Siberian Mathematical Journal, Vol. 46, No. 5, pp.

805–812, 2005
Original Russian Text Copyright
c 2005 Gatapov B. V. and Kazhikhov A. V.

EXISTENCE OF A GLOBAL SOLUTION TO ONE

MODEL PROBLEM OF ATMOSPHERE DYNAMICS
B. V. Gatapov and A. V. Kazhikhov UDC 517.9

Abstract: We consider a model problem of compressible viscous fluid dynamics in the two-dimensional
case. We prove a global existence theorem.
Keywords: compressible viscous fluid, shallow water approximation, a priori estimates, existence the-
orem

1. Introduction
Some sections of applied hydrodynamics, for example, meteorology and oceanology widely use a mod-
ified Navier–Stokes model, the so-called shallow water approximation (see [1, 2]). It is derived from the
general system by asymptotic analysis under the assumption that one size of the domain of motion
(vertical) is much smaller than the other (horizontal). In this case the shallow water approximation
consists in replacing the momentum equation for the vertical component of the momentum vector with
the hydrostatics equation. The details of the derivation of this model can be found, for example, in [3];
apparently, this model was first proposed by N. E. Kochin [4].
Questions of existence of solutions to boundary value problems for the Navier–Stokes equations
of a compressible viscous fluid have been studied by many authors; however, the state of satisfactory
completion has been reached only in the one-dimensional case (see [5, 2]). Most earlier results turn out
local in the case of the multidimensional model; in particular, the time interval is assumed sufficiently
small or the data of the problem are close to the steady state (see [6, 7]).
The shallow water systems are studied rather completely only for an incompressible fluid [8–10],
while no global existence theorems have been established for the model with compressibility.
In this article we prove global solvability of the initial-boundary value problem for the model system
of equations of a compressible viscous fluid in the shallow water approximation. This system applies to
modeling processes in the atmosphere:
∂ρ ∂ ∂
+ (ρu) + (ρv) = 0,
∂t ∂x ∂y




∂u ∂u ∂u ∂p ∂ ∂u ∂ ∂u
ρ +u +v + = ν1 + ν2 ,
∂t ∂x ∂y ∂x ∂x ∂x ∂y ∂y (1)
∂p
= −gρ,
∂y
p = c2 ρ, c2 = const > 0.
Here (t, x, y) are independent variables, t is the time, (x, y) are the horizontal and vertical spatial co-
ordinates, ρ is the density of the medium, u = (u, v) is the velocity vector, where u is the horizontal
component and v is the vertical component (the momentum equation for the latter is replaced with the
hydrostatics equation (13 )), p is the pressure, g = const > 0 is the free fall acceleration, and (ν1 , ν2 ) are
the turbulence viscosities in the horizontal and vertical directions.
The authors were supported by the Russian Foundation for Basic Research (Grant 05–01–00131).
Dedicated to a blessed memory of Tadeı̆ Ivanovich Zelenyak.

Novosibirsk. Translated from Sibirskiı̆ Matematicheskiı̆ Zhurnal, Vol. 46, No. 5, pp. 1011–1020, September–October,
2005. Original article submitted April 19, 2005.
0037-4466/05/4605–0805

c 2005 Springer Science+Business Media, Inc. 805
 Suppose that motion of the medium occurs in a rectangular domain Ω = {(x, y) | 0 < x < l,
0 < y < h} and the boundary conditions on the boundary of Ω are expressed by the relations
u|x=0 = u|x=l = 0,
v|y=0 = v|y=h = 0,
  (2)
∂u  ∂u 
= = 0.
∂y y=0 ∂y y=h
This is the simplest version of the boundary data; however, other boundary conditions can also be
considered.
We suppose finally that the distribution of the horizontal component of the velocity and the density
distribution are known at the initial time t = 0:
u|t=0 = u0 (x, y),
(3)
ρ|t=0 = ρ0 (x) exp (−gy/c2 );
moreover ρ0 (x) is a bounded strictly positive function: 0 < m ≤ ρ0 (x) ≤ M < ∞.
The proof of the existence theorem for (1)–(3) is based on some global a priori estimates. The
main idea enabling us to establish a priori estimates is connected, first, with the fact that hydrostatics
equation (13 ) can be integrated and, second, with a decomposition of the horizontal component of the
velocity into a sum of two summands one of which is the mean value (with a special weight) over the
vertical coordinate. We arrive eventually at a system of equations which coincides formally with the
equations for one-dimensional motion of a viscous gas [5]. So, we can use some methods and tools
developed for this model (see [5, 11]).
For simplicity, we assume l = 1, h = 1, and g = c2 . From the hydrostatics equation
∂p
= −gρ
∂y
we obtain p = ζ(x, t)e−y , and the continuity equation takes the form
∂ζ ∂ ∂
+ (ζu) + ey (ζe−y v) = 0.
∂t ∂x ∂y
Putting
1
ū = e−y u dy
0
and integrating the equation with respect to y, we obtain
∂ζ ∂
+ (ζ ū) = 0.
∂t ∂x
Similarly, integrate (12 ) with respect to y on denoting the difference u − ū by w. We then obtain the
relation

1 
1 
∂ ∂ −y 2 ∂ζ ∂2
(ζ ū) + ζ e u dy + c = ū + w dy ,
∂t ∂x ∂x ∂x2
0 0
1
where c = 0 e−y
dy = 1−e−1 > 0. Add and subtract ∂ 2
∂x (ζ(ū) )on the left-hand side of the above equality,
∂2
1
discard the last summand ∂x2 ( 0 w dy) on the right-hand side, and consider the following simplified
system of equations as a model system:
∂ζ ∂
+ (ζ ū) = 0,
∂t ∂x
  
1  (4)
∂ū ∂ū ∂ −y 2 2 ∂ 2 ū
ζ + ū + ζ c + e u dy − (ū) =
∂t ∂x ∂x ∂x2
0

806
which is also of interest in its own right. The authors intend to study the initial system (1) without
this simplification later. Note also that (4) arises without the simplification, if the horizontal turbulence
coefficient ν1 in (1) has the form ν1 = ν10 e−y , ν10 = const > 0.
Observe that the following relation holds in (4):
1
e−y u2 (x, y, t) dy − (ū)2 ≡ u2 − (ū)2 ≥ 0.
0
Formally, (4) looks as the usual Navier–Stokes system for a compressible fluid, where ζ plays the role
of density, ū is the velocity, and p ≡ ζ(1 + u2 − (ū)2 ) is the pressure. The fundamental difference from
the Navier–Stokes system is that the pressure depends not only on ζ but also on u2 and (ū)2 .
Thus, we consider as a model problem similar to (1) the system
∂ζ ∂ ∂
+ (ζu) + (ζw) = 0,

∂t ∂x  ∂z
∂u ∂u ∂u ∂ζ ∂2u ∂2u
ζ +u +w + = + 2, (5)
∂t ∂x ∂z ∂x ∂x2 ∂z
∂ζ
= 0.
∂z
Here the density is ζ(t, x), while u and w are the horizontal and vertical components of the velocity vector.
The domain of variation of x and z is Ω = (0, 1) × (0, 1). We consider (5) in the domain Q = Ω × (0, T ).
The boundary conditions have the form
u|x=0 = u|x=1 = 0,
 
∂u  ∂u 
= = 0, (6)
∂z z=0 ∂z z=1
w|z=0 = w|z=1 = 0.
By (3), for the velocity distribution and the density distribution at t = 0 we have
u|t=0 = u0 (x, z),
(7)
ζ|t=0 = ζ0 (x) (= ρ0 (x));
moreover, ζ0 (x) is a bounded strictly positive function:
0 < m ≤ ζ0 (x) ≤ M < ∞.
Definition 1. A weak solution to (5)–(7) is a collection (ζ, u, w) of functions such that ζ ≥ 0 and
 ∂ζ
ζ(t) ∈ L∞ 0, T ; W21 (0, 1) , ∈ L2 (0, T ; L2 (0, 1)),
∂t

u(t) ∈ L2 0, T ; W22 (Ω) ∩ W21 (0, T ; L2 (Ω)), w(t) ∈ L2 (0, T ; L2 (Ω)),
which satisfy (5) in the distribution sense; in particular, the integral identity
T  T  
2
(ζuϕt + ζu ϕx + ζwuϕz + ζϕx ) dxdzdt = − u∆ϕ dxdzdt + u0 ζ0 ϕ|t=0 dxdz (8)
0 Ω 0 Ω Ω

is valid for an arbitrary function ϕ ∈ L2 0, T ; C0∞ (Ω)
, with ϕ|t=T = 0. Relations (6) and (7) hold in the
sense of the traces of the functions of the above-indicated classes.
Theorem 1. Suppose that the initial data (ζ0 , u0 ) possess the following properties:
(ζ0 , u0 ) ∈ W21 (Ω), u0 |x=0 = u0 |x=1 = 0.
Then (5)–(7) has a weak solution; moreover, ζ(x, t) is a bounded strictly positive function.

807
 2. A Priori Estimates
The conservation law for (5) has the form
1 1
2  1 1
d u  2
ζ + (ζ log ζ − ζ + 1) dxdz + ux + u2z dxdz = 0. (9)
dt 2
0 0 0 0
This in particular implies the presence of the a priori estimates
1 1
1 
sup (ζ log ζ − ζ + 1) dx ≤ C and sup ζ(x, t) u2 (x, z, t) dz dx ≤ C.
0<t<T 0<t<T
0 0 0
Consequently,
1
ζ(ū)2 dx ≤ C,
0
which means in the Lagrangian variables (to be introduced below in (11)) that
1
sup (ū(η, τ ))2 dη ≤ C. (10)
0<τ <T
0
Here and in the sequel C stands for various positive constants depending only on the initial data and Ω.
The function ū is now expressed by an integral with respect to z without weight.
Similarly,
T 1 1 T 1
2
 2 2
∂ū
ux + uz dzdxdt ≤ C and ζ dηdτ ≤ C.
∂η
0 0 0 0 0

3. Upper and Lower Estimates for the Density

In line with [1], we can introduce the Lagrangian coordinates
x
η(x, t) = ζ(s, t)ds, τ = t, (11)
0
and so (4) takes the form

∂ 1 ∂ū
= ,
∂τ ζ ∂η

 (12)
∂ū ∂p ∂ ∂ū
+ = ζ ,
∂τ ∂η ∂η ∂η
where p ≡ ζ(1 + p0 ) and the function p0 = p0 (η, τ ) ≡ u2 − (ū)2 is nonnegative.
The problem is now to estimate ζ(x, t) from above and below. To this end, we use (12) and the
estimates of the previous section. From the first equation of (12) we obtain the equality
∂ū 1 ∂ζ ∂ log ζ
ζ =− ≡− .
∂η ζ ∂τ ∂τ
Insert it in the second and integrate the result with respect to τ :
∂ū ∂p ∂ ∂
+ =− log ζ,
∂τ ∂η ∂τ ∂η
τ 
∂ ∂ log ζ ∂ 
ū − u0 + pds = − + log ζ  .
∂η ∂η ∂η τ =0
0

808
Without loss of generality we may assume that ζ|τ =0 ≡ 1 and arrive at the equality

τ 
∂
log ζ + p ds = u0 − ū. (13)
∂η
0

Hence, by (10),


τ 

sup
log ζ + p ds
≤ c.
∂η

0<τ <T L2 (0,1)

0
From (12) we obtain

∂ 1 ∂ū
= .
∂τ ζ ∂η
Integrate this expression with respect to η from 0 to 1:
1
 1
∂ 1 ∂ 1
dη = ū|η=1 − ū|η=0 = 0, dη = 0.
∂τ ζ ∂τ ζ
0 0
Thus,
1 1
1 1
dη = dη ≡ 1.
ζ(η, τ ) ζ0 (η)
0 0
From here and continuity of the function ζ we find that
(∀τ ∃η0 = η0 (τ )) ζ(η0 (τ ), τ ) = 1.
Having this in mind, integrate (13) with respect to η from η0 (τ ) to an arbitrary η:
τ τ η
log ζ + p ds = p(η0 (τ ), s) ds + (ū0 − ū)dξ.
0 0 η0

Exponentiate this equality:

τ 
ζ(η, τ ) exp p(η, s) ds = Y (τ ) · B(η, τ ), (14)
0
where
τ  η 
Y (τ ) = exp p(η0 (τ ), s) ds B(η, τ ) = exp (ū0 − ū) dξ .
0 η0

The function B(η, τ ) is bounded above and below in view of the estimate for ū in L∞ (0, T ; L2 (0, 1))
(see (10)). Show that Y (τ ) is bounded too. Note that
T
max p0 (η, τ )dτ ≤ C. (15)
0≤η≤1
0
Indeed,
1   1  1 1 1  1 1
 ∂p0     2
max p0 (x, t) ≤   dx ≤ 2 uux dz − 2 u dz uz dz  dx ≤ c ux + u2z dxdz.
0≤x≤1 ∂x   
0 0 0 0 0 0 0

809
Multiply (14) by (1 + p0 (η, τ )) and integrate the result with respect to τ :
τ τ
p(η,s)ds
e0 =1+ (1 + p0 (η, s))Y (s)B(η, s) ds.
0

Consequently, (14) takes the form

τ 
ζ 1+ (1 + p0 )Y (s)B(η, s) ds = Y (τ )B(η, τ ). (16)
0

Putting η = η0 (τ ), where ζ(η0 (τ ), τ ) = 1, we obtain the following inequalities for Y (τ ):

0 < Y0−1 ≤ Y (τ ) ≤ Y0 = const < ∞.

Indeed, it follows from (16) that Y (τ ) ≥ B −1 (η, τ ) ≥ C > 0. The upper estimate results from (15)
and (16) by Gronwall’s lemma.
Now, (14) obviously gives some upper and lower estimates for ζ:

0 < ζ0−1 ≤ ζ(η, τ ) ≤ ζ0 < ∞.

From (13) and the inequalities for ζ we find the following estimate for the derivatives:

∂ζ

≤ c. (17)

∂η
∞ 2L (0,T ;L (0,1))

From the equations of (12) we straightforwardly obtain

T

ζτ (τ )
L∞ (0,T ;L2 (0,1)) ≤ c,
ut (t)
2 dt ≤ c.
0

From the continuity equation of (5) we immediately derive

∂w

≤ c.

∂z
2 2 L (0,T ;L (Ω))

The above estimates are sufficient for proving the existence of a weak solution to (5)–(7).

4. Proof of the Existence Theorem

To prove the existence theorem, we seek for a solution to (5)–(7) in the form of a sum

u(x, z, t) = ū(x, t) + ũ(x, z, t), (18)

1
where, as above, ū is the mean value ū(x, t) = 0 u dz over the vertical coordinate. The deviation ũ
satisfies the following condition by necessity:

H
ũ(x, z, t) dz ≡ 0, (19)
0

for all values of x and t.

810
 Using (19) and the equality u2 = ū2 + 2ūũ + (ũ)2 , we conclude that
1
2
u2 (x, t) − (ū) (x, t) = (ũ(x, z, t))2 dz.
0

Then the system (4) for ū(x, t) takes the form

∂ζ ∂
+ (ζ ū) = 0,
∂t ∂x



1  (20)
∂ū ∂ū ∂ ∂ 2 ū
ζ + ū + ζ c + (ũ)2 dz = .
∂t ∂x ∂x ∂x2
0

Inserting (18) in (5), we obtain

∂ ∂w
(ζ ũ) + ζ = 0,

∂x ∂z 
∂ū ∂ũ ∂ ∂ũ ∂ζ ∂ 2 ū ∂ 2 ũ ∂ 2 ũ
ζ + + (ū + ũ) (ū + ũ) + w + = + + 2, (21)
∂t ∂t ∂x ∂z ∂x ∂x2 ∂x2 ∂z
∂ζ
= 0.
∂z
We can rewrite the second equation as


1 
∂ū ∂ū ∂ ∂ 2 ū
ζ + ū + ζ c + (ũ)2 dz − 2
∂t ∂x ∂x ∂x
0


1 
∂ũ ∂ũ ∂ũ ∂ū ∂ũ ∂ 2 ∂ 2 ũ ∂ 2 ũ
+ζ + ũ + ū + ũ +w − ζ (ũ) dz = + 2,
∂t ∂x ∂x ∂x ∂z ∂x ∂x2 ∂z
0

and so (21) takes the form

∂ ∂w
(ζ ũ) + ζ = 0,
∂x ∂z


1 
∂ũ ∂ũ ∂ũ ∂ū ∂ũ ∂ 2 ∂ 2 ũ ∂ 2 ũ
ζ + ũ + ū + ũ +w − ζ (ũ) dz = + 2, (22)
∂t ∂x ∂x ∂x ∂z ∂x ∂x2 ∂z
0
∂ζ
= 0.
∂z
If we assume that the functions ζ and ū are known then (22) has the form of the hydrostatic model
for an incompressible fluid (see [9]). Indeed, if we pass to the Lagrangian coordinates η, with dη = ζ dx,
and replace ζ ũ with v then the first equation takes the form
∂v ∂w
+ = 0.
∂η ∂z
We can transform the second equation to the form

1 
∂ ∂ 2 ∂ 2 ∂ ∂
(ζ ũ) + (ζ(2ūũ + (ũ) )) − ζ (ũ) dz + ū (ζw) + (ζwũ) = ũxx + ũzz ,
∂t ∂x ∂x ∂z ∂z
0

811
if we note that
∂ ∂ ∂ ∂
(ζ ū) + (ζ ũ) + (ζ(ū)2 ) + (ζ(2ūũ + (ũ)2 ))
∂t ∂t ∂x ∂x

1 
1 
∂ 2 ∂ 2 ∂ ∂
+ ζ c + (ũ) dz − ζ (ũ) dz + ū (ζw) + (ζwũ) − ūxx = ũxx + ũzz .
∂x ∂x ∂z ∂z
0 0

This is equivalent to the equality


1 2 
∂v 1 ∂ v2 1 ∂ v ∂w ∂
+ 2ūv + − dz + ūζ + (wv)
∂t ζ ∂η ζ ζ ∂η ζ ∂z ∂z
0

2
1 ∂ 2 v ∂ζ ∂v 1 ∂ζ ∂2v ∂2ζ
= − − v+ − v.
ζ ∂z 2 ∂η ∂η ζ ∂η ∂η 2 ∂η 2

Construct a mapping A : ũ1 → ũ2 whose fixed point is a solution to (20) and (22).
Let K be a bounded set in L2 (0, T ; W 1,2 (Ω)) determined by the first a priori estimate for u and hence
for ũ:
K = {ũ |
ũ
L2 (0,T ;W 1,2 (Ω)) ≤ k = const < ∞}.
Given ũ1 ∈ K, construct a solution (ζ, ū) to (20). Formally, this system is the Navier–Stokes model of
motion for a compressible fluid; therefore, we can construct its unique solution as in [5].
Then insert (ζ, ū) and

1 
1 ∂
w=− ζ ũ1 dz
ζ ∂x
0

in (22). We thus obtain a parabolic equation for the function ũ = ũ2 which is considered as the result
of action of the mapping A : ũ1 → ũ2 . The solution ũ2 belongs to L2 (0, T ; W 2,2 (Ω)) ∩ W 1,2 (0, T ; L2 (Ω)),
and consequently to K. Thus, all hypotheses of the Schauder Fixed Point Theorem are satisfied and the
operator A has at least one fixed point, which implies existence of a solution.

References
1. Min Lu, Kazhikhov A. V., and Ukai Seiji, “Global solutions to the Cauchy problem of the Stokes approximation equations
for two-dimensional compressible flow,” Sci. Bull. Josai Univ. Spec. Iss., No. 5, 155–174 (1998).
2. Lions P.-L., Mathematical Topics in Fluid Mechanics. Compressible Models. Vol. 2, Oxford Univ. Press, Oxford (1998).
3. Lewandowski R., Analyse Mathématique et Océanographie, Masson (1997).
4. Kochin N. E., “On simplification of the equations of hydromechanics in the case of the general circulation of the
atmosphere,” Trudy Glavn. Geofiz. Observator., 4, 21–45 (1936).
5. Antontsev S. N., Kazhikhov A. V., and Monakhov V. N., Boundary Value Problems of Inhomogeneous Fluid Mechanics
[in Russian], Nauka, Novosibirsk (1983).
6. Nash J., “Le probleme de Cauchy pour les equations differentielles d’un fluide general,” Bull. Soc. Math. France, 90,
No. 4, 487–491 (1962).
7. Matsumura A. and Nishida T., “The initial value problem for the equation of motion of viscous and heat-conductive
gases,” J. Math. Kyoto Univ., 20, No. 1, 67–104 (1980).
8. Brech D., Gullen-Gonzales F., Masmoudi N., and Rodrigues-Bellido M. A., “On the uniqueness for the two-dimensional
primitive equation,” Differential Integral Equations, 16, No. 1, 77–94 (2001).
9. Bresch D., Kazhikhov A., and Lemoine J., “On the two-dimensional hydrostatic Navier–Stokes equations,” SIAM J.
Math. Anal., 36, No. 3, 796–814 (2004).
10. Gullen-Gonzales F., Masmoudi N., and Rodrigues-Bellido M. A., “Anisotropic estimates and strong solutions of the
primitive equations,” Differential Integral Equations, 14, No. 11, 1381–1400 (2001).
11. Ladyzhenskaya O. A. and Ural
tseva N. N., Linear and Quasilinear Equations of Elliptic Type [in Russian], Nauka,
Moscow (1973).

812