - vi
- 2b2v2y — g2(u) — v 2 ,
(3.92) (3.93)
where A\ = ai — s 2 &i = Ar(ai — s 2 ) — scr 62 = sAr + cr A2 — a2-\- h[a\ — s2) — a2 — 2sAr<7" 1)1 = vi — sti, 1)2 = kvi — ksu + v2 — au. Let
Vi(x) =
Vi(x)-8U(x)
V2(x) = Wi(x) - ksU(x) + V2(x) -
where (I7(x), Vi(#), V2(a?)) is the shock profile. Then
*U(x),
ax
(ai s2) = /l(CA)_
~ S
^
After this normalization, the functions /1 and g2 have the following estimates, which enable us to handle the transverse wave. L e m m a 3.2.9 \fi(U(x))\
< 0 ( l ) ( t i _ - ti+)
I^W*))! < 0(1)(«_ - «+)2(«_ - £/(*))
M^(*))|
Relaxation
97
The proof can be founded in [133]. Now, we decompose the solution as u = U(x + 5(y,t)) vi = Vi(* + S(y,t))
+
(px(x,y,t)}
+ fa{x, y,t) = h(u±) + fa,
h = V2{x + 6(y,t)) +
fa(x,y,t).
Hence, (<£>, fa, fa) satisfies the following equations2y = 0, V>lt - 2sfax - b2fay - S1p2y + t l W ^ y - sVjj^y +4iy>**+fciy?ary = / l ( ( 7 + <£>*) ~ fl(U) -fa, and ^2t - *^2* ~ 2b2fay ~ b2fax + V^t + ^ 2 ^ y - 262K>'Jy +&l *) - 02(tf) ~ ^2By a straightforward calculation, we arrive at
/o Q 4 \ V'**>
where F i = 2sSt + 26xJy 4- 2b2SySt - 6? - A282yy F2 = 2b2Syt — St — Stt 4- A2Syy. By writting # 2 (J7 4-
= 92(U)x8y
+ (9'2(U)
To integrate (3.94) with respect to x, with
+ 6(y,t))U'(x
+ 6(y,t))
= a(y,t)U'(x
and r
+ S(y,i)) + W,
with V and W in L 2 . Furthermore we require ay + /?y - F 2 = 0,
+ 6(y,t))
+
Vx(x.y,t),
98
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
which implies Stt - A2Syy + St - 2b2Syt + OCy + 0y = 0. Then we integrate (3.94) over (—00, x) to get
(3.95)
,
+Vy +/3U'5y + Wy +
_
-
.
It is also known that ¥>(*.», 0) = /
[uo(z, ff) - tf (* + S(y, 0)]dz,
J—oo
andl + (U- U-)8t + V^y + /
V>2y<^ = 0.
«/ —00
Thus, with the help of the formula on <$(y, 0) and 8t(y, 0), it follows that M*> y, 0) = -tf(*, y, 0) - [tf(a? + <S(y, 0)) - u-ft(y.O) - V2{x, y, Q)8y(y, 0) -/!oo^2y(*,y,0)
Relaxation
99
The next step is to perform- the energy and the weighted energy methods to get a time uniform estimate with which the stability theorem can be proved. Set N(t) = su P o < r < JIM-, r)||2 + | M - , r)\\l + | M - , r)\\2 + | | ^ ( - , r)\\l
+ll*(^)III+ll*(-,r)||a + sup 0 < r < t / fR2 \x\((pl +0 A2 - a2 + Ar2(ai - s2) - a2 - 2ska > 0 (k(a1-82)-8 +ooJt JR1 The study of the nonlinear stability of waves for a strictly hyperbolic system is still far from being complete. There are open problems even for the gas dy namics system. For instance, the stability of strong shocks, the decay rate in the stability of rarefaction waves, the study on systems for which the characteristic families lose convexity etc. C. Nonstrictly hyperbolic system (5.2) Many mathematical models in mechanics and fluid mechanics are hyperbolic conservation laws with parabolic degeneracy in certain parts of the state space (see Keyfitz and Kranzer [99] and Brio [8] for models in elasticity ; Wu [182] and Brio-Wu [9] for magnetohydrodynamics; Isaacson [85] and Temple [175] for multiphase flow, etc.). For example, let us consider the simple models ut + auux + bvx = 0 (5.54) vt + vvx = 0 where a and b are positive constants. This system has characteristics Ai = au, A2 = v, for which the corresponding characteristic directions take the form n = (1,0)' and r 2 = ( l , au j~ v) respectively. Thus, (5.54) is strictly hyper bolic for v ^ au and is parabolically degenerate along v = au. Because of the degeneracy, new wave phenomena occur. In general, non strictly hyperbolic systems may exhibit not only classical Lax shock waves, but also overcompressive, marginaly overcompressive, marginaly undercompressive and undercompresssive shock waves. Therefore, it would be more interesting to understand the differences and similarities between hyperbolic and viscous systems in terms of shock behavior. Introduce the corresponding viscous system for (5.54) as ut + auux + bvx = uxx (5.55) Vt +vvx = vxx. Liu and Xin carry out their analysis on (5.54)-(5.55) in [125] to study the behavior of the corresponding viscous waves, in particular the manner in which these waves are stable. Shock waves for the hyperbolic system (5.54) correspond to travelling waves, viscous shock waves, (u, v)(x,t) = (>, if>)(x - at), for the viscous system (5.55),
bl =
We give some estimates on a, /?, V and W in the next lemma, which enable us to handle the transverse waves. The proof of this Lemma can be found in [133]. L e m m a 3.2.11.
a2<0{l){u--u+)j\U'\
V2dxdy<0{l){u--u+)
P
//
W2dxdy <0(l)sup\r\2sup
-
f f
(fr
\x\
tp2xdxdy
here v = u~ — u+. With these preminary estimates, we can use energy and weighted energy methods to get the following time-uniform estimate. P r o p o s i t i o n 3.2.12. If v and v~2N(T)
are sufficiently small, then it holds
100 Quasilinear Hyperbolic Systems and Dissipative Mechanisms that
IM-,0111 + IM-,011! + IM-,*)lli +11^(^)111 +II*MII2 + II**MII2 + Ik* \x\{lv + ¥&)(*> V, i)dxdy
+ /o (irT'VII 2 + ii^m + ||^||i + IMH + ||^||l)rfr
(3.96)
+ /o(II^Hi + \\St\\l)dr
0
We prove Theorem 3.2.8 with the help of time-uniform estimate (3.96). In fact, if the initial perturbation N(0) is small enough, then (3.96) is true a little longer and hence forever. The smallness of N(0) can be insured by the assumptions in Theorem 3.2.8. According to the embedding theorem H3{R2) C C^R2), H2{R2) C C°{R2) and H^R1) C C2(R% we have y>x(.,t) G C\R2), ?**(•,<) € C°.(R2), 6(-,t) G C2{Rl) and J t (.,*) G C 1 ^ 1 ) for all t > 0. Then u(-,t) G C 1 ^ 2 ) , ti t (-,t) G
C°(tf2).
Next, we turn to prove the asymptotic stability. Due to / Jo
/ /
< +oo,
and
lo+0° \& nR*(pl(x>y>t)dxdy\
J Jl2
(pl(x.y,t)dxdy->0
as t —> -f oo. Similarly, we have
J
J2(Q,
as t —>■ + oo. On the other hand, su
P(*,y)ei 2 IM*>2/>*)| 2
< HkM
+ rfy)(*,y,*)^y]1/2[//Ea(^ + ^Ly)(*,y,0^]1/2,
Relaxation
101
in virtue of
=
fl002
< 2 [ / _ + ~
^x{x,y,t)dx]^
< [//«»(^ +^tf)(*,y,<)d*rf»]1/WKa(^, + ^)(«,y,*)d^»] 1 / 2 Since ff^i^lx then
+
is bounded uniformly with respect to 2,
sup
\
->0,
(*,y)GK 2
as t —» + 0 0 . The same argument leads to
sup (\
(ar,y)€R 3
+ \vxy{x,y>t)\)
-*0
as ^ -> + 0 0 . With the same argument as above, we arrive at sup | % , t ) | - » 0 , yGM 1
as £ —» + 0 0 . Because ipx(x,y,t)
= u{x,y,t) sup
— U(x +
8(y,t)),
|ti(jc,y,f)-{7(a:)| -► 0
(3.97)
(a:,y)GlR2
as t —>• -foo. Recall the transformation of the coordinates. (3.97) gives, in our original coordinates, that ^
sup
\u(x,y,t)-U{x-st)\->0
(3.98)
as t —> + 0 0 . Similarly, it turns that sup
\ux{x, y, t) - U'{x - st)\ -+ 0,
(3.99)
(a:,y)Gffi2
sup (x,y)GM 2
as 2 —» + 0 0 .
\uy{x,y,t)\^0
(3.100)
102 Quasilinear Hyperbolic Systems and Dissipative Mechanisms Let p0(x, y, t) = u(x, y, t) - U{x - st) pi(x,y,t)
=
v1(x,y,t)-Vi(x-st)
p2(x,y,t)
=
v2(x,y,t)-V2(x-st).
It is easy to get POt +Plx +P2y = 0,
Pit + axpx = f(U + p0) - f(U) - pi, P2t + a2py = g{U + p 0 ) - ^(^) ~ P2Hence ^ ( x ^ ^ j ^ e - V i ^ ^ ^ J + r e - ^ ^ i f / ^ + P o J - Z ^ - a x p o , } ^ . (3.101) By using (3.98)-(3.100), this implies s u p ^ ^ a |j9i(x, y, <)| -» 0 as t -> oo. The same argument leads to sup,^ y \ GH2 |/>2(#>2/>OI ~^ 0 as t —>- oo, which completes the proof of Theorem 3.2.8. The nonlinear stability of elementary waves is an important subject and some progresses have been made for relaxation approximation in one space di mensional case as discussed in Section 3.1. However, few results are known on the stability of elementary waves for relaxation approximation in multidimen sional case. The results introduced in this section have made a good progress. 3.3
The zero relaxation limit Consider the hyperbolic system of conservation laws with relaxation
f ut + f(u, v)x = 0 \
_L /
\
V*(*)-v
(3.102)
We are interested in the zero relaxation limit r(u) -> 0. In many physical situa tions certain stability criterion holds and no oscillation is expected to develop in the limit. And the solution would converge to that of the equilibrium hyperbolic conservation law tit+/(ti,V;(ti)) x = 0. (3.103) A stability criterion for (3.102), under which (3.103) is expected to be the limiting equation, is obtained by Liu in [122] through asymptotic expansion of the Chapman-Enskog type in the kinetic theory. The first order expansion of (3.102) is (3.103) which is a scalar conservation law with the equilibrium characteristic speed A*(u) = -4-f{u, K.(w)).
Relaxation
103
For the second order approximation, we set v\ to be the deviation from equilibrium and obtain from the equation (3.102)2 that v = K (u) + Vd vd =
r(u)(vt+g(u,v)x)
=
T(u)(V*(u)t+g(u,V*(u))x)
where we have made a priori hypothesis that Vd is small and its gradients are even smaller due to the dissipation induced by relaxation. By virtue of dt + K{u)dx = 0, it follows that vd S T(U)[-\*(U)V;(U)
+ gu(u, V*(u)) + V:(u)gv(u,
V.(u))]ux.
Pluging this into the first equation in (3.102), we arrive at the second order viscous approximation to (3.102) u>t + / ( t i , V*(u))x =
[f3{u)ux]x.
The viscosity f3(u) is related to the equilibrium speed A* and the characteristic speeds Ai, A2 for (3.102), as follows P(u) = r(ti)[A,(ti) - Ai(ti, K(ti))][A 2 (ti, Vi(ti)) - A*(ti)]. The stability criterion is that the viscosity f3(u) is positive, namely, the equi librium speed A* is subcharacteristic with respect to the frozen speeds Ai and A2: Ai
1 (*-/(«)), + ^ M H l
=
0)/(0) = 0, / <>0,r>0,
( 31 ° 5 )
(3.104) holds when 0 < \i < 1 > / ' M > 0. In fact, (3.105) can be rewritten as <(ut h
+ (v =- 0v x (/i +- f(u)) l)f(u)
[vt = ±£blj^Jf
_
,, .
.
-
7 - / ( w ) , / ih > 0 , r > 0b , g,v = ( c
for which the corresponding equilibrium equation is
(3.106) f jghbvjh
u , + (/*/(«)), (3.107) It is easy to see that the characteristic speeds =Ai0.and A2 of (3.106) take the form Ai = 0,
A2 = / ' ( « )
104
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
while the equibibrium speed A* = fif'(u).
Thus, the basic assumptions
0
/ » > 0
(3.108)
yield the stability criterion (3.104). It is proved by Chen and Liu in [15], for the stable case when (3.108) holds, that no oscillation is expected to develop for the system (3.106) in the zero relaxation limit and that solutions of (3.106) converge to those of (3.107). Consider the Cauchy problem for (3.106) with (u,v)(x,0) = (u0(x)1vo(x)) which satisfies 0 < v0(x) < f(u0(x))
< Mo < oo
(wo — u, vo — v) £ L°° fl L2(—oo, oo), for some equilibrium state (w, v) = (u, (/J, — The result in [15] states that
(3.109)
l)f(u)).
T h e o r e m 3 . 3 . 1 . The solutions (uT,vT) of (3.106) with initial data satisfying (3.109) converge to bounded measurable functions (t/, v)(x,t), a.e. in t > 0 as r —y 0. Moreover, u = u(x,i) is an admissible weak solution of the Cauchy problem of the equilibrium equation (3.107) with initial data u(x,0) = uo(x), and v(x,t) equals to the equilibrium value (fi — l)f(u(x,t)) for any t > 0. To prove the theorem, we first show the existence of L°° solutions of the Cauchy problem for (3.106). We construct the solution by the zero dissipation limit of solutions of the corresponding viscous system
{
u
t + (f(u) + v)x = euxx
vt+v-b(u,v){x,0)
WW =
(u<0(x)iv'0(x)).
L e m m a 3 . 3 . 2 . Suppose that (UQ(X),VO(X)) some smooth L2 approximation of (u0lvo), (u£,v£)(x,t) satisfying 0 < ve(x,t)
(3.110)
= £Vxx
satisfies (3.109). Let (u£0,v£0) be then (3.110) has global solutions
< f{u£(x,t))
<M
\\ve-{p-1)f{u<)\\L,
(3.111)
V^IIK,^),||L 2 <M, for some M independent of e and r . Moreover, there exists a sequence Sk -> 0, such that (u£k,v£k)(x,t)
-)■ (u,v)(x,t),
a.e.
Relaxation
105
and (u,v)(x,t) is a global weak solution for the Cauchy problem of (3.106) which has the same bound as the initial data in (3.109), 0 < v(x,t)
< f(u{x,t))
< M0
(3.112)
and satisfies the entropy condition r)(u, v)t + q{u, v)x + 77(tz, v)v V~W-
*) / W <
(3.113)
0?
T
and the estimate on the deviation from equilibrium \\v-(ii-l)f(u)\\L,
(3.114)
for some M independent of r where (77, q) is entropy pairs for (3.106). Proof. The initial data (WQJ^O ) ^s obtained through standard convolution with smooth modifiers j£ (x) : (u£0,v£Q)(x)=
/
^{u0{y),vo(y))je(x-y)dy. ~*
Suppose that there exists ue G (0,oo) with
(u-ue)f"(u)
> 0,
ue (0,oo)
then we can show that the following regions are invariant ones for the system in (3.110) for all e > 0 , r > 0 :
£ c
= {(",*) : 0 < v < ]—^C, f-'iv) V
where C > max<
P 1 _ °^ ^, f(ue)
f-\v
+ c)\ , )
\ (The definition of invariant regions is
given by Chuen, Conley and Smoller in [20].) The existence of global solutions to (3.110) follows then from usual local existence theory and the a priori sup norm estimate on the solutions by the in variant region argument. To show (3.111) we need certain estimates on entropy pairs for the model system (3.106). It is well-known that, for general conservation laws Ut + F(U)X = G{U), U € i T , the entropy pairs (77, g) satisfy V ? = V*7 V F,
(3.115)
106 Quasilinear Hyperbolic Systems and Dissipative Mechanisms so that the smooth solutions U of (3.115) satisfy v(U)t
+ q(U)x = \7vG(U).
An admissible weak solution U of (3.115) satisfies the entropy condition ri(U)t+q(U)x)r)uv = 0.
Thus, the general representation of entropy pairs is 17(11, *; H, G) = JU H(f(t)
- v)d£ + G(v),
f(u)-v
H(t)d£, / for any continuous functions H and G. Under the hypothesis that /'(0) ^> 1, the followings are particularly useful r)o(u,v;H)=
ij(u,v)=
f
H(f{{)-v)d£+
(/(o - «)<« + ft /
[
""
H(pf{Z))dt
/m,
•'/-'(jritr) •/ since the t; given above have the property that 7/w vanishes on equilibrium state v = (// — l)/(w) and that [^,,)/-^ ; l)/(")> C l [^-^-l)/(»)] 2 ,
j l ^ , , ) . ^ - ^ ; i)/(»)I < C2[y-(»-Ti)fW
(3-116)
for positive constants C\ and C2 depending only on the function H and M, which defines the bounded set {(u,v) : 0 < v < f(u) < M}. To show (3.111), we multiply the system in (3.110) by V*7*(w£> v£) with *7*(u, v) = rj(u, v) - rj(u,v) - Vrf(%v) I " _ ^ J , and obtain
+£^(we,ve)xx-£(w£,t;£)x v2^.(«c,«e) ("e J •
Relaxation 107 The estimate (3.111) follows by integrating the above equality over (—00,00) x(0,*) with the help of (3.116). Similarly, (3.112)-(3.114) follow from (3.111) and the above equality in the limit e —> 0. The convergence of (u£, v£) as e -> 0 can be proved by applying the theory of compensated compactness (see [15] for detail). Now, we turn to prove Theorem 3.3.1. For convergence and admissibility, we need to estimate the entropy inequal ity for the Cauchy problem of the equilibrium equalion ( « * + (f/(«)>• = ° {u(x10) = uo(x) based on the entropy condition (3.113), r)(uTjvT)t + q{uTlvT)x
(3.117) v
+ r)v(uT,vT) • —
;
*—— < 0 T
with V*(u) = (ii-l)f{u). It is not difficult to show that, given any entropy pair (77, q) for (3.106) with r)v(u, v) vanishing on the equilibrium curve v = V*(u) = (/i — l)/(u), the pair 17* (ti) = 77(1/, (fi - l ) / ( t i ) )
(3.118) g*(ti) = q{u, {fi- l)f(u)) forms an entropy pair for (3.107), namely,
+ q(uT, (1 -
/i)f(uT))g
= 7j(uTlVT)t + q(uT, VT)x + A r , where II A r HH-1
(n)
loc
/"{Mtlr,»r)-l?(«r,(l-/l)/(«r))]2
Jo Jn +[q{uT,Vr)-q{uT,{l-lL)f{uT))¥}dxdt I I [VT - (1 " fl)f(uT)]2dxdt Jo Jci < CT -* 0, as r -> 0.
108 Quasilinear Hyperbolic Systems and Dissipative Mechanisms Thus, the convergence of uT to u can be obtained by applying the theory of compensated compactness as in the proof of lemma 3.3.2. The admissibility of u can be guaranteed by the above estimate and (3.113). Finally, the convergence of vT(x,t),i > 0, to its equilibrium value (1 — fi)f(u(x,t)) follows from the convergence of uT to u and the formular below vT{x,t) = e-$vQ(x) +
/ e-l~r Jo
T
f(uT(xX))d(
which comes from the second equation in (3.110) and the convergence of (we,ve). Theorem 3.3.1 is then proved. Some new investigation on the model (3.105) is made by Luo and Natalini in [131]. They obtained the global existence of BV solutions by using Glimm scheme or modified Godunov scheme with a frictional step version in which the most significant part is the BV estimates on the approximate solutions by Godunov scheme. Since the BV and L°° estimates do not depend on relaxation time, the convergence to the equilibrium equation as r —> 0 is proved at the same time in [131]. By using the methods of compensated compactness, the rigorous justification of the relaxation approximation to equilibrium solutions containing shock waves has been verified by Chen, Liu and Levermore in [14] and [15] respectively with other special models. One such model is the following system f ut + vx = 0 \ vt +
- f{u))
(r > 0)
where a and / are some given smooth functions such that (T'(U) > v, (y > 0), /(0) = 0. For this model, the stability criterion (3.104) holds if |/»|
2
<
(3.119)
It is proved that the uniformly bounded sequences of perturbed solutions con verge strongly to some weak equilibrium solutions as the relaxation parameter tends to zero, for initial data close to the equilibrium, provided the subcharacteristic condition (3.119) is verified. The uniform boundedness of relaxing solutions is recovered by assuming the function / be constant out of a bounded set. However, a recent investigation on this model is made by Natalini in [150] to show that this uniform estimates are just a consequence of the stability con dition (3.119), without any further requirement on / . Another interesting model with relaxation effect is the Broadwell model in the kinetic theory of gases. It consists of the system of semilinear hyperbolic
Relaxation
109
equations
# - | l
= 2(/32-/i/2)
(3-120)
and derives from a six-velocity model when it is specialized to one-dimensional flows, for which the densities of particles moving in directions orthogonal to the flow are all equal. (The reader is refered to Broadwell [10] or Platkowski and Illner [161] for the derivation in the kinetic theory context). The function / = ( / i , hi h) is defined for ( i , i ) G l x M+ and describes densities of particles: / i for particles moving in the positive ^-direction, fa in the negative x-direction and fo in each of the positive or negative y-or z-directions. The parameter r stands for the mean free path, a measure of the average distance between successive collisions. The limit, when the mean free path approachs zero, is known as the fluid dynamic limit. For small mean free path the strong interactions of particles allow a macroscopic description of the flow to become meaningful. It is not difficult to identify the form of the induced macroscopic "Eular equations" as
S?+£(/»>=o §t(H + £(P9(«)) = o where g(u) := | [ 2 ( 1 + 3 « 2 ) 1 ' ' 2 - 1 ] ) and the macroscopic density and momentum of the fluid are given by p = Fi + 4 ( F i F 2 ) 1 / 2 + F 2 ,
m = pu = F1-
F2
with the notation (i<\, F2, F3) to denote the limit of / . By showing that a given piecewise smooth solution with noninteracting shocks of the limit fluid equations can be approximated by solutions of the Broadwell system as r —> 0, a recent study by Xin [185] gives a definitive answer to one direction of the zero relaxation limit problem. The converse problem, which shows that a given family of solutions to the Broadwell system converges globally in time to a fluid-dynamical solution, has not been solved yet. But, the approach of self-similar fluid dynamic limits is introduced by Slemrod and Tzavaras in [168] and further studies have been made by Tzavaras [176] [177], and Fan [42].
110 Quasilinear Hyperbolic Systems and Dissipative Mechanisms They consider a modified Broadwell system
^ - H
= H(/32-/1/2),
# = -2fe(/32-/1/2)> which preserves the invariance under dilation (x,t) —» (ax, at), a > 0. As far as the multidimensional case is concerned, a new progress is made by Katsoulakis and Tzavaras [91] recently for the scalar conservation law in sev eral space dimensions. They consider a class of semilinear hyperbolic systems of relaxation type which are contractive in the I, 1 -norm and admit invariant regions. The characteristic speeds of the conservation law satisfy, relative to the convective velocities of the relaxation system, a multidimensional analogue of the subcharacteristic condition. It is shown in [91] that solutions emanating from 51^-stable data are compact in a strong topology and converge, in the zero-relaxation limit, to a weak solution of the scalar multidimensional con servation law that satisfies the Kruzhkov entropy conditions. For solutions emanating from stable data in the L1 C\ L°°-norm, it is shown that the Young measure, describing the weak zero-relaxation limits, concentrates on the consti tutive manifold of the associated conservation law and gives rise to a measure valued solution.
Chapter 4 The influence of dissipation mechanism on the qualitative behavior of solutions For a system of hyperbolic conservation laws
mvnjbn where u = (i*i, • • •, un)> and the matrix /'(i«) has n real distinct eigenvalues Ai(u) < \2(u) < " < An(w), the nonlinearity of the system is reflected in the dependence of A,(u) on u and has a distabilizing effect on the qualitative behavior of the solution the result of which is the development of singularity at a finite time. The large time behavior of the solution for (4.1) with initial data u(ar,0) = uo(x) has been well studied also. For instance, suppose that (4.1) is a scalar equation, the flux function f(u) satisfies f"{u) > 0, and the initial datum uo(x) is a bounded measurable function, then it was shown by Diperna [33] that
IK,<)-^(-,0IUi W = O(r 1/2 ) if uo{x) has a compact support, where N(x,t) is called iV-wave which is intro duced by Lax in [102]; if UQ(X) takes the form as
{
u_, P(*). ti+,
x<-X -X <x<X x> X
with u- ^ u+, then it is proved by Dafermos [22] that
where UR(x,t) is the weak solution of Riemann problem with Riemann data u
, ,
«°W
=
fu_,
x< 0
{u+,
x>0.
Ill
112
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
Suppose that (4.1) is a 2 x 2 system which is genuinely nonlinear in the sense of Lax [102], and the amplitudes of initial data are small, then it is proved by Glimm and Lax [46] that ||ti(-,J) - U\\L°°(R) = 0{t~l) if the initial data uo(x) is a periodic function, where u is the mean value of uo(x); \\u(-,t) — U*\\L°°(R) = 0{t~1/2) if u0(x) -> u* as \x\ -+ oo. For a general system (4.1) which is genuinely nonlinear, more results have been established by Liu [113] [115]. For instance, with the initial data which has small total variation and compact support, the solution of the Cauchy problem satisfy 7Vti(s*) = 0 ( * - 1 / 2 )
IK^)-E^(^)r«(0)^w = °( rl/6 ) i
where JVt- is the iV-wave corresponding to the i-characteristic field and r,- is the right eigenvector corresponding to the i-characteristic speed. For more detailed information, we refer the reader to Hopf [62], Lax [102], Diperna [33], Dafermos [22] [23] for a scalar equation, and to Glimm and Lax [46], Diperna [33], Liu [113] [114] [115] for systems. The situation will be different when dissipation is introduced. There are different kinds of dissipation in which the common three kinds are the follows. A. Lower order term, such as relaxation and damping which have been discussed in the first three chapters of the book. B. Lower order term, such as the materials with memory. Consider the equations of motion of a one-dimensional body, with unit reference density and zero body force in Lagrangian coordinates f ftii(ar, t) - dxv(x, t) = 0 \dtv(x,t)-dx
,
. V'Z)
where u-deformation gradient, v-velocity, cr-stress. When the body is elastic, the stress at the material point x and time t is determined solely by the value of deformation gradient at (x,i) via a constitutive relation
f(u{x,t)).
Under the standard assumption f'(u) > 0, (4.2) yields a strictly hyperbolic system. It is interesting to compare the behavior of elastic bodies with the behavior of materials with fading memory in which the stress (r(x,t) at the materials point x and time t is determined by the entire history TIW(X, •) of deformation gradient at x. For concreteness, let us consider the simple model
f J—oo
k(t-
r)g{u{x,
r))dr
The influence of dissipation mechanism...
113
namely, dtu{x,i) — dxv(x,t)
{
—0
dtv{x, t) - dxf{u(x, t)) + j
k(t-
r)gx{u(x, r))dr = 0.
It is known that the memory exerts a rather weak damping influence if k(t) satisfies appropriate assumptions. It has been shown that, when the initial data are smooth and "small", dissipation prevails over the destabilizing action of nonlinear instantaneous elastic response. As a result, a smooth solution exists globally in time. On the other hand, when /(«) is nonlinear and the initial data are "large" then the destabilizing action of nonlinearity prevails and solutions break down in a finite time. We refer the reader to Dafermos [25], Renardy, Hrusa and Nohel [163], and the references there. C. Higher order term, such as viscosity and heat diffusion. For instance, the basic system for the one-dimensional gas flow without any dissipation takes the following form in Lagrangian coordinates ( vt — ux — 0 \ ut+px = 0 I Et + (pu)x = 0 where v denotes specific volume, u denotes velocity, E denotes total energy, ?/ E = e -h %-, e is internal energy, and p denotes pressure. These equations are known as the conservation laws of mass, momentum and energy respectively. If the viscosity and heat diffusion can not be neglected, we have the system 2
{
vt — ux =0 v>t+Px- (l*ux)x = 0 (e + \
)
+
fr™)* ~ \PUU*]* + 9x = 0
where the constitutive relations are given according to the materials in consid eration: e = e{v,0),p = p{vJ0),q =
q{v,6,0x),e>Q,q{v,0,0x)0x<<),
fi > 0 may depend on v and 0, where 9 denotes the temparature. The dissipation mechanism appears in the form of higher order term and they do influence the smoothness and the large time behavior of the solution. It is shown in Section 4.1 that the combined dissipation of viscosity and heat diffusion will preserve the smoothness of the initial data without the restriction of the smallness. For the case when there is only heat diffusion as the dissipation, namely, p, = 0, the dissipation is able to preserve the smoothness of the initial
114 Quasilinear Hyperbolic Systems and Dissipative Mechanisms data only when initial data are smooth and small. This is also discussed in Section 4.1. Another emphasis of this chapter is to understand the influence of dissipa tion on the large-time behavior of solutions. Various phenomena may occur on the large-time behavior of solutions corresponding to different materials in con sideration. For instance, with the same boundary condition as stress-free and thermally insulated, the solution may go to infinity as t -» oo with certain rate, may approach to a unique state exponentially fast or may have phase transition phenomena, according to different kind of constitutive relations to be concerned. This is discussed in Section 4.2. The last section of this chapter - Section 4.3 is devoted to the study on various behavior of solutions corresponding to different boundary conditions. 4.1
The influence of dissipation mechanism on the smoothness of solutions
The intent of this section is to elucidate the role of dissipation, particularly the stabilizing effect, in the continuum physics. Within the framework of thermomechanics, the referential (Lagrangian) de scription of the balance laws of mass, momentum and energy for one-dimensional materials with reference density po = 1 is
{
Vt — ux = 0 ut -
(43) jfhg
while the second law of thermodynamics is expressed by the Clausius-Duhem inequality
*+(f),>°
(4-4)
where v, u, e, <7, 77, $ and q denote deformation gradient (specific volume), veloc ity, internal energy, stress, specific entropy, temperature and heat flux, respec tively. Note that v, e and 0 may only take positive values. For one-dimensional, homogeneous, thermoviscoelastic materials, internal energy, stress, entropy and heat flux are given by the constitutive relations e = c(v, 0),0,
The influence of dissipation mechanism...
115
where ij) = e — Or) is the Helmholtz free energy. A typical problem is to consider a body with reference configuration the interval [0,1] whose boundary is stress-free and thermally insulated,
$;>:;<%=:
oS«»
<«,
and determine the thermomechanical process, described by the functions (v, w, Q)(x,t), 0 < x < 1, t > 0, under prescribed initial conditions v{x, 0) = vo(ar),
u(x, 0) = u0{x),
0(x, 0) = $0(x),
0 < x < 1.
(4.7)
In the absence of dissipation, when the body is a thermoelastic nonconductor, i.e., a = — p(v,0),e = e(v,0),g = 0, (4.3) reduces to a hyperbolic system of conservation laws (4.1) which does not generally possess globally defined smooth solutions, even when the initial data are very smooth. When the material is a thermoelastic conductor, it has been shown by Slemrod in [167] that smooth and "small" data generate globally defined smooth solutions to (4.3). However, it is generally conjectured that heat dissipation alone cannot prevent the breaking of waves of large amplitude and the forma tion of shocks. And it has been proved by Dafermos and Hsiao in [31] that this conjecture is indeed true. The question is whether the combined dissipative effects of viscosity and thermal diffusion, when the material is thermoviscoelastic, may counterbalance the destabilizing influence of nonlinearity and thus induce the existence of glob ally defined smooth solutions to the initial-boundary value problem (4.3) (4.6) (4.7) even for large initial data. This is first verified by Kazhikhov [97], Kazhikhov and Shelukhin [98] for the case where the material is an ideal, linearly viscous, gas with constant specific heats: e = c0,
— v
by means of an interesting analysis which depends crucially upon the special form of the above constitutive assumptions. For the case of ideal gas, the global existence of smooth solution to initial boundary value problems or Cauchy problem, with smooth and large initial data have been investigated also by Antontsev, Kazhikhov and Monakhov in [4], Kawashima and Nishida in [95], Nagasawa in [147] [148] and [149]. On the other hand, Dafermos and Hsiao in [29] and Dafermos in [24] consider the problem (4.3) (4.6) (4.7) for a fairly general class of solid like linearly viscous materials. They establish the global existence which will be discussed in this section. The global existence with a different boundary condition to (4.6) is obtained later by Jiang [86].
116
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
The corresponding results for the material of real gases have been obtained by Kawohl in [96], Jiang in [87], Luo in [128], [129], Pan in [156], and etc. As far as the thermoviscoelastic material in more than one dimension is concerned, the global existence of smooth solutions for general domains has been investigated only in the case of sufficiently small initial data. We refer the reader to Matsumura and Nishida [141] [142] and Valli [178] [179]. For the material of thermoelastic conductor in multi-dimension, a survey of the study can be found in the book by Racke and the references cited there [162]. For the small initial data problem of nonlinear coupled hyperbolic-parabolic systems, there are a lot of works on both the global existence and blow-up of solutions, for instance, the book by Zheng [194]. This book also discusses the global existence and asymptotic behavior of solutions with arbitrary initial data for certain coupled hyperbolic-parabolic systems. The problem (4.3) (4.6) (4.7) is investigated by Dafermos and Hsiao in [29] for a fairly general class of solid like linearly viscous material e = e(v,0),cr = -p(v,0)
+ fiuXlq —
q(v,0,0x).
In this paper, shear viscosity \LV is inversely proportional to density. Conse quently, its dissipative effect is so weak at high density that existence of global solutions is to be expected only when the initial energy is not too large. To overcome this limitation, Dafermos [24] considers the problem (4.3) (4.6) (4.7) for linearly viscous materials e = c(«, 0), a = -p{v, 0) + jj.(v)ux, q = -K{v,
0)0X
(4.8)
where the viscosity p(v)v is uniformly positive, that is fi(v)v > /i 0 > 0,0 < v < oo.
(4.9)
R e m a r k . In the case when viscosity may depend on temperature, a typical problem to understand the effect, caused by the dependence of viscosity on temperature, has been studied first by Dafermos and Hsiao in [30]. We assume that e(v,9),p{v,0),fi(v) and K(v,0) are twice continuously dif f e r e n t i a t e o n O < v < o o , O < 0 < o o and are interrelated by ev(v,0)
= -p(v10)
+ 0pe{v,0),
(4.10)
so as to comply with (4.5). Furthermore, we require that, at any temperature, the elastic part of the stress be compressive at high density and tensile at low density, i.e. there are 0 < v < V < oo such that P(w, 0) > 0,0 < v < v, 0 < 0 < oo, (4.H) p(v, 0) < 0, V < v < oo, 0 < 0 < oo.
The influence of dissipation mechanism...
117
Employing an idea of Andrews in [2], we will show later in proposition 4.1.3 that, as a consequence of (4.9) and (4.11), the deformation gradient is a priori confined in a bounded interval 0 < v < v(x,i) < V and hence no restrictions are necessary on the behavior of e(v, 9),p(v, 0), K(v, 9) at v = 0+ and v = oo. As regards growth with respect to temperature, we assume that for any 0 < v < V < oo_ there are positive constants i/, Ar0 and N, possibly depending on v and / or V, such that, for any v < v
*o < K(v,9)
e(v,0) > 0, v < e0{v, 9) < N(l + 0 1 / 3 ),
(4.12)
\Pv(v, 0)| < N(l + 0 1 + * ) , \pe(v, 0)| < N(l + 0 1 ' 3 )
(4.13)
< N,\Kv{v,0)\
< W , | £ , ( t ; , 0 ) | < N,\Kvv{v,0)\
< N.
(4.14)
(A explanation on the above growth rates can be found in [24]). The existence theorem established in [24] reads: T h e o r e m 4 . 1 . 1 . Consider the initial-boundary value problem (4.3) (4.6) (4.7) under conditions (4.8)-(4.14). Assume that v0(x), v'0(x), u0(x), u'0(x), u%(x), 0o(x), 0'0(x), and 9%{x) are all in C a [0,1] and v0(x) > O,0o{x) > 0,0 < x < 1. Furthermore, let the initial data be compatible with the bound ary conditions (4.7) at (0,0) and (1,0). Then there exists a unique solution {v(x,t), u(x,t),9(x,t)} on [0,1] x [0,oo) such that for every T > 0 the functions ^ •> ^X 5 lit 5 ^XX ) " ) "x , Uxx lhkjbmnkj are all in Ca,a^2(QT) and vtt, u>xt, 9xt are in nvhbgj
V, Vx , l^t , Vxt)
L2(QT). Moreover, 0(x,t) > 0,i7 < v(x,t) < V, for 0 < x < 1, 0 < t < oo, where v and V are positive constants depending on the initial data, C^fO, 1] denotes the Banach space of functions on [0,1] which are uniformly Holder continuous with exponent a, while Ca,a^2(QT) stands for the Banach space of functions on QT = [0,1] x [0,T] which are uniformly Holder continuous with exponent a in x and a / 2 in t. nvbhfgy The theorem can be proved by the procedure devised by Dafermos and Hsiao in [29], namely, solutions to (4.3) (4.6) (4.7) are visualized as fixed points of a m a p P on the Banach space B of functions {V(x,t)1U(xit),Q(x1t)} kgjhnbjh with V,U,UX,Q,QX in C 1 / 3 ' 1 / 6 ( Q T ) and existence is established by means of the Leray-Schauder fixed point theorem. mbnjghyftgryhfgtf The m a p P carries {V(x,i), U(x,t),S(x,t)} into the solution of a compli cated linear "parabolic" system obtaned by linearizing (4.3) about {V(x,t), U(x,t), 0 ( # , t)}. Due to the smoothing action of linear parabolic systems, P is completely continuous and its range is contained in the set of functions {v(x,t),u(x,t),9(x,t)} with v, vx, vu vxU u,uXluuuxx,9,9X,9U9XX in C a ' a / 2 (QT) The construction of P and the precise statements and proofs of its aforemen tioned properties can be found in [29]. Here we only show that any possible fixed point of P , i.e. any solution {v(x,t), u(x,t),9(x,t)} of (4.3)J4.6) (4.7) satisfies the admissibility conditions 9(x,t) > 0,0 < v < v(x,t) < V and is contained
118 Quasilinear Hyperbolic Systems and Dissipative Mechanisms in an a priori bounded set of B. This will complete the list of requirements for the application of the Leray-Schauder fixed point theorem. Turn to the initial data, we normalize uo(x), by superimposing a trivial rigid motion if necessary, so that / uo(x)dx = 0 (4.15) Jo Let {v(x,t),u{x,t),6(x,t)} be a fixed solution of (4.3)(4.6)(4.7) on [0,1] x [0,oo) in the function class indicated in Theorem 4.1.1. Now, we derive the a priori estimates. Integrating (4.3) over [0,1] x [0,tf] and using the boundary conditions (4.7), we obtain the conservation laws of total momentum and energy: / u{x,t)dx= Jo I
U(x,t) + -u2{x,t)\dx=
I u0(x,t)dx Jo I
= 0,0
\e(x,0) + -ul(x)\dxd=
Eo,
(4.16)
0
Substituting a from (4.8), we write (4.3)2 in the form ut+p{v,e)x
= \ji(v)ux]g,
(4.18)
while combining (4.3)3 with (4.3)2 and using (4.8), (4.10) and (4.3)i we get e0(v,O)Ot + epe(v,O)ux-ji(v)ul
= [K{v,0)0x]x.
(4.19)
Applying the maximum principle on (4.19), recalling that 00(x) > 0,0 < x < 1, one deduces Proposition 4.1.2. 0OM) > 0,0 < x < 1,
0 < t < oo.
Now, we derive bounds on the deformation gradient. Using (4.3)i, the equa tion (4.18) can be rewritten as ut + p{v,0)x = M{v)xt,
(4.20)
where M(v) =
/
ji{u)duj.
Due to (4.9), M(v) is a strictly increasing function which maps (0,oo) onto (—00,00).
The influence of dissipation mechanism...
119
Proposition 4.1.3. It holds that v< v(x,t) < V,0 < x < 1,0 < * < oo, where v = M- 1 (M(min{i;,mint;o(-)}) - v ^ o ) i V = f M-HM(max{V r ,maxi;o(-)}) + \ / 2 ^ ) Prcx>/. Integrating (4.20) over [0, y] x [«, r], 0 < y < 1,0 < s < r, and using the boundary condition (4.7)i, we get M{v{y,T)) = M{v(y,s))+
rT
p(y,t)dt +
Js
ry
u(xJr)dx-
JO
rv
u(x,s)dx.
(4.21)
JO
By (4.16), (4.17) and (4.12), it follows r ry l2 1 Z*1 2 / u(xJt)dx\
1 < -E0 2
0
(4.22)
This, with the definition of v and V together, implies v < vo(x) < V, 0 < x < 1. Thus, if v(x, t) > v is violated on [0,1] x [0, oo), there are r > 0 and y E [0,1] such that v(x,t) > v, for 0 < x < 1,0
> M{v0(y)) - yj2E~o-
In the latter case, (4.21) together with (4.11) and (4.22) yield M(v(y, T)) > M(v) - \/2Eo~. In either case, it follows from the definition of v that M(v(y, r)) > M(v) which is a contradiction to v(y, r) — v. This shows v(x, t) >v,0<x< 1,00 and restrict our solution to the rectangle QT = [0,1] x [0, T]. In the sequel, A will denote a generic constant which may depend at most on T, no,v,ko,N and the upper bounds of the C a [0,1] norm of vo, v0l «o, w0, w o> 0o, #0* ^0- W e will show that {v(x,t), u(x,t), 0(x,t)} is a priori bounded in the Banach space B, that is
120 Quasilinear Hyperbolic Systems and Dissipative Mechanisms Proposition 4.1.4. It holds that ll v llc i / 3 . i / 6 (Q T ) ^
A
'
||ti||Cl/3,l/6(QT) < A,||wa:||Cl/3,l/6(QT) <
A,
||0||C1/3,1/6(QT) < A , | | ^ | | C l / 3 , l / 6 ( g T ) <
A.
To prove this proposition, we need the following lemma 4.1.5 - lemma 4.1.11. The first observation is that, in view of (4.17) and (4.12), max / 0(x,t)dx < A. [0,T]7 0
(4.23)
"
We now proceed to get estimates which are motivated by the second law of thermodynamics and embody the dissipative character of viscosity and thermal diffusion. Lemma 4.1.5. It holds that / / 0-Al3e2xdxdt Jo Jo
< A,
(4.24)
f f 0s/3dxdt
(4.25)
/
(4.26)
rT
max0 5/3 (.,<)cft< A.
f9 Proof. Define H(v,9) = / £~ 1/3 e0(v, £)£. It is easy to see, on account of (4.12), that \H(v,0)\<2N(l
+ 0).
(4.27)
Moreover, we can show, with the help of (4.10) that He(vy0)=e-VZee(v,6)
where
These, with the setting H(x,t) e-1'3, imply
= H(v(x,t),6(x,t))
and multiplying (4.19) by
Ht + G(v, e)ux - e-1'5 ■ /*(„)«! - e-liz[k{v, e)ex)x = o.
The influence of dissipation mechanism...
121
By first integrating this equation over [0,1] x [0,T], and then integrating by parts with respect to x and recalling the boundary condition (4.7)2, we obtain / / JA(v)e"1f3uldxdt + l [ [ A > , 0 ) 0 " 4 / 3 -Oldxdt Jo Jo 3 Jo JO = / H{x,T)dxJo
f H{x,0)dx+ Jo
f f Jo Jo
G{v,0)uxdxdt.
By virtue of the definition of G(v, 0) and (4.13), it holds \G(v, 9)\ < 2N(l + 0). So, using (4.9), (4.14), (4.23), (4.27), Proposition 4.1.3, and the Cauchy-Schwarz inequality, we deduce fioV'1 [ f O-VSuldxdt+lko Jo Jo 3
I f Jo Jo
0-4'3'Oldxdt
< A + 47V 2 .\7^- 1 / / e^dxdt+^/ioW1 Jo Jo 2
f J0
f J0
Q-ll3u2xdxdt.
(4.28) On the other hand, it follows from (4.23) and the use of Cauchy-Schwarz in equality that / / Oldxdt Jo Jo
<\[ max05/3(-,*W* " Jo [0,1] < A+ A / / Jo Jo
e2f3\ex\dxdt
< A + 1 / / 08/3dxdt + A f f O-4/302xdxdt, Jo Jo Jo Jo (4.29) whence 4AT2F/i0"1 / / 07l3dxdt
/ / 0-Al302xdxdt. Jo Jo
(4.30)
Thus, (4.28) and (4.30) yield (4.24), and then (4.29) implies (4.25) and (4.26). As a corollary of Lemma 4.1.5, we have Lemma 4.1.6. / / u\dxdt < A. Jo Jo In the following lemmas we employ the bounds obtained in order to estimate by interpolation the square integral of various derivatives of the solution in terms
122 Quasilinear Hyperbolic Systems and Dissipative Mechanisms of low powers of 92{xA)dx,
Y = max / [0,T]7 0
Z = max / u2T(x,t)dx. XXK } [0,T]J0 It is not difficult to show, by (4.23) and Schwarz's inequality, that e3/2(y,t)
[ ^ ( ^ I M ^ O I ^ A + AY 1 / 2 , * Jo
0 < y < 1, (4.31)
whence m a x 0 < A + AY 1 / 3 .
(4.32)
QT
Similary, combining I u2x(x,t)dx + 2 [ Jo Jo
ul(y,t)<
\ux(x,t)\\uxx(x,t)\dx,0
with the standard interpolation estimate (see Agmon [1]) / u2x(x,t)dx<
108 I u2(x,t)dx+4S2<
j u2(x,t)dx\
I /
u2xx(x,t)dx\
and using (4.17), we arrive at max | ^ | < A + AZ 3 / 8 .
(4.33)
QT
L e m m a 4.1.7. max/
vl(x ,t)dx<
A + AY 1 / 9 .
Proof. Multiplying (4.20) by (M(v)x - u) and integrating over [0,1] x [0,*], 0 < t < T, we obtain \J^
[M(v{x,t))x-u{x,t)]2dx-^j
=
[M(vo(x))x-u0(x)]2dx
\PvVx +P$0x][M(v)x - u]dxdr. Jo Jo
By virtue of (4.13), the definition of M(v) and proposition 4.1.3, it follows / / pvVx[M(v)x I Jo Jo
- u)dxdr\ < A / / (1 + 6^3){[M{v)x I Jo Jo
- u}2 + u2}dxdr
The influence of dissipation mechanism...
123
5 3 < A / ( l + max0 / (-,r)l / [M(v(x, r))x - u(x, r)]2dxdr Jo I I0.1] J Jo
u2(x,T)dxdr.
+A / I l + max0(-,r)l j
Similarly, using (4.20) and applying Schwarz's inequality,
/ / p$0x[M(v)x - u]dxdr\ \Jo Jo I
f j l + m0 a1 x 0 5 / 3 ( . , r ) i / M(v(x, r))x - u(x, r))2dxdr Jo I I.! J JO
+ 7 v j l + max01/3j / ' /
0-^0\dxdr.
The lemma can be proved then by combining the above, applying GronwalPs inequality and taking account of (4.17), (4.24), (4.26) and (4.32). Lemma 4.1.8. Y < A + AZ 3 / 4 /
/
02dxdt
+ AZ3/4.
Jo Jo
Proof. Define
(M)=f / K(v,tW, Jo
set Q(x,t) = Q{v(x,t), 0(x,t)), multiply (4.19) by Qt and integrate over [0,1] x [0, t], 0 < t < T. After an integration by parts with respect to x we arrive at
/ / {ieOt + 0peux - {ml}Qtdxdr+ Jo Jo
/ / K0xQxtdxdt Jo Jo
= 0.
By using (4.12), (4.13), (4.14), (4.24), (4.25), (4.26), (4.33), Proposition 4.1.3, Lemma 4.1.6, and the facts of \QV\ < NO, \QVV\ < NO, each term in the above
124 Quasilinear Hyperbolic Systems and Dissipative Mechanisms equation can be estimated as follows rt
pi
pt
t>\
/ / edK0?dxdT >vk0 / B2tdxdr, Jo Jo Jo Jo / / \Jo Jo
ee6tQvUxdxdr\ I / 02tdxdr + A + AZ3'4,
\ / {Opdux \Jo Jo
fjtul}Qvuxdxdr\ I
I {\ + 07l* + ul}dxdT
< /
AZ3/\ / {0peux -
p,ul}K0tdxdT\ I
\Jo Jo
f f {1 + 0 8 / 3 + u2x}dxdr Jo Jo
( f e^dxdr+{AmaxuDJo Jo Q
<\vk0
I
j
efdxdr+AZ3'4,
[ I kex[KOx]tdxdr>h0 Jo Jo " I /■* fl - I / / K6xQvuxxdxdT\ \Jo Jo I
I
I Jo
0%(x,t)dx-A,
+ AZ3'4, K0xQvvuxvxdxdr\
< iV 2 {maxK|} {max**/*} { j f £ • | j f max0 5 / 3 (.,r)jf
t£(*,r)cte(ir}
^^dxdr}^
The influence of dissipation mechanism...
125
I f* f1 ~ /
/
\Jo
Jo
~
I
K0xKvvx0tdxdT\ I
< lisk0 I f 02dxdr+A f f Jo Jo Jo Jo in which the last term can be estimated as A l l [KOx]2vldxdr Jo Jo
[K6x]2v2xdxdr,
< A / m0 a1x [ / ^ t ] 2 / Jo I - ] Jo
v2x{x,T)dxdr
<(A + Ay!/9) / /
\kex\\(kex)t\dxdT
Jo Jo
< (A + AY1'9) { m a x * 2 ' 3 j j f*
fV4'3e2xdxdr\
'{Jo L KK°'M2dxdT} < A + ^k0Y + A | /
/
{{kex)x)2dxdT\
To estimate the last term in the above inequality, we use (4.19) and the facts of / / e2ee2dxdr<[A Jo Jo
+ AY2'9]
f I {Opeux - fiulfdxdr Jo Jo
f f 02dxdr, Jo Jo
< A+
AZ3/4.
Thus, A
{ / / HK°*)*]2dxdr}
^ A + YQkoY + ^ * ° / / °tdxdT +
Combining all of the estimates obtained above, we obtain \vk0 2
/
/
02tdxdr+\ko
Jo Jo
2
f
$Z(x,t)dz < A + h0Y
J0
4
which finishes the proof of lemma 4.1.8. Lemma 4.1.9. A + AZ11'12
m a x / u!(z,t)dx.< [o,T]y0
~~
/ / ultdzdtKA + AZ11'12. Jo Jo
+ AZ 3 / 4
AZV4
'
126 Quasilinear Hyperbolic Systems and Dissipative Mechanisms Proof. Differentiate formally (4.18) with respect to 2, multiply by ut and inte grate over [0,1] x [0,2],0 < t < T. Integrating by parts with respect to x we obtain u2{x,t)dx-~
\ \ +
u2(x,Q)dx +
£L'(v)uluxtdxdTJo Jo
/
fi{v)u2xtdxdr
/ {pvux + pe0t}uxtdxdr Jo Jo
=0
in which each term can be estimated, with the help of (4.9), (4.13), (4.25), (4.32), (4.33), Proposition 4.1.3, Lemma 4.1.6, Lemma 4.1.8, as follows: / / fi(v)ultdxdr Jo Jo
/
/
< \voV
/
/ ultdxdr + A < m a x t ^ f /
< \^V~l
I
I u2xtdxdr + A + A^ 3 / 4 ,
/
/
u2xdxdr
I pvuxuxtdxdr\ I
\Jo Jo
< \noV~1
f
I
< \voV~1
f
f u2xtdxdr + A + AZ 3 ' 4 ,
u2xtdxdr + A lma.xu2x\
I f* f1 /
u2xtdxdr,
ft{v)uxuxtdxdT\
\ fl f1 /
> fi0V~1 / / Jo Jo
/
f
f (1 + 08'3)dxdT
I pgOtuxtdxdT\ I
\Jo Jo
<\»oV~l
j
I
<\noV~1
f
f u2xtdxdT+A
These yield Lemma 4.1.9.
u2xtdxdT+AU
+ max02'3\ +
AZn/12.
f
f
$2dxdr
The influence of dissipation mechanism...
127
L e m m a 4.1.10. max / u2(x,t)dx
< A,
[0,T]7 0
/
/
u2xtdxdt
Jo Jo 2 max / uxxK (x,t)dx [o,T]Jo '
< A. -
Proof. Due to Lemma 4.1.9 and the definition of Z, it suffices to show Z < A. We employ (4.18) whose terms can be estimated, by virtue of (4.9), (4.13), (4.32), (4.33), Proposition 4.1.3, Lemma 4.1.8, as follows: / fL-2fL2v2xu2xdx< A J m a x t ^ i j / ii~2fvv2xdx
< A + AZ 5 / 6 ,
v\dx
< 2N2 | l + max0 8 / 3 j C v2xdx < A + AZ 3 / 4 ,
f fi-2p2ee2xdx < 2N2 | l + m a x ^ 3 | f
Q2xdx < A + AZ 1 1 / 1 2 .
These yield Z < A + AZn/12, which implies Z < A. L e m m a 4.1.11. rrnix I e2(x1t)dx
(4.34)
/ 92xtdxdt
(4.35)
/ JO
m a x ^ 02rOM)dz
(4.36)
Proo/. We differentiate formally (4.19) with respect to 2, multiply by eoOt, integrate over [0,1] x [0,tf],0 < t < T, and integrate by parts. After a lengthy sequence of estimations, the Lemma 4.1.11 can be obtained. The detail can be found in [24]. Now, we finish the proof of Proposition 4.1.4. By (4.34), (4.35) and Schwarz's inequality, 0(x,t) is uniformly Holder con tinuous in t with exponent ^ and (4.36) yields that 0x(x,i) is uniformly Holder continuous in x with exponent A. It then follows, from a standard interpolation
128
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
property (see Ladyzenskaja, Solonnikov and Ural'ceva [101]), that 0x(xjt) is also uniformly Holder continuous in t with exponent A. Hence ||^||c 1 / 3 ' 1 / 6 (Q T ) — A, which immediately yields | | 0 | | C I / 3 , I / 6 ( Q T ) < A. Similarly, Lemma 4.1.10 im plies that \\UX\\CI/*,I/*(QT)(qr) < A which gives that | | U | | C I / 3 , I / 6 ( Q T ) < A, | | V | | C I / 3 , I / 6 ( Q T ) < A. 4.2
T h e i n f l u e n c e of d i s s i p a t i o n m e c h a n i s m o n t h e l a r g e - t i m e b e havior of solutions
This section investigates the large-time behavior of globally defined smooth solutions established in Section 4.1. For the material of ideal gas, it has been proved by Nagasawa in [147] that the solution (v*,u*,0*) to the problem (4.3) (4.6) (4.7) satisfies u*M)>C*log(l-M),
C* > 0 .
However, totally different phenomena may occur on large-time behavior of solutions for other kind of constitutive relations. In the case of isothermal viscoelasticity, the solution may approach to a unique state exponentially fast as shown by Greenberg and MacCamy in [52], or phase transition may take place as discovered by Andrews and Ball in [3] with nonmonotone pressure. It is proved, in [3], that the large time behavior of strain is described by a Young measure whose support is confined in the set of zeros of pressure. The goal here is to extend the analysis used in [52] and [3] to the nonisothermal case — thermoviscoelastic materials. For simplicity, we consider the kind of solid-like materials with the following constitutive relations e = Cv0y
(T = -f{v)6+fi(v)uXl
q = -k(v)0x
where Cy is a positive constant, k(y) and f(v) t i a t e for v > 0 such that k(v) > 0 for f{v) > 0,
0< v < v
(4.37)
are twice continuously differen
v > 0
f(v) < 0,
V < v < +oo
(4.38)
for some fixed 0 < v < V < -foo as mentioned in (4.11), and the viscosity fi(v)v is uniformly positive, as assumed in (4.9), namely fi(v)v > /i 0 > 0,
0 < v < +oo.
(4.39)
Remark 4.2.0. The example proposed by Ericksen ([40]) shows that f(v) is not monotone in v and satisfies (4.38). For rubber, it is known that a good
The influence of dissipation mechanism...
129
model of pressure takes the form p(v, 0) = —~f0(v
«■),
7 is a positive constant
v namely, f(v) = —*f(v — \ ) , which satisfies (4.38) with v = V = 1. Now, we turn to assumptions on initial data. The initial velocity is normal ized as in (4.15) and the initial data are compatible with the boundary conditions (4.6). The global existence of (4.3), (4.6) and (4.7), under the assumptions of (4.37)-(4.39) and (4.15) has been obtained in Theorem 4.1.1., which concerns the solid-like material with more general constitutive relations than (4.37). The following results on large-time behavior of solutions have been estab lished by Hsiao and Luo in [76]. Theorem 4.2.1. Assume {u(x,tf), v(x,t),0(x,t)}, (x,t) solution of the problem (4.3), I. l|p(M)(-,t)||Li[o,i]
that (4.37)-(4.39) and (4.15) are satisfied. Let E [0,1] x [0,oo)) be the globally defined smooth (4.6), (4.7). Then = l|/W0(-,*)IUi[o,i] -> 0 as t -+ + c o ,
ll/(v)(->*)IU a [o,i] -> 0 as * -► +oo, IN-,t)|| L 2 [ 0 ) i] -> 0 as t -)- + o o , and 0(t) =
[ 6(x,t)dx Jo
-> ^ 5 . as t -> + o o ^v
f1 1 where E0 = / [Cv00 + -v$\(x)dx. Jo * II. There exists a family of probability measure {^ar}a:e[o,i] o n ^ (depending measurably on x) with suppi/ r C K - {z : f(z) = 0} such that if $ E C(M) and 0 * ( * ) d =
ae
-
then * ( v ( - , t ) ) A £<*>(•) in L°°[0,1] as * -> + o o . R e m a r k 4 . 2 . 2 . Theorem 4.2.1 extends the phase transition results in [3] to nonisothermal cases. Corollary 4 . 2 . 3 . Suppose the equation f(z) = 0 has exactly m roots, z\, z2, - • •, zm, m > 1. Then there exist nonnegative functions \i{ G £°°[0,1], 1 < i < m, such that m
*(«(-,'))
A
53*(Z*)W(-)
in
^°°[0,l], a s t - ^ + o o ,
130 Quasilinear Hyperbolic Systems and Dissipative Mechanisms for any $ G C(M). m
Furthermore Yjji,-(ar) = 1, a.e. . i=i
Corollary 4.2A. z = z\, then
Suppose the equation f(z)
v(-,t) —> z\
= 0 possesses only one root
L 9 (0,1)
strongly in
as t —>■ oo
for all ^, 1 < g < + o o , provided the conditions (4.37)-(4.39) and (4.15) hold. If f(v) is strictly monotone decreasing, namely, f'{v)<0
for
v€[t7,F]
(4.40)
it follows from (4.38) that there exists a unique v £ [£, V] such that f(v) = 0. Thus, we have further results in the next theorem. T h e o r e m 4 . 2 . 5 . Assume that (4.37)-(4.40) and (4.15) hold. Then there are positive constants / ? , T and A, independent o f t , such that IK-,*) - 1)11*1(0,1) + l|ti(-,t)||jyi (0> i) + ||0(-,<) - ^ - 1 1 ^ ( 0 , 1 ) < ^ " ^ for t > f. Theorem 4.2.5 generalizes the results obtained in [52] which discusses the case of isothermal viscoelasticity. Firs, we prove Theorem 4.2.1. ^From now on, {u(x, t), v(x, t),0(x, t)} will denote the solution described in the global existence theorem. It is known from Proposition 4.1.3 that o
v(x,t)
< V,
0(x,t)>O,
a? € [ 0 , 1 ] ,
*e[0,+oo).
(4.41)
(4.41) and (4.39) yield 0 < /ii < fi(v(x, t)) < ji 2 ,
x€ [0,1],
t e [0, + o o ) ,
*i < k(v) < k2
(4.42)
where //i,/i2, &i and k2 are positive constants, independent oft. In the sequel, A will denote a generic constant, independent oft. The conservation laws of total momentum and energy have been obtained in Section 4.1, namely (4.16) and (4.17). L e m m a 4.2.6.
jff Proof.
o
[^f
+ ^\{x,r)dxdr
<e[0,+oo).
Substituting
«« + lf(v)6]* = [£(»)««]«•
(4.43)
The influence of dissipation mechanism...
131
Combining (4.3)3 with (4.3)2 and using (4.37) we obtain CvOt + f(v)0ux - ft{v)u2x - {k{v)0x)x = 0.
(4.44)
Multiplying (4.44) by 0~l and integrating over [0,1] x [0, t], with the help of (4.6) and (4.3)i, we get jf £
P
^
+ ^ ]
(*>')***'
[J\og0{z,t)dz
- / \og0{x,O)dx\ + / G(v)(x,t)dxJo J Jo
f Jo
G(v)(x,0)dx
where G(v) = / /(£)<*£. This, with (4.41), (4.42), and the inequality log0 < J\F
0 — 1, for 0 > 0, implies the Lemma 4.2.6. Due to (4.16) and the mean value theorem, there exists a y(t) G [0,1] for every t > 0 such that ti(y(<),*)=0. Thus I />x
|ti(M)|= /
I
T /*1
i 1/2 r /»1
MU)dt\ < I e{x,i)dx\
"i 1/2
2
j ^(x,t)dx\
\Jy{t) I UO J VJo V which, combined with (4.17) and Lemma 4.2.6, yields / maxu 2 (-, r)dr < A, Jo [o,i]
t£ [0, +oo).
J
(4.45)
Lemma 4.2.7. / {u4 + 02){x,t)dx+ JO
f f [02x + u2u2x]{x,r)dxdT
*€[0,+oo).
2
Proof. Multiply (4.3) 3 with (CV0+ \ ) and integrate over [0,1] x [0,*]. With the help of (4.6), (4.37), (4.3), (4.4) and Young's inequality, we arrive at \ I
+^ <
[ C v 0 + y ] {x,t)dx + inj
j
u2u2x{x,r)dxdr
f f 0l(x,r)dxdT
2 A + A / maxu (-,r) / 02(x,r)dxdr 01 Jo t ' ! Jo
+A / / Jo Jo
u2u2x{x,r)dxdr. (4.46)
132 Quasilinear Hyperbolic Systems and Dissipative Mechanisms To estimate the term /
/ u2u2dxdr,
we multiply (4.3)2 by t/3, integrate the
Jo Jo
resulting equation over [0,1] x [0,tf], and use the boundary condition (4.6), (4.41), (4.42) and the Cauchy inequality. It then follows that \ I u4(x,t)dx + 2//i / J
°
\°
t 2
<
u2ul(x,r)dxdr
/
J J
(4.47) 2
A + A / maxu (-,r) / Jo [o>i] Jo
0 (x,T)dxdr. 2
TJT)2
By using the Cauchy inequality with the term (CyO + obtain /
62{x,t)dx+
Jo ft
<
/
/
u2ul(x,T)dxdr+
/
0l(x,r)dxdT
Jo Jo
fl
ft
fl
/ u2u2(x, r)dxdr-t- A / maxw2 ( , r ) /
A 4- A /
Jo Jo
+A / Jo
/
Jo Jo
in (4.46), we
Jo [M]
02(,T)dxdT
Jo
u4(x,t)dx.
(4.48) Multiplying (4.47) with a suitably large positive constant, and combing with (4.48), we get /
02(x,t)dx+
Jo
f u*{x,t)dx+ Jo
f
f u2u2x(x,T)dxdr+
Jo Jo
< A + A I maxw 2 (,r) /
e2x{x,T)dxdr
[ f Jo Jo
e2{x,r)dxdr.
(4.49) Applying GronwalFs inequality to (4.49) and using (4.45), we obtain the Lemma 4.2.7. L e m m a 4.2.8. / / ul(x,r)dxdr Jo Jo
< A.
Proof. Multiplying (4.3)2 by u and integrating over [0,1] x [0,<], it follows, with the help of (4.6), (4.41) and (4.42), that 2 J0 ^ M ^ + Z'iy J
u2x(x,T)dxdr
I
f
f{v)0uxdxdr.
The influence of dissipation mechanism...
133
Due to (4.3)i and (4.41), / / Jo Jo
f{v)0uxdxdr
=
11 f(v){0 -H)u*{x,T)dxdT + / / Jo Jo Jo Jo
<
^T I I U2x{x,T)dxdT+k Jo Jo
+ / / Jo Jo
f{v]6ux{x,T)dxdr (0-0)2{x,T)dxdT
I f Jo Jo
f{v)Bvt{x,T)dxdT
where 6(t) = / 0(x,i)dx,t 6 [0,-f-oo). Jo _ By the mean value theorem, there exists a z(t) £ [0,1] such that 0(t) = 0{z{t),i). Thus i
|0-0|(*,r) =
11*)
i
f
1
0x^T)dc\<\f el(xJr)dx\
\JZ{T)
I
,
L/o
r e [ 0 , + o o ) . (4.50)
J
Integrating (4.3)3 over [0,1] and using (4.6) and (4.37), we arrive at Cv([
0(x,r)dx\
= - ( /
\u2{x,r)dx\
,
re[0,+oo).
Namely, 0t(r) = - ^ - \J
i u 2 0 r , r)dx\
,
r e [0, +oo),
(4.51)
which, together with (4.17), implies 0(T) = ± . \ E
0
- J
\U2{X,
r)dx] ,
te[0,
+oo).
(4.52)
This, together with (4.3)i, (4.41), (4.17), (4.45), implies, upon integrating by parts and using Cauchy inequality, that / / Jo Jo
[f(v)9vt](x,r)dxdT
- A~JlJ! {f,{v)vvt [ih (E° ~ I + J!J!{f{v)v[^{J!^u2{x,r)d8)] = A - g . |jf \(v)(x,t)dx - jf A(»)(*,0)d*J
^2{s'r)ds)]}{x'r)dxdT }{x'T)dxdr
134 Quasilinear Hyperbolic Systems and Dissipative Mechanisms
+
I I y^VUx\J
+
& {jf [f{v)v Of r 2 ( M ) d s )] (x't)dx
- fQ
2u2^'T^ds)\^x'T^dxdT
[f(v)v Q f ' ^(s,
0)ds)] (*, 0)dx}
- ^ 7 j f fQ {[f'(v)v + /(«)] «• ( j f 5« 2 («, r)d«) | (*, r)dxdr <
A+^r / / ul(x,r)dxdr+Jo Jo
f 4 Jo
u2{x,r)dx
where
A(«)0M)= f *'' /'(OfdeJv Lemma 4.2.8 follows from (4.49) and all of the estimates obtained. Lemma 4.2.9. [f(v)6]2(x,T)dxdr
/ / Jo Jo
Proof. By integrating (4.3)2 over [0, x] for any x G [0,1] and using the bound ary condition (4.6), we obtain that [f(v)0](x,t) = [jji{v)ua!]{z,t)-(j
u(y,t)dy\
,
x G [0, l],t € [0, +oo).
(4.53) Multiplying (4.51) by f(v)0 and integrating it over [0,1] x [0,tf], we arrive at [f(v)0]2(x,T)dxdr
/ / Jo Jo =
1 1 JO
[fl{v)uxf{v)0}{x,T)dxdT
JQ
rt
~JoJo
/AKA\
fl
f r rx {[Jo
-i w(2/
'rH
We estimate each term separately.
>! / ( * ) ( ' - * ) } (*>r)d*dr
(4.54)
The influence of dissipation mechanism...
135
In view of (4.41), Lemma 4.2.8 and Cauchy's inequality, / / \Jo JO
[fi{v)uxf{v)e}{x,T)dxdr\ I
\jj[f{v)e)2{x,T)dxdT+K,
<
te [0,4-oo).
By (4.41), Lemma4.2.7, Lemma 4.2.8, (4.50), (4.53) and the Cauchy inequal ity,
III {[I^^H /(v)^-^)}(x>r)^r| u
(V'r)dy)
<
faj
JQ \(J*
] (*,r)dxdT + Aj
<
ij
I [ / W ^ ] 2 ( x , r ) ^ c f r + A,
J
Ol{x,T)dxdr
*e[0,+oo).
Integrating by parts and using (4.3)i, (4.17), (4.45), Lemma 4.2.8, (4.51), (4.52), (4.53), Holder's inequality and Cauchy's inequality, we obtain that \ j j <
\\j*
A+ A j
j
+
4
u{y,r)dy\ «*+/
/ ( » ) ? } (*,r)(tedr| u2(s,r)ds\{x,T)dxdT
u(y r)dy m
2(s r)ds {x r)dxdT
&+I/ r i [£ ' u{y r)dy\ trm+^u{y'r)dy nv)u \i ' A
* * ijf f {[or ' ) r or ' ) 2
.j
<
A
+ g /
-u {s,T)ds\{x,T)dxdr\
/
[f{v)0)2{x,T)dxdr.
Lemma4.2.9 follows from these estimates on each term in (4.54). Lemma 4.2.10. / u2(x, t)dx -* 0 as t -> +oo. Jo
*\
136
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
Proof.
It is clear from (4.45) that /•+OO
/ Jo
*i
u2{x,t)dxdt
/ Jo
Namely, / u2(x,t)dx etfdOt+oo)). Jo Multiplying (4.3)2 by u and integrating over [0,1], we obtain, with the help of (4.37) and (4.6),
I f1
I
/ (uut)(x,t)dx\ \Jo I < A [ [(f(v)0)2 + ul](x,t)dx. Jo Combined with Lemma 4.2.8 and Lemma 4.2.9, it implies rii / u2(x,t)dx\dx \dt Jo I
/ Jo
< A.
Thus Lemma 4.2.10 follows. Lemma 4.2.11. 0(i) = / 0(x, t)dx -*-?r, Jo &v
as t -> +oo
I I [/(")]2(*> r)dxdr < A, t e [0, +oo) Jo Jo \\f(v)(; t)\\L'[o,i] -* 0, as t -+ +00 ||p(»,tf)(-,OIU'[o,i] = ll(/(«)tf)(-.«)llL«[o,i]-+0, a s i - j . + c o .
(4.55) (4.56) (4.57) (4.58)
PTW>/. (4.55) follows from (4.52) and Lemma 4.2.10 directly. It is known from (4.55) that there exists To > 0 such that
^ ) > 2 § 7 as<>T0. Together with (4.41), Lemma 4.2.7, (4.50) and Lemma 4.2.9, it implies /
[f(v)]2(x,r)dxdr
/
JTo JO
<¥$-[' t\f{v)e]2{*,T)dxdT &0
<
A.
JTo JO
The influence of dissipation mechanism...
137
This estimate and (4.41) yield (4.56). To prove (4.57), we use (4.3)i, (4.41), Lemma 4.2.8 and (4.56) to make the following estimate
r+oo
<
A+ A / Jo
rl
/ Jo
u2x{x,t)dxdt
This with (4.56) together implies (4.57) directly. It is known that II(/(»)*)(-,<)||LI[O,I]
< (jT[/(f)]2(:M)
.
This, combined with Lemma 4.2.7 and (4.57), gives (4.58). Thus, part I of Theorem 4.2.1 is established by the above Lemmas. Next, we employ an idea of Andrews and Ball (see [3]) and the above results to prove part II of Theorem 4.2.1. Suppose tf G £ 2 [0,1] with tf > 0 and $ G C2([v,V]) satisfying *'(*)/(*) > 0 for z G [v,V]
(4.59)
where v and V are the lower and upper bounds of v. Let
^t):=IomJ^§dy
fr
° **W*°-
Multiplying (4.3)2 with
■jr--"[)T'«^f*]*-r-<"Lf'«w*]* - /
/
U XT
( ,)
\f
+ / V(x)$(v(x,i))dx Jo
*(y)(^y)
- / Jo
(v,T))ug(y,T)dy\dxdr
V(x)$(v0(x))dx
where ' denotes the differentiation with respect to v.
, (4.60)
138 Quasilinear Hyperbolic Systems and Dissipative Mechanisms To show the existence of the limit of the left hand side of (4.60) as t -> +00, we estimate each term on the right hand side of (4.60). For the first term, it is easy to see that
\f\{x>t)\f\{y)%^Mdy)]dx\ \Jo
[Jo
t*(v{y,*))
J I
< |K,t)||L»[0,l]- IWUW] • l l t / j ' / t f l II pyyy-'1)) IIL2[O,I] <
A|M-,t)|| L a [ 0 | i]
which tends to zero as t —> 4-00, due to Lemma 4.2.10. The third term can be treated as follows:
If1 /
\Jo
\r u{x,t)
/
[Jo
r$'Mi' tf(y)
—Vf v
[ V>\ ) J
< IM-,*)IMI»IIL» • [j$\ IIL
<
1 I {y,t))u*(y,t)dy\
(-,*)) J
A[||u(. I <)||2 3 + ||u,(.,<)lli a ]
J
dx\
I
• \MS)\\»
IIL°°
forall<>0.
Therefore, the limit of the third term as t -> +00 exists by (4.45), Lemma 4.2.8 and the dominated convergence theorem. It is obvious that the term fio j ^(x)^(v(xit))dx
Jo
_
is uniformly bounded in
t > 0 since v < v(x,t) < V. Thus, the above estimates imply that f* f1 &(v) / / {[f{v)0]V{x)^T-T}(vx> r)dxdr is bounded uniformly in t > 0. Jo Jo V>\ ) This, together with (4.59), (4.4) and 0 > 0, yields the existence of t lim
^°°Jo
/
/
[f(v)0V(x)Q'{v)](x,T)dxdT.
Jo
Furthermore, the existence of lim
/
tf(z)$(t;(x,*))cte,
t->+co JQ for all tf e L2[0,1] with tf > 0, is established since each term in (4.60), apart from / ^(ar)$(v(x,t))c?x, is either independent of t or tends to a limit as Jo t -> +00.
The influence of dissipation mechanism...
139
Therefore, it follows that *(«(•, * ) ) - ^ » ( 0
in
L2[0,1]
as t -> +00 for some g$ G £ 2 [0,1]. In view of ||(i;(-,*))||Loo < A, we can show"that <7*eL°°[0,l] and ^(•,0)A^(-)in^°°[0,l].
(4.61)
Now, let $ G C([i7,F]) be arbitrary and # G L1^, 1). By using the same method in [3] and the following Lemma 4.2.12, it is easy to verify that lim / - ° Jo
t fo
V(x)$(v(x,t))dx
exists for all # G Lx[0,1] and $ G C{[v, V]). Lemma 4.2.12 can be proved by the same argument as used for Lemma 3.1 in [3]. Lemma 4.2.12.
Let / G C(R) and let 0 < v < V. Then the set
S = span{$ G C2{[v,V}) : &{z)f(z)
> 0 if z G [v,V]}
is dense in C([v, V\). Thus, it turns that (4.61) holds for an arbitrary $ G C(\y, V]). The existence of probability measures vx follows from (4.61) and the Theorem 5 in Tartar's paper in 1979 ([174]). To prove that supply C K = {z : f(z) = 0} a.e., it suffices to show if $ is zero on K then (i/ r ,$) = 0 a.e.. But, if $ is zero on iv, then $(v(-,t)) —>• 0 in measure as / -> +oo due to (4.57). Therefore, $(i;(-,t)) —*■ 0 in L°°[0,1] as t —> oo, and hence {vXi$) — 0, a.e., as required. Thus, the proof of Theorem 4.2.1 is complete. Corollary 4.2.3 and 4.2.4 can be proved in the same way as in [3]. Next, we prove Theorem 4.2.5. In view of (4.40), (4.41) and the smoothness of f'{v)> there exists a constant 6 > 0 such that - / » > 6 > 0 for ve[v,V). (4.62) Lemma 4.2.13.
If (4.40) holds, then
I vl(x,t)dx+ Jo
I I (v2x + 0v2x){x,T)dxdT< Jo Jo
A,
*G[0,+oo).
140 Quasilinear Hyperbolic Systems and Dissipative Mechanisms Proof.
Define M(v) = I p>(£)d£ and consider M(v) as a function of x and Jv t , we may rewrite (4.3)2 as [u-M(v)x]t
=
[-f(v)0]x.
Multiplying the above equation with [u — M(v)x] and integrating then over [0,1] x [0,f], we arrive, with the help of the estimates obtained in the proof of Theorem 4.2.1, at I I ev2x(x,T)dxdT
/ vl(x,t)dx+ Jo
*£[0,+oo).
Next, it is known from (4.55) that there exists a To > 0 such that $(t) = / 0(z,t)dx = 0(z{t),t) > T^- > 0 Jo ^v
for t > T 0 .
Then, we can show by Holder inequality that
IP
, / r1 &
r
\
1/2
1
which implies
\tle2
OWfaQ+l
i1/2
-£(x,t)dx\ Uo & J
>A>0,
for*>T0.
By using Cauchy inequality, we have *(*>*)+/
-£(*>*)**> &>0,
Jo "
tG[T 0 > +oo).
Therefore, vl{x,t)
+ vl{x,t)
/ JO
-£(x,t)dx}9
*<E[r 0 ,+oo).
V
This, combined with Lemma 4.2.6 and (4.63), implies /
/
JTQJO
v%(x, r)dxdr < A,
te [T0, +oo).
(4.63)
The influence of dissipation mechanism...
141
This estimate and (4.63) yield Lemma 4.2.13. Lemma 4.2.14. If (4.40) holds, then there exists a T* > 0 such that / (u2x + 0l){z,t)dx
/ [u2xx + 02](x, r)dxdr < A for t > T
+ /
JO
JT* JO
and lim
/ (u2x + B2x){x, t)dx -» 0,
as t -> +oo.
t-++oo JQ
Proof. Multipling (4.3)2 with {-f{v)0 + ft(v)ux)x (4.41), (4.42) and Cauchy's inequality, we get \il{v)uxx)2{x,T)dxdT+^j^
<
\J
\jk{v)ul]{z,T)dz + £ j
+ J
J
+A / +A /
and using (4.3)i, (4.6),
{fL{v)u2x){x,t)dx
[±p(v)u*]{z,T)dzdT
[f{v)0uxt]{x,T)dxdT+^
J
J
ulx(z,T)dxdT
[max«2(.,r) / i^(x,r)
dr,
02x{x,r)dxdT
/
for any t > T > 0.
To estimate the terms above, we first give an estimate on / / ux(x, r)dxdr, Jo Jo which plays a key role in the following estimates. Due to W1*1 «->• L°°, it follows /
/ ux(x,r)dxdr
<
JT JO
/ / [u2x + u2xx] (x, r)dxdr,
< A sup / ul(x, r)dx re[T,t]Jo
UTJO
for any t > T > 0.
J
Similarly, we can show, with the help of W1,1 «->- L°° and Cauchy's inequality, that / maxu 2 (',T)dr [o,i]
JT
<
A(<$) /
/
JT JO
u2xdxdr + 6 f
f
u2xxdxdr,
JT JO
for any t > T > 0 and S > 0.
142 Quasilinear Hyperbolic Systems and Dissipative Mechanisms It reads from (4.50), (4.52) and Lemma 4.2.7, that
maxO(',t)
I 0l{x,t)dx) \Jo J
,
for t > 0.
Moreover, Lemma 4.2.7 and Lemma 4.2.13 yield
/ [62(x,t) + v%(x,t)]dx < A, Jo
for t > 0.
Using these estimates in which S can be chosen suitably, we can arrive, by the standard method (the details can be found in [76]), at
<
A / [(f(v)0)2 + ux){x, T)dx + A [ /
L/0
Jo
\f(v)0\(x, t)dx
+ (j[ \f{v)0\{x,t)dx^ ] + A / / [vl + ul + 0l](x,T)dxdT + A sup / JT JO re[T}t] Jo [£j\ul + !%■ £
+
ul(x,r)dx
ulx)dxdrj
6l(x,t)dx
+ Q- J* J' e?{x, r)dxdr.
Multiplying (4.44) by $t and integrating over [0,1] x [T,t], we get, with the
The influence of dissipation mechanism...
143
help of the obtained estimates and Cauchy inequality,
fy J\l(z,t)dx+!f f f 0*(*,T)dzdT <
Jo
+A I
JT
t0'1!
I uj.(x,T)dxdT + A I
JT JO
<
92{x,T)dx]dr
h- f 0*{x,T)dx + A f maxt|2(-,r) ( f
fyj
\Jo
J
e2x\ux\dxdr
I
JT JO
92x{x,T)dx+!±J
J
u2xx{x,T)dxdr+\j
J
u2x{x,r)dxdr
+A / / Oldxdr + A sup | / {u2x + 02x){x,r)dx\ JT JO T€[T,t] Uo J \j
j (u2x + u2xx)(x, r)dxdr\
for any t > T > 0.
The above two inequalities imply /
(ul + 0l)(x,i)dx+
Jo
f
(u2xx+0?)(x,T)dxdT
[
JT JO
< A / [ul + $l + (f{v)0)2](x,T)dx+ Jo + ( J T ' \f{v)0\{x,t)dz}
f Jo
\f(v)0\(x,t)dx
+ j f J \ v l + Ul+
0Z){X,
+ sup f / {u2x + e2x){x,T)dx\ iff (ul + re[T,t] Uo J UT JO for any
T)dxdT
kgjnvhjg
ulx){xtT)dxdr\ J
i > T > 0.
Due to Lemma 4.2.13 and the estimates obtained in the proof of Theorem 4.2.1, it is known that r+oo
/ Jo
rl
/ {[f{v)9]2 + ul +
vl+0l}dxdr
Jo
and / \f(v)0\(xyt)dx Jo
-+ 0 as
t-++oo.
144 Quasilinear Hyperbolic Systems and Dissipative Mechanisms Therefore, for any e > 0, there exists a T\ > 0 such that it holds A f{[f{v)9)2 Jo A< / A /
+ ul+OlUx^dx
\f(v)0\(x,i)dx
+ I/
< e |/(w)0|(x,t)
(4.65) ><e
for any t > Tx
2 [ [vl + v% + 0x}dxdT < e for any t > 7\.
(4.66) (4.67)
For convenience, we assume that the constant A in (4.64)-(4.67) satisfies A > 1. We choose e so small that e< ^. (4.68) It follows from (4.65) that
Jjul+eDix^dxK^Ke. Let us define T2 = sup \t : sup / (u2x + $l)(x, r)dx < be > { re[Tut]Jo J and show T2 = +oo. Suppose that T2 < +oo. By taking t = T2 and T = T\ in (4.64) and using (4.65)-(4.68), we have / (0l + ul)(xyT2)dx + \ I 2 / u2xxdxdr+ Jo * JTI JO <
[
2
/
02dxdr
JTX JO
2
3e + be < 3e + § < 4e.
Namely, / (0* + t4)(z,T 2 )cte < 4s, which contradicts the definition of T 2 . Jo Then, T2 = +oo. This implies that / (el + u2x){x,t)dx < be for t > Tx. Jo It yields, due to the arbitrary smallness of e, that / (0l + «£)(*, t)dar -> 0 as t -► +oo.
(4.69)
The influence of dissipation mechanism...
145
The first estimate of Lemma 4.2.14 can be obtained from (4.64), (4.69) and the above arguments. Thus, the proof of Lemma 4.2.14 is finished. Now, we prove Theorem 4.2.5. It is proved that ||u(-,*)||#i[o,i] -» 0 as t - K + O O . It is known from the part I of Theorem 4.2.1 that / [fiv)]2 (x> t)d* -> 0 as t -> +oo. Jo On the other hand, using (4.62) and the mean value theorem, we can show that / [f{v)]2{x,t)dx>b2 Jo
f Jo
{v-v)2(x,t)dx
where v is the unique root of f(v) = 0 in [v,.t^]. Thus, f (v- v)2(x, t)dx -► 0 as t -> +oo. Jo
(4.70)
/.+OO
Furthermore, it follows from Lemma 4.2.13 that / Jo -foo, while Lemma 4.2.14 and (4.3)i imply r+ool
/ JT*
J
/»1
I
i»+00/»l
T" / vl(x,t)dx\dx
r+OO/-I
/ u2xxdxdt < A < +oo.
/ vldxdt + A
I
rl
/ v2dxdr < A < Jo
JT* JO
JT* JO
Therefore, /
t£(s,t)dx-+0ast-»+oo.
(4.71)
Jo (4.70) and (4.71) yield
IK,*)-*ll!rt->o. It is known from part I of Theorem 4.2.1 that f1 0(t)=
0(x,t)dx-+ Jo
E —5.,
ast-H-oo.
^y
This, combined with lemma 4.2.14, implies F [ f1 l1/2 E ||*(-,t) - —HL 2 [O,I] < / 0l(x,t)dx\ + \\0(t) - - S - | | L a -» 0 as t -+ +oo cv Uo J ^v
146 Quasilinear Hyperbolic Systems and Dissipative Mechanisms and furthermore ||0(.,*)
-\\Hi cy
-> 0
as t -> + 0 0 .
So far, we have proved that all of (v — v),u and (9 — W2-) become small in i^-norm for large t. Thus, the arguments similar to those used by Okada and Kawashima in [153] can be used to obtain the exponential convergence of {v, w, 6} to the constant state {i), 0, T ^ - } as t -> +oo. We omit the detail. A similar result on large-time behavior of solutions has been established by Hsiao and Jian in [67] to deal with the following boundary condition
*(0,t) = iu{0,t),tr(l,t)
=
-iu(l,t), (4.72)
0(0,*) = 0(1,*) = T0,t> 0 where 7 = 0 or 7 = 1, and To > 0 is the reference temperature. Physically, the case with 7 = 1 represents that the endpoints of the interval [0,1] are connected to some sort of dash pot. The global existence of classical solutions to (4.3), (4.7), (4.72) for the same kind of solid-like materials as in [24] has been obtained by Jiang [86]. 4.3
Various large-time behavior corresponding to different bound ary conditions
The difference in the large-time behavior caused by the difference in consti tutive relations has been discussed in the last section. This section is devoted to understand the different phenomena caused by different boundary conditions. Such understanding is quite important not only in mathematical theory but also in applications. For instance, the outer pressure problem, namely q{0,t) = q(l,t) = 0, (4.73)
=
where the outer pressure R(t) is a given function, possesses a completely dif ferent large-time behavior comparing to the problem with boundary conditions g(0,t) = g(M) = 0, a(0,t) =
The influence of dissipation mechanism...
147
The global existence result for the outer pressure problem has been obtained by Nagasawa [148] and Luo [129], corresponding to the case of an ideal gas and real gas respectively. The asymptotic behavior of solutions to this problem, in the case of an ideal gas, has been established in [149] where e = c0,
V
(4.74) V
c, R, fi and K are positive constants. A real gas is well approximated by an ideal gas within moderate ranges of 6 and v. However, (4.74) become inadequate for high temperature and densities, since the specific heat, conductivity and viscosity vary with temperature and density. Our objective here is to establish the result on the asymptotic behavior of solutions to (4.3) (4.7) (4.73) with the following constitutive relations which are more realistic than (4.74) for a real gas.
+ !±l2Ji.ux
(4.75)
= -k±^lex v e = i(v,0)
(4.76)
q
(4.77)
where e(v, 0),p(v, 0), //(v, 0) and k(vy 0) are twice continuously differentiate on 0 < v < +oo and 0 < 6 < oo. Since internal energy may only take non-negative values, we assume c(v, 0 ) > O
0
O<0
(4.78)
For compatibility with the second law of thermodynamics (4.4), we need l*{v,0) > 0 , * ( M ) > 0 a nd (4.10). Finally, we impose upon e(v, 0), p(v, 0) and k(v, 6) the following growth con ditions: There are exponents r £ [0,1], s > 2 + 2r and positive constants ^,&0,p,-(i = 1)2,3,4), and there are positive constants N(v) and ki(y) for any v > 0, such that for any v > v and 0 > 0 *(1 + 0r) < ee(v, 0) < N(v){l + 0r) + e'+r}^^,^^ P2[i + (i-i)e 2 V
(4.79) 1+r [ / + ( i / ) 0 + 0 ] Pl
SPv\V,V)S ■
2
'
V
I = 0 or / = 1 (4.80) 1+r 1+r p 4 p + ( i - Q g + g ] < p{Vj e) < PsP + ( i - Q g + g ] ?
148 Quasilinear Hyperbolic Systems and Dissipative Mechanisms / = 0 or / = 1 \Pe(v,O)\)
(4.81) (4.82) (4.83)
IMw, e )\ + \kw{v,e)\ < *i(a)(l + **)•
(4-84)
The above assumptions are motivated by the facts in [7] and [192]. In these papers, it is pointed out that e grows as 0 1 + r with r « 0.5 and /: increases like 05 with 4.5 < s < 5.5. For technical reasons and simplicity, we only discuss the case of fi(vi0) = no>O.
(4.85)
We now turn to the assumptions on the outer pressure R(t) and the initial data. We assume that mR =
inf
R(t) > 0
(4.86)
t€[0,+oo) r+oo
TV(R) = / \Rf(t)\dt < +oo. Jo (4.87) implies that there exist R and MR such that
(4.87)
R = lim R(t) > 0 and 0 < MR =
sup
R(t) < +oo.
t€[0,+oo)
Without loss of generality, we may normalize, by superimposing a trivial rigid motion, the initial velocity as (4.15) in the previous section, namely / uo(x)dx = 0. Jo We assume that the initial data are compatible with the boundary conditions (4.73). The global existence of smooth solutions to the initial boundary problem (4.3) (4.7) (4.73) has been established in [129] under the above assumptions. Namely, T h e o r e m 4.3.1. Assume v0(x), v'0(x), tio(#), u'o(#), u" 0 (x), 0o(s), 0'o(x), 0"o(x) are in C a [0,1] for some 0 < a < 1. Moreover, v0(x) > 0, 00(x) > 0, 0 < x < 1. Under conditions (4.75)-(4.87), (4.15), there exists a unique solution {«(*,*), v(x,t), 0(x,t)} to (4.3) (4.7) (4.73) on [0, l]x[0,oo)such that, for every T > 0, v, vgJ vtj vxt, «, uXl ut, uXXj 0, 0Xi 0U 0Xx are in C a ' a / 2 ( Q T ) and vtt, uxt,
The influence of dissipation mechanism...
149
0xt are in L2(QT), QT = [0,1] x [0,T). Furthermore, 0{xyt) > 0, v{x,t) > 0, for xe [0,1], te [0,oo). The following theorem on the asymptotic behavior of solutions is established in [77] T h e o r e m 4.3.2. Under the same assumptions made in Theorem 4.3.1, it holds, for the globally defined solution of (4.3) (4.7) (4.73), that 0
forz E[0,l],*E[0,oo),
(4.88)
where v and v are positive constants independent of tf, / u(y,t)dy -> 0 as t -)■ oo uniformly with respect to x E [0,1], Jo
(4.89)
< v(x,t) — / v(x1t)dx > —> 0 as t —>» oo uniformly with respect to x E [0,1], (4.90) /
R'(r)
lv{x,t)-&
v(x,r)dxdr
4
|
Y(r)e(v,e)(x,T)dr
\
/*oV(*)
|
+^
ii
l_>£l
as t —^ oo uniformly with respect to x € [0,1], where Ei=f
[i(vo(x), 60(x)) + ^ M +
Y{t)=exp
{£^rdT}>
i? (4.91)
R(0)Vo(x)]dx
'*°-
Moreover,
j f IFW "(x' ^k^' *)(a5'r)" R{r)]dT~* °
(4-92)
as t —> oo uniformly with respect to x E [0,1]. Remark 4.3.3. In contrast with the case of R(t) = 0, in which the specific volume v satisfies v(x,t) > c(log(l +t))k with some constants c> 0 and k > 1 for the ideal gas [147], (4.88) discloses that v(x,t) is uniformly bounded even for the ideal gas if the boundary condition is (4.73) instead of (4.6). (4.91) describes the asymptotic relation between v(x,t),0(x,t) and R(t) as t
~*00' Y(T) (4.92) can be understood as follows. With a weight function y W v ( a : , r ) , the weighted average of \p(v, 0)(x, T)-R(T)] over [Q,t] tends to zero as t -)• +oo,
150 Quasilinear Hyperbolic Systems and Dissipative Mechanisms uniformly in x £ [0,1]. For the weight function, it is known that vexp[-Mr(t Remark 4.3.4.
Y(T)
- r)] < -y77}v(x>T) ^ vexp[-mR(t
- r)].
It follows from (4.15) that /
u(x, t)dx = 0 for t > 0.
(4.93)
Jo For the proof of Theorem 4.3.2, we need a sequence of estimates in which c denotes a generic constant that does not depend on the time t. We integrate (4.3)3 over Qt = [0,1] x [0,tf] and use the boundary conditions (4.73) and the initial conditions (4.6) to get pi
tt
2
rl
/ [(cMH-5-)(x 3 r)+R(t)v(x,t)]dx = E1 + R'{T) v{x,r)dxdr. Jo * Jo Jo Using GronwalPs inequality and the assumptions on R(t), we obtain Lemma 4.3.5.
r1
(4.94)
u2
/ [e{v,6) + — + Jo *
v]{x,t)dx
Lemma 4.3.6.
Proof. Let i>(v,6) = e(v,0) — 0r){v,Q) denote the Helmholtz free energy function. It is known from (4.5) that
tl>v(v,0) =
*
# M )
=
(495)
_ £ ^ .
Define E(v, 0) = 1>{v, 0) - tf(l,1) - ^ ( 1 , l)(v - 1) - ^ ( w , 0)(0 - 1). On account of (4.3), (4.75) and (4.95), we deduce, after a straightforward cal culation, that
The influence of dissipation mechanism...
151
Integrating this equation over Qt and using (4.73), (4.85), (4.86) and Lemma 4.3.5, we arrive at
| V,•) + $(..,)*+ Jl jf [<£ + ^ » > ] (.,r)** <-R(t)
R'(T) v(x,T)dxdr + p(l,l) v(x,t)dx + c < c0 Jo Jo Jo (4.96) where Co > 0 is a constant. In view of (4.79), (4.80) and (4.95), it follows from Taylor's theorem that Jo
v(x,t)dx+
E(v, 0) -
= ( ^ - i ) 2 / (i - c)iMi + C(«-i), i K >o Jo #,»)^Ki)-(«-i)^(f,»)
r i/(6» — log^ — 1),
= ^-logtf-l) + J
r =0
_L_,T
l l T T +i L H
, r>0.
With the above two inequalitites, we end up with the estimates E(v,0)>v{0-log0-l),
r=0
(4.97)
or E(v, 9)>c6 + c01+r - c + i/(0 - log0 - 1), 0 < r < 1. This, together with (4.96), yields lemma 4.3.6. Lemma 4.3.7. Let a and (3 be the two positive roots of equation
(4.98)
0-log0-l =— where Co is the same positive constant as in (4.96). Then, there exists a a(t) 6 [0,1] for each t > 0 such that 0< a<0(a(t),t)0 that
a< I 0(x,t)dx 3. Jo
152 Quasilinear Hyperbolic Systems and Dissipative Mechanisms Proof.
(4.96)-(4.98) yield
v I [0{x,t) - logO(x,t) - l]dx < c 0 , t > 0. (4.99) Jo This implies that for each t > 0 there exists a a(t) £ [0,1] such that 0(a(t),t) — \og6(a(t)1t) — 1 < £*, from which the first estimate of Lemma 4.3.7 follows. Using (4.99) and applying Jensen's inequality to the convex function U — log U — 1, we deduce / Jo
0(x,t)dx - log / 6(x1t)dxJo
1 < —, v
which implies the second estimate of Lemma 4.3.7. Lemma 4.3.8. There exist positive constants v > 0 and v > 0, independent of t, such that for any (#, t) £ [0,1] x [0, oo) it holds v < v(x,t) < v. Proof.
We integrate (4.3)2 over [0,x] x [0,tf] and use (4.3)i to get rX
i»X
I u(y,t)dyJo =
I Jo
u0(y)dy
/ [-P{v, e){x> T) + R(r)]dr + fio log v(x, t) - /i 0 log v0(x). Jo
Furthermore, taking exponent to the above equality, we deduce that
^exp{^J0p{v'e){x'T)dT} where
Y(t) =exp| ±j*R(T)dTy Due to this expression, it follows that
^ e x p {^/^ M ) ( x ' r ) r f r }
The influence of dissipation mechanism...
153
Intergrating the above equation over [0,2], we arrive at exp
^o /
P(M)(s,r)dr-l
Thus, it follows 1 vo(x) H w
«'■"-
t* / £(s,r)Y(r)p(v,0)(a?,r)t;(x,r)dr
*
B(..oy(o
'
(4100)
Next, we establish the upper bound of v. With the help of Lemma 4.3.6, it is easy to estimate B(x,t) as follows 0 < c _ 1 < B(x,t)
for (x,t) G [ 0 , l ] x [0,oo).
Using Lemma 4.3.5, Schwarz's inequality and the condition s > 2 + 2r, we can show that
I^OM) - <^(a(t),i)| = f ^Mv,*)**
a
1
er~l B2 \ ^ 2 / r1 —?-dx\
I
K«(*) v(x,t)dx)
where V{t) = f ( ^ + ^ * f ^2)0M)
I
\1/2
(4.101)
f o r w h i c h lt h o l d s t h a t
/ ^ ( r ) d r < c for any < > 0, Jo on account of Lemma 4.3.6. This and Lemma 4.3.7 yield, ^ -
(4.102)
" cV{t) < 0r+1 (x, t) < 2/? r+1 + V{t)
(4.103)
for ( x , t ) € [ 0 , l ] x [0,oo). Using (4.81), we deal with vp(v,0) to get p4[l+{l-l)e+01+r]
/ = 0 o r / = l. (4.104)
154 Quasilinear Hyperbolic Systems and Dissipative Mechanisms Finally, (4.100)-(4.104) imply
v(x,t)
jexp(^y
< c + c I exp
fc^(t
T
E(O^H
- T)] dr + c f [exp [ : = - ^ ( * - r)l | V{r)dr < c.
(4.105) Now, we estimate the positive lower bound of v. First, there exists b(t) G [0,1] for any t > 0 such that / 0(x1t)dx = 6(b(t),t). Jo Thus, due to Lemma 4.3.5, Lemma 4.3.7 and (4.105),
ou>(x,t)= ew{b{t),t)+r
O
r1
> Namely, 0 2(x,t) ^
+
^TT^v
Jb(t) 20 i ^(t/,*)
x'2(t)
\1/2 T
/ rl f)2 \ 1 / 2 1
c{l-Vll2{i)). > c. This gives,
cV 9(x,t)>c(l-V{t)).
(4.106)
(4.100), (4.102)-(4.104) and (4.106) yield, v(x,t)>
cJ >
c f
| e x p (1- JT R(S)dA 1 {p4[l + (1 - 1)9 + el+r]} (*, r)dr Lxp\-^-(t-r)]\{c-cV(T)}dT. (4.107)
In view of (4.102), we claim that
<"~ / {6XP [ ^ ~ T)] } V(T)dr = °-
(4 108)
'
The influence of dissipation mechanism...
155
In fact,
which tends to zero as t -» oo. It is proved in [129] that t>0M) > v ( t ) > 0 .
(4.109)
On account of (4.107)-(4.109), we end up with v{x,t) >v>0
for any (x,t) G [0,1] x [0,oo).
Lemma 4.3.9. / vl(x,t)dx+ Jo Proof.
f f Jo Jo
{l^01+r)vl{x,T)dxdr
With the help of (4.3)i, we rewrite (4.3)2 into the form {u-fio—
)t+Pv{v,0)vx
= -p9(v,0)0x.
(4.110)
v Multiplying (4.110) by (u - ^o^f") and integrating over [0,1] x [0, t], we get, due to (4.80), (4.82), lemma 4.3.6 and the condition s > 2 + 2r, that / {^(log v)x - u}2dx + ex f I [I + (1 - 1)0 + 01+r]v2xdxdr Jo Jo Jo <
*L [ [ [l + Q-ty Jo Jo
+
+ oV+^vldxdT
+ c f [ {l + 61+r)v2xdxdr Jo Jo
% I [ (l + °1+r)vldxdT+c(4.111)
By virtue of (4.93), there exists d(t) G [0,1] for any t > 0 such that u(d(t),t) = 0.
156 Quasilinear Hyperbolic Systems and Dissipative Mechanisms Hence, u(x,i) can be expressed as u(x,t) = / ux(y,t)dy. Jd(t) This, together with Lemma 4.3.6 and Lemma 4.3.8, yields
/
max u2(x,r)dr
JO *€[0,1]
<
I [ /
ux(y, r)dy J dr
Jo \Jd(t)
J
*i:i(i»i>¥
(4n2)
< c. Combining (4.112) with Lemma 4.3.6, we obtain that, /
/ (l + Or)u2dxdr
J
)t , fi v (4.H3) < / max u2 f / (l + Or)dx ) dr < c. ~ Jo *€[0fl] \JoK ) ~ For the case of / = 1 in (4.80), the desired result of Lemma 4.3.9 follows, on account of Lemma 4.3.6, Lemma4.3.8, (4.111) and (4.113). The case of / = 0 can be dealt with as follows. Using Lemma 4.3.6 and recalling that s > 2 + 2r, we have / fpe{v,e)Ox^o'—dxdT\
I {l + 0r)\vx\\0x\dxdT Jo Jo
e j f {0 + 01+r)v2xdxdT + c Jo Jo
<
(4.114) where e is a small positive constant. By a similar approach as in (4.111), with the help of (4.113) and (4.114), we obtain, / {^o(logx;)x - u}2dx + d / / {0 + Jo Jo Jo <
f
01+r)v2xdxdr
/ / {0 + 01+r)v2xdxdT + e f f (0 + 01+r)v2xdxdr + c Jo Jo Jo Jo
which implies / v2xdx+ f f {0 + 01+r)v2xdxdr
(4.115)
The influence of dissipation mechanism...
157
In view of (4.106) 0{x,t) + cV{t) >c. Thus, vl{x,t)
+
cV{t)vl{x,t).
This, together with Lemma 4.3.6 and (4.115), gives / / v2x(x,t)dx
I I 6v2xdxdr + c f V{r) f v2x{x,T)dxdr < c Jo Jo Jo Jo
which, combined to (4.115), implies the Lemma 4.3.9. Lemma 4.3.10. / u(y,t)dy —> 0 Jo Proof.
as t —> oo, uniformly with respect to x G [0,1].
Integrating (4.3)2 over [0,x] and using (4.7), we obtain that \[Xu(y,t)dy\ Uo
Multiplying (4.116) by /
=[-^,l?) + ^ ) ] + /io-.
(4.116)
v
It
u(y,t)dyy and integrating over [0,1], we arrive at
Jo
* {jfGT ■*)'*}. =
[ [-P{v,0) + R]([
udy\dx
+ no I Y ( /
ud
y)dx-
Since / udy vanishes at x = 0 and x = 1 due to (4.93), the second term Jo of the right-hand side turns into Uo Jo
— /
udydx = -fi0
v J0
/ J0
—dx -f /i 0 / V
Jo
—s- / V
udydx,
Jo
with the help of integrating by parts. Combining this expression with the esti mate /
udy\ < ( / u2dx j
158 Quasilinear Hyperbolic Systems and Dissipative Mechanisms we get
dy dx +c l 2dx {fAFr J2 }t ~ fr u2dx\
{ / (1 + 01+r)dx + f (v2x+ u2)dx\ .
This, with lemma 4.3.6 and the following inequality / u2dx > I
[ I udy 1 dx,
yields
{l!Uoudy)
}t^cJ!Uoudy)dx
1/2
x
<
dx
2
\ 2
c(fu dx)
j l + / {u +
jkfhbcgf ( 4
-U7)
2
v x)dx\.
Define
z(t) = I ( I u2dy\ dx. (4.117) means z'(t)
+ cz(t)
I u2dx\
11 + / (u2 + v2x)dx\ .
Multiplying the above inequality by exp(ctf), and integrating over [0, t], we obtain z{t) < e" c t z(0)+c / e - ^ " ^ ( f u2dx] Jo \Jo )
dr+c [ e " ^ " ^ / {u2+v2x)dxdr. Jo Jo
With the help of (4.112), Lemma 4.3.9 and the following Lemma 4.3.11 estab lished by Nagasawa in [149], it can be easily proved that z(t) —>• 0 as t —> oo. Namely, I
(
u2dy\
cte->0 a s * - > o o .
(4.118)
Lemma 4.3.11. Let X(t)(> 0) and w(t) be continuous functions satisfying that there exist positive constants Ci(i = 1, 2, 3,4) such that d e C 2 ^ - r ) < exp | / u(Z)dt\
< c3ec^-T\
for 0 < r < t.
The influence of dissipation mechanism...
159
Denote A(t) by A(r)dr. Then c" 1 Jim A(t) < Jim / e x p i - / <*{€)
t-+oo JO
<
K
JT
J
TmT / exp | - / w ( 0 d 6 | A(r)dr
< d i m A(*). Now, we are ready to finish the proof of Lemma 4.3.10. Using embedded theorem W 1 ' 1 ^ , 1) *-+ L°°(0,1), we have sup ( / u(y,t)dy) *€[0,1] WO
u(y,t)dy)
\./o
dx /
\2 \l/2 n1
frwr
+cl /
I JO
( / u(y,t)dy\
dx\
\1/2
• f / ti2(x,f)cfxj
(4119)
.
The proof is completed from (4.118), (4.119) and Lemma 4.3.6. Lemma 4.3.12. v(xjt) — I v(x,i)dx —> 0, as t —> oo uniformly in x G [0,1], ./o
/ Y(r) f
'<*•*>+ •*■ /•oo
-*
#1+ / ^
e(v(x,T),0{x,T))dxdr
JSL
-TOYW)
rl
# ( r ) / v(z,r)cteciT ==P , as t -> oo uniformly in x G [0,1],
uniformly in x G [0,1]. /• + OO
Remark 4.3.13.
The existence of integral / Jo lows from (4.87) and lemma 4.3.8.
/»1
R'{T) I v(x,r)dxdr Jo
fo\-
160 Quasilinear Hyperbolic Systems and Dissipative Mechanisms Proof.
We rewrite (4.100) as
B^ki+^J!PMV{^T)[^-mr)dT
«<*■*> -
(4.120) Because of the facts that c" 1 < B(x,t) < c and Y(t) ->■ +oo as t -> +oo, the first term on the right-hand side of (4.120) tends to zero as t —> oo uniformly i n x G [0,1]. We now estimate the second term. By virtue of Lemma 4.3.10, there exists a non-negative constant T = T(s) for any e > 0, such that
l l f e l H h h P { ^ I [«(V.*)-Ii(y,r)]dy}.l| <
5 for T{e)
(4m)
In view of (4.81) and (4.103), it holds that |p(v, # H < c(l + 0 1 + r ) < c(l + V(t)).
(4.122)
(4.122), (4.102) and the fact of lim Y(t) = +oo yield
^ i^k) r *•■ ^ r ) \W) -1\nT)dT -> ° as t —> oo, uniformly in x G [0,1]. Moreover, (4.121) and (4.122) imply
Since y £ j = exp I /
—^-d£ I, it follows that
The influence of dissipation mechanism...
161
This, together with (4.102) and Lemma 4.3.11, gives
^w)£Y{r)v{r)dT=oOn the other hand, it holds that
W)IrY(T)d^C£Due to the arbitrariness of e, we can prove that the second term on the right-hand side of (4.120) tends to zero as t -> oo, uniformly in x £ [0,1]. To estimate the third term, we rewrite it as
-L^ =
Jj{v,9)v(x,r)Y(r)dr
~toh?j J Wv'e)v{x' T)~ I p { v ' 9 ) v { x ' r ) d x ] y ( r ) r f r +
<4-123)
-^fonT)fQP{v,e)v{X,T)dXdT.
For any r > 0, there exists y(r) G [0,1] such that / p(v, 0)v(x, r)dr = p(t>(t/(r), r), % ( r ) , r))v{y(r), r). Jo Hence, on account of (4.80)-(4.82) and Lemma 4.3.6, we deduce \P(V,$)V(X,T)
— I
p(v,Q)v(x,T)dr\
Jo <
I \\p{v,e)v{x,r)]x\dx Jo
(4.124)
\1/2
(r1
r r1
i1/2
< cl / (l + 01+r)v2xdx) / (1 + O2r)0ldx\ + c ^From Lemma 4.3.6 and Lemma 4.3.9, it is known that Jo and / (l + 02r)tf*ds €£*((),oo). Jo
.
162 Quasilinear Hyperbolic Systems and Dissipative Mechanisms This, combined with (4.124) and Lemma 4.3.11, yields —YffT I [p(v, 0)v(x, r)WV'lJo
I p(v, $)v(x, Jo
r)dx]Y(r)dr
(4.125)
—> 0 as t —> oo, uniformly in x £ [0,1]. In order to estimate the second term on the right-hand side of (4.123), we multiply (4.116) by v(x,t) to get v[ u(y,t)dy]t = -p{v,6)v + R(t)v + fi0ux. Jo It implies
iJri)J!YiT)J!p{v'e)vix'r)dxdT =
~sfefy(r) I! Uou{y' +
Yji)
y
(r) /
r)dy v(x r)dxdT \t ' jfhbvgjhf
u
x{x,r)dxdT.
Integrating by parts, and using (4.3)i, we arrive at
=
^ / *(*>*) I u(y,t)dydx
Lemma 4.3.8 and Lemma 4.3.10 show that / v(x, t) / u(y, t)dydx -» 0 as t -+ oo, uniformly in x E [0,11. Jo Jo To deal with the second term on the right-hand side of the above equality, we use the facts that
M rWi i
jf *'
d
ve
2£ CV /2(T)
Gf ^ 'T Gf ) "^° Gf 5*) "
" '
The influence of dissipation mechanism...
163
V(r) eZ,°°(0,+oo),and / u(y,r)dy\ < c. \Jo I These, together with Lemma 4.3.11, imply Y(i\
/
/
( /
u y
( > T)dy)
uxY{T)dTdx -> 0 as t -+ oo,
uniformly in x G [0,1]. Furthermore, Lemma 4.3.11 and (4.112) yield 1 f1 f* \ f* u , 1 vRMYlr) , J —77777 / / / dy\ ———t-drdx -> 0 as t -> oo, PoY(t) Jo J0 [Jo
J
Po
uniformly in x G [0,1]. Therefore, it is proved that the first term on the right-hand side of (4.126) tends to zero as t —> oo, uniformly in x G [0,1]. With a similar approach, we can claim that the third term in (4.126) tends to zero as well as t —> oo, uniformly in x G [0,1]. ^From the above arguments, we end up with
<x^--QY(t)J
y(r)i
* ( r ) / v(*,T)dxdr^Q
as t -> oo, uniformly in x G [0,1]. Therefore,
/ v(x,t)dx Jo
—-r / Y(T)R(T) PoY(t) Jo
v(x,r)dxdT^0,
as t -* oo.
Jo
Thus, it follows that v(x,t) — j v(x,t)dx -» 0 as t -¥ oo, uniformly in x G [0,1]. Jo We now turn to the second estimate in Lemma 4.3.12. Due to (4.94), it holds that
164 Quasilinear Hyperbolic Systems and Dissipative Mechanisms Using L'Hospital's theorem, we have lim — L - r I Y(T) \E1+ t-yoo fl0Y(t)
Ex+ =
Jo
I
R'(T) I JQ
f
R'(i) f v(x,Z)dzdi]
Jo
Jo
dr J
v(x,r)dxdr
_Jo
R On the other hand, Lemma 4.3.11 and (4.112) yield —777-T / Y(r)
/
—dxdr -> 0 as t -» oo.
Thus, it follows that +oo
/
l
+
/.l
K(T) /
""" ^ m X «'"/, •>••"-"—
v(x,T)dxdT
$
- T
(4.127) similar approach as used in the proof of (4.125), it can be easily as tWith —> oo,a uniformly in x £ [0,1]. verified that ——r-r I Y(r) / e(v, 0)dx — e(v, 0) cfr ->• 0, as tf -» oo. A*o*(*) Jo Uo J Combining with (4.127), it implies r+oo
.t
/
^•')+wmi "«*(•.•)(«.')*-A
/.l
R'(T) /
*
v(x,r)dxdr
-i>
as tf -> oo, uniformly in x £ [0,1]. The last estimate in Lemma 4.3.12 can be obtained using a similar argument as used in (4.126).
Chapter 5 Vanishing viscosity and nonlinear stability of waves
Consider the Cauchy problem for a general system of viscous conservation laws f ut + f(u)x = e(B(u)ux)x, - o o < x < oo,t > 0,£ > 0 ,. ^ \ u(x,0) = uo(#), — oo < x < oo ^ ' ' where u is an ra-vector, f(u) is a smooth ra-vector-valued function, B(u) is a smooth n x n matrix which is either positive definite or positive semi-definite, and uo(x) is a given n-vector-valued function with uo(x) —> u^ as x -t =poo. We are interested in the relation between the solution of (5.1) and the distributional solution of the corresponding hyperbolic system of conservation laws without viscosity: ut + f{u)x = 0. (5.2) It is conjectured that general distributional entropy solutions of (5.2), with the same initial data (5.1)2, are strong limits, as e —> 0, of solutions of (5.1). Section 5.1 is devoted to this topic, namely, it discusses the zero dissipation limit problem. Another closely related topic is the study on the similarities and differences of the large-time behavior of solutions for these two systems. In particular, the comparism the large time behavior of solutions of (5.1) with the solution of the Riemann problem for (5.2), namely
f ut + f{u)x = 0
\*^={Z: HI
(53)
which involves shock waves, rarefaction waves and contact discontinuities, de pending on the relative position of u- and u+ in the phase space. The nonlinear stability of waves will be discussed in Section 5.2. The last Section 5.3 contains an investigation on the above two topics for multi-space dimensional system. 165
166
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
5.1
Vanishing viscosity - the zero dissipation limit
The conjecture that general distributional solutions of (5.2), satisfying the entropy condition, are strong limits, as e —» 0, of solutions of (5.1) with the same initial data has been proved by Kruskov [100], Oleinik [154] and Volpert [180] for scalar equations, namely, u £ M 1 . It is also established for some special 2 x 2 systems for which a weak maximum principle holds. This is investigated by Diperna [35] and Chen [13] for the system of isentropic gas dynamics by the method of compensated compactness, where the Cauchy problem for the system (Pt + (pu)x = 0 \ (pu)t + [pv? + p(p))x
v ( kfjg nbj = 0, p(p) = * V ,
7
> 1
[ V
°'
and the corresponding parabolic system f Pt + [py)x - epxx \ (pu)t + {pu2 + p{p))x = e{pu)xx,e > 0 are studied. Here p and u denote the density and the velocity. Using the theory of compensated compactness and the viscosity method, a global existence theorem is established in [35] for 7 = 1 + <> . 1 , ra > 2 integers, in [13] for 1 < 7 < § . Recently, progress on the above problem for any 7 > 1 has been made by Lions, Perthame and Souganidis [108], with the help of a kinetic formulation. We will explain the zero dissipation limit result by the method of compensated compactness in the first part of this section. The common feature of the above mentioned results is that the initial data are required to have the same limit value at x = =foo. Although this class of initial data is fairly broad, it excludes Riemann data, which are building blocks for the inviscid system. Moreover, the abstract analysis yields little information on the qualitative nature of the viscous solutions. Another kind of result on the inviscid limit problem has been obtained by Hoff-Liu [60], and Xin [188] for the one-dimensional Navier-Stokes equation of compressible isentropic gases ( \
Vt Ut
- ux = 0 + p(v)x = e (^-)x
,xeR\t>0
(5'5>>
where u, v and p denote the fluid velocity, the specific volume, and the pressure in a compressible fluid, e > 0 is the constant viscosity coefficient, and the pressure p is assumed to be a smooth function of v > 0 satisfying p'(v) < 0 < p"(v)
for
v > 0.
Vanishing viscosity and nonlinear stability of waves
167
Consider the Cauchy problem for (5.5) with initial data (ti,t;)(s,0) = (tig, *§)(*), JjmTOK(z)X(*)) =
xER1
(«=F>V=F)
in which u^ and v^(> 0) are given constants. We shall study the relation between the Navier-Stokes solutions (u£, v£)(x,t) and the solutions (tz, v)(x,t) of the corresponding inviscid Euler equation
{
ut - ux = 0 ut+p{v)x = Q {u,v)(x,0)
( 5>6 )
= (uo,vo){x),
lim {u0,v0)(x)
x—► ipoo
=
(uT,vT).
The zero-dissipation limit problem is studied in [60] for the case that the un derlying inviscid flow is a single shock wave, and in [188] for the case that the corresponding Euler equation have rarefaction wave solutions. For more general systems of hyperbolic conservation laws with small viscos ity of the form < + f{ue)x = euexs,ue eRn,x£R\t>0,e>0
(5.7)
it is shown by Goodman and Xin [50] that if the solution of the problem with zero viscosity is piecewise smooth with a finite number of noninteracting shocks satisfying the entropy condition, then there exist solutions to the correspond ing system with viscosity that converge to the solutions of the system without viscosity away from shock discontinuities at a rate of order e as e —> 0. We will describe the above kind of result in the second part of this section. Let us consider the system (5.4) of isentropic gas dynamics in Euler coor dinates, with k2 — I/7 for convenience and the initial data {p, u)\t=o = {po{x),u0{x)).
(5.8)
The system (5.4) can be written as
{
Pt + rnx = 0 W H + ( ^ + P ( P ) )
=0,p(p)=pV7,7>l.
For smooth solutions, (5.4) or (5.4)' are equivalent to vt + V / ( v ) w * = 0 where m = pti, v = I M
and v/(*>) = ( _2z£ + p '( p )
2m j .
(5*4)'
168 Quasilinear Hyperbolic Systems and Dissipative Mechanisms We construct the viscosity approximations for the Cauchy problem (5.4) (5.8) first, namely, show the existence of global solutions for the following parabolic problems Pt + mx — epxx
{ m+i^f+Pipf)
=emXXlp(p)=pyll7>l
(p,rn)\t=o = (p£o(*),™o(*))
(5 9)
'
(5-10)
or
{
vt + f(v)x = evxx,
(5.9)'
v\t=o = v£o (*) (5.10)' where VQ = (po(x), m£0(x)) is an approximating sequence of the initial data v0(x) = (po(x),po(x)u0{x)). T h e o r e m 5.1.1. Suppose that the initial data (po(x), uo(x)) satisfy
eL2nL°°,
(po(x)-p,u0(x)-u)
(5.11)
p0(x) > 0. Then there is an approximating sequence VQ (X) satisfying VQ(X) -> VQ(X),
v£0(x) -ve
in
L2
C£(-oo, oo),
0
(5.12)
|ti5(*)| < Mo,
such that there exist global solutions (p e , u£) to the Cauchy problem (5.9) (5.10) satisfying
(f(;t)-J>,u'{;t)-ii)€C1nH1, 0
|ue|<M,
(5.13)
where both M0 and M are constants, independent of s. To prove this theorem, we first extend the pressure function p(p) to p(p) such that P(P)\P>O =
p{p)
P(p)£C\ and establish the following lemmas. L e m m a 5.1.2. (local solution). Let (p£Q(x), u£0(x)) satisfy (5.12). Then, for any M > M 0 , there is a constant T0(M) such that there exists a unique weak
Vanishing viscosity and nonlinear stability of waves
169
solution (p, u) E C 1 , 1 to the Cauchy problem (579) (5.10), which is the same as the Cauchy problem (5.9)(5.10) except that p(p) is replaced by p(p), on IIT 0 , satisfying
|p£0M)| + K ( M ) | < M ,
where n T o = {(x1t),-oo< x < o o , 0 < * < T 0 }. The proof of this lemma is based on the Banach theorem and the contractivity of the operator:
P : C^CIITO) x C^(UTo) -+ C^l(UTo) x
c^l{nTo)9
(p,m) -* (p,m), defined by the following integral equalities: p(x,t) = y ~ G(x - y,t)po(y)dy+[
dr j T
9G{x
"*'* ~
T)
m{y,r)dy,
OO
/
G(x-2/,<)m0(y)rfy ■CX)
*lirL-L^—-\r*'Mr
on a small time interval (0,To), where G(z,t) is the elementary solution of the heat equation ut — euxx = 0, that is,
G(M)= exp
^ (-£)-
Furthermore, using the invariant regions theorem established by Chueh, Conley and Smoller [20], it follows that L e m m a 5.1.3. The regions £) = {(p, m) : w < wo, ^ > zQ)w - z > 0} C {(/>, w) : p > 0} are invariant under the solutions defined by (5.9) (5.10), that is, if the initial data lie in ]T], the solutions of the Cauchy problems (5.9) (5.10) lie in J^ as well. Here w and z are the Riemann invariants of (5.4), defined by m , 2pX*~ y
__ m
2p^
Combining Lemma 5.1.2 with Lemma 5.1.3, we can extend the local solution in Lemma 5.1.2 to the global solution with (5.11) for any fixed e > 0 and obtain Theorem 5.1.1 with the help of the following lemma.
170 Quasilinear Hyperbolic Systems and Dissipative Mechanisms Lemma 5.1.4. Let f(x) -JeL2D L°°(-oo, oo)J(x) approximating sequence f£ (x) such that
> 0. Then there is an
/»(*)-+/(*)(£*), /*(*)-/€ ^(-ocoo), 0
(M
(5.14)
independent of
e).
As a matter of fact, we can construct f£(x) as follows oo
/
fl{y)h{x-y)dy -oo
where
fo(y) = {L(x)'
xe
H*}>
[ /, otherwise. £ It is easy to check that f (x) satisfies (5.14). So far, we have constructed the viscosity approximations. To prove the con vergence of these viscosity approximations, we apply the theory of compensated compactness to the system (5.4). T h e o r e m 5.1.5. Suppose that the approximate solutions vk(x,t)
= (pk(x,t),mk(x,t))
-
(pk(x,t),pk(x,t)uk(x,t))
to the Cauchy problem (5.4), (5.8) satisfy the following conditions: (i) There is a constant C > 0 such that 0 < Pk(*,t) < C,
\uk(x,t)\
< C,
a.e.
(5.15)
(ii) On any bounded domain Q C M+ and for any weak entropy pair (77, q), the measures r}(vk)t + q{vk)x compact in H^(Q). (5.16) Then there exists a subsequence vk (still use the same subscript) such that (pk(x,t),mk(x,t))
-> (p{x)t),m{x1t)))
a.e. .
The proof can be found in [13]. Now one only needs to show that the viscosity approximations satisfy the above assumptions in order to obtain a global solution to the Cauchy problem (5.4) (5.8).
Vanishing viscosity and nonlinear stability of waves
171
Theorem 5.1.6. Suppose that the initial data vo(x) = (po{x), po(x)uo(x)) satisfy 0<po(x)<M, /
|ti 0 (*)|< M,
{n+(v0(x)) - T)*(v) - Vn*{v)ivo{x)
jghnvbhn
- v))dx < M,
./—oo
for some constant state v and 77* = rjpu2 H—/ _ i \ ? t n e mechanical energy, which is a weak convex entropy of (5.4) (A discussion on general weak entropies can be found in [13]). Then the viscosity approximations v£(x,t) determined by Theorem 5.1.1. satisfy the compactness framework (5.15) (5.16). We refer the reader to [13] for the proof. Next we turn to the system (5.5) and (5.6), where the pressure p is assumed to be a smooth function of v > 0 satisfying p'{v)
<0
for
v>0.
(5.18)
It is conjectured that a general distributional entropy solution of the Euler equa tions should be a strong limit, as dissipation goes to zero, of solutions of the corresponding Navier-Stokes equations with the same initial data. In the spe cial case that the underlying inviscid flow is smooth, the zero-dissipation limit question can be resolved using classical methods (see [13]). In the case that the underlying inviscid flow is a single shock wave, Hoff and Liu have proved the global existence of solutions for the system (5.5) with shock data and have shown that the solution converges to the inviscid shock wave as the viscosity tends to zero, uniformly away from the shock. Moreover, the qualitative properties of the solution are given in [60]: the solution of (5.5) contains a convective discon tinuity which decays exponentially in time and vanishes in the zero-dissipation limit; for intermediate time, the solution is approximated well by solutions of the Burgers equation, appropriately lifted to phase space; for large time, the solution coalesces with a viscous shock, which is a travelling wave solution of the Navier-Stokes equations. Consider Riemann initial data
"<■••>-[#,o)H£: i l l
<"•>
where Ui and Ur are assumed to satisfy the Rankine-Hugoniot condition
=p{vr)
-p(vt)
172 Quasilinear Hyperbolic Systems and Dissipative Mechanisms with shock speed a < 0 and to satisfy the entropy condition -p'{vr)
>
>-p'{vi).
Then, the corresponding inviscid system (5.6) admits the entropy-satisfying shock-wave solution
v ™={%. in
<5-2i>
On the other hand, it is well-known [169] that there is a travelling wave solution of (5.5)
satisfying $(+oo) = Ur
and
$ ( - o o ) = Ui
(5.22)
and the compressibility condition ^ P W ) ) < 0-
(5-23)
Since a translate of $ also satisfies (5.22) and (5.23), we fix such a translate by requiring that oo
/
[9(x)-U(x,0)]dx
= 0.
(5.24)
-OO
The first result in [60], which asserts the global existence of a solution of the Cauchy problem (5.5), (5.18), (5.19), establishes its large-time behavior, and specifies its regularity class, is T h e o r e m 5.1.7. Given U\ with v\ > 0, there exist constants C = C(Ui,p,e) and C\ = Ci(Ui,p) such that, if S = \Ur — Ui\ < 1/Ci, where Ui and Ur satisfy the Rankine-Hugoniot condition (5.20), then a solution U{x,t) of (5.5), (5.18), (5.19) exists for all tf > 0 and satisfies the following: (a) Um ||W(-,*) - *(-,*)IU~(«) = 0, t—>oo
v
'
(b) u is continuous for t > 0; ux, v, vx, and vt are uniformly Holder continuous in the sets {x < 0,2 > r } and {x > 0,t > r} for any r > 0 ( in particular, these quantities have one-sided limits at x — 0 for t > 0); and utluxx and vxt are Holder continuous on compact sets in the complement of {x = 0} U {t = 0}, (c) The following jump conditions hold at x = 0: [log<;(*)] = pogtKOjK^/o a^ds where [w(t)] denotes the jump w(Q+,t) - w(0-,t)
"'"-isSr
(5.25)
in a given quantity w,
(526)
Vanishing viscosity and nonlinear stability of waves
173
and
\p(v)) = [ ^ ]
(5.27)
for t > 0. Moreover, both quantities in (5.27) are bounded in absolute value by Cie-'/c*e. A straightforward scaling argument applied to (a) above yields the second result in [60] concerning the inviscid limit of the solution U, whose dependence on e is expressed by writing U = U£. T h e o r e m 5.1.8. The solutions U£ described in Theorem 5.1.7 converge to the shock wave (5.21) uniformly on sets of the form {(x, t) : \x - st\ > a } , for any positive number a. The analysis involved in the proofs of these theorems can be partitioned into three parts, corresponding to three diffferent time scales. The first part is used to prove local existence and regularity, and derive the required information on the evolution of j u m p discontinuities in the solution. The proof of local existence parallels closely the result of Hoff [57] in which existence is proved for much less regular intial data, but data having the same limits at x = ± o o . The second part is to show that U remains close to a certain approximate solution constructed from solutions of the Burgers equation, at least for times up to time O 0(6~2~d), where 0 is a small positive constant. The analysis relies heavily on the ideas and techniques of Liu [121], in which stability of travelling wave solutions for more general systems, and for more general, but smooth, perturbations, is proved. Finally, the third part of the analysis, proving that the solution U coalesces with the travelling wave $ for very large time, is carried by way of energy estimates, using as initial time the time scale of the second part of the analysis. We refer the reader to [60] for details. In the case that the underlying inviscid flow is either a centered rarefaction wave or a smooth rarefaction wave, Xin [188] has shown that the solution of system (5.5) with weak centered rarefaction data exists for all time, and con verges to the inviscid centered rarefaction wave as the viscosity tends to zero, uniformly away from the initial discontinuity. It is also proved in [188] that for a given centered rarefaction wave of the Euler equation (5.6) with finite strength, a viscons solution to the Navier-Stokes equations (5.5) with initial data depending on the viscosity can be constructed such that the viscous solu tions approach the centered rarefaction wave as the viscosity goes to zero, at the rate e ^ ^ l n e l , uniformly for all time away from t = 0. In the case that the underlying inviscid flow is a smooth rarefaction wave, then the solution of (5.5) with the same initial data as that of the smooth rarefaction wave exists and converges to the rarefaction wave at the rate e 1 / 2 , uniformly for all time. To descibe the result in [188] more precisely, we first recall the concept of a centered rarefaction wave. Let us concentrate in the first family of rarefaction waves for the Euler equation (5.6), which are characterized by the fact that the 1-
174 Quasilinear Hyperbolic Systems and Dissipative Mechanisms Riemann invariant, u-\- I Ai(s)c/s, is constant in (x,t), and the 1-characteristic speed Xi(v)(x,t)(\i(v) = —\/—p'{v) < 0) is increasing in x. Suppose that the end states (u±, v±) of the initial data satisfy rv+ u++
\1(v)dv = u-,
Ai(v+) > Ai(v-).
(5.28)
JvThen the state (w_, v~) can be connected to the state (w+, v+) by 1-rarefaction waves. The centered 1-rarefaction wave connecting (u_,t>_) to (u+,v+) is the self similar solution (vr ,vr)(x/t) of (5.6) defined by
«=T, *
fAi(«_), t<\i(v-), A!K(0) = U *i(v-)\i(v+), /•«r(€) « r (£) = - / Ai(s)rfs + « _ .
(5.29)
(5.30)
This is uniquely determined by the system (5.6) and the initial rarefaction wave data
(«o,«d)(«) = K , i « ) ( « ) S { ^ ; ^ ;
x xtl
(5.3i)
It is noted that a centered rarefaction wave is only Lipschitz continuous away from its center. We now solve the Navier-Stokes equations (5.5) with the fixed rarefaction wave initial data: («$(*),«§(*)) = K,t;5)(*),
xGR1,
(5.32)
and study the asymptotic behavior of its solution. The result in [188], which yields the global existence of solutions to the problem (5.5), (5.18), (5.32), and asserts that its asymptotic behavior is completely determined by the underlying centered rarefaction wave, is stated as Theorem 5.1.9. Let the constant states (u±,v±) (with v± > 0) be connected by a centered 1-rarefaction wave, (ur (j) ,vr ( § ) ) , defined by (5.29) and (5.30), with suitably small strength. Then the Navier-Stokes equation (5.5) with the rarefaction wave initial data (5.32) has a unique, global, piecewise smooth solution (u e , v£)(x,t) such that (i) u£(x, t) is continuous for t>0',uexive, and v£x are uniformly Holder continu ous in the sets {x < 0, t > r} and {x > 0, t > r} for any r > 0; and u\, uexx, and
Vanishing viscosity and nonlinear stability of waves v£xt are Holder continuous on compact sets in {{x,t),x the j u m p s in v£(x,t) and u£x(x,t) at x = 0 satisfy
175
^ 0,* > 0}. Moreover,
IK(0,t)]|,|K(0,t)]|0, uniformly away from t = 0, i.e. for any positive number /i, lim sup -*°xeRl,t>h
e
|(ti e ,v e )(a!,t) - ( u r , t ; r ) ( z / * ) | = 0.
(iii) For fixed viscosity e > 0, the solution (u e , v e )(a:, J) approaches the centered rarefaction wave (ur, vr)(x/t) uniformly as time t goes to infinity, i.e., lim
sap\(ify)(x,t)-(u',vr)(j)\
= 0.
The proof of this theorem can be reduced, by using a natural scaling argu ment, to the nonlinear asymptotic stability analysis of weak rarefaction waves for the Navier-Stokes equations (5.5) under nonsmooth initial perturbations. The nonlinear asymptotic stability of centered rarefaction waves under small smooth initial perturbation has been established for fairly general systems of conservation laws which will be discussed in Section 5.2. For the Navier-Stokes equation (5.5), it is pointed out by Hoff in [56], Hoff and Smoller in [61], that the initial discontinuities in the specific volume v propagate along the particle paths, but the discontinuities will decay exponentially fast. With the help of this fact, the nonlinear asymptotic stability of rarefaction waves for (5.5) under discontinuous initial perturbation is proved by combining the local existence and regularity result in [60] with a priori energy estimate. For details, see [188, section 3]. The above theorem 5.1.9 applies only to weak waves and has no rate of convergence involved. The following theorem applies to centered 1-rarefaction wave (ur,vr) {x/t) of arbitrary strength and studies the rate of convergence. However, the initial data for the Navier-Stokes equations are allowed to depend on the viscosity. T h e o r e m 5 . 1 . 1 0 . Let (ur,vr) ( j) be the centered 1-rarefaction wave defined by (5.29)-(5.30), which connects two constant states (u±, v±) satisfying (5.28) with v± > 0. Then there exists a positive constant SQ such that for each £, 0 < e < So, we can construct a global smooth solution (w e , v£) of (5.5) with the following properties (i)
(ue - ur, v£ - vr) 6 C°(0, +oo; L2),
176 Quasilinear Hyperbolic Systems and Dissipative Mechanisms
(u<,v%eC°(0,+oo;L2), u£xxeL2(0,+oo;L2). (ii) As viscosity e -» 0, (u£, v£)(x, t) converges to (ur, vr) ( j ) pointwise except at (0,0). Furthermore, for any given positive constant A, there is a constant C(h) > 0, independent of e, so that sap\\(u'y){;t)
- (ur,vr) ( j ) \\Lao <
C(h)e^\\ns\.
Kt/
t>h
The proof consists of two essential steps. First, one approximates the cen tered 1-rarefaction wave (ur1vr) ( ? ) by a carefully constructed smooth rarefac tion wave (Ug,Vg)(x,t), depending on the viscosity e, so that (ur£,v£) converges to (ur,vr) at a rate in an appropriate topology as e —>• 0. Second, one shows that the smooth rarefaction wave (u£1 vre) dominates the behavior of the solution (ur,vr)(x,t) to (5.5) with the same initial data as that of (u£,t;£), by decom posing the solution (u£,v£) as a small perturbation of the smooth rarefaction wave {ur€,vre) and choosing appropriate scales for the perturbation. The next theorem concerns the case that the underlying inviscid flow is a smooth rarefaction wave. By a smooth 1-rarefaction wave (ur ,vr)(x1t), we mean the unique solution r (t/ , vr)(x,t) to the Euler equation (5.6) with sufficiently smooth initial data (ur,vr) (x,0) satisfying —Ai (vr'(ar, 0)) > 0 for any ox
x G R1,
and
rvr(x,0)
ur(x,0)+
\i(s)ds
= u-,
for any
xER1.
For technical reasons, we further assume that as x -± ±oo, dl vr(x,0)^v±i—[vr(x,0)->0 ox
(1 = 1,2,3)
sufficiently fast, and the rates of decay for -^-jvr(xy0)(l = 2,3) as x -» =poo, dx are not less than that of -8-vr(x, 0). T h e o r e m 5.1.11. Let (ur, vr)(x, t) be a smooth 1-rarefaction wave as descibed above. Then there exists an e0 > 0 such that for each e G (0,eo], the NavierStokes equation (5.5) with initial data
K,«5)(«) = («r ,«■•)(*, 0),
x€R\
Vanishing viscosity and nonlinear stability of waves has a unique smooth solution (u£ ,v£)(x,t) £
r
£
177
satisfying
r
{u - u , v - v ) e C°(0, +00; H1) {u£ - ur)x G L 2 (0, +00; H1), (v£ - vr)x E L 2 (0, +00; L 2 ), and \{u£ - ur,v£ -vr){x,t)\
sup
< Ce 1 ' 2 ,
where C is a positive constant independent of e. The proof is carried out by combining a finite time estimate (applying for arbitrary smooth solutions of (5.6)) with a large time estimate, obtained by an energy method with the help of the geometrical structure of the rarefaction wave. The details are given in [188, Section 5]. We finish this section by describing the result obtained by Goodman and Xin [50] on the zero-dissipation limit problem for more general system of hy perbolic conservation laws with small viscosity in the form of (5.7) for which the corresponding system of conservation laws without viscosity takes the form ut + f(u)x - 0,
ueRn,xeR1,t>0.
(5.33)
It is proved in [50] that the piecewise smooth solutions u of (5.33), with finitely many noninteracting shocks satisfying the entropy condition, are strong limits as e —> 0 of solutions u£ of (5.7). For simplicity, we only discuss the case in which it is a single-shock solution of (5.33), i.e., a distributional solution smooth up to a single shock satisfying the entropy condition (A more precise definition of a single-shock solution can be found in [50]). Theorem 5.1.12. Suppose that the system (5.33) is strictly hyperbolic and the p-th characteristic family is genuinely nonlinear. There exist positive constants 770 and £0 such that if u(x,t) is a single-shock solution up to time T with J2 f l
[\%u(z,t)\*dxdt
\u{s(t) + Q,t)-u(s(t)-0,t)\
Moreover, for any given 77 G (0,1), sup 0
sup 0.n
/ \u{x,t) - u£(x,t)\2dx
< C^e",
|u(a:,tf) — u£(x,t)\ < Cne,
178 Quasilinear Hyperbolic Systems and Dissipative Mechanisms where Cv is a positive constant depending only on 77, x — s(t) is the shock of u(x, i) which is a smooth curve for 0 < t < T. We refer the reader to [50] for the interesting proof which uses a matched asymptotic analysis and an energy estimate related to the stability theory for viscous shock profiles. 5.2
Nonlinear stability of waves
In this section we study the system (5.1) with e fixed, say e = 1 for definiteness, namely ut + f(u)x = (B(u)ux)x, with initial data u(x,0) = t/o(z),
xeR\t>0
(5.34)
x e R1, (5.35)
lim uo(x) = u=.
The corresponding inviscid problem is the hyperbolic system of conservation laws (5.2) ut + f(u)x = 0, xeR1, t>0 with Riemann data (5.3)
u(X) o)
= „*(*) = I u + >
x>Q
The Riemann problem has played an important role in the theory of hyper bolic conservation laws, including the study on the asymptotic behavior of the solutions for the system (5.2). Many physical phenomena are governed by systems (5.34) among which some important examples have been discussed in Chapter 4 where one can get the inviscid hyperbolic equations (5.2) from the viscous system (5.34) by neglecting the dissipative effects such as heat conductivity and viscosity, and one uses the inviscid system as a first approximation to the real viscous system. Therefore, it is natural to expect that the asymptotic behavior of the solution for the viscous problem (5.34) (5.35) is closely related to the Riemann problem (5.2)(5.3). Namely, suppose that the initial data uo(x) in (5.35) are suitably close to the initial data of elementary waves for the Riemann problem (5.2) (5.3), the question is whether or not the solution for the viscous problem exists globally in time and converges to the elementary waves. Ilin and Oleinik [82] first investigated this problem for single convex con servation laws with positive constant viscosity. They showed that both viscous shock profiles and rarefaction waves are nonlinearly stable in the Loo-norm by using a maximum principle.
Vanishing viscosity and nonlinear stability of waves
179
The study on this problem has been developed intensively since then and we only give a survey of it in this section, corresponding to the cases where (5.2) is a scalar equation, a strictly hyperbolic system and a nonstrictly hyperbolic system. A. Scalar equation. We start with the stability of viscous shock profiles for a scalar equation (5.34) where B is a positive constant. Suppose / " > 0, u- and u+ are related to the shock speed a by the RankineHugoniot condition
<
which is equivalent to u+ < u_. A viscous shock wave for (5.34) is a traveling wave solution with speed a which interpolates the asymptotic values u^ as x = =Foo, i.e. u(x,i) = {/(£)>£ — x — at, lim U(£) = u T . f-^oo
It is proved by II'in and Oleinik [82] [83] that if the initial data uo(x) are bounded, measurable such that | /
(UQ(X) — u-)dx\ < oo
Jx <0 f
| /
(5.36) (u0(x) - u+)dx\ < oo,
Jx>0
where U(x — at) is the viscous profile uniquely determined by oo
(uo(z) - U(x))dx = 0, /
-OO
then max|u(x,t) -U{x-
at)\ -^ 0 as t -)■ oo.
(5.37)
x£R
It is also shown that if the initial data decay exponentially as e~a^(a > 0) for |x| -> oo, then the solution approaches the traveling wave solution at an exponential rate e~7*(7 > 0) as t —t oo. Instead of using the maximum principle, Peletier [159] gave an alternative proof for the result (5.37) by using the energy method, while Sattinger [164] obtained a result similar to the above exponential decay with the help of the spectral analysis for the linearized operator. Another important contribution was made in 1985 by Kawashima and Matsumura [92] who used the weighted energy method to show the decay rate in Loo-norm: if the initial disturbance is
180 Quasilinear Hyperbolic Systems and Dissipative Mechanisms of an algebraic order 0(\x\~a)(a > 0), for \x\ -> oo, then the solution converges to the traveling wave solution at an algebraic rate 2~ 7 (7 > 0) as t ->■ oo. When B is not a constant, the nonlinear stability of weak viscous shock waves is a special case of Liu's work [120] where there is no decay rate in time.Xin [189] extends Kawashima and Matsumura's result [92] to the case where B(u) is a positive smooth function to get a similar algebraic rate of time decay in the Loo-norm. As far as Instability is concerned, Osher and Ralston [155] proved the In stability of viscous profiles in 1982 by using semigroup methods. The result in [155] says that under the assumption / " > 0, u+ < «_, the initial data u0(x) are such that B(u0) is uniformly Lip-continuous,
Jf
\u0(x) - u^\dx < oo,
x<0
/
\UQ(X) - u+\dx < oo,
Jx>0
and there exists a constant xo such that [ {uo(x)-U{x
+ xo))dz = 0,
JR
then lim / \u(x,t)-
U(x + x0 -
This Ll-stability result is improved recently by Serre [166] who gets rid of restrictions such as / " ^ 0 etc. The stability of shock profiles for a scalar equation which is non-convex (/" changes sign), has been studied by Kawashima-Matsumura [93], Jones-GardnerKapitula [90], Serre [166], Matsumura-Nishihara [146], and Liu [110]. The case when f(u) satisfies uf"(u)>0 0 for i i ^ O is studied in [93] by a weighted energy method. A much more general case than (5.38) is treated in [90] by a spectral analysis to obtain both the stability and the time decay rate in Loo-norm. The L1-stability of shock profiles for general non-convex / is obtained in [166] based on the I^-contractiveness of solutions. The improvement of the results in [93] and [90] have been made in [110] and [146] by removing the condition f'"(u) ^ 0 and allowing the flux to have more than one inflection point and by sharpening the time decay rate. Suppose (ri+-ti_)/"(ti) < 0, the solution of (5.34) (5.35) is expected to tend toward the solution of the corresponding Riemann problem — rarefaction wave solution uR (j) . The discussion to show the nonlinear stability of rarefactions has been made by Hopf [62], IFin-Oleinik [83], Hattori-Nishihara [55], Liu [109] etc., where there are no smallness assumptions for \u+ — u_ | and the initial data.
Vanishing viscosity and nonlinear stability of waves
181
The result in [55] is obtained for Burgers' equation, using an explicit formula, the work in [109] is for general / , with the help of a maximum principle. The decay rate is shown in [55] for Burgers' equation where the solution decays as t~ll2 in a neighborhood of the edges of the rarefaction uR (f-j ; as t~l otherwise. A detailed decay rate for general / with convexity is obtained in
[109]. B. Strictly hyperbolic system (5.2) For systems, the situation becomes quite complicated. In the case that u+ = u _ , the existence of a small amplitude solution and its large time behav ior toward the constant state for viscous problem (5.34) (5.35) have been in vestigated intensively by many authors. When the system (5.34) is a uniformly parabolic system, large time behavior of the solution to the Cauchy problem (5.34) (5.35) have been studied in L*(l < p < oo) by Chern-Liu [17] [18], Liu [123] and the asymptotic solution has been identified as a superposition of linear and nonlinear diffusion waves, solutions to the heat equation and the Burgers equation. The optimal rate has been obtained in [123] also. When the viscosity matrix B is only positive semi-definite, hence the system (5.34) is hyperbolic-parabolic, such as the system for compressible isentropic viscous flow and the full Navier-Stokes equations, the global existence of solution to the problem (5.34) (5.35) has been established by Kawashima [94] where the L\ stability has also been demonstrated by examining the nonlinear stability of diffusion waves; the Lp stability with optimal rate has been established by Zeng [193], and Liu-Zeng [126] for general systems recently. We start with the strictly hyperbolic system (5.2) for which each charac teristic field is either genuinely nonlinear on linearly degenerate, i.e., for each i £ {1,2, • • - , n } , either VA,- • rj(u) ^ 0
for all
u
VA; • ri{u) = 0
for all
u,
or
where A,- and r,- are the i-eigenvalue and the corresponding right-eigenvector. It is well-known that the solution of the Riemann problem for this kind of strictly hyperbolic system (5.2) consists of shock waves, rarefaction waves and contact discontinuities (see [102], [169] and [12]). The study of asymptotic stability of weak shock waves for viscous conserva tion laws was started independently by Goodman [47] on general n x n system with a positive definite B(u) and by Matsumura-Nishihara [143] on the isen tropic Navier-Stokes equations which is a 2 x 2 hyperbolic-parabolic system. They considered the nonlinear stability of viscous weak shock waves by applying an elementary energy method under the condition that the initial data
182 Quasilinear Hyperbolic Systems and Dissipative Mechanisms uo is a local perturbation of the viscous p-shock wave U(x — at) with U(±oo) = u±, satisfying / (u0{x) - U(x))dx = 0.
(5.38)
JR
By a p-shock, we mean that its speed a satisfies \p(u+) < a < A p (u_). Based on a similar energy method as used in [47] and [143], the general com pressible Navier-Stokes equations are studied by Kawashima and Matsumura in [92]. Important progress for nonzero total mass perturbation without the restric tion (5.38) was made for weak viscous shock waves by Liu in [120], where the stability of multi-mode shock waves is studied also, the viscosity matrix B(u) for (5.34) satisfies that there exists a choice of U(u) and r,(w),i = 1,2, • • -,n, so that B(u) = ftM, • • •, ln(u))T - B(u) • (n(ti), • ■ •, rn(u)) > 0 i.e., positive definite for each u under consideration, where rt-(t*)(Z,-(u)) is the right (left) i-th eigenvector of f'(u). Liu observed that initial data, which are nonzero total mass perturbations of a shock, generate a translation of the shock together with n — 1 so called diffusion waves in other wave families. The trans lation 8 and the masses 0Oi of the diffusion waves are determined by the initial datum through oo
/
[uo - U(x)]dx = 8{u. - , i + ) + £ 0 O i r O i ,
(5.39)
where r0- = r,(w±) if ±(i — p) > 0. The decoupled diffusion waves are solutions of 0i + (\ot0i + i V Ao, • r0t6f j
- 0 , ^ = 0,
i^p
and are initially Dirac measures with the prescribed masses #0i- Using the diffusion waves, Liu shows that u tends asymptotically to the translated shock wave for certain nonlocal perturbations that decay slowly when x tends to ±oo : ci|*|~* < M * ) - U{x)\ < c 2 | * r * .
(5.40)
By refining the approach in [120], Liu established this result to the general compressible Navier-Stokes equations in [121].. To treat general local nonzero total mass perturbations without the restric tion (5.40), a linear coupled diffusion wave 77 with zero mass is introduced by Szepessy and Xin in [171] where the viscous conservation laws take the form ut+f{u)x
=uxx,
u(x,t)eRn
(5.41)
Vanishing viscosity and nonlinear stability of waves
183
with strictly hyperbolic flux / . Denote 0 = 2^0,-r,-, which solves the equation (5.41) with an error called Ex. The r) wave describing the coupling of the diffusion waves is defined by f*lt + (f'{U +
0)TJ)X
-
TJXX
= Ex,
U(-,o) = o.
(- 4 9 v
[bA2
>
A careful estimate on rj is obtained in [171] with which the following theorem is established there. T h e o r e m 5.2.1. Suppose that the system (5.2) is strictly hyperbolic, and each characteristic field is either genuinely nonlinear or linearly degenerate. Let U = U{x—ai) be the weak viscous jD-shock wave solution to (5.41) corresponding to the p-th genuinely nonlinear characteristic field; let 8 G R be the translation uniquely determined by the excess mass
1[UQ{X)
— U(x)]dx through (5.39).
There are positive constants ci,C2 and C3, depending only on / , so that if u0{x)-U{x)€H1(R) and e = \u+ — w_| < ci, / (UQ(X) - U(x - S))dx\ < c2£,
[[1 + (x - S)2]\u0{x) - U(x - S)\2dx < c 3 , / — u0(x) - ^ - ^ ( ^ )
dx
< +°°i
then there exists a unique global (in time) smooth solution, u{x,t), to (5.41) satisfying lim / / |ti(x, r) - U(x -ar-6)0(x, r) - r)(x, r)\2dxdr = 0, > *- +°° it JR lim / \u(x,t) - U(x -
lim sup \u(x,t) — U(x — at — 6)\ = 0. t-++coxeR
An interesting observation by Goodman, Szepessy and Zumbrun [49] is that the estimate on 77 in [171] is sharp for the case f(ui,u2) = {u\ - u\, u2). The result in [143] has been refined by Liu and Wang in [111] by determining exactly the optimal shock strength in the sense that the coupling energy method does not work beyond this strength.
184
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
So far, the results mentioned for the system (5.34) show that a weak viscous shock profile is nonlinearly stable in the L^ and Li norm in the sense that a small initial disturbance, under suitable restrictions, will die out as time tends to infinity. However, there is no rate of convergence given (except for the scalar case n = 1). One of the important problems of both mathematical and physical interest, which remains open, is the Zq-stability of viscous shock profiles [155]. To this end, it seems that an optimal rate of convergence to viscous shock waves in Li and L^ norms will be needed. As a step in this direction, the linearized stability of traveling wave solutions for the viscous hyperbolic system (5.41) is studied by Xin in [187] where it is shown that for a given traveling wave with shock profile from any characteristic family, there exists an appropriate weighted norm space such that the traveling wave is exponentially stable in this space. As a consquence, if the initial distrubance has average zero and decays exponentially fast as \x\ —>• oo, then the corresponding solution of the linearized equation decays to zero exponentially fast in t on any compact interval in x. The study on linearized stability of shock profiles for systems of conservation laws with viscosity is made also by Pego [157]. Now we turn to the case when u+ can be connected to w_ by a weak rarefaction wave. It is also expected that the solution of (5.34), (5.35) in a neighborhood of the rarefaction wave exists and approaches asymptotically the rarefaction wave determined by u^. However, due to the fact that rarefaction waves are expansive, the analysis of the stability of rarefaction waves would be very different from that for viscous shocks. The stability of weak rarefaction waves for a barotropic model of compress ible viscous gas ( 2 x 2 system) is studied by Matsumura and Nishihara [144] under a smallness condition on the initial data. The stability result for com pressible Navier-Stokes equations (3 x 3 system) is obtained by Liu and Xin [124]. For a general system (5.34) with n = 2 and positive viscosity, the asymp totic behavior towards rarefaction waves of the solution is investigated by Xin in [183] and [184] where the system is genuinely nonlinear and strongly cou pled. Suppose that the 2 x 2 matrix (df/du) has real and distinct eigenvalues Ai (u) < \i{u), with corresponding right and left eigenvectors r,- and /,• satisfying ?fcri{u) = li(u)^ li(u)rj(u)
\i(u)n(u),
= \i(u)li(u)1 = Sij,
(5.43) i,j = 1,2,
and the system (5.34) is strongly coupled in the sense that j ^ - j ^ O ,
V«GQ
(5.44)
Vanishing viscosity and nonlinear stability of waves
185
where ft is some region in R2. The result obtained in [183] is stated as T h e o r e m 5 . 2 . 2 . Suppose (5.43) and (5.44) hold, and the Arth characteristic field is genuinely nonlinear (k = 1 or 2). Then for each fixed te_ G ft there exists a positive constant e, such that if u+ G Rk(u-) and \\u0 — ti^H^a + llwOa:||#i 4||u+ — u__|| < e, then the problem (5.34) (5.35) has a unique global solution w(x,t),t > 0, satisfying t*-ti£ ux
GC°(0,+oo;L2) GC°(0,+oo;#2)
and lim sup \uR(x,t) *->+°° xeR1
— u(x,t)\
= 0.
To prove this theorem, the first step is to approximate uR(x,t) by a smooth function $ ( # , £ ) , then derive energy estimates on the perturbation of $ . The method used is strongly motivated by the interesting work in [47] for nonlinear stability of viscous shock waves. It should be pointed out that in the analysis of the stability of viscous shock waves (cf [47]), due to the fact that viscous shock waves are compressive, if one works on the antiderivative of the perturbation of the viscous shock wave, then the hyperbolic part would not cause much difficulty in the energy estimates. However, since rarefaction waves are expansive, the treatment of the hyperbolic part becomes the main difficulty in the analysis which is overcomed with the help of the condition (5.44). For a general system (5.41) with n > 2, a result which discribes the time asymptotic behavior and the point wise convergence, toward an approximate "Burgers" rarefaction wave is obtained recently by Szepessy and Zumbrun [172]. For a rarefaction wave which is not necessarily weak, the study of stability has been made by Matsumura and Nishihara [145] for the isentropic viscous system with p(v) = v ~ 7 , 1 < 7 < 2. So far, we have discussed the nonlinear stability of shock profiles and rarefac tion waves. Another important elementary wave for the system of hyperbolic conservation laws is the contact discontinuity. The large time asymptotic be havior towards contact discontinuities for the 1-D compressible Euler equation with uniform viscosity has been investigated recently by Xin [190]. Consider the 1 — D compressible Euler equations in Lagrangian coordinates:
{
vt - ux = 0, ut+px = 0,
xGR\t>0
(5.45)
Et 4- [up)x = 0, where v,u,E and p are specific volume, velocity, total energy, and pressure, respectively. Denote by e the internal energy. Then E = e + ^u2. We will only
186
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
consider polytropic gases, i.e., we assume that the equation of state is given by p(v,e)
= -.
The contact discontinuity solution of (5.45) is given by U = (v,u,E) that
fiS0,*SM(«,<) = { £ HI
such
(5.46)
where vi and vr are positive constants, and v\ ^ vr. We are interested in the large time behavior of solutions towards the inviscid discontinuity given in (5.46) for the corresponding viscous system of the form
{
Vt — ux = evxx, ut +Px =euxx, Et + {up)x
xeR11t>0
(5.47)
=eExx,
with initial data (ti, w, E){x, 0) = (v0(x), tio(x), E0(x))
= U0{x),x £ R1,
(5.48)
where e is a positive constant. A simple calculation for the linear heat equation shows that an inviscid con tact discontinuity is only a metastable state for the viscous problem. Thus, despite the nonlinear modes in (5.47), one would expect that a similar phe nomenon arises for the problem here. Indeed, the main result in J190] states, roughly speaking, that although the inviscid contact discontinuity U(x,t) is not an asymptotic state for (5.47) and (5.48), one can construct a smooth viscous contact wave, U(x,t), with the following properties (i) U(x,t) solves (5.47) exactly. (ii) the estimate \\U(;t)-U(;t)hrM0,p > 1. (iii) U(xit) is nonlinearly stable for the problem (5.47) and (5.48). Furthermore, the asymptotic structure of the solutions to (5.47) and (5.48) is described by a shifted viscous contact wave superimposed on the nonlinear diffusive waves in the sound wave families, and the shift and the masses carried by the nonlinear diffusion waves can be determined a priori from the distribution of the initial excess mass. To_describe the asymptotic ansatz, we first introduce the viscous contact wave, U(x,t), as the solution to the problem: UM) S 0,
P(^)=P^I,
{l\=:i)*=Vl
+ {Vr_Vl)H{x),
(5-49)
Vanishing viscosity and nonlinear stability of waves where H(x) is the Heaviside function. It is clear from (5.49) that U{x,t) the following properties: (a) U(x,t) solves system (5.47) exactly;
187 has
(b) \\U(.,t) - U{;t)\\L, = 0(l)(e(l + * ) ) * for p > 1; (c) vx< 0 ( > 0) if vr > vi(vr
^-wrA^m^)with c(y) =
<"■»■ <"•>
gr 1 (exp(a i m t /2) - l)exp(-t/ 2 )
Too V T F + (exp(a t m,/2 - 1) / exp(-£ 2 )d£
Jy where Ai, A2 = 0, and A3 are the eigenvalues of the Jacobian matrix of the flux in (5.45) with corresponding left and right eigenvector r,- and U{i = 1,2,3) respec tively, U\Q = Ui and C/30 = Ur,oti = v M ^ * o ) • n(^»o), and m,- = / 0i(xJt)dx is the mass carried by 0,-. It should be noted that 0i(x,t) Burgers' equation (with speed A,(v,o))- We set 0(x1t)
= e1(x1t)r1(Ul)
solves the convected
+ e3(x1t)r3(Ur).
(5.51)
To handle the interactions between decoupled diffusive sound waves and the viscous contact wave, we need to introduce a coupled diffusion wave, D(x1t)i as the solution to the following problem I du — d2x — sd\xx
=
(Si)x
d2t + (A(d 3 - di))9 - ed2xx = ( 5 2 ) , - (V W ) ) ( * , D)x I dst + d2x — ed$xx
y=(58), - (2&(*+o 3 )+(y^e 3 - y j ^ ) i(d3 - dx))^ with
D{x,0) = Do(x) = 0, where
(Sl,S2,S3) = ( J ( a i » ? + d3fli), - J ( " i * ? r i ( t f i ) + a 3 0§ra(tf r ), §(*!*? + *3*§)) ,
188
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
is the major souce term due to wave interactions. By a delicate approximate parametrix method, the following optimal global pointwise estimate for the cou pled diffusive waves is obtained, namely, P r o p o s i t i o n 5.2.3. Let D(x,t) be the coupled diffusive wave. Denote by R(U) =
(r1(U),r2(U),r3(U))
the matrix of the right eigenvectors of the Jacobian matrix of the fluxes in (5.45). Set D = R{U)X£ + R(U)£X = (R(U)O*- Then, for suitably chosen N, fO(l)exp{-c(|x|+*)},
(lal6
' ' ) " on) ( K ]
|x|>e'(* + l)
*»+"
I (min|x-A,t| + v^)
^
M'+fl 1/2
'>/rTTmin(|^~A^| + V^) otherwise
) 1/2
I'
where /,(*) = {x\\x - Xi{Uio)(t + 1)| < CNy/tlog(l
+*)}.
The initial excess mass can be decomposed as follows +oo
/
Q(Ur
(UQ - U){x,0)dx
- Ui)r2{U) + miri(Ui)
+ m 3 r 3 (t/ r ),
(5.52)
= x
-oo
from which #o,wii and m 3 can be determined uniquely when S = \Ur — Ui\ is small. The asymptotic ansatz for the solution of (5.47) and (5.48) is defined then as Ua{x, t) = U(x + xo, t) + 0(x, t) + D(x, t). (5.53) The main theorem states that there exists a unique solution to the problem (5.47) and (5.48) in the neiborhood of Ua(x,t) whose asymptotic behavior is completely determined by Ua(x,t). More precisely, T h e o r e m 5.2.4. Let Ua{x)t) be the function constructed above, and e be a fixed positive constant. Then there exist positive constants &i and S2 such that if _
(i)£/o-tf e # W , (ii)*=|£/r-0i|<*i, / (Uo -U)(x)dx\ + f (1 + x2)\U0 - U\2dx < <S2, 1 WR I JRI then there exists a unique global solution,C/(x,t), to the Cauchy problem (5.47) and (5.48) with the following properties: (ajCZ-^GCaC+oo),^1^1)), (iii)
Vanishing viscosity and nonlinear stability of waves
189
(b) U - Ua £ C([0, +00); H^M1)) H L2([0, +OO), H1^1)), (c) ^ \\U(.,t) - U(- + x0,t)\\Lp = 0 for all p > 2, (d) lim / / |C/(x,r)-*7a(z,r)|2dx
190
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
satisfying
~*~ "^ where a is the speed of the shock and (t/±, v±) are the limiting states
(
;
*
)
=
(
:
)
<
-
>
(5.56)
■
Viscous shock waves are classified by the relative position of its speed a and the characteristic speeds Ai, A2 of (5.54): Ai = aw,
A2 = v.
(5.57)
The second equation in (5.54) is the Burgers equation. By the classical shock wave theory, it is known that a viscous shock wave is always compressible with respect to A2 : A 2 (u_,v_) = v_ > a > v+ = A2(t«+, v+).
(5.58)
The wave is overcompressive if it attracts characteristics on both sides Ai(u_,v_) >Ai(u+,v+),
(5.59)i
marginaly overcompressive if Ai(i/_,v_) = a > \i(u+,v+)
or
Ai(u_,v_) > a = Ai(t/+,v+),
(5.59)2
or
Ai(w_,v_) < a = Ai(ti + ,v + ),
(5.59)3
marginaly undercompressive if Ai(w_,t;_) = a < Ai(tt+,v+) and a classical Lax shock if a > Ai(u_,t/_),Ai(u+,r;+)
or
a < Ai(u_, v_),cr <
\i(u+,v+). (5.59)4 For strictly hyperbolic systems, classical Lax waves satisfy the stability cri terion, introduced by Lax in [102], and are saddle-node connections of the equations (5.56). Given end states {u±,v±) there is only one travelling wave (<£, V>) except the translation in the physical variable x — at. A perturbation of
Vanishing viscosity and nonlinear stability of waves
191
a classical Lax shock gives rise to a translation of the wave and the diffusion wave of other characteristic family, as discussed previously. The amount of the translation and strength of the diffusion wave are determined through the time-invariants
£ ( ( 0 <•■«> -(*)<-•") * =
£ ((«) (*,o) ~ (*) w ) d x '' - °-
(56o)
'
However, nonclassical waves behave distinctly from the classical Lax shock waves. Overcompressive viscous shock waves are node-node connections of (5.56). It is shown in [125] that there is a one-parameter family of such waves for given end states (u±1v±). Overcompressibility, (5.58) and (5.59)i prevent diffusion waves with characteristic speed from leaving the shock. A perturbation of an overcompressive viscous shock wave gives rise to another wave profile within the one-parameter family, properly translated. The translation and the new wave profile are uniquely determined by the conservation laws (5.60). For the model (5.55), marginal overcompressive waves satisfy the first inequality in (5.59)2 but not the second. Due to the fact that (u_, v_) is a degenerate critical point of (5.56), there are two types of wave profiles, a monotone profile and a one-parameter family of nonmonotone profile. The L\(x) distance between the monotone one and any nonmonotone is infinite; while the L\{x) distance among nonmonotone ones is finite. Thus the conservation laws (5.60) dictate that a perturbation of the monotone one yields only the translation of the wave; while a perturbation of a nonmonotone one gives rise to another nonmonotone wave. New difficulties arise for the stability analysis since waves may be nonmono tone, or, in the case of monotone marginaly overcompressive waves, there is no natural way to make use of the conservation laws (5.60) because there is no diffusion wave leaving such a travelling wave. These difficulties are overcome through an intricate combination of characteristic-energy and weighted-energy methods. We refer the reader to [125] for details. There is extensive literature on shock structure and stability in magnetohydrodynamics, see, for example, Chu [19] and Germain [45] and the references there. The dynamic stability and evolutionary conditions (linearized stability) of viscous shock profiles have been studied. However, the rigorous study on the nonlinear stability of viscous shock profiles has started only recently. Freistuhler and Liu study the nonlinear stability of overcompressive shock waves in a rotationally invariant system of viscous conservation laws in [44]. In physical systems of conservation laws, rotational invariance typically arises due to natu ral isotropy. Systems such as those describing magnetohydrodynamic or elastic plane waves display rotational symmetry in its generic form, which induces an
192
Quasilinear Hyperbolic Systems and Dissipative Mechanisms
inviscid model as described in [44], namely ut + {\u\2u)x where u(x,t) laws is
= 0
(5.61)
G M n (n > 2). The corresponding system of viscous conservation ut + (\u\2u)x
= fiuxx,
fi>0.
(5.62)
It is proved in [44] that overcompressive shocks for (5.62) can be stable against small perturbations, i.e., given the profile *('/AO is small (in an appropriate sense), then the solution u of (5.62) with initial data wo converges time-asymptotically to another profile > : lim sup \u(x,t) — <j>(x —
(5.63)
This situation is thus similar to that of small shocks associated with genuinely nonlinear modes, discussed by Liu in [120]. However, there are two important differences between the stability of overcompressive shocks and that of classi cal Lax shocks. It is well known that from any viscous shock wave u*(x,t) =* (x ~nat)
,*{±oo) = i / * , ! / - ^ u + , any phase shift x >-> x + 8,S G M,
trivially produces another (just shifted) shock wave with the same end states. For classical Lax shock waves, the profile <j> is vice versa uniquely determined modulo such phase shifts. By contrast, in the case of overcompressive shock discussed in [44], whole families of orbitally different profiles exist. Correspond ingly, the asymptotic profilein (5.63) generically difffers from the profile * of the unperturbed solution not only by a phase shift, but also by a true change in shape. The second important difference concerns the fact that although it is generally believed that the stability of Lax shocks is uniform in the viscosity //, this is no longer true for the overcompressive shocks considered. As far as an undercompressive shock is concerned, though there is strong numerical evidence for the stability of undercompressive shocks, their stability has not been verified analytically. We present the first result given by Liu and Zumbrun recently in [127] where the authors study the nonlinear stability of smooth travelling wave solutions associated with undercompressive shock waves to viscous complex Burgers equation. In contrast to the nonlinear stability theory of shock profiles associated with classical Lax shocks, the asymptotic form — the mass carried by the diffusive waves in the transversal directions and the phase shift on the profile-can be determined a priorily by the initial distribution of the excess mass and the conservation laws. The precise form of
Vanishing viscosity and nonlinear stability of waves
193
the asymptotic ansatz, which plays an important role in the pointwise estimate, has to be determined dynamically. This is achieved in [127] where the analysis is by direct calculation of the Green's function for the linearized equations, combined with an elaborate iteration argument. Both global L\ and pointwise estimates are obtained.
5.3
The multidimensional case
In this section, we first introduce the global existence of weak solutions for Navier-Stokes fluids in two and three space dimensions, with fairly general initial data. The nonlinear stability of planar shock fronts and rarefaction waves for the viscous scalar conservation law is discussed in the second part. A. The global existence of weak solutions for Navier-Stokes fluids in multidimensions. Consider the Navier-Stokes equations
i> t + div(H = o \ {pu3)t + div(ptiJti) + P(p)Xj = e A u> + Adivti*,.
jghvbfg l ,0)
°
in two and three dimensions. Here p(x,t)1u(x1t) t) = (w 1 (x,t), • • •, un(x,t))(n n= = 2 or 3), and P = P(p) represent the fluid density, velocity, and pressure, e > 0 and A > 0 are viscosity coefficients. We are interested in the global existence of weak solutions for the Cauchy problem for (5.64) with initial data (po, uo) which are small but possibly discontinuous. The global existence of weak solutions for the case of P(p) = Kp has been obtained by Hoff in [58]. To extend the result to the case when P(p) = Kp1,7 r > 1, an essential difficulty occurs. To be specific, solutions are obtained in the case of P(p) = Kp as limits of smooth solutions (ps ,u6), corresponding to mollified initial data p*0 = js */>o> ^0 = Js * u o with the mollifier js. A number of important a priori estimates, indepedent of <J, are obtained, which are sufficient to yield limiting regularity properties for us; but the only estimates for ps are the uniform pointwise bounds. It therefore follows only that ps —^ p weakly, and this is insufficient to pass the limit under a nonlinear pressure term. In particular, it could not be concluded that the limiting pair (/?, u) is a weak solution in the required sense, except in the case P(p) — Kp. Hoff gives an additional argument in [59] which proves that, when P(p) = Kp1,7 r > 1, the convergence ps —^ p is in fact strong. It then follows that P{ps) -» P(p), so that the pair is indeed a weak solution in the required sense. Given a velocity field u(x,t), we define the corresponding vorticity matrix u by ujj>k = ui
— ut .
194 Quasilinear Hyperbolic Systems and Dissipative Mechanisms The pair (p, u) is called a weak solution of (5.64) with Cauchy data (po,uo) if p}u,\/u e L}oc(t > 0) and r + oo
n
/ JRn
poj;{-,0)dx+ JO
I pQUJ0i/;(',0)dx+ / Jmn J0 r + OO
/•
/ JRn
(pil>t+fm'V4>)dxdt
= 0,
/ [puju-\7ijj + P(p)ipXj)dxdt JRn
(5.65)
jfhbcgf v
/-
'
—- I / fc V uj ' V ^ + \(divu)ipXj]dxdt Jo JR* j = 1, • • •, n for all test functions tp (E S)(IRn x [0, +oo)). (The solutions obtained below are somewhat more regular than the one required here). Fix constants e > 0 and A > 0, and assume X/e < 6 n , where 6n =
\ J ( 3 + V2T), » = 3.
nvhbfgj (5 67)
'
The pressure is given by P(p) = if/? 7 , where A' is a positive constant and 7 > 1, and the initial data (po> v>o) are measured in the norm Co = ||po - pllicjM.) + /
[(Po(x) - pf + |uo(x)| 2 ](l + |a:| V d * .
Here p is a positive constant, /3 is positive but arbitrarily small when n = 2, and /? = 0 when n = 3. (This weighting of the L 2 norm serves to compensate for the growth of the fundamental solution of the Laplacian at infinity in two space dimension). The material derivative S? = •, is defined by dw dw w = — = - - - 4- V™ -u = wt + wJx.uJ at at (summation over repeated indices is understood). Given a density-velocity pair (p, u), we define the corresponding effective viscous flux F by F = (e + \)divu - P(p) + P(p). (5.68) The important role played by F will be discussed below, following the statement of the theorem. Finally, we employ the standard notation for Holder norms, {wr=
sup
H«)-«(y)l
Vanishing viscosity and nonlinear stability of waves
195
for functions w : Rn -+ Rm, and
for w : Q C Rn x [0, +00) -> M m . The principal result is T h e o r e m 5.3.1. Let n = 2 or 3, fix £, A,p and f3 as above, and let N > 0 be given (N may be arbitrarily large). Then there are positive constants a and C, depending on K, 7, p, /?, TV, and on upper bounds for £,A, and (6n£ — A)" 1 (see [59] ), and there is a positive constant 0, such that, for given initial data (PO,UQ) satisfying Co < a, ||U 0 ||L». < N, (5.69) the Cauchy problem for (5.64) with data (po,uo) has a global weak solution (p, u) for which sup [[\u(x,t)\2
u(x,t)\2dx
+
t>0J
+esssup t > 0 f{p{x,t) J r + OO
+ 1
- p)2 + *{t)n(\u{x,t)\2
+ Iv
w(x,t)\2)dx (5.70)
r
I [| V «| 2 +
(5.71)
The solution (p, u) satisfies the following regularity conditions: \H;t)\\i°°
t>r>0.
M ^ S g < C{a, r)C0, r > 0,
(5.72)
(5.73)
for all a G (0,1) when n = 2; and
<«>i/.x1[^+oo)<^)cS.'->o. when n = 3; ti(.,t)€ff1(Rn), 1
n
w(.,t),F(-,*)e# (R ), a
n
F(.,f)GC (IR ),
*>0,
for almost all t > 0,
for almost all * > 0,
(5.74) (5.75) (5.76)
196 Quasilinear Hyperbolic Systems and Dissipative Mechanisms where a = 1/2 when n = 2 and a = ^g when n — 3. Finally, (/>, w) -» (/?, 0) as tf -> +oo in the sense that, for all q G (2, oo), lim [\\p -
/5||L~(EMT,OO))
+ IK*, ^lU^ffin) = 0.
(5.77)
T-++00
The effective viscous flux F of (5.68) plays an important role in the present analysis. The quantities p and P evidently may be as rough as /?o> and the velocity V w i s n o * smoother than an L2 function. However, (5.75) and (5.76) indicate that sigularities in P(p) and (e + A)divw exactly cancel, so that F is somewhat smoother than either P or divu. We shall exploit this fact to show that, while the approximating quantities P{ps) and divu6 may converge only weakly, the approximate effective viscous fluxes Fs in fact converge strongly. This result will be derived as a consequence of the equation for F, which is derived as follows for smooth solutions (p, u). We first subtract w-7 times the mass equation (5.64)i from the momentum equation (5.64)2 and rewrite the viscosity terms to get puj + P{p)Xj = (e + \)divuXj
ukx.)Xk.
+ e(v?Xk -
Then it is easy to show AF = div(pu). We now give a brief outline of the essential ideas in the proof of the theorem. Again let (p6, us) be the approximate solutions corresponding to the mollified initial data (P60,UQ). It is shown in [58] that the bounds (5.70)—(5.73) hold uniformly in 6. Thus there exists a subsequence S —> 0 for which u5 —> u locally and uniformly in {t > 0}, and ps —^ /?, s/u6 —^ V w weakly in Lfoc({t > 0}). Now, it was shown in [58] that, even for very general equations of state P, including the isentropic pressure P(p) = Kp1 considered here, the limiting pair (p, u) satisfies (5.70)—(5.77), with the exception that the weak momentum equation (5.66) holds with the pressure P(p) replaced by the weak limit P* of the approximate pressure P(ps). Therefore we only have to show that ps -> p a.e., so that P(p) = P*. For this purpose, we define P* = wkhmP(p6),
r* = wklimip6)1^,
Since the functions p -> P(p) = Kp
1
l
r*>p ~\
/ = strlimF*.
11
and p-+ p ' P*>P{p)
are convex, we know that a.e..
(5.78) 1 1
The proof would be complete if we can show that r* = p " . In fact, by a simple Taylor series expansion, it can be shown that
J J [(P*)1-'1 - P^dxdt f f
i
ft
(5J9)
S
> y y a - T K v - p)dxdt+c- jj (P - pfdxdt,
Vanishing viscosity and nonlinear stability of waves
197
for any compact set S C {t > 0}. The integral on the left-hand side and the first integral on the right-hand side converge to zero as S -» 0 by the weak convergence of ps to p and {p6)1'1 to r* = p1""*7. This shows that ps -> p strongly in Lfoc, as required. To show that r* = p 1 - 7 we pursue an idea of Lions and DiPerna in [36], namely, certain nonlinear changes of dependent variables are admissible in the mass equation (5.64)i, even for weak solutions, due to the fact that this equation is linear in p. Therefore, we are able to compare the quantities p 1 " 7 to r* with the help of their respective differential equations. It is easy to see that the smooth approximations (ps ,u6) satisfy m
(5.80)
which shows, by taking the limit S —> 0, that r* is the weak solution of the equation r? + div(r*u) = j^\[r*(f - P) + Kp], (5.81) where P = P(p). Observe that the specific nonlinear function p1""7 is chosen so that the right most terms in (5.79) and (5.80) are linear in p. Note also the crucial use of the strong convergence of Fs to / . One is able to show, by a more subtle argument based on mollifiers, that p 1 - 7 is the weak solution of the equation: (p^)t
+ d i v ( / - ^ ) = - X _ ( p i - T r ( F -P)
+ Kp}.
(5.82)
Subtracting (5.82) from (5.81) and applying (5.78), we find that the difference w = r* — p 1 - 7 is the weak solution of the following differential inequality wt+diy{wu)
< (—^TJ
w{divu - P*)
(5.83)
with zero initial data. One therefore expects w < 0 a.e. in {t > 0}, which together with (5.78) then shows that w = 0 a.e.. Therefore p 1 " 7 = r* = w & l i m ^ ) 1 " 7 , as required. A number of delicate issues arise in the execution of the above program. Of particular interest are the nonlinear change of variables leading to (5.82), the behavior of weak solutions of the differential inequality (5.83), especially in light of the fact that the regularity of its coefficients degrades near the initial layer t = 0, and finally the precise sense in which the variables p,pl_1, and r* achieve their initial values at t = 0. These issues are dealt with in a sequence of technical lemmas which can be found in [58] or [59].
198 Quasilinear Hyperbolic Systems and Dissipative Mechanisms The global existence result explained above is given by Hoff in [59]. The following viscous polytropic ideal gas equations in IRn (n = 2 or 3) in Euler coordinates are studied recently by Jiang [88].
| £ + div(H* = 0 p [fjf + (v • V)"] =AiAt; + (A + / i ) V (divv) - R v {pO) Cvp(%
RpO(divv) + A(divv)2 + 2pD • D.
+ {v- V)0J =kA0-
Here p,0 and v = (vi,---v n ) are the density, the absolute temperature and the velocity respectively, R, Cv and k are positive constants; A and p. are the constant viscosity coefFficients, p > 0, A+2/i/n > 0; D is the deformation tensor
It is proved that the smooth solutions for the above equations tends to a constant state as the time tend to infinity if the initial data are small perturbations of this constant state. The problem concerning limiting behavior when the viscosity cofficients tend to zero is still open for Navier-Stokes equations in 2 or 3 dimensions. B. Nonlinear stability of planar shock fronts and rarefaction waves for viscous scalar conservation laws in multidimensions. We have discussed the stability of elementary waves for the viscous conser vation laws in one-dimension in Section 5.2. For this topic in multidimensions, we only discuss the case of 2 — D, the other cases being handled similarly. Consider equations of the form u
t + f(u)x + g(u)y = uxx + uyy
(5.84)
where f(u) is a strictly convex function of «, satisfying f"(u)
> a > 0.
(5.85)
The planar viscous shock wave considered here is a solution of (5.84) in the form u(x, y,t) = <j)(x — st), where <j> satisfies
(x) —> m <j)(x) —tur
as as
x —>• —oo x —> H-oo,
Vanishing viscosity and nonlinear stability of waves
199
with ui and ur satisfying the Rankine-Hugoniot condition and the Lax shock condition, i.e., (ui -ur)s = f(ui) - f(ur) Ul >
ur.
Thus <j>'{x) < 0 for all x. The following stability result of planar viscous shock waves is given by Goodman in [48]. T h e o r e m 5.3.2. If ui — ur is small enough, for any p > 1 there is a C > 0 so that if 2 a
Y, I I (i + x nd u(x,y,o)-Hx))?dxdy
\a\<2J
•/K2
then
f [ (u{x,y,t)-«*-*))>d*dy->0
as
t^+oo.
This theorem can be proved by using the energy method. To write the pertur bation as u — (j) — Ux, with U 6 £2, we need a shift &{y,t). Since +00
/
r+00
(u(x,y,t)
- <j)(x -S))dx
-00
= /
(u(x,y,t)
-(x))dx + 8 • (m - ti r ),
J—oo
it is possible to choose S(y,t) so that +00
(u(x, y, t) —(x — S(y, t))dx = 0 for all /
y, t.
-00
Thus, the program is to decompose the solution as u - <j>(x - S(y,t)) + Ux(x, y,t) and to seek L2 estimates for 8 and U. For the special case of s = 0 and gr = 0, it is easy to get
iULu2dxdy+Ls2dy) + 11 (U* + U?)dxdy+?- j j W{x-8)\U2dxdy+ f S2ydy<0. For the general case, one needs some normalization. Normalization 1. Set s = 0. Choose t' =t,x' = x — st, j / = y.
200
Quasilinear Hyperbolic Systems and Dissipative Mechanisms In the primed variables, f{m) = f(ur),
which implies, for weak shocks, that
f({x)) = 0(ui - ur)
for all
x.
Normalization 2. Make g" small. The substitution t' = t,x' = x,y = y + px with p = — g"(<j>(0))/f"(<j)(0)), transforms (5.84) into (dropping the primes) u
t + f(u)x
+ g(u)y
= Uxx + 2pUxy + (1 + P2)u>yy,
(5.86)
where g"({x)) = 0(ui — ur) for all x, which is small for sufficiently weak shocks. Normalization 3. Make g(<j>(x)) G £2 0R)Use t' = t, x' — x, y' = y — crt where a, defined by a — (g(ui) —g(ur))/(ui — t//), is the transverse wave speed. Under this transformation, the new g satisfies (dropping the primes once more) WW)|
for all
x.
(the details can be found in [48]). With these normalizations, we are able to decompose u as u(x,y,t)
=(x - S(y,t)) +
Ux(x,y,t),
then - M + Utx + (f'{)Ux)x + {qUl)x - g{<j>)xSy +W(4>)U)yx - {9,,{d>) ,U)y + (rUS)y = Uxxx + 2pUxyx + (1 + p2)Uyyx - 2p "6y
(5.87)
"(1+P2)^y + (1+P2n2. Here, q and r satisfy f(<j> + Ux) = f() + f'(