Вы находитесь на странице: 1из 25

GLOBAL SOLUTIONS OF THE COMPRESSIBLE

NAVIER-STOKES EQUATIONS WITH LARGE


DISCONTINUOUS INITIAL DATA

Gui-Qiang Chen
Department of Mathematics
Northwestern University, Evanston, IL 60208
gqchen@math.nwu.edu

David Hoff
Department of Mathematics
Indiana University, Bloomington, IN 47405
hoff@indiana.edu

Konstantina Trivisa1
Department of Mathematics
Northwestern University, Evanston, IL 60208
trivisa@math.nwu.edu

Abstract
We prove the global existence of weak solutions to the Navier-Stokes equations for compress-
ible, heat-conducting flow in one space dimension with large, discontinuous initial data, and
we obtain a-priori estimates for these solutions which are independent of time, sufficient to
determine their asymptotic behavior. In particular, we show that, as time goes to infinity,
the solution tends to a constant state determined by the initial mass and the initial energy,
and that the magnitudes of singularities in the solution decay to zero.

1991 Mathematics Subject Classification. 35B40, 35D05, 76N10, 35B45.


Key words and phrases. Navier-Stokes equations, compressible flow, global discontinuous
solutions, large-time behavior, large discontinuous initial data, uniform bounds.
1
Current Address. Mathematics Department, University of Maryland, College Park, MD
20742-4015.
1
1. Introduction

We prove the global existence of weak solutions to the Navier-Stokes equa-


tions for compressible, heat-conducting flow in one space dimension with large,
discontinuous initial data, and we obtain a-priori estimates for these solutions,
which are independent of time, sufficient to determine their asymptotic behav-
ior.
The equations under consideration express the conservation of mass and
momentum and the balance of energy:
vt − ux = 0,
ut + p(v, e)x = uvx x ,

(1.1)  
u2
e + 2 + (up(v, e))x = uuxv+λex x .

t

Here v, u, e, and p represent respectively the specific volume, velocity, specific


internal energy, and pressure;  and λ are fixed positive viscosity parameters;
and x is the Lagrangian coordinate, so that x =constant corresponds to a
particle trajectory. We shall assume that e, v, and p are related by the equation
of state of an ideal, polytropic fluid:
e = pv/(γ − 1) = cv θ,
where θ is the temperature, and γ > 1 and cv > 0 are constants.
In the present paper we study the initial boundary-value problem for (1.1);
thus 0 < x < 1 without loss of generality, and the boundary conditions
(1.2) u(i, t) = 0, ex (i, t) = 0, i = 0, 1,
are to hold for t > 0. On the other hand, our existence results can be extended
with little difficulty to the Cauchy problem, as in [17] and [18].
We now give a precise formulation of our results. Let initial data
(1.3) (v, u, e)|t=0 = (v0 (x), u0 (x), e0 (x)), 0 ≤ x ≤ 1,
be given, satisfying
(1.4) C0−1 ≤ v0 (x) ≤ C0 , e0 (x) ≥ C0−1 , ku0 kL4 + ke0 kL2 + TV(v0 ) ≤ C0 ,
for a constant C0 > 0. Define the following functionals for weak solutions of
(1.1):
E(t) = sup σ(s)kux (·, s)k2 + σ 2 (s)kex (·, s)k2

0≤s≤t
Z t
(1.5) + (kux (·, s)k2 + kex (·, s)k2 + σ(s)kut (·, s)k2 + σ 2 (s)ket (·, s)k2 )ds,
0
F(t) = sup σ 2 (s)kut (·, s)k2 + σ 3 (s)ket (·, s)k2

0≤s≤t
Z t Z t
2 2
(1.6) + σ (s)kuxt (·, s)k ds + σ 3 (s)kext (·, s)k2 ds,
0 0
where σ(t) =min{t, 1}, and k · k denotes the norm in L2 (0, 1). The following
theorem then gives the main result of this paper.
Theorem 1.1 (Well-Posedness and Large-Time Behavior). Given initial data
(v0 , u0 , e0 ) satisfying (1.4), there is a global weak solution (v, u, e) such that
v, u ∈ C([0, ∞); L2 ), e ∈ C((0, ∞); L2 ) with e(·, t) * e0 weakly in L2 as
t → 0. Furthermore, there is a constant M depending on C0 , but independent
of t > 0, such that the following hold:
(1.7) M −1 ≤ v(x, t) ≤ M, M −1 ≤ e(x, t) ≤ M σ −1 (t),
(1.8) TV[0,1] (v(·, t)) ≤ M , kv(·, t0 ) − v(·, t)k ≤ M |t0 − t|1/2 ,
(1.9) kux (·, t)k ≤ M σ −1/2 (t), kex (·, t)k ≤ M σ −1 (t),
(1.10) ku(·, t)kL∞ ≤ M σ −1/4 (t), E(t) + F(t) ≤ M.
Finally, the solution tends to a constant as t → ∞ in the sense that
(1.11) k(v − v∞ )(·, t)kL∞ (0,1) + k(u, e − e∞ )(·, t)kH 1 (0,1) → 0,
R1 R1 u20 (x)

where v∞ = 0 v0 (x)dx and e∞ = 0 e0 (x) + 2 dx.

The construction of solutions proceeds in two steps. In the first, we assume


that v0 is piecewise H 1 , having a finite number of points of discontinuity. So-
lutions for this case are obtained as limits of approximate solutions derived
from a suitable semidiscrete difference approximation to (1.1). Then in the
second step we exploit the uniform total variation estimate (1.8) to complete
the solution operator for the more general case that v0 ∈ BV . This entire con-
struction parallels the analysis of Hoff [10] and [11], in which global existence
was obtained for small initial data. Our results thus improve upon [10] and [11]
by allowing for large, discontinuous initial data, and by achieving estimates
(1.7)-(1.10) which are independent of time, sufficient to obtain the large-time
behavior (1.11) of solutions. The crucial step here is to obtain pointwise upper
and lower bounds for the specific volume v. Our derivation of these bounds
starts from the idea of Kazhikov and Shelukhin [19], but elaborates upon it
in a new and significant way. We therefore give the complete details of this
argument (Lemma 2.2 below).
The pointwise bounds in (1.7) for the specific volume (equivalently, the den-
sity) show that neither vacuum states nor concentration states can occur, no
matter how large the initial data is. This is one of several important differ-
ences between the Navier-Stokes equations and the inviscid Euler equations,
for which vacuum states may in fact occur for large initial data and for certain
equations of state (cf. [2, 4]). It is also relevant in this regard that solutions
of the Navier-Stokes equations show certain instabilities when vacuum states
are allowed (cf. Hoff and Serre [14]).
Once these pointwise bounds (1.7) for v have been established, the higher-
order regularity assertions of Theorem 1.1 follow very much as in [10] and
[11]; we therefore omit most of the details of their proofs. The asymptotic
behavior (1.11) can then be derived as a consequence of the time-independent
bounds (1.7)-(1.10), together with the weak form of the equations (1.1). This
argument is given in Section 3.
The construction of solutions for the intermediate case that v0 is piecewise
H 1 plays a largely auxiliary role in the present paper. Still, there are interesting
and important observations that can be made concerning the propagation of
singularities for this case. These observations are made in [10] and [11] as well,
but their range or validity is now extended to the case of large initial data.
Thus let v0 be piecewise smooth, having jump discontinuities at isolated points
y1 < · · · < yN . Then, by applying the Rankine-Hugoniot conditions to (1.1)
(together with the hypothesis that u and e are continuous in positive time) as
in [10] and [11], we find at the heuristic level that discontinuities in v, p, ux ,
and ex occur only at x = yi , and satisfy the following jump conditions:
h ux i he i
x
(1.12) p(v, e) − = 0, = 0,
v v
where [w] denotes a jump of the function w across x = yi : [w(yi , t)] = w(yi +
0, t) − w(yi − 0, t). Now, combining (1.12) with the first equation in (1.1), we
find that  
−1 γ−1 1
[log v]t =  [p] = e .
 v
Since e is positive and 1/v is a decreasing function of log v, we therefore antic-
ipate exponential decay of the magnitudes of these discontinuities as t → ∞.
This is indeed the case, and its proof depends crucially on the fact that the
pointwise bounds for v and e of Theorem 1.1 are independent of time. The
precise statement is as follows.
Theorem 1.2 (Discontinuities and Large-Time Behavior). Assume, in addi-
tion to the hypotheses of Theorem 1.1, that v0 is piecewise H 1 , having isolated
jump discontinuities at points y1 < · · · < yN . Then, for t > 0, each of the
quantities v(·, t), p(·, t), ux (·, t), and ex (·, t) has one-sided limits at each yi , and
the jump conditions (1.12) are satisfied in a strict pointwise sense. Moreover,
 Z t 
−1
(1.13) [log v(yi , t)] = [log v(yi , 0)] exp − αi (s) ds ,
0

where
[(1/v)(yi , t)]
(1.14) αi (t) = −(γ − 1)e(yi , t) .
[log v(yi , t)]
Finally, there is a constant M depending on C0 , but independent of t and N ,
such that
|[(v, p, ux , ex )(yd , t)]| ≤ M min exp{−M −1 t1/2 }, σ(t)−3/2 exp{−M −1 t}


→ 0, as t → ∞.
Uniqueness is a rather delicate issue for solutions which are as general as
those of Theorem 1.1, owing to the absence of uniform regularity in the initial
layer near t = 0. By imposing slightly stronger conditions on u0 and e0 ,
however, we can improve the smoothing rates implicit in the definitions of
E(t) and F(t) sufficiently to prove that solutions are in fact unique and depend
continuously on their initial values. The following theorem is established in
Hoff [10] and [13] for small solutions and can be extended to the present case
with little difficulty.
Theorem 1.3 (Regularity and Stability). Assume, in addition to the hypothe-
ses of Theorem 1.1, that
(1.15) T V (u0 ) + T V (e0 ) ≤ C0 .
Then there exists M > 0 independent of t such that the solution of Theorem 1.1
satisfies the additional estimates:
(1.16) kux (·, t)k ≤ M σ −1/4 (t), kex (·, t)k ≤ M σ −1/4 (t),
(1.17) TV[0,1] (u(·, t)) ≤ M σ −1/4 (t), TV[0,1] (e(·, t)) ≤ M σ −1/4 (t).
Moreover, solutions satisfying (1.7)-(1.10) and (1.16)-(1.17) are unique and
depend continuously on their initial data in the sense that, if (v1 , u1 , e1 ) and
(v2 , u2 , e2 ) are any two such solutions, and if S(t) is defined by
S(t) = k(v2 − v1 )(·, t)k + k(u2 − u1 )(·, t)k−α + k(e2 − e1 )(·, t)k−β ,
where α and β are small and positive (k · k−r denotes the norm in the negative
Sobolev space H −r (0, 1)), then given T > 0, there is a constant C(T ) such
that, for 0 ≤ t ≤ T ,
S(t) ≤ C(T )S(0).
We remark that the results of Theorems 1.1 and 1.2 can be converted to
equivalent statements for the Navier-Stokes equations in Eulerian coordinates
(cf. [3]):
ρt + (ρu)x = 0,
(ρu)t + (ρu2 + p)x = (ux )x ,
(1.18)
(ρE)t + (u(ρE + p))x = (uux )x + (λex )x ,

where E = e + u2 /2 is the total specific energy. Corresponding statements


concerning uniqueness and continuous dependence are more subtle, however,
owing to the fact that the change of variables from Lagrangian to Eulerian co-
ordinates is solution–dependent, and our solutions are only minimally regular.
The Navier-Stokes system (1.1) has been studied by a great many authors in
a large variety of contexts. Besides the results of Hoff [10, 11] and Kazhikov and
Shelukhin [19] referred to earlier, the reader may consult Amosov and Zlotnick
[1], Jiang [16], Matsumura and Yanagi [21], and Fujita-Yashima, Padula and
Novotny [7], and the references contained therein. As far as we are aware,
the present paper is the first to give time-independent estimates for large,
discontinuous initial data, for the full nonbarotropic system (1.1). See also
Hoff [12] for a global existence result for the corresponding multidimensional
system, but with small, discontinuous initial data, and Lions [20] for the global
existence of weak solutions for the isentropic Navier-Stokes equations in the
presence of vacuum.
The approach developed here has also been applied to establishing the global
well-posedness and large-time behavior of discontinuous solutions with large
initial data to the Navier-Stokes equations for a compressible reacting flow,
describing dynamic combustion (see [3, 6, 8, 9, 22]). These results will be
presented in our forthcoming paper [5].

2. Difference Approximations and A-Priori Estimates


In this section we construct semidiscrete difference approximations of solu-
tions to (1.1)-(1.4), and we derive various a-priori bounds for these approxi-
mations required for the subsequent analysis.
Let h be an increment in x such that Kh = 1 for K ∈ Z+ , xk = kh for
k ∈ {0, 1, · · · , K}, and xj = jh for j ∈ { 21 , 32 , · · · K − 12 }. Approximations
vj (t), uk (t), ej (t) to v(xj , t), u(xk , t), e(xj , t) are then constructed as follows:
(2.1) v̇j = δuj ,
 
δu
(2.2) u̇k + δpk = δ ,
v k
(δuj )2
 
δe
(2.3) ėj + pj δuj =  + λδ .
vj v j
vk+ 1 +vk− 1
Here pj = p(vj , ej ), ej = e(vj , ej ), vk is taken to be the average vk = 2
2
2

1 3 1
with j ∈ { 2 , 2 , · · · , K − 2 }, k ∈ {0, 1, · · · , K}, and δ is the operator defined
by
wl+ 1 − wl− 1
2 2
δwl = , l = k, or j.
h
For the time being we assume only that initial data (vj (0), uk (0), ej (0)) for the
above ordinary differential equations has been specified and satisfies
(2.4) u0 = uK = 0, δe0 = δeK = 0,
and
X X
(2.5) C0−1 ≤ vj (0) ≤ C0 , ej (0) ≥ C0−1 , u4k (0)h + e2j (0)h ≤ C0 .
k j

We also assume that there are distinguished points 0 < xk1 < xk2 < · · · <
xkN < 1, N = N (h), N 4 h ≤ 1, such that
X X
(2.6) |[vki (0)]| + |δvk (0)|2 h ≤ C0 .
k=ki k6=ki
Now, standard theory of ordinary differential equations applies to show that
the initial-value problem (2.1)-(2.5) has a unique solution (vj (t), uk (t), ej (t)),
defined at least for small time. The a-priori bounds to be derived in this
section will show that these solutions exist globally in time, and will provide
sufficient compactness both to extract limiting solutions as h → 0 as well as
to determine their asymptotic behavior.
In the following, M will denote a generic positive constant independent
of t and h. We begin with certain basic energy estimates for the difference
approximations.
Lemma 2.1. The solutions (vj (t), uk (t), ej (t)) of (2.1)-(2.5) satisfy the fol-
lowing relations:
X
(2.7) vj (t)h = 1,
j
X 1X 2 X 1X 2
(2.8) ej (t)h + uk (t)h = ej (0)h + uk (0)h,
j
2 k j
2 k
Z t
(2.9) E(t) + V (s)ds = E(0) < ∞,
0

where
 P 1
P 2
 E(t) = j ((ej − 1 − log ej ) + (γ
 − 1)(v j −1 − log v j )) (t)h + 2 k uk (t)h,
P  (δuj )2  2
 V (t) = j vj ej (t)h + k v eλ(δe1ke) 1 (t)h.
P
k k+ 2 k− 2

Proof: Step 1. We obtain from (2.1) that


d X X
(2.10) vj (t)h = δuj (t)h = uK − u0 = 0.
dt j j

Step 2. Summing over j in (2.3) and integrating with respect to t, we obtain


Z tX Z tX
X X (δuj )2
(2.11) ej (t)h + (pj δuj )hds = ej (0)h+ hds.
j 0 j j 0 j
v j

We obtain in a similar way from (2.2) that


Z tX Z tX
1X 2 1X 2 (δuj )2
(2.12) uk (t)h − (pj δuj )hds = uk (0)h − hds.
2 k 0 j
2 k 0 j
vj

Then (2.8) follows from (2.11) and (2.12).


Step 3. Differentiating the energy E = E(t), we obtain
d X 1 1
 X
E(t) = (ėj − ėj ) + (γ − 1)(v̇j − v̇j ) h + uk u̇k h.
dt j
ej vj k
Then applying (2.3) yields
(  !)
d X X (δuj )2  X λ(δek )2
E(t) = (ėj + (γ − 1)v̇j )h − h+ h
dt j j
vj ej k
vk ek+ 1 ek− 1
2 2
X
+ uk u̇k h.
k

Integrating over [0, t] and using (2.10), we then get


!
X 1X 2 X 1X 2
E(t) − E(0) = ej h + u h− ej (0)h + u (0)h
j
2 k k j
2 k k
Z t (X  ! )
(δuj )2 λ(δek )2
 X
− h+ h ds.
0 j
v j ej
k
v k ek+ 1e
k− 1
2 2

The result (2.9) then follows from this and (2.8). 


Next we derive pointwise upper and lower bounds for the specific volume
v = vj (t) as well as a pointwise lower bound for the internal energy e = ej (t).
These bounds constitute the most important part of the entire analysis: the
fact that they give time-independent estimates insures that subsequent esti-
mates for higher-order terms are time-independent as well, and it is this time-
independence that allows for the determination of the asymptotic behavior of
solutions.
Lemma 2.2. There exists a constant M > 0, independent of t and h, such
that
(2.13) M −1 ≤ vj (t) ≤ M < ∞.
Proof: Step 1. We claim that, for any t > 0, there exist j1 (t) and positive
constants α, β with α < β such that
α ≤ vj1 , ej1 ≤ β.
To see this, observe that (2.9) implies
X
(f (vj ) + f (ej ))hds ≤ M,
j

where f (s) = s − 1 − log s ≥ 0. Therefore, there exists j1 = j1 (t) such that


(2.14) f (vj1 ), f (ej1 ) ≤ M.
The result now follows from the properties of function f .
Step 2. We claim that, for any t ≥ 0,
Rt
1 + γ−1
 0
P (s)Qj (s)ej (s)ds
(2.15) vj (t) = ,
P (t)Qj (t)
where
( Rt
P (t) = vj1 (0)exp{ 1 0 pj1 (s)ds},
(2.16)
Qj (t) = vj (0)v1 j (t) exp{ 1 (j1 ,j) (uk (0) − uk (t))h},
P
1

and
P
X   j1 <k<j , j1 < j,
j1 = j1 (t), = 0, j1 = j,

− P
j<k<j1 , j1 > j.
(j1 ,j)

To prove this, we let L = log v and rewrite (2.2) in the form


(2.17)  δ L̇k = u̇k + δpk .
Integrating with respect to t, we obtain
Z t
 δLk (t) =  δLk (0) + uk (t) − uk (0) + δpk (s)ds,
0
and summing over j,
(2.18) (Lj (t) − Lj1 (t))
X Z t
= (Lj (0) − Lj1 (0)) + (uk (t) − uk (0))h + (pj (s) − pj1 (s))ds.
(j1 ,j) 0

We divide both sides of (2.18) by  and take the exponential to obtain


 
 Z t 
vj (t) vj (0)  1 X  1
= exp (uk (t) − uk (0))h exp (pj (s) − pj1 (s))ds ,
vj1 (t) vj1 (0)    0
(j1 ,j)

which implies
Rt
exp{ 1 0 pj (s)ds}
(2.19) vj (t) = .
P (t)Qj (t)
n o
Multiplying (2.19) by (γ−1)
ej (t) and rearranging then give

γ − 1 t ej (s)
 
(γ − 1)
Z
d
(2.20) exp ds = ej (t)P (t)Qj (t).
dt  0 vj (s) 
Finally, integrating (2.20) with respect to t and taking into account that
ej
pj = (γ − 1) ,
vj
we obtain
 Z t
γ−1 t
 Z
1
exp pj (s)ds = 1 + P (s)Qj (s)ej (s)ds,
 0  0

which implies (2.15).


Step 3. We claim that, for every t > 0, there exists j2 = j2 (t) such that
Z t !
X X X
uk (0)h + σj2 (t) (s)ds = vj (0) uk (0)h h
k<j2 (t) 0 j k<j

u2j+ 1 u2j− 1
!
Z tX +
(2.21) − 2 2
+ (γ − 1)ej hds + O(h1/2 ),
0 j
2
δu

where σj = v j − pj , and O(h1/2 ) denotes terms bounded by M h1/2 .
To prove this, we let
X Z t
Ψj (t) = uk (0)h + σj (s)ds,
k<j 0

so that Ψ̇j = σj . Now,


Z t    Z t
δu
δΨk = uk (0) + δ − δpk ds = uk (0) + u̇k (s)ds = uk (t),
0 v k 0

and therefore
vj vj vj γ−1
δ 2 Ψj = δuj = σj + pj = Ψ̇j + ej
   
1 1 γ−1
= (vj Ψj )t − Ψj δuj + ej ,
  
that is,
(vj Ψj )t = δ 2 Ψj + Ψj δuj − (γ − 1)ej .
Letting
Ψk+ 1 + Ψk− 1
2 2
Ψk = ,
2
we obtain
uj+ 1 − uj− 1
2 2
Ψj δuj = Ψj
h
uj+ 1 Ψj+ 1 − uj− 1 Ψj− 1 Ψj − Ψj+1 Ψj−1 − Ψj
2 2 2 2
= + uj+ 1 + uj− 1
h 2 2h 2 2h
uj+ 1 δΨj+ 1 uj− 1 δΨj− 1
= δ(uΨ)j − 2 2
− 2 2
,
2 2
which implies
u2j+ 1 + u2j− 1
!
(vj Ψj )t = δ 2 Ψj − 2 2
+ (γ − 1)ej .
2
Applying the boundary conditions
δΨ0 = u0 = 0, δΨK = uK = 0,
we then obtain
K− 12 K− 12
u2j+ 1 + u2j− 1
Z tX !
X X
(vj Ψj )(t)h = (vj Ψj )(0)h − 2 2
+ (γ − 1)ej hds.
0 2
j= 21 j= 12 j

Let
X
Ψ(t) = (vj Ψj )(t)h.
j
P 0 0 00 00
Then, since j vj (t)h = 1, there exist points j = j (t), j = j (t) such that
Ψj 0 (t) ≤ Ψ(t) ≤ Ψj 00 (t), for fixed t.
In fact, there exists j2 = j2 (t) such that, either
Ψj2 (t) ≤ Ψ(t) ≤ Ψj2 +1 (t),
or
Ψj2 +1 (t) ≤ Ψ(t) ≤ Ψj2 (t).
On the other hand, for any j = j(t),
!1/2
X
|Ψj+1 − Ψj | = h|δΨj+ 1 | ≤ h1/2 u2k h ≤ M h1/2 ,
2
k
and so
|Ψj2 (t) − Ψ(t)| ≤ M h1/2 .
We conclude that, for each t > 0, there exists j2 (t) such that
!
X X
Ψj2 (t) = vj (0) uk (0)h h
j k<j

u2j+ 1 + u2j− 1
Z tX !
(2.22) − 2 2
+ (γ − 1)ej hds + O(h1/2 ).
0 j
2
This proves (2.21).
Step 4. We now give a second representation for v. Specifically, we shall show
that, for any t ≥ 0,
u2l+ 1 + u2l− 1
( Z ! )
tX
1
vj (t) = (1 + O(h1/2 ))Dj (t)exp − 2 2
+ (γ − 1)el hds ×
 0 l 2
( ( Z 2 2 ! ) )
1 + O(h1/2 ) t(γ − 1)ej (s) 1 sX ul+ 12 + ul− 21
Z
1+ exp + (γ − 1)el hdτ ds ,
 0 Dj (s)  0 l 2
where   ! 
1 X X X X 
Dj (t) = vj (0)exp  uk (t)h − uk (0)h + vl (0) uk (0)h h ,
 
k<j
(j2 (t),j) l k<l
and j2 (t) is as in (2.21).

To prove this, we first write (2.2) in the form

u̇k (t) + δpk (t) = (δ L̇)k (t),

where again L = log v. Summing and integrating yield


X Z t
(uk (t) − uk (0))h + (pj (s) − pj2 (s))ds
(j2 ,j) 0

= (Lj (t) − Lj (0) − Lj2 (t) + Lj2 (0)),

where j2 = j2 (t). Rearranging and applying the relation


 
d δu
Lj2 (t) (s) = (s),
ds v j2 (t)

we then obtain
Z t
Lj (t) − pj (s)ds
0
Z t (  )
δu X
= (s) − pj2 (t) (s) ds + (uk (t) − uk (0))h + Lj (0)
0 v j2 (t)
(j2 (t),j)
Z t X
= σj2 (t) (s)ds + (uk (t) − uk (0))h + Lj (0),
0 (j2 ,j)

which by (2.21) implies


Z t
Lj (t) − pj (s)ds
0
K− 12
u2l+ 1 + u2l− 1
! Z tX !
X X
= vl (0) uk (0)h h − 2 2
+ (γ − 1)el hds
0 2
l= 12 k<l l
X X
+ uk (t)h − uk (0)h + Lj (0) + O(h1/2 ).
(j2 (t),j) k<j

Dividing by  and taking the exponential, we get


 Z t 
1
vj (t)exp − pj (s)ds
 0
( 2 2 ! )
1 t X ul+ 12 + ul− 12
Z
= exp − +(γ − 1)el hds Dj (t)(1 + O(h1/2 )),
 0 l 2
so that
 Z t 
1 1
exp pj (s)ds
vj (t)  0
u2l+ 1 + u2l− 1
( Z ! )
tX
1
exp  2 2
+ (γ − 1)el hds
0 l
2
(2.23) = (1 + O(h1/2 )).
Dj (t)
 γ−1

Multiplying (2.23) by 
ej (t) and rearranging, we obtain
γ−1  t ej (s)
n R o Z
1 t 1/2
exp  0 pj (s)ds = 1+ 1 + O(h )
 0 Dj (s)
(Z 2 2 ! )
1 sX ul+ 12 + ul− 12
× exp + (γ − 1)el hdτ ds.
 0 l 2

Substituting this back into (2.23), we then obtain the second representation
(2.23) for vj (·).
Step 5. We complete the proof of the lemma by obtaining the required point-
wise bounds for vj (·). Thus define
mv (t) = min v(x, t), me (t) = min e(x, t),
x∈[0,1] x∈[0,1]

Mv (t) = max v(x, t), Me (t) = max e(x, t).


x∈[0,1] x∈[0,1]

Now, by (2.9), there exists M0 > 0, independent of t and h, such that


M0−1 ≤ Dj , Qj ≤ M0 .
Observe that f (s) = s − 1 − log s is a convex function of s. Thus, by (2.9),
!
X X
f ej h ≤ f (ej )h ≤ M,
j j

which implies
X
(2.24) α≤ ej (t)h ≤ β.
j

This inequality, together with (2.9), gives that


u2j+ 1 − u2j− 1
!
1X
0 ≤ a1 ≤ (γ − 1)ej + 2 2
h ≤ a2 ,
 j 2

where
(γ − 1)α 1
a1 = , a2 = ((γ − 1)β + 2E(0)).
 
The second representation (2.23) for vj (·) implies
 Z t 
vj (t) ≤ M exp{−a1 t} 1 + Me (s)exp{a1 s}ds ,
0

that is,
 Z t 
(2.25) Mv (t) ≤ M exp{−a1 t} 1 + Me (s)exp{a1 s}ds .
0

At this point we need to obtain bounds for Me (·) and me (·). Note that
 
1/2 1/2
X 
1/2 1/2

1/2
X δe k
ej = ej2 + ek+ 1 − ek− 1 = ej2 + 
1/2 1/2
 h.
(j2 ,j)
2 2
(j2 ,j) ek+ 12
+ ek− 12

Using the fact that a + b ≥ 2 ab, we obtain
|δek | |δek |  1/4
≤ e 1e 1
1/2
ek+ 1 + ek− 1
1/2 2(ek+ 1 ek− 1 )1/2 k+ 2 k− 2
2 2
2 2
2
(δek )  
≤ + M ek+ 1 + ek− 1 ,
ek+ 1 ek− 1 2 2
2 2

so that
 
X
 |δek | h
1/2 1/2
(j2 ,j) ek+ 1 + ek− 1
2 2
!
(δek )2
X X X
≤ Mv (t) h+M ek+ 1 h + ek− 1 h
k
ek+ 1 ek− 1 vk k
2
k
2
2 2
X (δek )2
≤ M + Mv (t) h.
k
ek+ 1e
k− 1 vk
2 2

Thus
(2.26) Me (t) ≤ M (1 + Mv (t)B(t)),
(2.27) me (t) ≥ M −1 (1 − Mv (t)B(t)) ,
where !
X (δek )2
B(t) = (t)h.
k
ek+ 1 ek− 1 vk
2 2

Substituting (2.26) into (2.25), we then obtain


 Z t 
Mv (t) ≤ M exp{−a1 t} 1 + exp{a1 s} (1 + Mv (s)B(s)) ds .
0

Now define
A(t) = Mv (t) exp{a1 t}.
Then
 Z t 
A(t) ≤ M 1 + (exp{a1 s} + A(s)B(s))ds
0
 Z t 
≤ M exp{a1 t} + A(s)B(s)ds .
0

Therefore,
Z t  Z t 
A(t) ≤ M exp{a1 t} + M B(s) exp M B(τ )dτ + a1 s ds.
0 s

Applying the fact that


Z t
B(τ )dτ ≤ M,
0
we obtain
A(t) ≤ M (1 + exp{a1 t}),
which implies
(2.28) Mv (t) = exp{−a1 t}A(t) ≤ M.
Next, (2.27)-(2.28), together with the second representation (2.23) for vj (·),
imply that
 Z t 
−1
mv (t) ≥ M exp{−a2 t} + me (s) exp{−a2 (t − s)}ds
0
 Z t 
−1
≥M exp{−a2 t} + exp{−a2 (t − s)}{1 − M B(s)}ds
0
 Z t 
−1
≥M 1 − exp{−a2 t} − M exp{−a2 (t − s)}B(s)ds .
0

Notice that
Z t
lim exp{−a2 (t − s)}B(s)ds
t→∞ 0
t
!
Z
2
Z t
= lim exp{−a2 (t − s)}B(s)ds + exp{−a2 (t − s)}B(s)ds
t→∞ 0 t
2
!
Z ∞ Z t
a2
≤ lim exp{− t} B(s)ds + B(s)ds = 0.
t→∞ 2 0 t
2

Therefore, there exists t0 > 0 such that


M −1
(2.29) mv (t) ≥ , t ≥ t0 > 0.
2
On the other hand,
Rt !
γ−1
X X 1+ e
0 j
(s)P

(s)Qj (s)ds
P (t) = P (t)vj (t)h = h
j j
Q j (t)
Z t !  Z t 
X
≤M 1+ P (s)( ej (s)h)ds ≤ M 1 + P (s)ds ,
0 j 0

which gives
P (t) ≤ M exp{M t}.
Therefore, from (2.15),
M −1
vj (t) ≥ ≥ M −1 exp{−M t} ≥ M −1 exp{−M t0 }, when 0 ≤ t ≤ t0 .
P (t)
Combining this last relation with (2.29), we then obtain the required bounds
for vj . This completes the proof of Lemma 2.2.

In the following lemma, we derive a lower bound for the internal energy
ej (·).
Lemma 2.3. There is a constant M , independent of t and h, such that
1
(2.30) ej (t) ≥ .
M (t + 1)
Proof: Setting ωj = 1
ej
and multiplying (2.3) by {−e−2j }, we obtain

ωj2 (δuj )2
 
pj λ δe
ω̇j − 2 δuj = − − 2δ ,
ej vj ej v j
which implies
ωj2 (δuj )2
 
(γ − 1)ωj δuj δω
ω̇j ≤ − + λδ
vj vj v j
 2 2
ωj (δuj ) 2 ωj (δuj )2
2
  
1 (γ − 1) δω
≤  + − + λδ .
vj  vj vj v j
Therefore,
 
M δω
ω̇j ≤ + λδ ,
vj v j

so that
dωj2q qM ωj2q−1
 
δω
≤ + 2qλωj2q−1 δ .
dt vj v j
Summing by parts, we obtain
!
d X 2q X ωj2q−1 X 2q−2 δωk
2

ωj h ≤ qM h − 2qλ (2q − 1)ωk h
dt j j
vj k
vk
( )2q−1( )1
X 2q−1 1 X 2q 2q X −2q 2q
≤ qM ωj h ≤ qM ωj · vj h .
j
v j j j

Define !2q1
X
D(t) = ωj2q (t)h .
j
Then we have
!2q1
M X −2q
D2q−1 (t)D0 (t) ≤ D2q−1 (t) vj (t)h ,
2 j

which implies
!2q1
M X M
D0 (t) ≤ vj−2q (t)h ≤ ≤ M.
2 j
2mv (t)
Therefore
D(t) ≤ D0 + M t.
Now let q → ∞ to obtain
kω(t)kL∞ ≤ kω(0)kL∞ + M t,
which implies
ω(t) ≤ M (1 + t).
This completes the proof. 
Remark: Lemmas 2.1-2.3, together with the bounds
! !
X X
ei (t) ≤ ej (t)h h−1 ≤ M h−1 , u2i (t) ≤ u2j (t)h h−1 ≤ M h−1 ,
j j

show that the initial-value problem (2.1)-(2.5) is solvable for all t > 0 with
fixed h > 0.

In the following lemma we derive three auxiliary estimates relating to the


discrete piecewise-H 1 norm of v and to the evolution in time of the jumps in
v. In particular, we shall show that the magnitude of the jump discontinuity
[vki ] = vki + 1 − vki − 1
2 2

at time t is bounded by the magnitude of the corresponding jump at time


t = 0.
Lemma 2.4. The following estimates hold for all t > 0:
(a). For ûj defined by u3j+ 1 − u3j− 1 = 3û2j (uj+ 1 − uj− 1 ),
2 2 2 2

!
X  Z t X X
(ej − 1)2 + u4j+ 1 h + (δek )2 + û2j e2j + û2j (δuj )2 hds ≤ M.

2
j 0 k j

(b). For the distinguished discontinuity points

0 < xk1 < xk2 < · · · < xkN < 1,

|[vki (t)]| ≤ M exp{−M −1 t1/2 }|[vki (0)]| + h1/2 .



(2.31)

(c). Away from the distinguished discontinuities,


!
X Z t X X
2
sup (δvk ) h+ (1 + ek )(δvk )2 + (δuj )2 hds ≤ M (1 + N h1/2 t).
t 0
k6=ki k6=ki j

Proof: The first result follows from routine energy estimates together with
the L2 and pointwise estimates of Lemmas 2.1–2.3; and the third is derived
exactly as in [10], pp. 29–30. We shall therefore give the details only for the
second estimate, (2.31).
Suppressing the subscript ki , we have
   
v̇ δu
(2.32) [L]t = = = [p] + hu̇.
v v
Taking αi is as in (1.14), we can solve this first-order differential equation to
obtain
  Z t 
1
|[vki (t)]| ≤ M exp αi eki ds |[vki (0)]|
 0
Z t  Z t  1/2 )
2
(2.33) +h1/2 exp αi eki dτ ds + h3/4 .
0  s

Rt
We need to derive a lower bound for the term 0 ek ds appearing here. Observe
that
Z t Z t 1/2 Z t 1/2
−1
t= ds ≤ ek ds ek ds ,
0 0 0

so that
t
t2
Z
ek ds ≥ R t .
0 0
ek −1 ds
However,

t t
!
Z
1
Z
1 X 1
ds ≤ P + δ
e h ds

0 ek 0 i ei h k k
Z tX
|δek |
(2.34) ≤ Mt + hds.
0 k
ek+ 1 ek− 1
2 2

Also, from Lemma 2.3, we have


!
1 1 1 1 (1 + t)
(2.35) = + ≤M .
ek+ 1 ek− 1 2ek ek− 1 ek+ 1 ek
2 2 2 2

Combining (2.34) with (2.35), we then obtain

t Z tX
|δek |
Z
1
ds ≤ M t + M (1 + s)hds
0 ek 0 k
ek

Z tX !1/2 Z 1/2
t
(δek )2 2
≤ Mt + M hds (1 + s) ds
0 k
e2k 0

≤ M t + M (1 + t)3/2 ≤ M (1 + t)3/2 .

Hence, for t ≥ 0,

t
t2
Z
ek ds ≥ ≥ M −1 t1/2 ,
0 M (1 + t)3/2

which implies
Z t 
≤ M exp −M −1 t1/2 .

exp αi ek ds
0

This completes the proof. 

In the following lemma we derive certain discrete higher-order estimates for


u and e. These will be crucial both for the passage to the limit as h goes to
zero, as well as for the determination of the large-time behavior.
Lemma 2.5. Define σ(t) = min{t, 1}. The solutions (vj (t), uk (t), ej (t)) of
(2.1)-(2.5) then satisfy the following estimates:
! Z
X t X
(a). sup σ(t) (δuj )2 (t)h + σ(s) u̇2k (s)h ds ≤ M (1 + N h1/2 t),
t 0
j k
!
X Z t X
(b). sup σ 2 (t) (δek )2 (t)h + σ 2 (s) ė2j (s)h ds ≤ M (1 + N h1/2 t),
t 0
k j
!!
X X
(c). sup σ 2 (t) u̇2k (t)h + (δuj )4 (t)h
t
k j
Z t X
+ σ 2 (s) (δ u̇j )2 (s)h ds ≤ M (1 + N h1/2 t),
0 j
!
X Z t X  (δ ėj )2 
3
(d). sup σ (t) ė2j (t)h + 3
σ (s) (s)h ds ≤ M (1 + N h1/2 t).
t
j 0 j
vj

Proof: The estimates (a) − (c) follow from fairly tedious, but routine energy
estimates, and are similar to those given in [10], pp. 30–34 (except that the
estimates given here are time-independent, owing to the fact that the pointwise
bounds of Lemmas 2.2–2.3 are time-independent). We therefore omit their
proofs and give only the details of the proof of (d).
We differentiate (2.3) with respect to t, multiply the result by {σ 3 ėj }, and
sum and integrate to obtain
Z t X Z t X Z t X
3 3
σ (ėj ëj ) h ds + σ (ėj ṗj (δuj )) h ds + σ3 (ėj pj (δ u̇j )) hds
0 j 0 j 0 j
t t
X (δuj )3
Z  Z X 
(δuj )(δ u̇j )
3 3
= 2 σ ėj hds −  σ ėj h ds
0 j
vj 0 j
vj2
Z t X  ˙ ! Z t
2 δe X
+λ σ ėj δ h ds + 3 σ 2 σ̇ ė2j hds.
0 j
v j 0 j

Applying the boundary conditions (1.2), we then obtain


! Z
X t X  (δ ėj )2 
3 2 3
sup σ ėj h + σ h ds
t
j 0 j
v j
Z t X
≤ M σ3 ė2j (δuj ) + ej ėj (δuj )2 + ej ėj (δ u̇j ) + (δuj )(δ u̇j )ėj + ėj (δuj )3 h ds

0 j
Z t X  (δ ėj )(δej )(δuj ) 
3
+M σ h ds.
0 j
vj2
It follows that
! Z
X t X  (δ ėj )2 
3 2 3
sup σ ėj h + σ h ds
t
j 0 j
v j
Z t X 
2 2 2 (δ ėj )
≤ M kσ(δuj )kL∞ σ ėj + ėj (δuj ) + ej ėj (δuj ) + (δej ) + (δ ėj )ėj h ds
0 j
vj
Z t X
+M σ3 (ej ėj (δ u̇j )) hds.
0 j

Next, from the momentum equation (2.2),


 
2 δuj
|δuj | ≤ M k − pj k∞ + kpj k∞
vj
(  2 )
X   δu X
(δuj )2 + p2j h + kek2∞

≤M δ −p h+
k
v k j
( )
X X X
≤M 1+ (δek )2 h + u̇2k h + (δuj )2 h .
k k j

Therefore, by Lemma 2.5 (a) – (c),


( )1/2
X X X
kσδuj kL∞ ≤ M σ 2 + σ 2 (δek )2 h + σ 2 u̇2k h + σ 2 (δuj )2 h
k k j
1/2 1/2
≤ M (1 + N h t) .
The result (d) now follows.


3. Proof of Theorems 1.1 and 1.2


The existence and regularity statements in Theorems 1.1 and 1.2 can now
be derived by the technique of Hoff [10]–[11], with the aid of Lemmas 2.1-2.5.
Briefly, we begin with initial data as in Theorem 1.2, that is, with v0 piecewise
H 1 . Difference approximations (vj (t), uk (t), ej (t)) are constructed as in Sec-
tion 2, and these mesh functions are used to construct approximate solutions
(v h (x, t), uh (x, t), eh (x, t)) by a suitable interpolation procedure. The estimates
of Lemmas 2.1–2.5, which are uniform in h, then apply to show that these ap-
proximate solutions are appropriately compact, that their limits are indeed
weak solutions, and that these weak solutions inherit all the properties (1.7)–
(1.15) asserted in Theorems 1.1 and 1.2. The uniform total variation estimate
(1.8) for v, obtained for the case that v0 is piecewise H 1 , can then be applied to
complete the solution operator to the more general data of Theorem 1.1, that
is, data for which v0 is of bounded variation. This entire construction requires
a fairly lengthy, but straightforward analysis. The details for the present case
are nearly identical to those of [10]–[11], the important differences being that
all of the estimates given here are independent of time and are valid for large
initial data. These details are therefore omitted.
The large-time behavior result of Theorem 1.1 can now be derived as an a
posteriori consequence of the weak form of the equations (1.1) and the esti-
mates (1.7)–(1.10) of Theorem 1.1. A similar result is given in Hoff [12] for the
multidimensional case under the assumption that the initial data are small;
this restriction is required only for the derivation of time-independent bounds,
however. Once these bounds have been established, as in Theorem 1.1, the
derivation of the large-time behavior is the same. We therefore give here just
a brief sketch.
The energy estimates (1.10) show that the function a(t) = kux (·, t)k2L2 is
integrable and of bounded variation on [1, ∞). It follows that
lim a(t) = 0,
t→∞

that is, that u(·, t) → 0 in H01 . A similar argument shows that


kex (·, t)kL2 → 0,
and routine bounds based on these results and on (1.10) then show that
k(e + u2 /2)x (·, t)kL2 → 0.
This fact, together with the conservation law
Z 1
(e + u2 /2)(x, t) dx = e∞ ,
0

and the convergence of u(·, t), then shows that e(·, t) → e∞ in H 1 as t → ∞.


The argument for v is somewhat more involved and depends upon the same
dissipative structure responsible for the damping of singularities. We will give
a heuristic sketch of the proof; a somewhat delicate smoothing and renormal-
ization analysis is
Z required to makeZthe argument precise (see [12] for details).
1 1
u2 (x)
We define v∞ = v0 (x)dx, e∞ = (e0 (x) + 0 )dx, p∞ = p(v∞ , e∞ ), and
0 0 2
the so-called effective viscous flux
ux
F = − p(v, e) + p∞
v   
ux e − e∞ 1 1
= − (γ − 1) + e∞ − .
v v v v∞
Observe from (1.1) that Fx Z= uZt . Now, by applying the energy estimates
∞ 1
(1.10), especially the bound u2xt dxdt < ∞, we can show that
1 0
kut (·, t)kL2 → 0, as t → ∞.
It follows that
kF (·, t) − F̄ (t)kL∞ → 0, t → ∞,
where F̄ is the space average of F . This in turn, together with the previous
results that ux , e − e∞ → 0 in L2 , substituted into the definition of F , then
shows that Z 1
1 dx
k − kL2 → 0,
v(·, t) 0 v(x, t)
which implies Z 1
dx −1
kv(·, t) − ( ) kL2 → 0.
0 v(x, t)
Notice that
Z 1  Z 1  Z 1 
dx −1 dx −1
v∞ = lim v(x, t) − ( ) dx + ( )
t→∞ 0 0 v(x, t) 0 v(x, t)
 Z 1 −1
dx
= lim ,
t→∞ 0 v(x, t)

so that v(·, t) → v∞ in L2 as t → ∞. Then, by taking space averages in the


definition of F , we find that F̄ (t) → 0, and therefore that
kF (·, t)kL∞ → 0.
Finally, to show that v converges in L∞ , we define w(x, t) = (log v(x, t) −
log v∞ ) and compute formally from (1.1),
ux
wt = = F + p − p∞
v   
e − e∞ 1 1
= F + (γ − 1) + e∞ − .
v v v∞
Define
(γ − 1)e∞ (1/v(x, t) − 1/v∞ )
β(x, t) = − ,
(log v(x, t) − log v∞ )
so that
γ−1
wt + βw = F + (e − e∞ ) .
v
Evidently C −1 ≤ β(x, t) ≤ C, so that this ordinary differential equation is
strictly dissipative. Thus
Rt
− 1 β(x,s)ds
|w(x, t)| ≤ e w(x, 1)

Z Z t! R 
t/2 
t γ − 1
+ + e− s β(x,τ )dτ F (x, s) + (e(x, s) − e∞ ) ds

1 t/2 v(x, s)
≤ Ce−t/C + Cte−t/C + C sup (kF (·, s)kL∞ + ke(·, s) − e∞ kL∞ )
s≥t/2
→ 0, as t → ∞ .
This completes the proof of Theorems 1.1–1.2.
Acknowledgments
Gui-Qiang Chen’s research was supported in part by the National Science
Foundation under Grants DMS-9971793 and DMS-9708261. David Hoff’s re-
search was supported in part by the National Science Foundation under Grant
DMS–9703703.

References
[1] A. A. Amosov and A. A. Zlotnick, A semidiscrete method for solving equations of
the one-dimensional motion of a non-homogeneous viscous heat-conducting gas with
nonsmooth data, Izv. Vyssh. Uchebn. Zaved. Mat. 1997, 3–19 (Russian); transl. in
Russian Math. (Iz. VUZ) 41 (1997), 1–17.
[2] T. Chang and L. Hsiao, The Riemann Problem and Interaction of Waves in Gas Dy-
namics, Longman Scientific & Technical, Harlow; copublished in the United States with
John Wiley & Sons, Inc.: New York, 1989.
[3] G.-Q. Chen, Global solutions to the compressible Navier-Stokes equations for a reacting
mixture, SIAM J. Math. Anal. 23 (1992) 609-634.
[4] G.-Q. Chen and J. Glimm, Global solutions to the compressible Euler equations with
geometrical structure, Comm. Math. Phys. 180 (1996) 153–193.
[5] G.-Q. Chen, D. Hoff, and K. Trivisa, Global discontinuous solutions of the Navier-
Stokes equations for compressible reacting flow, In preparation.
[6] P. Collela, A. Majda and V. Roytburd, Theoretical and numerical structure for re-
acting shock waves, SIAM J. Sci. Stat. Comput. 7 (1986), 1059–1080.
[7] H. Fujita-Yashima, M. Padula, and A. Novotny, Équation monodimensionnelle d’un
gaz vizqueux et calorifére avec des conditions initiales moins restrictives, Ricerche Mat.
42 (1993), 199–248.
[8] J. Glimm, The continuous structure of discontinuities, Lecture Notes in Physics 344
(1989), 177–186.
[9] E. Godlewski and P. Raviart, Numerical Approximation of Hyperbolic Systems of
Conservation Laws, Appl. Math. Sc. 118, Springer-Verlag: New York, 1996.
[10] D. Hoff, Global well–posedness of the Cauchy problem for nonisentropic gas dynamics
with discontinuous initial data, J. Diff. Eqs. 95 (1992), 33-74.
[11] D. Hoff, Discontinuous solutions of the Navier–Stokes equations for compressible flow,
Arch. Rational Mech. Anal. 114, (1991), 15–46.
[12] D. Hoff. Discontinuous solutions of the Navier-Stokes equations for multidimensional
heat-conducting fluids, Arch. Rational Mech. Anal. 139 (1997), 303–354.
[13] D. Hoff. Continuous dependence on initial data for discontinuous solutions of the
Navier-Stokes equations for one-dimensional, compressible flow, SIAM J. Math. Anal.
27 (1996), 1193-1211.
[14] D. Hoff and D. Serre, The failure of continuous dependence on initial data for the
Navier-Stokes equations of compressible flow, SIAM J. Appl. Math. 51 (1991), 887–898.
[15] N. Itaya, On the Cauchy problem for the system of fundamental equations describing
the movement of compressible fluid, Kodai Math. Sem. Rep. 23 (1971), 60–120.
[16] S. Jiang, Global smooth solutions to the equations of a viscous, heat-conducting, one-
dimensional gas with density-dependent viscosity, Math. Nachr. 190 (1998), 169-183.
[17] Y. Kanel, On a model system of equations of one–dimensional gas motion, Diff. Eq. 4
(1968), 374–380.
[18] A. V. Kazhikhov. On the theory of initial–boundary–value problems for the equations of
one–dimensional nonstationary motion of a viscous heat-conductive gas. Din. Sploshnoi
Sredy 50 (1981), 37–62 (Russian).
[19] A. V. Kazhikhov and V. V. Shelukhin, Unique global solution with respect to time
of initial-boundary-value problems for one–dimensional equations of a viscous gas, J.
Appl. Math. Mech. 41 (1977), 273–282.
[20] P. Lions, Mathematical Topics in Fluid Mechanics, Vol. 2, Oxford University Press:
New York, 1998.
[21] A. Matsumura and S. Yanagi, Uniform boundedness of the solutions for a one-
dimensional isentropic model system of compressible viscous gas, Comm. Math. Phys.
175 (1996), 259–274.
[22] F. A. Williams, Combustion Theory, Addison-Wesley, Reading, MA, 1965.

Вам также может понравиться