A GENERALIZATION O F A LEMMA OF BELLMAN

AND ITS APPLICATION TO UNIQUENESS PROBLEMS

OF DIFFERENTIAL EQUATIONS
By
1. BIHARI ( B u d a p e s t )

(Presented by

P; TuR~,N)

1. T h e original l e m m a in question is as follows:

Let Y(x) and F(x) be positive continuous functions in a <_ x <: b and
let be k > O, M > O, then the inequality
Y(x) <: kq- M i F ( t ) Y ( t ) d t

(1)

~
( (a< x <b).

involves the other inequality

Y(x) < ke "

(2)

It was first made use of this lemma in discussions of stability problems

of differential equations.' Later~ it was applied to investigations of uniqueness
and dependence of the solutions of differential equations and systems on
the initial conditions and parameters, but only for the case of Lipschitz
condition with the exponent 1.
One can deduce by-means of this lemma the lemma of-GRONWALL~ tOO.
The purpose of this paper i s to establish
by means of the Bellman
lemma and its generalized form
a bound for the difference of the solutions
of the equations y ' = f ( x , y ) @ e I and y ' - - f ( x , y ) - k e 2 (~1 and e2 are constants)
with the same or different initial conditions, provided that f(x,y) satisfies
the ,,Osgood condition"

If(x, y2)--f(x,

Yl) < t o ( Y s - - Y z t )

in a domain G where, of course, to(u) is subjected to certain conditions.

Further we shall givesimple proofs for the uniqueness theorems of NAGUMO,
OSGOOD etc.
I . BELLMAN~The stability of,solutions of linear differentia! equations, Duke Math.
10 (i943), pp. 643--647. H o w e v e r , the lemma holds for arbitral- continuous Y(x)
and non-negative continuous F(x). The value k ~ O is also possible.
_o E. g. B. B. H e M hi L~K n f i n B. B. C T e n a ~ o B, KaqecTBe~Ha~ TeopH~ ~aqb~epeutlnanbn~,ix ypaBuennfi (MocKBa, 1947), p. 19.
a E. g. E. KAMKe, Diffet'entialgleichungen reeller Funktionen, p. 93.

Journal,

Acta /~lathematica VII,:1

82

L BIHAI{I

2. T h e u n i q u e n e s s t h e o r e m o f N a g u m o ~ is the following :

If.the function f(x, y) is continuous in a domain G(x, y) and atl points~, ~ of G have a neighbourhood where
(3)
lx--~] If(x, Y2)--f(x, Y~)I ~ ]yo.-Ty~t
holds whenever th e points (x, Yl) and (x, y~) belong to this neighbourhood, then
the equation
(4)
y" : f(x, y)
has at most one solution q)(x) satisfying the initial condition 9v(~):-9.
PROOF: Assuming there exists another solution ~p(x) too with the same
'initial condition g,(.~)=~?, we should have
11r

,~:

q~(x) = rl + ff(t, qD(t))dt,

W(X) = ,2 + .if(t, g,(t))dt

whence

I (x) -- ',p (x) J =<

iJf(t,

at

(x

Taking (3) into account we get

x

(5)

_-<

for sufficiently small x - - ~ > 0. Here the integrand has a limit for t = ~ , since
lira 9(t)--p(t) _ lim r
t--g
i. e. the function cf(t)~p(t)

'(t) _f(~, ~)-Tf(~, r~)= O,

may be completed to a continuous function

by taking 0 for its volue at t~---~. Therefore one can determine to all , > 0
a number d > 0

such that

~(t)_gj(t)
~

<,

if

[t--.~l<d

and

~+d<x.

Then we get from (5)

4+0

+ f~ q~(t)--~(t,
~TT~-~

dt =< ,'.d+ f [fc(t)--ip(t)l t ~ ~ dt.

* M. NAewO, Eine hinreichende Bedingung ft'lr die Unit~* der LiSsung yon Different{algleichtmgen erster Ordnung, Japanese Journal of Math., 3 (1926), p p . 107--I12.

A GENERALIZATION OF A LEh~MA O F ~ELLMAN

8~

Applying the Bellman lemma we obtain

l~(x)--~(x)I

<-~.~e~

--

--

for a certain right band neighbourhood of .~ and any positive number ~.

x

For x~_.~ we conclude in the same way by replacing ~]u(t)dtl~'l by xfU(t)dt

(u(t)~--O). In this neighbourhood of ~ we have q~(x)~-w(x) and - - in a

known way - - we can conclude that this holds for their whole existence
intervals? Q. e. d.
Let us remark that if instead of condition (3) we supposed

Ix

~l lf(x, Y~) f(x, Y~)I -- MlY~.--Y~t

with

M>I,

then b y a similar calculation we should obtain

,~(x)

~(x) < ~.01

x~'

ix--~t '~
-- ~ 0,I_~

and since this depends on d, the further conclusion is impossible. Therefore

the constant M - - 1 cannot be increased ~ in this way.

3. Generalization of the Bellman

lemma.

Let Y(x), F(x) be positive continuous functions in a < x <=b and

k > O, M >--O, further ~o(u) a non-negative non-decreasing continuous function
for u > O. Then the inequality

(6)

Y ( x ) < k + M i F ( t)~ ( V (t) ) d t

(a ~ x <

a)

implies the inequality

(7)

Mj F(t) at)

r(x) =<

(a <-- x =< b' :< b)

where

(8)

" ,dt

~(u) = j ~ / )
9

This procedure may be generalized: If in (1)

(u~ > O, u ~ O)
k=O, F(t)

is continuous in

a<x~b

and lira F(x)Y(x)~A exists and lim6ea+6

-z- -a+0
6-*0
,

~_K(x),then Y(x) >_-K(x).

fA
9

6 0 . PE~ROS, Eine hinreichende Bedingung ffir Unit~it der L6sung yon Differentialgleichungen erster Ordnung, Math. Zeitschrift , 28 (1928), pp. 216--2191 PERRO~has shown
that M ~ 1 cannot be increased at all.
5"

~4

L BIHARI

and ..Q-~(u) means the inverse function of -q(u) (_Qq(u) exists certainly owing
to the monotonity of ~2(u)). Of course,, x must be in a subinterval (a, b') of
(a,b)so that the argument $2(k)+ M [ F ( t ) d t be within the domain of deftd

nition of ~2-~(u). Therefore (7) holds for a < x < b' ~ b with a certain b'.
PRoov. Denoting the right hand member of (6) by V ( x ) = V, we have
Y < V for o)(Y)<= (o(V), being (o(U) non-decreasing. This may be written
as follows
Mc,~(Y)F(x) <_ MF(x)
o~(v)
By making Use of the notation (8)

V'

or

-< MF(x).

~o(v) =

d~(v) < MF(x)

dx
and integrating between a and x

~2(V(x))--~2(V(a)) <=M F(t) at.,

or, since V(a) = k,
a~

~o..(V(x)) <=.~(k) + Mj F(t) at

whence, being ~ ~(u) increasing also,

V(x).~ 0-'(0

(k) q- M.['F(t)at);
c~

finally, from (6), we o b t a i n

Y(x)~-l(~(k)+Mj'f(t)dt)

(a<--_x<b'~b).

The boundary furn'ished by (7)!s independent Of Uo because, taking Oo in (8)

instead of Uo, we get
9(.

$2(u)=

~to

:dr ~ Q ( u ) : - - 6 where a~-, o ) ~

and ~-'(u)

(u+6).

Thus
3 - ' ( ~ (k) + A) = ~-~ ( ~ (k) -7 6 + A) -- S2-'(~ (k) + A).
It is easy to see that,the theorem holds als O if we suppose only~tha~ g(x)
is continuous and F ( x ) i s non-negative and. continuous. For the. function
~" Y @ I Y i

> Y (6) and conseq/mntly (7) are valid too, therefore (7) holds
_

A OENERALIZATION

OF A LEMMA OF BELLMAN

85,

for I7. Putting F ( x ) - F ( x ) + ~, the theorem is true for F(x) a n d by the pas,
sage of limit ~--, + 0 we can see its validity for F(x) too.
4. T h e u n i q u e n e s s t h e o r e m of O s g o o d 7 is the following:

If in a domain G(x, y) the function f(x, y)satisfies the condition

(9)
If(x, y2)--f(x, Y,)i <~ e2({y~,--y~!)
where o)(u) is continuous for u >=O, ~,)(u) > 0 for u > 0 and o ) ( 0 ) ~ 0 , fur-

ther if

-d(t(t) is divergent for u > O, then the equation y" = f ( x , y) has at

o
most one solution ff(x) in G with 9~(~)= ~ where (~, 7~) is a point of G.
In the proof we shall restrict ourselves to the case where re(u) is nondecreasing.
PROOF.s Suppose there exist two different solutions 9(x),~p(x) with
r f ( ~ ) = ~ f , ( ~ ) ~ . That is to say, there exist points ~-(e.g. ~ - ~ ) w h e r e
~ ( ~ ) : : ] : ~ ( ~ ) . Let the lower bound of these ~ be ~0.. We have then ~0 >'~=.,,
and ~(~o)~P(_~o), but q~(x)dy,,p(x ) for .~o< x <= -~0q-7 with a certain number 7 > 0 .
By virtue of our hypotheses

~' ( x ) - ~" (x) ~- f(x, ~ (x) ) - f ( x , ~q,(x) ) <=~,,(I 9 (x) - ,~ (x) {)

whence
~:

aT,

lw(x)--~(x)l <= }9~ , ( [ r

or

A N f ,,,(3)at

V(x)= v>o..

Hence

co(A) < 1

V'

or

(o(v) =

<1

,~)(v) = "

Introducing the notation

at __S~(u)
,o(t)

(u,~ > 0, u ->_ 0)

tt o

we h a v e
dx

--

whence by integration

(lo>

Sa ( V(x) ) -~ ~ ( v(~o + 0)) + x - - (5~+ O)

d>0

and ~0+6 < x.

W.F. OsoooD, Beweis der Existenz einer L6sung der Differentia!gleichungy'-f(x, y)

ohne Hinzuname tier Cauchy--Lipschitz Bedingung, Monatshefle f. Math. u. Phys., 9 (1898),
pp. 331--345.
s We make use of the procedure applied to prove the generalized BelJman lemma.

86

.L g~HA~

If 6 ~ + O, then

v(~ + 6) --, 0; .e(v(~o+o))-~.

l"

dt

040-

(u0>0),

but O(V(x)) is a finite number for x > ~O, i. e. (10) leads to a contradiction. We obtain the same contradiction for x < ~. Therefore q ~ ( x ) ~ ( x ) in
some neighbourhood o f _~, consequently, also in their whole existence intervals too. Q. e. d.
*g,

If/--d@~, is convergent for u > 0 , then we get from ( 1 0 ) f o r d ~ + O

]

sa(v) < s~(O)+ x--~o.

~-

Assuming. e. g. uo = 0 ,

it is
S?(O) = 0 ,

.~(V(x)) <=x - - ~

whence
~=

[~#(x)--,~,(x)l =< V ~ ~~(x--=~o)

(x _> ~,,).

Therefore, if uniqueness does not hold, the difference of two sok~tions is

subdue to this estimation.

5. If rp(x) and ~o(x) are the solutions of the equations

y'~-f(x,y)+~l and y ' = f ( x , y ) + e ~
(~i~0,~,2>=0)
satisfying the initial conditions ~(~) --= ~h, gJ(~.2)-~ ~22and existing in a <=x ~ b,
furthe r if
lf(x, Y~3" f(x, Y,)] ~- ~)(iY2--Y~I) and if(x, y)l ~ A in G,
then

PRone? We have
,1~

,(x)=,~+.i'f(t, ~(t))dt+,,(x--~d, ~(x)

3:

~i~+ tf( t,'p(t))dt+e~(x

~),

whence

+ jIf(t,,

rp(t))!dt + Cf]f(t' ~(t))--f(t, ,!,(t))ldt

(x -> ~);

~ Here 0(u) is subjected to the same conditions as in 3 and ~?(u) is also the same
~uncfion as in 3.

A OENERALIZATION OF A LE~'~MA OF BELLNAN

~;

further
a:

'i~(x)"

~ ( x ) l < [,,~,

:-~,~IA f ~o(!qD(t)~gJ(t)[)dt

~'
(x -> ~)

or

A <~ tc-~ j'a)(d)dt.

Hence, applying the generalized Bellman ]emma with F ( t ) ~ l, M ~ I ,
,conclude
~- ~_ ~' ( ~ (k) + x - - ~ )
(x ~ ~).~~

i"

dt

This formula holds also if .. on(t)

0

is divergent ( u > O ) .

( ~ , r/o)--.(~, "~) and.~.~,,--~O, then k---.O, s (k) ---. - - ~

In t h i s case,

we

if

and

-' ( ~ (k) + x - ~) -~ s~ -~( - - ~ ) = o,

i. e. A--~O. Therefore, if the the function f(x, y) satisfies the ,,Osgood Condition" in all points of G, the solution of the equation y" f(x,y) is a c o n tinuous function of the initial values ~, ri and of the parameter 2.

f{u)=o
Fig. 1
From ibis fact one can deduce - - as known - - the following general
dependence theorem too:

Let f(x, y) satisfy (9) #l G(x,y) and let f~(x, O, go, ~0) be the ,(uniquO
solution of y' ~ f ( x , y ) w i t h cp(~o,O, ~o, ~o)-~- ~lo, existing in a ~ x ~ b, :and
r
2, ~, ~) the unique solution of y ' ~ f ( x , y ) - + - ~ with ~(~, ~,~, ~ ) = ~ and
a < e < # < b, then q~(x, e, ~, ~) exists in ~ <=x _<=~ for sufficiently small
10 A similar formula holds for x ~<

88

L B|HARI

, > O and i-~--'~ol + [~r

further 5c(x,~; ~, .~) tends to 5c(x, O, ~o, r~o) uni"
formly in ~ <-- x ~ fl if ~, ~ --} ~.o, ~?o and s --+ O.
T h e same theorem may be proved1' if the right hand member of the
differential equation depends on an arbitrary parameter ,u, i.e. we have the
equation y" = f ( X , y, F~) and we have f(x, y, ~) -+f(x, y, tr uniformly in G if
~-~P0. If, especially, we take o ) ( u ) = M . u ~ where M > 0 , 0 < a < 1, then
d(t)
J ~ ) ( t ) (u > 0) i s convergent, therefore we cannot conclude that the Lipschitz
o
condition with the exponent a
]f(x,y~)--f(x,y~)] ~Miy,2--y~i ~'

(M>0,

0<a<

1)

implies uniqueness (moreover, one can find easily examples where this condition is satisfied and there are more solutions).
By the above results we obtain as application the following bound
estimation :
IZ r
l ~ , ( x ) - ~ ( x ) i _<- (k~-~ ~ M(1--~)lx--.,,~l)

1
1-ct

(a -~ x <= b)

where
Assuming identical initial condi}ions and ~.~=~._,=0, we have
1

[~(x)--~(x)i

=< ( M 0 - - ~ 0 1 x - - . ~ l )

~-~

(0 4 ~ < I, a =< x =< 0).

E. 'g. the so-called maximal and minimal integrals '~ have a difference subdue
to this bound estimation. If is.easy to see also that the limit of the right
hand member is 0 for a~-1=--0.
6, Consider now the condition
[x--~[ if(x,y~)--f(x,y~)i <= ,o(iy.,_--Y,l)

(0<a<

1)

in place of

If(x, ya)--f(x, Yl)i ~ (o(iY~--Yl I)"

We assume that this condition is valid for a certain neighbourhood of alI
points _~,~ of G. The function t o ( u ) i s t h e s a m e as in the theorem of OS~OOD
but we assume further that ~o'(0) exists. In this case (one equation and identical initial conditions)
,~im re(z/) = lira D+m(Iq~(t)--g'(t)i) = - t- o ~ ' ( o ) [ + i~'(~)--~'(_~)l].o = o
~+~+o ( t - . ~ ) ~
,+~+o
~(t--_~)~: '
~a K ~ E , 10c. cit., p. 87.
a~ KA~aSE, IOC. cit., p. 78.

A OENERALIZATION (~F A LEMMA OF B E L L M A N

89

(D_~ means the right hand derivative), that is, the function (t(~(A)
~)" ~ may be
completed to a continuous function for t = ~ too.
By a similar w a y a s above we get (two equations and arbitrary initial:
conditions)
;c

(t--~-2) ~ dt

for sufficiently small

x--~.2 ~ 0

( k = l~--_~,iA + i,~--~i +(,~ +~) (b a)).

The hypotheses of the generalized Bellman 1emma are not satisfied
here ( e . g .

1 ) ~ is not continuous in ~ =< t c< x ) .a However,

n
= one
(t_~

easily obtain - - by a little modification of the procedure (as i n the case of

the NAGUMO theorem) - - the following inequality"
J <= .Q~ ~ ( k ) -t

1--e

'

i. e. also condition (11) assures the uniqueness of the solution under identical initial conditions and e~--~a=-0. If these are not identical or ~'~4=~;,
~t

or e l s e j~-co(t)dt is convergent for u > 0, then we obtain a bound for A in

0

the neighbourhood o f ~ in which ( l l ) is valid.

The uniqueness theerem discussed here is r~ot a special case Of PE~RON'S
theoremJ 3

7. Uniqueness theorem of Perron.

Let the function f(x, y) - - defined in the domain ix-- ~I < a, iY - - rl I < b
(domain G(x, y)) - - satisfy the condition
(12)

tf(x,y.z)--f(x, yO[ <~_e)(Ix--~l, ly~--y:l)

f o r points (x, yz) and (x, y2) of G, and let ~o(x, u) be a continuous function
for 0 ~ x < a, u >=0. If ~(x) and ~(x) are integrals of the equation

(13)

y' - f(x, y)

with q~(2)= ~(~) = '2 and belonging to G(x, y), then

!~(x)-~(x)l < Z(x),

~ O. PERRON,Uber Ein- und Mehrdeutigkeit des Integrals eines systems Yon Dirge,
renfialgleichungen, Math. Ann., 95 (1926), pp. 98--101.

90

L maA~

w'here Z(x) means the maximal integral of the equation

04)
u ' = ,.(x--~, u)
with Z(~):-0. --. Therefore, if o(x, 0 ) - - : 0 and u ~ O is the unique solutio~
of (14) with u(~)=O, then ~(x):~o(x) in. G(x,y).
E. BONPIANI TM has found this theorem before PERRON but has made use
of the restriction that w(x, u) is non-decreasing function of u. We shah make
use of the same resirictioa but the proof will be very simple.

PROOF. Being

,/ (x)-~.y (x) =f(x, ,p(x) ) - f ( x , ~/,(x) ) <=~o((x-~!, ~(x)-'~(x)l)

(x > ~)

we have

i~(x)

~p(x)l_-.<[o)(t-~, I~p(t)-~4t)l)~t

(x ~ ~)

Or

05)

~e)(t--~,~)dt

where

J~2(x)=!q)(x)--'~(x)!.

Denoting the right member of (15) by V(x)= V, we have d--< V. But

V'(x) = ~, (x--~, ~(x))

and, on account of the monotonity of o(X, u) in u,
v' (x) ~ ~ ( x - - ~, v(x) )
i. e. V(x) is a lower function of the equation

06)

.' =o,(x--~,.)+~

(x >=~),

(~>o)

for x ~ ~ and V(~)--0, and therefore ~:'

V(x) <=z~(x)

(x ~ ~_)

where z~(x) means the minimal integral of (16) for x ~ ~ with Z~(~)~O and,
as known ~e - - z~(x) tends uniformly to Z(x) if e--~,+O.
Consequently

o ~ A(x) ~ V(x) ~ Z(x).

Similar.proof holds for x ~< ~.
1~ E. Bor~maNb Un teorenia di confronto ed un teorema di. unicita per l'equazione
differenziale y' = f ( x , y), Rendiconti dell'Accad, dei Lincei, Classe di Scienze Fisiche, (6),
1 (1925), pp. 298--302.
~50. PE~oN, Ein neuer Existenzbeweis ffir die ~ntegrafe der Differentialgleichung
y" ---f(x, y), Math. Ann., 76 (1915), pp. 471--484, especially pp. 473 and 4'79.
16 KAr~tKE,lOC. cit., p. 83.

A C~ENERAL~ZAT[ON O F A LEN[MA OF ~ELLN[AN

91'

It is easy to see t.hat the theorem of OSGOOD is a specia~ case of, this
theorem. - - Taking once again the case of paragraph: 5 we have

A<=k@:!e)(t--~,A)dt~
provided that (12) holds with ~ = ~

for

x~,

and ~ ( x ) a n d ~p(x) are in G. Then

O <=A(x) ~ 2(x) i n

where [Z(x) means the maximal integral of (14) with 2 ( ~ ) ~ k o

These results may be extended to systems of differential equations too.
(Received 1 December 1955)

8. Finally consider a theorem .due to TAMARKIN~e and corrected b y

LAVRENTIEV :18

Let f(x, y) be continuous for Ixl < a, IY[ < b (domain D) with a > O, b > 0
satisfying the condition

(17)

Jr(x, y ) - f ( x ,

)l >_ (ly- o(x) l)

where ~ (x) is an integral curve (passing through the origin and defined for
lxi < a) of the differential equation

08)

y'--f(x, y).

and oJ(u) is an increasing continuous function for u >=0, farther e~(O)-~-0

and the integral

dt

is convergent for u > O:

Then the equation (18) has at least two (and thus an infinite numbe r of)
integral curves passing through the origin. Moreover - - a s LAVRENTmV~8
notes - - the same is valid concerning all the points of the curve cp(x) (in D).
We give a simple proof of the above theorem, further we give a lower
b o u n d estimation for the difference of t h e maximal integral G(x) and the
minimal integral g(x) of (18) with the initial conditions G ( 0 ) = g ( 0 ) - - 0 .
On account of (17) the function h ( x , y ) = f ( x , y ) - - f ( x , q ) ( x ) ) can not
vanish in the connected domain tx i < a,y > 5o(x) arid therefore has a constant
sign. We distinguish case a) (h(x, y) > 0) and case b) (h(x, y) < 0).
Let a fixed integral of the equation

(l 9)

y' = f(x, y) + ,

> o)

17 J. TANARKINE, Stir te th6or~me d'unicit6 des solutions des 6quations diff6rentielles

ordinaires, Malh. Zeitschrift, 16 (1923), pp,: 207--212.
~8 M. LAV~EnTmV,Sur une 6quation diff6rentielle du premier erdre,,Matl~. Zeitschrtfr
23 (i925), p p : 197--198.

92

~. ~HARI

passing through the origin be denoted by %(x). As k n o w n y one can determine to all a~ < a a number ~ such that ~p~(x) exists for tIx I <.a~ and satisfies
Igadx) l < b, and we have ~

%(x) > G(x) >=~(x)

(0 < x < a,),

V,~(x) < g(x) <= ~(x)

(-<

< x < o).

We obtain immediately
a:

g,, (x)-- ,p (x) ~- f [f(t, % (t))--f(t, q;(t.)) ] d t -k ~x.

(20)

Take now the case a) and 0 < x < al. Then

%(x)--~f(x) >

f [f( t, ,p,(t))---f(t,

q~(t))l d t >= f ,o(,p~(t)--9(t))dt

(O<x <a~)

whence applying lemma of paragraph 3

i

V,~(x)--~(x) > s~'~(x) ( o < x < <)

where . Q ( u ) - ( d t

-3o-~(t)

consequently (since ~p,(x)--> G(x) for ~--*%0 and 0 =< x < a i 2~

0 (x)--,p (x) >= 5~-~(x)

(o <- x < <)

for all 0 < a~< a and thus for 0 ~ x < a. We have a fortiori
~ ( x ) - g ( x ) >__~-~ (x)
(o <= x < a).
Considering the case b) and again 0 < x < ai, equation (20) gives
0 < ~ o ( x ) - - ~ ( x ) < ~x < ~< < ~'a

(0 < x < <)

whence

G (x)--~# (x)=_ 0

(0 <= x < a).

Regard now both cases a) and b) for - - a < x < 0. By the linear transformation x = - - ~ equation (18) will have the form
08, )

dy
d~. - -

f(--~-' y(~) )~- F(~, y),

and here the function F(~,y) satisfies the condition (17) and H(~.,y)=
F(~, y)--F(~, 9o(--~)) has an opposite sign as h(x, y) ~-f(x, y)--f(x,,q)(x))~
Therefore we have in case a)
G ( x ) ~ ?(x)
( - - a < x ~ 0)
and in case b)
~(x)-,~(x) =
> s2 ~(--x)
(--a < x =
< o).
13 KnM~E, iOC. cir., p. 82, Satz 1.
~o KnMKE, 1OC. cir., p. 83.

A GENEPALIZATION OF A LEMMA OF BELLMAN

93

Instead of/he domain ixi < a,y > of(x) (domain D~) we Could have considered
the domain lxl < a, y < 9(x) (domain D~) wi;thout any change in the above
reasoning.
We have four cases corresponding to the signs of h(x, y)in D~ and
D~, resp.:

1. G--q) >=62-~(x) (0 ~ x < a), q)--g>=g2-~(--x)(--a < x < = O ) ( + + ) ,

2. G . ~ > = s e - ~ ( x )
.
, ~o--g~o
,,
(+--),
3. G - - ~ e ~ o
,,
, ~--g ~ sa-~(--x)
,,
(-- + ) ,
4. O - - q ~ 0
,,
, ~--g~0
,,
(-).
E. g. in the fourth case a n d for 9 ( x ) ~

G(x)~g(x)

iI g(O
(o
< O,
G(x)(--a<x~O)uniqueness

follows.

(Added in proof 24 May 1956)

OBOBIJ~EHHE O~IHOH JIEMMbI BEJIJIMAHA H HPHMEHEHHE Is BOHPOCAM

E~!HHCTBEHHOCTH PELLIEH!4t~I ~I,H~bq~EPEHIIHMIbHblX YPABHEHI4I)
H. B n x a p .

(By~anemw)

(P e a ~o ~ e)
B pa6oTe o6o6taaeTc~ c~e~ymma~ ne~,~a:
HyCTL Y(x) ~ F(x) rmao~axem, m,m senpeps~m,m 0pya~ann na owpea~e x o <~ x < X~,
k ~ 0, M ~ 0, x0 < a ~ x < Xo. Tor~xa na nepaueacwBa

7(x) < ~ + M [_ F(O r(t) at

c~

cae~ye7 ~epaBeacvso
a7

~rf F(t)a,

Y(x) -< ke ~

Ecau k ~ 0, TO, KouenH% Y(a) ~ O.

DTa ae~,~a arpaeT 6o~bmy~o po~s npH sccne~xosan~H yCTO~BOCTH perueHnfi anqbq)epena~aab~b~x ypaBuenn~i n 3aBHCHNOCTI4HX OT Haqa.lbnBIx yCaOBnfi n napaMeTpos, ecnn
BhlIIO,rlHgleTCfl yCJIOBHe fl~IlHgl~llIa C noKaaaTeaeM a-~-1.
3Ta aem~a MO~eT 6s~rb o6o6~ena cae~yFoatnM o6paa9a:
FlycxL Y(x), F(x), k, M, xo, X~, a Ta~OBL~ >xe, ~aK n Bslme, dpya~Un~ o(u) soapacTamma~, neoxpnaaTenbna n nenpep~,~Baa np~ U ~ 0. Tor~ta na nepaBencTsa

a,

94

I. B]HARt: A (3ENERAL,IZATIONOF A LE/flMAOF BELL/~'~AN

cneayeT ~epaBencT~o

3~eCb ~ Q ( / 0 =

f --g(~-)
dt (U0, t / > 0) n ,Q-l(/1) eCTI~ I~yH~:tl,n~, o6pav~a~ qbynKnnn .Q(u). Ecan
%

k = 0, To g ( a ) ~

0.

B pa60Te aortaaMBaroTc~,

c ne~oTopMra~ o r p a n n n e n n a ~ n ,

TeopeNb~ o e~,nncxgennoN

O c r y ~ a n I I e p p o n a, ~ ~taeTca o~en~a OT~C,aonennn pemenn~ 99(x) ~ ~(x), y;~oBaeTnop~mmnx O~nnaK0nbiM nan paaannnb~M rpannqn~lM y c a o B n ~ , ypannenn~ y' = f ( x , y ) + e1
n y' : f ( x , z) + ,o, ecru Bbmona~eTc~ ycnosne ~qnntuuua c noKaaaTaae~ 0 < ~ < I naa
YCa0~ne OcryAa
(1)
(2)

(f(x, y~)--f(x, Y0) ~ ~ ( IY~--Yx [)Eca., uanpnmep, ~bmoaneno yc.,~oBne

[f(x, yo)--f(x,y~)l~Mlyo--y~{ ~

(0<a<

LM>0),

HflqaJIbHble yc:~oBn~ cogna~amT ~ q ~- r~ ~ O, TO

I

I~o(x)-- e(x)l ~ (M(I

~)tx--.~l)~: ~

(a~x~b),

r~e (a, b) npoMea<y~o~, r~e cy~ecT~ymT ~(X) ~ g,(x), ~ ~(_~)~pff). Ecnn ~--~ 1 - - 0 , ~o
~er~o no~aaa~b, qxo 0Ta rpann~a CTpe/~lIITCfl I~ nyam.
Ec~n ycno~ne (2) aa~ennT~ ycaoBne~ (1), TO noayqaevcn cne~ymmee nepa~encTBo:

l~(x)'~(x)t.~ ~-!(Ix~l)
(uo=0).
Pa6oTa c0~ep>ggT eme ;~o~aaaTenbcTno y e o p e ~ o e~nncTBeHttocYrl H a r y ~ o ,
,~cnon~aymu~ee ncxo/lnylo : ~ e ~ y ~, na~o~ea, ;laeT npocToe ~o~a~aTeJ~bCTBO o/lnOl~ n3 TeopeN
T a ~ a p ~ n n a, ~: OT~OC~ule~c~~ ~H0roaBaqnocTn petuenn~ /JjJOpqbepemmaa~nux ypaBHenn~
~epsoro n o p ~ a .