GEORGIAN MATHEMATICAL JOURNAL: Vol. 5, No.

4, 1998, 341-360

ON SOME BOUNDARY VALUE PROBLEMS FOR AN

ULTRAHYPERBOLIC EQUATION

S. KHARIBEGASHVILI

Abstract. The correct formulation of a characteristic problem and

a Darboux type problem in the special weighted functional spaces for
an ultrahyperbolic equation is investigated.

In the space of variables x1 , x2 , y1 and y2 we consider the ultrahyperbolic

equation

uy1 y1 + uy2 y2 ux1 x1 ux2 x2 = F. (1)

Denote by D : y1 < x1 < y1 , 0 < y1 < +, a dihedral angle bounded

by the characteristic surfaces S1 : x1 y1 = 0, 0 y1 < +, and S2 :
x1 + y1 = 0, 0 y1 < +, of equation (1).
We shall consider a characteristic problem formulated as follows: in the
domain D find a solution u(x1 , x2 , y1 , y2 ) of equation (1) by the boundary
conditions

u | Si = f i , i = 1, 2, (2)

where fi , i = 1, 2, are given real functions on Si and (f1 f2 ) |S1 S2 = 0.

Characteristic problems formulated similarly were considered in [13].
Let G = {(x1 , 1 , y1 , 2 ) R4 : y1 < x1 < y1 , 0 < y1 < +;
< i < +, i = 1, 2}.
Denote by k (D), k 2, the space of functions u(x1 , x2 , y1 , y2 ) of the
class C k (D) whose partial Fourier transforms u b(x1 , 1 , y1 , 2 ) with respect to
the variables x2 and y2 are continuous functions in G together with partial
derivatives with respect to the variables x1 and y1 up to kth order inclusive
and satisfy the following estimates: for any natural N there exist positive

1991 Mathematics Subject Classification. 35L20.

Key words and phrases. Ultrahyperbolic equation, characteristic problem, Darboux
type problem.
341
c 1998 Plenum Publishing Corporation
1072-947X/98/0700-0341$15.00/0
342 S. KHARIBEGASHVILI

integers CN = CN (x1 , y1 ) and KN = KN (x1 , y1 ) not depending on 1 , 2

such that the inequalities
i +i
1 2 /xi1 y i2 u N (|1 |+|2 |)
, 0 i1 + i2 k, (3)
1 1 b(x1 , 1 , y1 , 2 ) CN e

are fulfilled for y1 x1 y1 , 0 y1 < + and |1 | + |2 | > KN . Note

that
0
CN (x1 , y1 ) = sup CN (x01 , y10 ) < +,
(x01 ,y10 )I(x1 ,y1 )
0
KN (x1 , y1 ) = sup KN (x01 , y10 ) < +,
(x01 ,y10 )I(x1 ,y1 )

where I(x1 , y1 ) = {(x01 , y10 ) R2 : 0 y10 x10 y1 x1 , 0 y10 + x01

y1 + x1 } is the closed rectangle.
The spaces k (Si ), i = 1, 2, are introduced similarly. Note that the trace
u |Si of the function u from the space k (D) belongs to the space k (Si ).
When considering a solution of problem (1), (2) in the space k (D),
k 2, it will be assumed that F k1 (D), fi k (Si ), i = 1, 2.
If u is a solution of problem (1), (2) of the class k (D), then after the
Fourier transformation with respect to the variables x2 and y2 , equation (1)
and conditions (2) can be rewritten as
vy1 y1 vx1 x1 + (12 22 )v = F0 , (4)
v |lj = gj , j = 1, 2, (5)
where v, F0 , g1 , g2 are respectively the Fourier transforms of the functions
u, F , f1 , f2 with respect to the variables x2 and y2 , i.e.,
Z Z
1
v(x1 , y1 , 1 , 2 ) = u(x1 , x2 , y1 , y2 )eix2 1 iy2 2 dx2 dy2 ,
2

Z Z
1
F0 (x1 , y1 , 1 , 2 ) = F (x1 , x2 , y1 , y2 )eix2 1 iy2 2 dx2 dy2 ,
2

Z Z
1
gj (x1 , y1 , 1 , 2 ) = gj (x1 , x2 , y1 , y2 )eix2 1 iy2 2 dx2 dy2 ,
2

(x1 y1 ) lj , j = 1, 2,
and l1 : x1 y1 = 0, 0 y1 < +, l2 : x1 + y1 = 0, 0 y1 < +, are the
beams lying in the plane of variables x1 , y1 and outgoing from the origin
O(0, 0).
Remark. Thus by the Fourier transformation with respect to the variables
x2 and y2 the spatial problem (1), (2) is reduced to the Goursat plane
 BVP FOR AN ULTRAHYPERBOLIC EQUATION 343

problem (4), (5) with the parameters 1 and 2 lying in the domain D0 :
y1 < x1 < y1 , 0 < y1 < + of the plane of variables x1 , y1 . It is easy to
see that in the class of functions defined by inequalities (3), this reduction
is equivalent, i.e., problem (1), (2) is equivalent to problem (4), (5).
Using for the functions v, F0 , gi the previous notation, in terms of the
new variables
1 1
x= (y1 + x1 ), y= (y1 x1 ), (6)
2 2
problem (4), (5) takes the form

2v
+ (12 22 )v = F0 , (7)
xy
v |j = gj , j = 1, 2. (8)

Here the solution v = v(x, y, 1 , 2 ) of equation (7) is considered in the

domain 0 : 0 < x < +, 0 < y < + of the plane of variables x, y, which
is the image of the domain D0 for the linear transform (6), and 1 : y = 0,
0 x < +, and 2 : x = 0, 0 y < +, are the images l1 and l2 for the
same transform (6).
With our assumptions for the functions F0 , g1 , g2 , problem (7), (8) has
a unique solution v of the class C 2 (0 ) which has the form [4]

v(x, y, 1 , 2 ) = R(x, 0; x, y)g1 (x, 1 , 2 ) + R(0, y; x, y)g2 (y, 1 , 2 )

Zx
R(, 0; x, y)
R(0, 0; x, y)g1 (0, 1 , 2 ) g1 (, 1 , 2 )d

0
Zy Zx Zy
R(0, ; x, y)
g2 (, 1 , 2 )d+ d R(, ; x, y)F0 (, , 1 , 2 )d, (9)

0 0 0

where g1 (x, 1 , 2 ) = v(x, 0, 1 , 2 ), g2 (y, 1 , 2 ) = v(0, y, 1 , 2 ) are the Gour-

sat data for v, and R(, ; x, y) is the Riemann function for equation (7).
For simplicity, we omit the dependence of R on the parameters 1 and 2 .
It is well known that the Riemann function R(, ; x, y) for equation (7)
can be expressed in terms of the Bessel function J0 of zero order as follows:
q
R(, ; x, y) = J0 ( 4(12 22 )(x )(y )) =

X (x )k (y )k
= (22 12 )k . (10)
(k!)2
k=0
344 S. KHARIBEGASHVILI

1
R
The representation [6] J0 (z) = 2
exp(iz sin )d readily implies
(
1 for |1 | |2 |,
|R(, ; x, y)| p (11)
exp 4|12 22 |(x )(y ) for |1 | < |2 |.

Using the equalities

Z
z
J00 (z) = cos2 exp(iz sin )d,
2

Z
dJ0 (2 x) 2
= cos2 exp(i2 x sin )d, > 0,
dx

we obtain
(
dJ (2x) 22 for = 0,
0

dx 2||2 exp 2|| x for Re = 0.

Hence for derivatives of R we easily derive the estimates

2 2
R(, 0; x, y) 2(1 2 ) y forp|1 | |2 |,

2(12 22 ) y exp 4(12 22 )(x )y (12)


for |1 | < |2 |,

2 2
R(0, ; x, y) 2(1 2 )x forp|1 | |2 |,

2(12 22 ) x exp 4(12 22 )x(y ) (13)


for |1 | < |2 |.

Next, we assume without loss of generality that for the functions F0 , g1 ,

g2 estimates (3) hold with respect to 1 , 2 , with the same integers CN ,
0 0
CN , KN , and KN . Then, using (11)(13), for the solution v(x, y, 1 , 2 )
0
of problem (7), (8) written in form (9) we obtain for |1 | + |2 | > KN the
following estimates:

|v(x, y, 1 , 2 )| |g1 (x, 1 , 2 )| + |g2 (y, 1 , 2 )| + |g1 (0, 1 , 2 )

q q

exp 4|12 22 |xy + 2|22 12 | y exp 4|22 12 |xy
Zx q

|g1 (, 1 , 2 )|d + 2|22 12 | x exp 4|22 12 |xy
0
Zy q Zx Zy
2 2
|g2 (, 1 , 2 )|d + exp 4|2 1 |xy d |F0 (, , 1 , 2 )|d
0 0 0
 BVP FOR AN ULTRAHYPERBOLIC EQUATION 345
q
0 0
2CN exp(N (|1 | + |2 |) + CN exp 4|12 22 |xy
q

exp[N (|1 | + |2 |)] + 2|22 12 | yx exp 4|22 12 |xyCN 0

q
exp[N (|1 | + |2 |)] + 2|22 12 | xy exp 4|22 12 |xyCN 0

q
0
exp[N (|1 | + |2 |)] + xy exp 4|22 12 |xyCN
2 2

exp[N (|1 | + |2 |)] 3 + 2( xy + yx)|2 1 | + xy
0
CN exp 2 xy(|1 | + |2 |) exp[N (|1 | + |2 |)]. (14)

Now choose a least natural number N0 = N0 (x, y) such that

N0 > N + 2 xy + 1. (15)
Taking into account that

AN (x, y) = sup 0
CN 0
3 + 2( xy + yx)|22 12 | + xy
(1 ,2 )R2

exp (|1 | + |2 |) < +,
0
from (14) and (15), for |1 | + |2 | > KN 0
, we obtain
0
|v(x, y, 1 , 2 )| 3 + 2( xy + yx)|22 12 | + xy CN 0
exp 2 xy
0

(|1 | + |2 |) exp[N (|1 | + |2 |)] CN 0
3 + 2( xy + yx)|22 12 | +

+xy exp 2 xy(|1 | + |2 |) exp[(|1 | + |2 |)] exp 2 xy(|1 | + |2 |)

exp (N0 N 2 xy 1)(|1 | + |2 |)
exp[N (|1 | + |2 |)] AN (x, y) exp[N (|1 | + |2 |)]. (16)
The latter inequality implies estimate (3) for i1 = i2 = 0. The proof of
this estimate for i1 + i2 > 0 is similar. Thus we have
Theorem 1. For any F k1 (D), fi k (Si ), i = 1, 2, problem (1),
(2) is uniquely solvable in the class k (D), k 2.
We denote by D1 : k2 y1 < x1 < k1 y, 0 < y1 < +, 0 < ki < 1, i = 1, 2,
a dihedral angle bounded by the surfaces S10 : x1 k1 y1 = 0, 0 y1 < +
and S20 : x1 + k2 y1 = 0, 0 y1 < +.
Let us consider a multidimensional variant of the second Darboux prob-
lem fomulated as follows: in the domain D1 find a solution u(x1 , y1 , x2 , y2 )
of equation (1) by the boundary conditions

(M1 ux1 + N1 uy1 + Mf1 ux + N
e1 uy + Se1 u) 0 = f1 , (17)
2 2 S1

(M2 ux1 + N2 uy1 + M e2 uy + Se2 u) 0 = f2 ,
f2 ux + N (18)
2 2 S 2
346 S. KHARIBEGASHVILI

where Mi , Ni , M fi , N
ei , Sei , i = 1, 2, are given real functions on S 0 ; the
i
f
coefficients Mi , Ni , Mi , N ei , Sei , i = 1, 2, depend only on the variables x1
and y1 .
Note that some multidimensional variants of the second Darboux problem
for a wave equation are investigated in [7, 8].

Denote by k (D1 ), k 2, 0, the space of functions u(x1 , x2 , y1 , y2 )
of the class C k (D1 ) for which

i1 +i2 /xi11 y1i2 u(0, x2 , 0, y2 ) = 0,

< x2 < +, < y2 < + 0 i1 + i2 k,
and whose partial Fourier transforms u b(x1 , 1 , y1 , 2 ) with respect to the
variables x2 and y2 are continuous functions in

G1 = (x1 , 1 , y1 , 2 ) R4 : k2 y1 x1 k1 y1 ,

0 y1 < +; < i < +, i = 1, 2
together with partial derivatives with respect to the variables x1 and y1 up
to kth order inclusive and satisfy the following estimates: for any natural
N there exist positive integers CN = CN (x1 , y1 ) and KN = KN (x1 , y1 ) not
depending on 1 , 2 such that for k2 y1 x1 k1 y1 , 0 y1 < +, and
|1 | + |2 | > KN we have the inequalities
i +i
1 2 /xi1 y i2 u k+i1 i2 N (|1 |+|2 |)
1 1 b(x1 , 1 , y1 , 2 ) CN y1 e , (19)
0 i1 + i2 k,
with
0
CN (x1 , y1 ) = sup CN (x01 , y10 ) < +,
(x01 ,y10 )I1 (x1 ,y1 )
0
KN (x1 , y1 ) = sup KN (x01 , y10 ) < +,
(x01 ,y10 )I1 (x1 ,y1 )

where I1 (x1 , y1 ) = (x01 , y10 ) R2 : k2y10 x01 k1 y10 , y10 x01 y1
x1 , y10 + x01 y1 + x1 , k2 y1 x1 k1 y1 is the closed rectangle.

The spaces k (Si0 ), i = 1, 2, are derived in a similar manner. Note that

the trace u|Si0 of the function u from the space k (D1 ) belongs to the space

k (Si0 ).

Remark. When considering problem (1), (17), (18) in the class k (D1 ),
fi ,
we require of the functions F , fi , i = 1, 2, and the coefficients Mi , Ni , M
e e
Ni , Si , i = 1, 2, in the boundary conditions (17), (18) that

k1
F (D1 ), fi k1 0
(Si ), i = 1, 2;
 BVP FOR AN ULTRAHYPERBOLIC EQUATION 347

fi , N
Mi , Ni , M ei , Sei C k1 (S 0 ), i = 1, 2.
i

Using the notation uy1 = v1 , ux1 = v2 , uy2 = v3 , ux2 = v4 , we reduce

equation (1) to the first-order system

uy1 = v1 , (20)
v1y1 + v3y2 v2x1 v4x2 = F, (21)
v2y1 v1x1 = 0, (22)
v3y1 v1y2 = 0, (23)
v4y1 v1x2 = 0, (24)
and write the boundary conditions (17) and (18) in the form

(M1 v2 + N1 v1 + M e1 v3 + Se1 u) = f1 ,
f1 v4 + N (25)
S1

f2 v4 + N
(M2 v2 + N2 v1 + M e2 v3 + Se2 u) = f2 . (26)
S2

Along with conditions (25), (26) we shall also consider the boundary condi-
tions

(ux1 v2 )S 0 S 0 = 0, (27)
1 2
(ux2 v4 )S 0 S 0 = 0, (28)
1 2
(uy2 v3 )S 0 S 0 = 0. (29)
1 2

Clearly, if u is a regular solution of problem (1), (17), (18) of the class

k (D1 ), then the system of functions u, v1 , v2 , v3 , v4 will be a regular

solution of the boundary value problem (20)(29), where vi k1 (D 1 ),
i = 1, . . . , 4. Conversely, let the system of functions u, v1 , v2 , v3 , v4 of the

class k1
(D 1 ) be a solution of problem (20)(29). We shall prove that
then v1 = uy , v2 = ux1 , v3 = uy2 , v4 = ux2 and therefore the function u is a

solution of problem (1), (17), (18) in the class k (D1 ). Indeed, by equality
(22) we have
(ux1 v2 )y1 = (uy1 )x1 v2y1 = v1x1 v2y1 = 0,
which by virtue of the boundary condition (27) implies that v2 ux1 in D1 .
Further, the use of equality (23) gives
(uy2 v3 )y1 = (uy1 )y2 v3y1 = v1y2 v3y2 = 0,
which by virtue of the boundary condition (29) implies that v3 uy2 in D1 .
Finally, on account of (24) and (28) we find by a similar reasoning that
v4 ux2 in D1 .
348 S. KHARIBEGASHVILI

Thus problem (1), (17), (18) in the class k (D1 ) is equivalent to the prob-

lem of finding a system of functions u, v1 , v2 , v3 , v4 of the class k1
(D 1 )
satisfying (20)(29).
If u, v1 , v2 , v3 , v4 are a solution of problem (20)(29), then, after the
Fourier transformation with respect to the variables x2 and y2 , equations
(20)(24) and boundary conditions (25)(29) take the form

by1 = vb1 ,
u (30)
vb1y1 vb2x1 + i2 vb3 i1 vb4 = Fb, (31)
vb2y1 vb1x1 = 0, (32)
vb3y1 i2 vb1 = 0, (33)
vb4y1 i1 vb1 = 0, (34)

(M1 vb2 + N1 vb1 + Mf1 vb4 + N b)l = fb1 ,
e1 vb3 + Se1 u (35)
1

f e
(M2 vb2 + N2 vb1 + M2 vb4 + N2 vb3 + S2 ue b) l2 = fb2 ,
(36)

ux1 vb2 )l1 l2 = 0,
(b (37)

(b b)l1 l2 = 0,
v4 i1 u (38)

v3 i2 u
(b b)l1 l2 = 0, (39)

where u b, vbj , j = 1, . . . , 4, Fb , fb1 , fb2 are respectively the Fourier transforms

of the functions u, vj , j = 1, . . . , 4, F , f1 , f2 with respect to the variables x2
and y2 ; l1 : x1 k1 y1 = 0, 0 y1 < +, l2 : x1 + k2 y1 = 0, 0 y1 < +,
are the beams lying in the plane of the variables x1 , y1 and outgoing from
the origin O(0, 0). Here i = 1.
Thus, after the Fourier transformation with respect to the variables x2
and y2 , the spatial problem (20)(29) is reduced to the plane problem (30)
(39) with the parameters 1 and 2 lying in the domain 1 : k2 y1 < x1 <
k1 y1 , 0 < y1 < +, of the plane of variables x1 , y1 . It is easy to verify that

in the class k (1 ) of functions defined by inequalities (19) this reduction
is equivalent.
Assuming v = (b v1 , vb2 ) and v5 = (b v3 , vb4 ), we rewrite system (31), (32) in
the vector form

vy1 Avx1 + A1 v5 = F1 , (40)

where
0 1 i2 i1 Fb
A= , A1 = , F = .
1 0 0 0 0
We easily see that with respect to v system (40) is strictly hyperbolic
and its characteristics L1 (x01 , y10 ) and L2 (x01 , y10 ) passing through the point
 BVP FOR AN ULTRAHYPERBOLIC EQUATION 349

(x01 , y10 ) are the straight lines defined by the equations

L1 (x01 , y10 ) : x1 y1 = x01 y10 , L2 (x10 , y10 ) : x1 + y1 = x01 + y10 .
If

1 1 1 0
B= , = , (41)
1 1 0 1
then, as is easy to verify,

1 1 1 1
B = , B 1 AB = . (42)
2 1 1
Therefore by replacing v = Bw we can rewrite system (40) as
wy1 wx1 + B 1 A1 v5 = B 1 F1 . (43)
By virtue of (41)(43), after replacing v = Bw, problem (30)(39) is rewrit-
ten as
by1 = w1 w2 ,
u (44)
1 1
w1y1 w1x1 + i2 vb3 1 vb4 = F2 , (45)
2 2
1 1
w1y1 + w1x1 i2 vb3 + 1 vb4 = F3 , (46)
2 2
vb3y1 i2 (w1 w2 ) = 0, (47)
vb4y1 i1 (w1 w2 ) = 0, (48)

[(M1 + N1 )w1 + (M1 N1 )w2 + M f1 vb4 + N b]l1 = fb1 ,
e1 vb3 + Se1 u (49)

[(M2 + N2 )w1 + (M2 N2 )w2 + M f2 vb4 + N b]l = fb2 ,
e1 vb3 + Se2 u (50)
2

ux1 (w1 + w2 ))l1 l = 0,
(b (51)
2

(b b)l1 l2 = 0,
v4 i1 u (52)

v3 i2 u
(b b)l l = 0, (53)
1 2

where (F2 , F3 ) = B 1 F1 .
It is easy to see that the characteristic L1 (x10 , y10 ) : x1 + y1 = x01 + y10 of
equation (45) passing through the point (x01 , y10 ) 1 intersects the beam
l1 at a single point whose ordinate is denoted by 1 (x01 , y10 ). In a similar
manner, the characteristic L2 (x01 , y10 ) : x1 y1 = x10 y10 , (x01 , y10 ) 1 , of
equation (46) intersects l2 at a single point with ordinate 2 (x01 , y10 ). Clearly,
i (x10 , y10 ) C (1 ), i (x01 , y10 ) y10 , i = 1, 2. (54)
It should also be noted that a segment of each characteristic Li (x10 , y10 ),
i = 1, 2, (x10 , y10 ) 1 , drawn from the point (x01 , y10 ) towards decreasing
350 S. KHARIBEGASHVILI

ordinate values to the intersection with one of the beams l1 or l2 , admits a

parametrization of the form

x1 = zi (x01 , y10 ; t), y1 = t, i (x01 , y10 ) t y10 , (55)

where z1 (x01 , y10 ; t) = x01 + y10 t C [1 (x01 , y10 ), y10 ], z2 (x01 , y10 ; t) = x01
y10 + t C [2 (x01 , y10 ), y10 ].
Remark. The functions u b, w1 , w2 , vb3 , vb4 , F2 , F3 , fb1 , fb2 depend not only
on the independent variables x1 and x2 , but also on the parameters 1 and
2 . For simplicity, these parameters will be omitted below. For example,
instead of u
b(x1 , y1 , 1 , 2 ) we shall write u b(x1 , y1 ).

b(0, 0) = 0, for u k (D1 ) we obtain
Using (44), (51) and the fact that u

Zy1 Zy1
b(x1 , y1 ) = (k1 ux1 + uy1 )(k1 t, t)dt = [(1 + k1 )w1 +
u
0 0
+ (1 k1 )w2 ](k1 t, t)dt, (x1 , y1 ) l1 , (56)
Zy1 Zy1
b(x1 , y1 ) = (k2 ux1 + uy1 )(k2 t, t)dt = [(1 k2 )w1 +
u
0 0
+ (1 + k2 )w2 ](k2 t, t)dt, (x1 , y1 ) l2 . (57)

For (x1 , y1 ) 1 and x1 > 0, by integrating equations (44), (47), (48)

with respect to the variable y1 and taking into account the boundary con-
ditions (52), (53) and equalities (56), (57) we obtain

k11 x1
Z
u
b(x1 , y1 ) = [(1 + k1 )w1 + (1 k1 )w2 ](k1 t, t)dt +
0
Zy1
+ (w1 w2 )(x1 , t)dt, (58)
k11 x1

k11 x1
Z
vb3 (x1 , y1 ) = i2 [(1 + k1 )w1 + (1 k1 )w2 ](k1 t, t)dt +
0
Zy1
+ i2 (w1 w2 )(x1 , t)dt, (59)
k11 x1
 BVP FOR AN ULTRAHYPERBOLIC EQUATION 351

k11 x1
Z
vb4 (x1 , y1 ) = i1 [(1 + k1 )w1 + (1 k1 )w2 ](k1 t, t)dt +
0
Zy1
+ i1 (w1 w2 )(x1 , t)dt, (60)
k11 x1

and for (x1 , y1 ) 1 and x1 0 we have

k11 x1
Z
b(x1 , y1 ) =
u [(1 k2 )w1 + (1 + k2 )w2 ](k2 t, t)dt +
0
Zy1
+ (w1 w2 )(x1 , t)dt, (61)
k21 x1

k21 x1
Z
vb3 (x1 , y1 ) = i2 [(1 k2 )w1 + (1 + k2 )w2 ](k2 t, t)dt +
0
Zy1
+ i2 (w1 w2 )(x1 , t)dt, (62)
k21 x1

k21 x1
Z
vb4 (x1 , y1 ) = i1 [(1 k2 )w1 + (1 + k2 )w2 ](k2 t, t)dt +
0
Zy1
+ i1 (w1 w2 )(x1 , t)dt. (63)
k21 x1

Now, by integrating equations (45) and (46) along the respective charac-
teristics from the point (x1 , y1 ) 1 to the intersection points of these
characteristics with the beams l1 and l2 we obtain

w1 (x1 , y1 ) = w1 (k1 1 (x1 , y1 ), 1 (x1 , y1 )) +

Zy1
1
+ i (1 vb4 2 vb3 )(z1 (x1 , y1 ; t), t)dt+ Fe2 (x1 , y1 ), (64)
2
1 (x1 ,y1 )

w2 (x1 , y1 ) = w2 (k2 2 (x1 , y1 ), 2 (x1 , y1 )) +

352 S. KHARIBEGASHVILI

Zy1
1
+ i (2 vb3 1 vb4 )(z2 (x1 , y1 ; t), t)dt+ Fe3 (x1 , y1 ), (65)
2
2 (x1 ,y1 )

where
Zy1
Fe2 (x1 , y1 ) = F2 (z1 (x1 , y1 ; t), t)dt,
1 (x1 ,y1 )
Zy1
Fe3 (x1 , y1 ) = F3 (z2 (x1 , y1 ; t), t)dt.
2 (x1 ,y1 )

We set
= w1 |l1 , = w2 |l2 .
b, w1 , w2 , vb3 , vb4 from (58)
On substituting the obtained expressions for u
(65) into the boundary conditions (49), (50) and taking into account that
1 k2
1 (x1 , y1 )|l1 = y1 , 1 (x1 , y1 ) = y1 ,
1 + k1
1 k1
2 (x1 , y1 ) = y1 , 2 (x1 , y1 )|l2 = y1
1 + k2
we obtain
1
(M1 + N1 )(y1 ) + (M1 N1 )(2 y1 ) + (M1 N1 ) i
2
Zy1
f1 i1 + N
(2 vb3 1 vb4 )(z2 (k1 y1 , y1 ; t), t)dt + (M e1 i2 + Se1 )
2 y1
Zy1
[(1 + k1 )w1 + (1 k1 )w2 ](k1 t, t)dt = fb1 (y1 ) (M1 + N1 )(y1 )
0

Fe2 (k1 y1 , y1 ) (M1 N1 )(y1 )Fe3 (k1 y1 , y1 ), (66)

1
(M2 + N2 )(1 y1 ) + (M2 N2 )(y1 ) + (M2 + N2 ) i
2
Zy1
(1 vb4 2 vb3 )(z1 (k2 y1 , y1 ; t), t)dt + (M e2 i2 + Se2 )
f2 i1 + N
1 y1
Zy1
[(1 k2 )w1 + (1 + k2 )w2 ](k2 t, t)dt = fb2 (y1 )
0

(M2 + N2 )(y1 )Fe2 (k2 y1 , y1 ) (M2 N2 )(y1 )Fe3 (k2 y1 , y1 ). (67)

 BVP FOR AN ULTRAHYPERBOLIC EQUATION 353

Now, since 0 < ki < 1, i = 1, 2, we have

1 k2 1 k1
0 < 1 = < 1, 0 < 2 = < 1, 0 < = 1 2 < 1. (68)
1 + k1 1 + k2
It will assumed below that

(M1 + N1 )l 6= 0, (M2 N2 )l 6= 0.
1 2

Since by assumptions the functions Mj , Nj , j = 1, 2, do not depend on

the variables x2 and y2 , these conditions are evidently equivalent to the
conditions

(M1 + N1 )S 0 6= 0, (69)
1
(M2 N2 )S 0 =6 0. (70)
2

Solving system (66), (67) with respect to and , we obtain

(y1 ) a(y1 )( y1 ) = (T1 (w))(y1 ) + (T2 (v5 ))(y1 ) + g1 (y1 ), (71)

(y1 ) b(y1 )( y1 ) = (T3 (w))(y1 ) + (T5 (v5 ))(y1 ) + g2 (y1 ), (72)

where

a(y1 ) = (M1 N1 )(M1 + N1 )1 (y1 ) (M2 + N2 )(M2 N2 )1 (2 y1 ),

b(y1 ) = (M2 + N2 )(M2 N2 )1 (y1 ) (M1 N1 )(M1 + N1 )1 (1 y1 ),

and Tj , j = 1, . . . , 4, are linear integral operators of the form

Zy1
(T1 (w))(y1 ) = (E11 1 + E12 2 + E13 ) w(k1 t, t)dt +
0
Z2 y1

+ (E14 1 + E15 2 + E16 ) w(k2 t, t)dt,

0
Zy1
(T2 (v5 ))(y1 ) = (E21 1 + E22 2 ) v5 (z2 (k1 y1 , y1 ; t), t)dt +
2 y1
Z2 y1

+ (E23 1 + E24 2 ) v5 (z1 (k2 2 y1 , 2 y1 ; t), t)dt,

y1
Zy1
(T3 (w))(y1 ) = (E31 1 + E32 2 + E33 ) w(k2 t, t)dt +
0
354 S. KHARIBEGASHVILI

Z1 y1

+ (E34 1 + E35 2 + E36 ) w(k1 t, t)dt,

0
Zy1
(T4 (v5 ))(y1 ) = (E41 1 + E42 2 ) v5 (z1 (k2 y1 , y1 ; t), t)dt +
1 y 1
Z1 y1

+ (E43 1 + E44 2 ) v5 (z2 (k1 1 y1 , 1 y1 ; t), t)dt.

y1

Here v5 = (b v3 , vb4 ), Eij are the completely defined 1 2 matrices of the

class C k1 , and g1 , g2 are functions expressed in terms of the well-known
functions F and f1 , f2 .
As mentioned above, in equations (58)(65), (71), (72) the unknown
functions and the right-hand sides depend on the parameter = (1 , 2 ),
which we omit for simplicity.
Remark. As follows from the above reasoning, if conditions (69), (70)

are fulfilled, then in the class k (D1 ) problem (1), (17), (18) is equivalent
to the problem of finding a system of functions u b, w1 , w2 , vb3 , vb4 , , and
from equations (58)(65), (71), (72), where

k1
b, w1 , w2 , vb3 , vb4 k1
u k1
(1 ), [0, +), [0, +)

and

Fe2 , Fe3
k1 k1
(1 ), g1 [0, +), g2 k1
[0, +).

Let Q(x10 , y10 ) 1 . Denote by P1 and P2 the points of intersection of

the characteristics L1 (x01 , y10 ) : x1 y1 = x01 y10 and L2 (x01 , y10 ) : x1 + y1 =
x01 + y10 from system (40) with the beams l1 and l2 , respectively. Denote by
1Q 1 a rectangle with vertices at the points O(0, 0), P1 , P2 , and Q.
We set
h (M N )(M + N ) i
1 1 2 2
= a(0) = b(0) = (0),
(M2 N2 )(M1 + N1 )
where a(y1 ) and b(y1 ) are the coefficients in equations (71), (72).

Lemma 1. Let conditions (69), (70) be fulfilled. Then for k + >

log ||/ log + 1 the boundary value problem (44)(53) is uniquely solvable

in the class k1 (1 ); the domain of the dependence of the solution of this
problem for the point Q(x10 , y10 ) 1 is 1Q .

Remark. As mentioned above, the boundary value problem (44)(53)

is equivalent to the system of integro-functional equations (58)(65), (71),
 BVP FOR AN ULTRAHYPERBOLIC EQUATION 355

(72). When problem (44)(53) is considered in the domain 1Q , it is suf-

ficient to investigate equations (71) and (72) on the segments [0, d1 ] and
[0, d2 ], where d1 and d2 are the ordinates of the points P1 and P2 .
Let us consider the functional equations

(K1j ())(y1 ) = (y1 ) aj (y1 )( y1 ) = 1 (y1 ), 0 y1 d1 , (73)

(K2j ())(y1 ) = (y1 ) bj (y1 )( y1 ) = 2 (y2 ), 0 y1 d2 , (74)

where aj (y1 ) = j a(y1 ), bj (y1 ) = j b(y1 ), j = 0, 1, . . . , k 1. Note that if

we differentiate j-times the expression (K10 ())(y1 ), which is the left-hand
side of equation (71), with respect to y1 , then in the obtained expression
the sum of the terms which contain the function (y1 ) with its derivative
(j) (y1 ) gives (K1j ((j) ))(y1 ). A similar statement also holds for K2j .

Let in equations (73), (74) the right-hand sides p (y1 ) k1+j [0, dp ],

p = 1, 2. It is assumed that k [0, dp ] = [0, dp ] for k = 0. In that case,

by the definition of the space k1+j [0, dp ], for any natural number N
there exist positive numbers Cp = Cp (y, N, p ), Bp = Bp (y1 , N, p ), not
depending on = (1 , 2 ), such that for 0 y1 dp and || = |1 |+|2 | > Bp
the inequality
|p (y1 )| Cp y1k1+j eN ||
is fulfilled and

Cp0 (y1 ) = sup Cp (y10 ) < +, Bp0 (y1 ) = sup Bp (y10 ) < +.
0y10 y1 0y10 y1

Lemma 2. Let conditions (69), (70) be fulfilled. Then for k 1 + >

log ||
equations (73) and (74) are uniquely solvable in the spaces k1+j
log

[0, d1 ] and k1+j [0, d2 ], and for || > B1 and || > B2 the estimates
1
(K ( ))(y1 ) = |(y1 )| 1 C1 y k1+j eN || , (75)
1j 1 1
1 k1+j
(K (2 ))(y1 ) = |(y1 )| 2 C2 y e N ||
(76)
2j 1

hold, where the positive constants 1 and 2 do not depend on N , , and the
functions 1 , 2 .

The proof of this lemma repeats the reasoning from [9].

To prove Lemma 1, we shall solve the system of integro-functional equa-
tions (58)(65), (71), (72) with respect to the unknowns

k1 k1
b, w1 , w2 , vb3 , vb4 k1
u (1Q ), [0, d1 ], [0, d2 ]

by the method of successive approximations.

356 S. KHARIBEGASHVILI

We set

b0 (x1 , y1 ) 0, w1,0 (x1 , y1 ) 0, w2,0 (x1 , y1 ) 0,

u
vb3,0 (x1 , y1 ) 0, vb4,0 (x1 , y1 ) 0, 0 (y1 ) 0, 0 (y1 ) 0,
1
Z x1

u
bn (x1 , y1 ) = [(1 + )w1,n1 + (1 )w2,n1 ](t, t)dt +
0
Zy1
+ (w1,n1 w2,n1 )(x1 , t)dt, (77)
1 x 1
1
Z x1

vb3,n (x1 , y1 ) = i2 [(1 + )w1,n1 + (1 )w2,n1 ](t, t)dt +

0
Zy1
+i2 (w1,n1 w2,n1 )(x1 , t)dt, (78)
1 x1
1
Z x1

vb4,n (x1 , y1 ) = i1 [(1 + )w1,n1 + (1 )w2,n1 ](t, t)dt +

0
Zy1
+i1 (w1,n1 w2,n1 )(x1 , t)dt, (79)
1 x1
(
k1 for x1 > 0,
= 1 x1 y1 for (x1 , y1 ) 1Q ,
k2 for x1 0,
w1,n (x1 , y1 ) = n (1 (x1 , y1 )) +
Zy1
1
+ i (1 vb4,n1 2 vb3,n1 )(z1 (x1 , y1 ; t), t)dt + Fe2 (x1 , y1 ), (80)
2
1 (x1 ,y1 )

w2,n (x1 , y1 ) = n (2 (x1 , y1 )) +

Zy1
1
+ i (2 vb3,n1 1 vb4,n1 )(z2 (x1 , y1 ; t), t)dt + Fe3 (x1 , y1 ). (81)
2
2 (x1 ,y1 )

To define the functions n (y1 ) and n (y1 ), we shall use the equations

(K10 (n ))(y1 ) = (T1 (wn1 ))(y1 ) + (T2 (v5,n1 ))(y1 ) + g1 (y1 ), (82)
(K20 (n ))(y1 ) = (T3 (wn1 ))(y1 ) + (T4 (v5,n1 ))(y1 ) + g2 (y1 ), (83)
 BVP FOR AN ULTRAHYPERBOLIC EQUATION 357

where v5 = (b v3 , vb4 ).
If the conditions of Lemma 2 are fulfilled, then using estimate (75), (76)
for j = 0 and applying (54), (68), also the inequality 1 x1 y1 for
(x1 , y1 ) 1Q , Q(x01 , y10 ) 1 , and taking into account the Volterra struc-
ture of the integral operators contained in equalities (77)(78), for || > e
we obtain by the method of mathematical induction the inequalities:
Mn
ubn+1 (x1 , y1 ) ubn (x1 , y1 ) M (1 + ||)n eN || y1n+k1+ , (84)
n!
n
wp,n+1 (x1 , y1 ) wp,n (x1 , y1 ) M M (1 + ||)n eN || y n+k1+ , (85)
1
n!
p = 1, 2,
M n
vbq,n+1 (x1 , y1 ) vbq,n (x1 , y1 ) M (1 + ||)n eN || y n+k1+ , (86)
1
n!
q = 3, 4,
n
n+1 (y1 ) n (y1 ) M M (1 + ||)n eN || y n+k1+ , (87)
1
n!
n
n+1 (y1 ) n (y1 ) M M (1 + ||)n eN || y n+k1+ , (88)
1
n!
where the positive numbers B= e B(x
e 0 , y 0 , N, f1 , f2 , F ), M=M (x0 , y 0 , 1 , 2 )
1 1 1 1
0 0
and M = M (x1 , y1 , N, f1 , f2 , F, 1 , 2 ) do not depend on ; 1 and 2 are
the constants from (75) and (76).
Remark. As follows from equalities (77)(79), for x1 > 0 and x1 0
the functions ubn , vb3,n and vb4,n are defined by different formulas. This does
not cause discontinuities of the functions u bn , vb3,n and vb4,n and their partial
derivatives with respect to x1 and y1 up to order (k 1) inclusive along
the axis Oy1 : x1 = 0, since it is assumed that the functions Fe2 , Fe3 , g1 , g2
from (72)(83) and their derivatives up to order (k 1) inclusive are equal
to zero at the point O(0, 0).
After differentiating equalities (77)(83) with respect to x1 and y1 and
using the above estimates (84)(88) and (75), (76), for || > B ej +j we
1 2
obtain by the method of mathematical induction the inequalities
j ,j n
Mj1 +j2
[ 1 2 (b un+1 u bn )](x1 , y1 ) Mj1 +j2 (1 + ||)n eN ||
n!
y1n+k+j1 j2 1 , (89)
j ,j n
Mj1 +j2
[ 1 2 (wp,n+1 wp,n )(x1 , y1 ) Mj +j (1 + ||)n eN ||
1 2
n!
y1n+k+j1 j2 1 , p = 1, 2, (90)
n
Mj1 +j2
vq,n+1 vbq,n )(x1 , y1 ) Mj1 +j2
[ j1 ,j2 (b (1 + ||)n eN ||
n!
358 S. KHARIBEGASHVILI

y1n+k+j1 j2 1 , q = 3, 4, (91)
h dj1 +j2 i n
Mj
1 +j2
( n+1 n ) (y1 ) M
j1 +j2 (1 + ||)n
dy1j1 +j2 n!
eN || y1n+k+j1 j2 1 , (92)
h dj1 +j2 i n
Mj
1 +j2
( n+1 n ) (y1 ) M j1 +j2 (1 + ||)n
dy1j1 +j2 n!
eN || y1n+k+j1 j2 1 , (93)

where j1 ,j2 = j1 +j2 /x1j1 y1j2 , 0 j1 + j2 k 1; B

ep , M , and Mp ,
p
p = 1, . . . , k 1, are positive integers not depending on .
By (89) we find that for 0 j1 + j2 k 1 the series

X
bj1 ,j2 (x1 , y1 ) = lim [ j1 ,j2 u
u bn ](x1 , y1 ) = j1 ,j2 (b bn1 ) (x1 , y1 )
un u
n
n=1

converges uniformly in 1Q and for the sum of this series we have the esti-
mate

ubj1 ,j2 (x1 , y1 ) Mj1 +j2 eMj1 +j2 (1+||)y1 eN || y1k+j1 j2 1 . (94)

This estimate implies that ubj1 ,j2 k1+j1 j2 (1Q ), since, as is easy to
verify, the operator of multiplication by the function

eMj1 +j2 (1+||)y1

transforms the space k1+j1 j2 (1Q ) into itself. This in turn implies

b1(x1 , y1 ) u
that u b0,0(x1 , y1 ) k1 bj1 ,j2(x1 , y1 ) j1 ,j2 u
(1Q) and u b1 (x1 , y1 ).
By (90)(93) we find in a similar manner that the series

X
wp1 (x1 , y1 ) = lim wp,n (x1 , y1 ) = wp,n (x1 , y1 )wp,n1 (x1 , y1 ) , p = 1, 2,
n
n=1
X

vbq1 (x1 , y1 ) = lim vbq,n (x1 , y1 ) = vq,n1 (x1 , y1 ) , q = 3, 4,
vbq,n (x1 , y1 )b
n
n=1

converge in the space k1
(1Q ), and the series

X
1 (y1 ) = lim n (y1 ) = n (y1 ) n1 (y1 ) ,
n
n=1

X
1 (y1 ) = lim n (y1 ) = n (y1 ) n1 (y1 )
n
n=1
 BVP FOR AN ULTRAHYPERBOLIC EQUATION 359

k1
converge in the spaces k1 [0, d1 ] and [0, d2 ], respectively. Hence by
virtue of (77)(83) it follows that the limit fnctions u b1 , w1 , vb31 , vb41 , 1 , and
1
satisfy the system of equations (58)(65), (71), (72). Thus to prove
Lemma 1 it only remains to show that the system of equations (58)(65),
(71), (72) has no other solutions in the classes considered above. Indeed,
let us assume that the functions u b , w , vb3 , vb4 , , and from the classes
proved above satisfy the homogeneous system of equations corresponding
to (58)(65), (71), (72), i.e., for

Fe2 = 0, Fe3 = 0, g1 = 0, g2 = 0. (95)

To this system with the homogeneous conditions (95) we apply the me-
thod of successive approximations, assuming that the functions ub , w , vb3 ,
vb4 , , and are zero approximations. Since this system of functions
satisfy the homogeneous system of equations, each of the following approx-
imations will coincide with it, i.e.,
bn (x1 , y1 ) u
u b (x1 , y1 ), wn (x1 , y1 ) w (x1 , y1 ), vb3,n

(x1 , y1 ) vb3 (x1 , y1 ),

vb4,n (x1 , y1 ) vb4 (x1 , y1 ), n (y1 ) (y1 ), n (y1 ) (y1 ).
e we obtain
By the same reasoning as for estimates (84)(88), for || > B
fn
|bu (x1 , y1 )| = |b f M (1 + ||)n eN || y n+k+1 ,
un (x1 , y1 )| M 1
n!
f n
|wp (x1 , y1 )| = |wp,n
(x1 , y1 )| Mf M (1 + ||)n eN || y n+k+1 , p = 1, 2,
1
n!
f n
vq (x1 , y1 )| = |b
|b
vq,n (x1 , y1 )| M f M (1 + ||)n eN || y n+k+1 , q = 3, 4,
1
n!
fn
| (y1 )| = |n (y1 )| M f M (1 + ||)n eN || y n+k+1 ,
1
n!
fn
| (y1 )| = |n (y1 )| M f M (1 + ||)n eN || y n+k+1 .
1
n!

Hence, passing to the limit as n , we find that

b 0, w1 0, w2 0, vb3 0, vb4 0, 0, 0,
u
which completes the proof of the lemma.
Let Q0 (x01 , x02 , y10 , y20 D1 . Denote by
D1Q0 : k2 y1 < x1 < k1 y1 , y1 x1 < y10 x10 , y1 + x1 < y10 + x01
a subdomain of the domain D1 bounded by the surfaces S10 , S20 and the
characteristic hyperplanes Se1 : y1 x1 = y10 x01 and Se2 : y1 + x1 = y10 + x01
passing through the point Q0 .
360 S. KHARIBEGASHVILI

Since problem (1), (17), (18) is equivalently reduced to problem (44)

(53), Lemma 1 immediately implies
Theorem 2. Let conditions (69), (70) be fulfilled. Then for k + >

log ||
+ 1 problem (1), (17), (18) is uniquely solvable in the class k (D1 )
log

for any F k1 k1 0
(D 1 ) and fi (Si ), i = 1, 2; the domain of the de-
pendence of the solution u of this problem for the point Q0 D1 is contained
in D1Q0 .
Note that if conditions (69), (70) or the inequality k + > log ||
log + 1
are invalid, then problem (1), (17), (18) may turn out to be formulated
incorrectly.

References
1. J. Hadamard, Lectures on Cauchys problem in partial differential
equations. Yale University Press, New Haven, 1923.
2. J. Tolen. Probleme de Cauchy sur le deux hypersurfaces caracteris-
tiques secantes. C. R. Acad. Sci. Paris, Ser. AB 291(1980), No. 1,
4952.
3. S. S. Kharibegashvili, On a characteristic problem for the wave equa-
tion. (Russian) Proc. I. Vekua Inst. Appl. Math. 47(1992), 7682.
4. A. V. Bitsadze, Some classes of partial differential equations. (Rus-
sian) Nauka, Moscow, 1981.
5. R. Courant, Partial differential equations. New YorkLondon, 1962.
6. L. Olver, Introduction to asymptotics and special functions. Academic
Press, New YorkLondon, 1974.
7. S. Kharibegashvili, On a spatial problem of Darboux type for a second-
order hyperbolic equation. Georgian Math. J. 2(1995), No. 3, 299311.
8. S. Kharibegashvili, On the solvability of a noncharacteristic spa-
tial problem of Darboux type for the wave equation. Georgian Math. J.
3(1996), No. 1, 5968.
9. S. S. Kharibegashvili, On the solvability of one boundary value prob-
lem for second-order hyperbolic equations. (Russian) Proc. I. Vekua Inst.
Appl. Math. 19(1987), 122172.

(Recieved 29.12.1995)

Authors address:
Department of Theoretical Mechanics (No. 4)
Georgian Technical University
77, Kostava St., Tbilisi 380075
Georgia