ARKIV FOR MATEMATIK Band 6 nr 22
1.65 12 1 Communicated 13 October 1965 b y L. GXI~DING and L. CARLESO~
M i n i m i z a t i o n p r o b l e m s f o r t h e f u n c t i o n a l
supxF(x,f (x),f'(x)). (II)
By Gu~NAR AnoNsso~
Introduction
The present paper is a continuation of the paper [1]. This means t h a t we shall t r y to minimize the functional
/t(/) =sup~/~(z,/(x),/'(x))
over the class :~ of all absolutely continuous functions/(x) which satisfy the boundary conditions/@1)
=Yl
and/(Xz) =Y2-I n Chapter 1 we consider the question of the existence of a minimizing function, which was left open in [1].
I n Chapter 2 some of the previous results are generalized to a wider class of func- tions
F(x,y,z):
the previous condition~F
~z
> 0 for z > 0
[!o orz> x
= 0 for z = 0 is changed into ~-z 0 for z = ~o(x, y),
< 0 for z < 0 , 0 for
z<o:,(x,y).
An existence proof for absolutely minimizing functions is given in the general case.
I n Chapter 3 we examine further the properties of a.s. minimals. We also consider uniqueness questions for a.s. minimals and minimizing functions. For reasons of simplicity, this chapter is restricted to the case
o(x,y)=-0.
I n Chapter 4, finally, we give some examples in order to illustrate the previous exposition.
We shall now introduce notations for some conditions on
F(x,y, z).
They are as- sumed to hold for x E J and for all real y and z. Here J is an interval which will be stated explicitly in most cases. If it is not explicitly given, then the choice of J will be clear from the context.1. F(x,y,z)
is continuous and~F/~z
exists.~F(x, y, z)
is2A. ~z
> 0 if
= 0 if
< 0 if z > 0 , Z ~ 0 , z < 0 .
409
c. AROrCSSO~,
Minimization problems. I I
2 B . There is a continuous f u n c t i o n
co(x, y)
such t h a tOF(x, y, z)
Oz
f!
O ifz>o~(x,y),
is 0 if z = o ~ ( x , y ) , 0 if
z<co(x,y).
3. liml~l_,~r
F(x,
y, z ) = + oo if x a n d y are fixed.T h e conditions 1 a n d 3 will be assumed to hold t h r o u g h o u t this paper, t o g e t h e r with 2 A or 2 B . 1 2 A is a s s u m e d to hold in C h a p t e r 1 a n d Chapter 3; in Chapter 2 we shall use t h e m o r e general condition 2 B. Xt is easy t o see (using t h e H e i n e - B o r e l covering theorem) t h a t
lim ( inf
F(x, y, z))= +
I z l ~ ~x~<~
l y K K
for every c o m p a c t interval [:r c J a n d for every K > 0.
Chapter 1. The question of the existence of a minimizing function T h e assumptions on
F(x,y,z)
in this c h a p t e r are m a i n l y t h e same as in [1]. W e shall assume t h a t t h e following conditions are satisfied t h r o u g h o u t t h e c h a p t e r : 1, 2 A a n d 3. T h e interval J will be chosen asXl<~X<~X ~.
1. W e shall n o w give an example which shows t h a t t h e existence of a minimizing function does n o t follow f r o m these properties of
F(x, y, z).
A counterexample.
ChooseF(x, y, z) =z2/(y 2 +
1) 2 - cosx - x ( 2 ~ - x ) c o s x - ~ ( y ) (2~ - ~ ( y ) ) ,where~(y)=arctgy+~/2;
x l = 0 a n d
I x2=2~'
Yl = 0 [ Y2 = 0.
L e t /E :~. W e h a v e
H(I)
>I F ( ~ , / ( ~ ) , 0) = 1 + ~ 2 - ~ ( / ( ~ ) ) [ 2 ~ - ~ ( / ( ~ ) ) ] > 1,since 0 < ~ ( / ( ~ ) ) < z . H e n c e H ( / ) > I for all /E:~. Suppose n o w t h a t 0 < ~ < g / 2 a n d consider t h e function
1 T h e r e is one exception: E x a m p l e 5 in C h a p t e r 4.
ARKIV FOR MATEMATIK. Bd 6 nr 22 tg x
tg (2 :~ - x )
I t is easy to verify t h a t
for 0 ~ < x - ~ - ~, for 2 - - (~ ~ < x ~ + (5, for 3 - ~ + ~ ~<x~<27~.
H(g~) = F ( z , g~(ze), O) = 1 + ~2.
This proves t h a t infr~ ~ H ( / ) = 1 and t h a t there is no minimizing function. Compare the r e m a r k to Theorem 1.1.
2. I t follows from this example that, in order to ensure the existence of a mini- mizing function, we m u s t impose suitable extra conditions on F(x,y,z). We shall describe a few such conditions (they are separately sufficient), but we are not aiming at a complete discussion of the question.
I n each existence proof, we shall need
Lemma 1.1. Let {/~(x)}~ be a sequence o//unctions in :~ such that supl.<~<~ H(/~) <
and suppose that ]~(x) -+/(x) uni/ormly /or x I <<. x <~ x 2 . Then / 6 :~ and H(/) <~ lim~_~ ~ H(/r).l Proo/. Since ]/r(x)] ~<K1 a n d H(/~)<~K 2 for all v, it follows from the properties of F t h a t ]/:(x)] ~<K a. Hence all/~(x) a n d / ( x ) satisfy a uniform Lipschitz condition.
Clearly /6:~, and the rest of the proof follows easily. (Use the fact t h a t if /'(xo) exists, then we have
~(xo, /(xo), z) >~ F(Xo, /(xo), /'(xo))
for all z >~/'(xo) or for all z <<./'(xo), together with a continuity argument.)
L e t us now return to the minimization problem and introduce the notation M 0 =inf~E~ H(/). Suppose t h a t {/~}F is a minimizing sequence, i.e. lim~_~ ~ H ( / , ) = M 0.
I f the f u n c t i o n s / , ( x ) are uni/ormly bounded, then it follows (as above) t h a t t h e y are equicontinuous, and then there is a uniformly convergent subsequence. L e t the limit function b e / ( x ) . I t follows from the l e m m a t h a t H ( / ) = M 0.
Hence, in order to prove t h a t there is a minimizing function, it is sufficient to prove t h a t there is a minimizing sequence of uniformly bounded functions.
I n m a n y cases, it can be seen from the function F(x, y, 0) t h a t there is a minimizing function. This possibility is illustrated b y the following theorem.
Theorem 1.1. Suppose that there is a number Y1 >~max (Yl, Y2) such that m a x F(x, Yl, O) = m a x (F(~, Yl, 0), F(/~, Y1, 0))
/or all :r and fi satis/ying xl <~<fl<~x ~. Suppose also that there is a Y2 ~< rain (Yl, Y2) with the same property. Then there is a minimizing/unction.
number 1 In other words, the functional H(]) is lower semi-continuous.
411
G. AROrr
Minimization problems. I I
Proo/.
Y1 if / ( x ) > Y1, L e t / ( x ) e ~ and form
g(x)
= /(x) ifY2 <~/(x) <~ Y1,
[Y2 if
/(x)<Y2.
Clearly
H(g)<.H(/)
and the rest of the proof is obvious.I t is not difficult to find different conditions on
F(x,
y, 0) which lead to the same result. (For instance,F(x,
y, 0) increasing in y for y >/]71 and all x, decreasing in y for Y ~< Ye and all x, are also sufficient conditions for the existence of a minimizing function.)3. Suppose that there is a number
M > M o
(M0=infrE~H(/))
such t h a tH ( / ) < M
implies t h a t maxI/(x) l<~g,
where K depends on M but not on/(x). Then, clearly, every minimizing sequence is bounded and there is a minimizing function.I t is evident t h a t every number M > M 0 has this property if limlzl_~
F(x, y, z) = +
uniformly for all x and y. But there m a y exist numbers M with this property even if the limit is not uniform. We can describe the result of this section thus: if the roots zl and z2 of the equationF(x, y, z ) = M
do not increase too fast in absolute value when [Yl -+c~, then the attainable setE(M)
is bounded, and then there is a mini- mizing function. For instance, it will follow from Theorem 1.2' t h a t there is a minimizing function if liml~l_~rF(x,
y, y) = + ~ uniformly in x.Let us denote b y
El(M)
the set of points (x, y) such thata) xl<~x<~x~,
b) there is an absolutely continuous function
g(t),
joining the points @1, Yl) a n d(x,y),
such t h a tsup
F(t, g(t), g'(t)) <~ M.
t
The set
Ee(M )
is defined analogously, but with(x2,y~)
instead of(xl, Yl).
We areinterested in the set
E(M) = EI(M ) N E2(M ).
I t is clear that
H(/) <~ M
implies t h a t maxI/(x) I <~ K
if and only ifE(M)
is bounded, and in t h a t case the constant K is determined byE(M).
We shall first examine
E(M)
in the caseF = F(y, z).
Later, we shall t r y to carry over some of the results to the general case.Using some results from [1], we can easily prove the following theorem (concerning
dPM(t), ~pM(t)
and the integrals below, see section 2 of [1]):Theorem 1.2.
Assume that F is independent o / x and that M > Mo.
A)
E(M) is unbounded above (i.e. it contains points with arbitrarily large ordinates) i/and only i/the integrals
f ~ dt and 12= f~ dt
11
J~, O ~ ( t ) J ~ -~,(t)
are well-de/ined, convergent and
I i + I ~ < x 2 - x 1.
(1)A R K I V FOR M A T E M A T I K . B d 6 n r 2 2
B)
E(M) is unbounded below i / a n d only i/the integrals
i3=f _ - V,~(t) dt and i4=fu _ opt(t) dt are well-de/ined, convergent and
I3 + I4 < x 2 - x 1.
(2)Proo].
We restrict the proof to the first p a r t of the theorem. Assume first t h a tB(M)
is unbounded above. L e t (x0, Yo) CE(M)
and Yo > m a x(Yl,
Y2). I t follows from Theorem 2 in [1] t h a tf yV~ dt
1 ~ ~< x ~ xl a n df y . dt
~<x 2 - x 0., - ~ , ~ ( t )
I f we add these inequalities and let Y0 tend to infinity, then we have proved one p a r t of the assertion.
Next, suppose t h a t
I i + I 2 < x 2 - x 1.
I t follows from the discussion of " t h e at- tainable cone" in section 2B of [1] t h a t(x, y ) E E I ( M )
if x 1 + I l < x < x 2 andY>~Yl.
Similarly, it follows t h a t
(x, y)EE2(M )
ifx 1 <x<~x 2 - I 3
andY>~Y2.
Clearly, all points (Xl + I1, y), where y ~> m a x (Yl, Y2), belong to
E(M).
This completes the proof of Theorem 1.2, since the reasoning in the case B is analogous.A consequence of the theorem is t h a t
E(M)
is bounded if a n d only if neither (1) nor (2) is satisfied.L e t us now consider the general case
F = F ( x , y, z).
I t can be"reduced"
to the previous case simply b y neglecting the dependence on x. This m a y be considered to be a crude method, but, nevertheless, it works in m a n y cases.The functions
~gM(y)
and~fM(Y)
w e r e defined in [1]. Clearly, we can introduce here the corresponding functionsO(x,y,M)
and ~(x, y, M). Now fix y and M and suppose t h a t U ={x IF(x,
y, 0) ~<M} is non-empty.De/inition: I ~(y'
M) = max~~ v O(x, y, M),
[ fl(y,
M) = minx ~ v ~f(x, y, M).I t is easy to verify t h a t these functions are continuous and t h a t they are monotonic in M, as are (I)~(y) and ~v~(y). Finally, an argumentation, similar to the one used in the special case F =
F(y,z),
leads to the following result:Theorem 1.2'.
Assume that M > M o.
A)
I / E ( M ) is unbounded above, then the integrals dt
I1= , a(t,M) and
dt 12=
~ - f l ( t , M )are well.de/ined, convergent and I 1 + 12 <<-x2 -x l .
B)I / E ( M ) is unbounded below, then the integrals
413
c. ARO~SSOr~, Minimization problems. I I
i3=f dt
_ - f l ( t , M ) and Y~ dt
I ~ = _ a(t, M ) are well-defined, convergent and I a + I a <~ x 2 - x 1,
An application of this theorem is given in Example 1 in Chapter 4. Compare also the counterexample.
Chapter 2. A more general case. P r o o f of the existence of absolutely minimizing functions
]. I n this section we shall generalize some of the previous results to a more general class of functions F(x, y, z).
We are not going to change the condition limN_~cF(x, y, z ) = + c o , which has turned out to be very useful and which is a rather natural condition. However, the assumption t h a t F(Xo, Yo, z) has a minimum at z = 0 for all (x0, Y0), is more restrictive, and we shall assume instead t h a t F(x0, Yo, z) has a minimum at z=eo(x0,yo) , where oo(x,y) is a continuous function. Compare Example 5 in Chapter 4.
Thus, let us assume t h a t the following conditions hold throughout this chapter:
F(x, y, z) E C1; 2B and 3.
The minimization problem is meaningful, since we have H(/) >~ m a x [F(Xl, Yl, O~(Xl, Yl)), F(x2, Y2, c~
for every admissible function/(x).
We shall now carry over some of the results of the paper [1] and of the previous chapter to this more general case.
The analogues of the Theorems 5 and 6 in [1] are Theorem 5'. L e t / ( x ) be an admissible/unction such t h a t : a) /'(x) is continuous on I (x 1 <~x <~x2),
b)
F z ( x , / ( x ) , / ' ( x ) ) > ~ O on I , o r < O on I ,c) F(x,/(x),/'(x))
= M on I .T h e n / ( x ) is a m i n i m i z i n g / u n c t i o n 1 (not necessarily unique).
Theorem 6'. Suppose that/(x) E ~ satisfies:
a) /'(x) is continuous on I (x 1 ~x<~x~), b) F z ( x , / ( x ) , / ' ( x ) ) # 0 on I ,
c) F(x,/(x),/'(x))
= M on I .T h e n / ( x ) is a unique m i n i m i z i n g / u n c t i o n . 1
The proofs are analogous to those of the case ~o(x, y)=~0 and they are omitted.
Lemma 4 in [1] holds in this case without changes. The proof is simple.
1 It also follows that ](x) is an a.s. minimal on I.
ARKIV FOR MATEMATIK, B d 6 n r 22 A c o n s e q u e n c e of t h i s l e m m a is t h a t we can n e g l e c t sets of m e a s u r e zero. This m e a n s t h a t we can j o i n t w o f u n c t i o n s a t a f i n i t e n u m b e r of p o i n t s w h e r e t h e y a r e e q u a l b y t a k i n g one or t h e o t h e r on successive i n t e r v a l s . T h e n t h e s e p o i n t s will n o t affect t h e v a l u e of t h e f u n c t i o n a l for t h e r e s u l t i n g f u n c t i o n . (This s i t u a t i o n will o c c u r in t h e n e x t section.)
:Next, we m a k e a j u m p t o T h e o r e m 9, w h i c h is c a r r i e d o v e r w i t h o u t changes:
T h e o r e m 9'. I / ] ( x ) is an a.s. minimal on the interval 1 (x 1 ~ x <~x2), then:
1) /(x)EC 1 on I,
2) the differential equation
dF(~,/(x),/'(~))
Fz(X, / ( x ) , / ' ( x ) ) = 0 dxis satisfied on I in the sense that i / F ~ ( x , / ( x ) , / ' ( x ) ) =#0 on a sub-interval I 1 c I, then F ( x , /(x), /'(x)) is constant on 11.
W e shall p r o v e t h e a s s e r t i o n s 1) a n d 2) u n d e r t h e following, w e a k e r a s s u m p t i o n s
on/(x):
I) /(x) satisfies a L i p s c h i t z c o n d i t i o n on [Xl, x2],
I I ) t h e r e is a c o n s t a n t K > 0 such t h a t if [tl, t2] is a n a r b i t r a r y s u b - i n t e r v a l of I , t h e n t h e r e a r e two (not n e c e s s a r i l y different) m i n i m i z i n g f u n c t i o n s u(x) a n d v(x) be- t w e e n ( t i , / ( t i ) ) a n d (t2,/(t2) ) such t h a t [u(x)[ ~ K , Iv(x)] ~ K a n d u(x) <~/(x) <~v(x) for t 1 ~< x ~< t 2.
(These a s s u m p t i o n s a r e s a t i s f i e d if /(x) is a n a.s. m i n i m a l , since we c a n choose K = m a x z ~ ]/(x) l a n d u ( x ) = v ( x ) = / ( x ) in t h a t case.)
T h e r e a s o n w h y we shall p r o v e t h i s m o r e g e n e r a l r e s u l t is t h a t i t will be useful in t h e n e x t section.
Proo]. 1 Consider a n a r b i t r a r y p o i n t x o on t h e i n t e r v a l x 1 ~ x < x 2 a n d s u p p o s e t h a t t h e u p p e r a n d lower r i g h t - h a n d d e r i v a t i v e s of /(x), :r a n d /~, a r e d i f f e r e n t a t x 0.
(Clearly, t h e y a r e finite.) Choose a n u m b e r 7 ~:o)(Xo,/(Xo)) such t h a t g > ~ >ft. T h e i n i t i a l - v a l u e p r o b l e m F(x, g(x), g'(x))=constant, g(xo)=/(x0), g'(xo)=~, h a s a solu- t i o n ~(x) in a n e i g h b o u r h o o d of x 0. Since ~ > ~ >fi, t h e r e a r e p o i n t s x > x o, a r b i t r a r i l y close t o x0, w h e r e /(x)=~0(x). B u t since ~, ~=eo(x0,/(xo)), i t follows f r o m t h e condi- t i o n I I a n d f r o m T h e o r e m 6' t h a t / ( x ) =q~(x) on [Xo, x0+~t ] for s o m e ~ > 0 . T h e con- t r a d i c t i o n shows t h a t / ( x ) h a s a r i g h t - h a n d d e r i v a t i v e . S i m i l a r l y , i t follows t h a t / ( x ) h a s a l e f t - h a n d d e r i v a t i v e for x 1 < x ~<x 2.
O u r n e x t s t e p is to p r o v e t h a t t h e o n e - s i d e d d e r i v a t i v e s a r e equal. Consider a p o i n t x0, a n d a s s u m e t h a t o:=/'(xo)+4fi=/'(xo)-. Choose a n u m b e r y b e t w e e n a n d ft. C l e a r l y
F(x0,/(Xo), y) < m a x [F(xo,/(xo) , o:), F ( x o, ](Xo), fl)].
S e l e c t a sequence of i n t e r v a l s [Pn, q~] such t h a t p~ <xo <q~, q~-p~-->O
a n d /(q~) - / ( P ~ )
q~ - p~
1 The reasoning will be similar to the one used in [1], but not identically the same.
2s: 4 415
G. ARONSSO~,
Minimization problems. II
I t is obvious t h a t
lim
M(pn, qn; /(P,), /(q,)) < F(xo,/(xo), ~).
n ---> o v
(The n o t a t i o n M(...) was i n t r o d u c e d in [1].) L e t us assume, for example, t h a t g > y > f l a n d
F(xo,/(xo) ,
~) > F(x0,/(xo),
y). L e t the functionsUn(X)
correspond to t h e intervals [p~, q~] according to I I . I t follows t h a t lim~_~r (SUpxu'~(x))>~ ~z.
ThusConsequently
lira
H(un) ~ F(xo, /(xo), ~).
n ---> ~ r
lim
H(u~)
> liraM(pn, qn; /(Pn), /(qn)),
n .--> o r n ---> ~
which is a contradiction.
The reasoning is analogous in the other cases. This proves t h a t / ' ( x ) exists.
Suppose n o w t h a t
Fz(xo,/(xo) ,/'(xo) ) #0.
T h e n we assert t h a t / ( x )E C 2
in a neigh- b o u r h o o d U of x 0 a n d t h a tF(x, /(x), /'(x))
is c o n s t a n t in U. This is p r o v e d in t h e same w a y as T h e o r e m 8 in [1]. W e omit the details.N o w t h e only thing left to prove is t h a t / ' ( x ) is continuous a t those points where
Fz(X, /(x), /'(x))
= 0 . Assume t h e n t h a t]'(Xo)=o~(x
o,/(x0) ). Assume f u r t h e r t h a t there are n u m b e r s ~n->x0, ~ > x 0 such t h a t/'(~)>~/'(Xo)+~.
I f the functionsUn(X)cor-
respond to the intervals x 0 ~ x ~ according to I I , t h e n we h a v e sup~u'~(x) >~/'(xo) § ~.
H e n c e l i m n ~
H(Un) ~ F(Xo,/(Xo) ,/'(Xo)
+~). B u t this contradicts the obvious relation limM(xo, ~n; /(xo), /(~n)) <~ F(xo, /(Xo), /'(Xo)).
n - - > o ~
T h e other cases can be t r e a t e d similarly. This completes the proof.
W e h a v e t h u s established some i m p o r t a n t properties of a.s. minimals. B u t we h a v e n o t e x a m i n e d the properties of minimizing functions in general. T h e reason for this is evident f r o m E x a m p l e 2 in Chapter 4.
I n t h e previous c h a p t e r we t r e a t e d the question of the existence of a minimizing function for the case
co(x,y)~O.
I t is easy to verify t h a t L e m m a 1.1 holds in t h e present case. This means t h a t if there is a minimizing sequence of u n i f o r m l y b o u n d e d functions, t h e n there is a minimizing function. I n particular, this is t r u e ifE(M)
is b o u n d e d for someM > M o . 1
W e shall n o t discuss generalizations of the other criteria in Chapter 1.
2. A problem which we h a v e n o t y e t discussed, is t h a t of t h e existence of absolutely minimizing functions. W e shall give an existence proof for the general case.
Theorem 2.1.
Assume that F(x, y, z) E C 1 and satisfies the conditions 2 B and 3 / o r X I <~x<~X 2 and all y, z. (Thus J=[X1,X2]. )
Assume/urther that, /or every choice o/(~1,~1) and (~2,~2), there is a minimizing/unc- tion between these points, and, i/ there are several such/unctions, that they are uni/ormly bounded. (The bounds will depend on ~1, ~h, ~ and ~2")
1 A sufficient condition for this is that limlz[-~r
F(x,y,
z) = + c~ uniformly for all x and y.ARKIV FSR MATEMATIK. B d 6 n r 22
Then there is an absolutely m i n i m i z i n g / u n c t i o n / o r every minimization problem in the strip X I <~ x ~ X2, - ~ < y < ~ .
Proo/.
I n order to m a k e the proof more lucid, we shall divide it into several parts.l) Consider the minimization problem between
(Xl,
Yl) and (x2, Y2). L e t ~ be the class of minimizing functions, which, b y assumption, is not empty.P u t
[
u(x) ~ inf [(x) a n dlv (x) = sup/(x).
f e T n
I t is clear t h a t
u(x)
andv(x)
belong to ~ .Consequently there are always a smallest minimizing function
u(x)
and a greatest minimizing functionv(x).
2) We shall introduce two notations in order to simplify the expressions later on.
L e t
p(x)
be a function on an interval I . Let [Xl,X2] be a sub-interval of I and consider the minimization problem between (Xl,p(xl) )
and(x2, p(x2)).
L e tu(x)
be the smallest andv(x)
the greatest minimizing function, as above. We shall agree to say t h a tp(x)
has the p r o p e r t y A on I ifp(x)>~u(x)
on Ix1, xe]/orevery
sub-interval [xi, x2] of I and t h a tp(x)
has the p r o p e r t y B on I ifp(x) <~v(x)
on[xl, xeJ/or every
sub-interval Ix1, x2] of I . (Compare subharmonic and superharmonic functions.)3) Consider our minimization problem on an interval I with some b o u n d a r y values. I t is obvious t h a t the greatest minimizing function
v(x)
is not less t h a n thegreatest
minimizing functionVi(X )
between a n y two points (tl,v(ti) )
and(t2, v(t2) ).
Similarly, the smallest minimizing function
u(x)
is not greater t h a n thesmallest
minimizing functionu1(x )
between a n y two points(tl, u(tl) )
and(t2, u(t2) ).
4) We shall now construct an a.s. minimal.
From now on, we consider a /ixed minimization problem, namely between
(Xl, y~) and (x2, Y2). L e t G be the class of those minimizing functions which have the p r o p e r t y A on[xl,x2]. G
is not empty, since the greatest minimizing function belongs to G.We introduce the/unction
h(x)
= infg(x)
geG
and we assert that h(x) is an a.s. minimal.
I t is clear t h a t
h(x)
is a minimizing function.5) Now let us prove t h a t
h(x)
has the properties A and B.I) Choose an a r b i t r a r y interval it1, t2] a n d let
u(x)
be the smallest minimizing func- tion between(tl, h(tl) )
a n d(t2, h(t2) ).
L e tg(x)CG.
We assert t h a tg(x)>~u(x)
on t 1 ~<x ~ t 2. I f this were not true, then there would be an interval s 1 ~<x ~<s 2 such t h a t U(81)=g(81) , U(S2)=g(82) andu(x)>g(x)
for81<x<82.
Letul(x )
be the smallest minimizing function between (Sl,u(sl) )
and(se, u(s2) ).
I t follows from the definition of G t h a tg(x)~>Ul(X
). B u t we also haveu ( x ) ~ u l ( x ).
(See p a r t 3 of the proof.) This givesg(x)>~ul(x ) ~ u ( x )
and we have a contradiction.Thus
u(x) <~g(x),
and the inequalityu(x) <~h(x)
follows from the definition ofh(x).
This proves t h a t
h(x)
has the p r o p e r t y A.I I ) Assume t h a t
h(x)
has not the p r o p e r t y B. This means t h a t there is an interval It1, t2] such t h a t the corresponding greatest minimizing functionv(x)
does not satisfy 417c . AROr~SSOrr Minimization problems. II
t h e inequality
v(x)>~h(x).
I t follows f r o m p a r t 3 of the proof t h a t we m a y assume t h a th(x)>v(x)
for t I < X < t 2. L e t us f o r m the functionh(x) for
x<~q, p(x)=]v(x)
fortl<x<t2,
[h(x) for
x ~ t 2.I f we can prove t h a t
p(x)E G,
t h e n this will contradict the definition ofh(x),
since we h a v ep(x) < h(x)
on (tl, t2). I t is clear t h a tp(x)
is a minimizing function on [xl, x~], a n d it remains to prove t h a tp(x)
has t h e p r o p e r t y A. So let us consider a n interval Is1, s2] a n d let the corresponding smallest minimizing function beul(x ).
W e h a v eul(sl)=p(sl)<h(sO
a n dul(s2)=p(%)<~h(% ).
Sinceh(x)
has t h e p r o p e r t yA,
thisimplies t h a t
h(x)~>Ul(X
). If p ( ~ ) < u l ( ~ ) for some ~E(Sl, S2), t h e n we m u s t h a v ep(~)=v(~)<ul(~ ).
Consequently there m u s t be an interval [rl, r2] such t h a tv(rl)
= u l ( r 0 ,v(r2) -ul(r2)
a n dv(x) < ul(x )
for r 1 < x < r 2. L e tu2(x )
a n dv2(x )
corre- spond to this interval a n d these b o u n d a r y values in the usual manner. T h e n we h a v eudx)
<u2(x) <v2(x) < v(x).B u t this contradicts t h e inequality
v(x)<ul(x
). This proves t h a th(x)
has the pro- p e r t y B.6) I t is obvious t h a t
h(x)
satisfies the conditions I a n d I I which, as we h a v e seen, g u a r a n t e e t h a t t h e assertions in T h e o r e m 9' are true. Thush(x)E C 1
a n d the differential equationdE(x, h(x), h'(x))
dx 9 Fz(x, h(x), h'(x)) = 0
is satisfied in t h e sense described there.
7) W e are n o w in a position to prove t h a t
h(x)
is an a.s. minimal. Consider an a r b i t r a r y interval t 1 < x < t 2. L e t M 0 be t h e m i n i m u m value of the functional, as usual, a n d p u tM = m a x
F(x, h(x), h'(x)).
t x ~ x ~ t ~
W e w a n t to prove t h a t M = M 0.
I f
F(t 1,
h(tl), h'(t 0) = M, t h e n t h e result follows at once, sinceh(x)
has the properties A a n d B, a n d t h e same is true for t 2. So let us assume t h a tF(t 1,
h(tl),h'(tl))<M
a n d t h a t
F(t~, h(t~), h'(t2))<M.
I t follows f r o m p a r t 6 t h a t there is a n u m b e r2, tl<~<t2,
such t h a t h'(~)=r h(~)) a n d F(~, h(~), h'(~)) = M . L e t us assume t h a t M > M oThis means t h a t there is no minimizing function co(x) for which 0)(2)=h(~). (I.e.
(2, h(~)) CE(Mo). )
L e t K be the class of those minimizing functions a)(x), for which 0)(2)<h(~). (K is n o t e m p t y , see p a r t 5.) P u t k ( x ) = s u p ~ K co(x). I t is clear t h a t k ( x ) E K a n d t h a t k(~) < h(~).
P u t
/ 81 = sup {x ix < ~,
h(x)
= k(x)}a n d )
[ s2 = inf {x ] x > ~,
h(x) = k(x)}.
ARKIV FOR MATEMATIK. B d 6 n r 22 W e introduce the function
h(x) for
x < s 1, q(x)= jk(x)
fors l < x < s 2,
[h(x) for x ~ s 2.
If we can prove t h a t
q(x)E G,
t h e n this will c o n t r a d i c t the definition of h(x), sinceq(x)
<h(x) on (s 1, s2). (Compare p a r t 4.)So all we have to prove is that q(x) has the property A.
Consider t h e n a n a r b i t r a r y interval [rl, r2] a n d letu(x)
be t h e smallest minimizing f u n c t i o n between(rl, q(rl) )
a n d(r2, q(r2) ).
W e h a v eu(rl)=q(ri)<~h(rl)
a n d
u(r2)
=q(r2)<~h(r2).
Sinceh(x)
has the p r o p e r t y A, it follows easily t h a th(x) >~u(x)
for r~ ~< x ~< r e.So if we assume t h a t
u(xo)>q(x0)
for some x0, t h e n we m u s t h a v es 1 <xo<s 2
a n dq(xo) =k(x0).
Consequently there m u s t be a n interval (~h, ~2) such t h a t s 1~<~1 <~2 ~<s2, u(~h) = k(~l), u(~2) = k(~2) a n d
u(x) > k(x)
for ~1 < x <V2.If
H(k; ~ , ~2) <~H(u; ~ , ~2),
t h e n we get a contradiction to the definition ofu(x).
Suppose t h e n t h a t L e t us form the function
H(k; ~1,
~ 2 ) > H ( u ; ~1,~2)"
[
~(X) for Z ~ l ,kl(x ) = j u ( x )
for ~ 1 < x ~ 2 ,[k(x) for x ~ 2 .
Clearly, kl(X )
is a minimizing function between (tl,h(tl) )
a n d (t~,h(t2) ).
F u r t h e r , we h a v e kl(~)~<h(~ ) (for, as we m e n t i o n e d above,u(x)~h(x)
for all x, whereu(x)
is defined). H e n c e kl(x ) EK. :But this contradicts the definition of k(x). Hence we get a contradiction in a n y case.This completes the proof of T h e o r e m 2.1.
3. The m e t h o d of a p p r o x i m a t i n g a m a x i m u m b y a sequence of integral m e a n values is well known. I n our case it means t h a t we should consider t h e functional
H(/)
as the limit of the sequence of functionalsH n ( l ) = 1 ,
[F(x,/(x),f(x))]ndx] 11~,
n = 1 , 2 , 3 , . . . .W e have used this a p p r o a c h in [1] to derive t h e differential e q u a t i o n
(dF/dx). F z = O.
I t can also be used to give a different existence proof for a.s. minimals. This is v e r y n a t u r a l since a minimizing function in the calculus of variations a u t o m a t i c a l l y is minimizing on e v e r y sub-interval. However, we shall n o t c a r r y t h r o u g h this proof since it is more complicated t h a n the one already given, a n d since it requires stronger conditions on
F(x, y, z).
A n o t h e r result which can be p r o v e d with this "integral m e t h o d " is the following:
if/o(X) is t h e only a.s. minimal for
H(/)
a n d if/~(x) minimizesHn(/)
for n = 1, 2, 3, ..., t h e n limn_~ ~/~(x) =/o(X) uniformly.419
c. ARONSSON, Minimization problems. II
Chapter 3. Further examination of absolutely minimizing functions.
Uniqueness questions
For reasons of simplicity, this chapter will treat only the case
(o(x,y)=-O.
N o doubt, the results can be modified for the general case.1. The following theorem gives us further information on the structure of absolutely minimizing functions:
Theorem 3.1.
Assume that F(x, y, z) satis/ies these conditions:
F(x, y, z) is an analytic/unction o/ x, y and z in a complex neighbourhood o/the set o/ values
{
X l ~ X ~ X 2y, z real;
2 A and 3.
Assume/urther that/(x) is an absolutely minimizing/unction on the compact interval xl ~ x ~ x 2.
Then:
a)
/(x) EC 1 on
[xl, x2](already proved);
b)
the set {x I F~(x, /(x), /'(x)) # 0 } consists o / a f i n i t e number o/intervals (to be proved now);
c)
the di//erential equation
dF(x, /(x), /'(x)) Fz(x, /(x), /'(x)) = 0 dx
is satis/ied in the sense that F(x, /(x), /'(x)) is constant on each o/these intervals (al- ready proved).
Proo/.
I n order to facilitate the further references, we divide the proof into several parts.1) Assume t h a t
] ' ( x ) # 0
forp < x < q
and t h a t / ' ( p )=/'(q)=0.
Fx(p,/(p), o) < o,
Then /
[ Fx(q,/(q), O) >10.
To see this, we recall t h a t / ( x )
EC 2
on(p,q)
and t h a tF(x, /(x), /'(x))
is constant there (Theorem 8 in [1]). So we haveFx(X, /(x), /'(x)) + F~(...) /'(x) + Fz(...)/"(x) = 0
for
p < x < q .
Clearly, there m u s t be points arbitrarily close to p, where/'(x)
a n d/"(x)
have the same sign. At such a pointFz(...)/"(x)>0,
which givesFx(x, l(x), l'(x)) + ~'y(...) /'(x)
< 0 . 420ARKIV FOR MATEMATIK. B d 6 n r 2 2 Since l i m x - ~ v / ' ( x ) = O , t h i s p r o v e s t h e first i n e q u a l i t y , a n d t h e s e c o n d is p r o v e d s i m i l a r l y (there a r e p o i n t s a r b i t r a r i l y close t o q w h e r e
Fz(...)/"(x ) <0,
etc.).N o w s u p p o s e t h a t t h e r e is a n u m b e r
t>q
such t h a t / ' ( t )4=0.
T h e n i t follows f r o m t h e a b o v e r e s u l t t h a t we can definer=inf{xlx>~q,
F~(z,/(z), 0)=0},a n d i t follows t h a t
q<r<t.
I fq<r,
t h e n we o b v i o u s l y h a v e/'(x)=O
forq~x<~r
( a n d if q = r , t h e n t h i s is t r i v i a l l y true).I f t h e r e is a n u m b e r
u < p
such t h a t]'(u)4=0,
t h e n we define s = s u p {x; x < p , F ~ ( x , / ( x ) , 0) = 0 } . W e h a v e / ' ( x ) = 0 fors ~ x ~ p .
F i n a l l y , we o b s e r v e t h a t
{ Fx(x,/(x),O)~O
fors<~x~p, Fx(x,/(x),
0)~>0 forq<~x<~r.
2) W e shall give a n i n d i r e c t p r o o f of t h e t h e o r e m . So we a s s u m e t h a t t h e set {x[ x 1 < x < x2,
/'(x)
4 0 } is t h e u n i o n of a n i n f i n i t e n u m b e r of d i s j o i n t o p e n i n t e r - vals. These i n t e r v a l s m u s t h a v e a l i m i t - p o i n t a n d we m a y a s s u m e t h a t i t isx=O.
W e m a y also a s s u m e t h a t / ( 0 ) = 0 a n d t h a t t h e r e a r e i n f i n i t e l y m a n y such i n t e r v a l s i n e v e r y i n t e r v a l 0 < x < e. C l e a r l y / ' ( 0 ) = 0 a n d Fx(0, 0, 0) = 0.
T h e f u n c t i o n
Fx(x,
y, 0) will b e i m p o r t a n t for t h e r e s t of t h e proof. I t m a y be i d e n t i c a l l y zero or not. T h i s gives us t w o cases a n d we shall s t a r t w i t h t h e s i m p l e s t of t h e m .3) A s s u m e t h a t Fx(x, y, 0) ~ 0. T h i s m e a n s t h a t we h a v e
F(x, y, O) = F(O, y, O) =of(y),
w h e r e ~c(y) is a n a l y t i c in a (complex) n e i g h b o u r h o o d of y = 0 .A) ~c(y)--- c o n s t a n t , i.e.
F(x,
y, 0) --- c o n s t a n t . Consider a n i n t e r v a l(p, q)
as in 1). P u tt=89 +q).
T h e n we h a v eF(t, /(t), /'(t)) > F(t,/(t), 0) =
F(p, tip),
0),w h i c h gives a c o n t r a d i c t i o n .
B) ~c(y) ~ c o n s t a n t .
I t follows f r o m T h e o r e m 10 in [1] 1 t h a t / ( x ) is m o n o t o n i c , for i n s t a n c e n o n - d e c r e a s - ing.
B u t ~ c ' ( y ) # 0 for 0 < y < ~. H e n c e ~c(y) is s t r i c t l y m o n o t o n i c for 0 ~ y ~ . Consider a n i n t e r v a l
(p,q)
such t h a tO</(p)</(q)<~.
T h e n we m u s t h a v ecf(/(p))=cf(/(q))
w h i c h gives a c o n t r a d i c t i o n .1 I t is c l e a r t h a t t h e c o n d i t i o n o n F z i n T h e o r e m 10 n e e d o n l y b e a s s u m e d t o h o l d for z = 0.
I n f a c t , t h e c o n d i t i o n of T h e o r e m 1.1 i n t h i s p a p e r is s u f f i c i e n t if i t h o l d s f o r a l l y.
4 2 1
G. AROr~SSOr~, Minimization problems. II
4) ~qow let us assume t h a t
F~(x,
y, 0) ~ 0. ClearlyF~(x, y,
0) is a n analytic function a n d we can a p p l y Weierstrass' p r e p a r a t i o n t h e o r e m (see [2], p. 89). I n our case it says t h a tF~(x, y, O) =x~o~(x, y)W'(x, y)
in a n e i g h b o u r h o o d U of the origin. Here # is an integer ~>0;
~o(x,y)
is analytic in U a n d # 0 ; ~F(x, y) iseither =- 1 or
of t h e f o r mvt~(x ' y) =ym + A~(x)y,~-i + A2(x)ym-2 + ... + A,~(x),
where t h e functions
Ak(x)
are analytic in a (sufficiently small) n e i g h b o u r h o o d ofx=O
a n d A k ( 0 ) = 0 for all k.W e k n o w f r o m 2) t h a t
F:~(x,
y, 0) - 0 at t h e points (r,/(r)) a n d(s, ](s)).
Therefore we can exclude t h e case ~F ~= 1 and, instead, we inquire a b o u t the zeros ofuz.(x,y ) =ym + A~(x)yr~-I + ... +Am(x).
L e t us, for the present, replace x a n d y b y the complex variables ~ a n d ~]. W e shall t r y to determine ~ as a function of $ f r o m t h e relation ~F(~,~) = 0 . W e are only in- terested in the solutions of this e q u a t i o n in a n e i g h b o u r h o o d of ~ = 9 = 0 .
Consider a complex n e i g h b o u r h o o d V of $ = 0 a n d c u t it along t h e negative real axis. T h e n every root of ~ can be defined as a regular function in V, a n d the roots
~k of ~F(~, ~]) = 0 m a y be written
= ~ C i~l/n~
for k = l , 2 , . . , m . ,~,o, j , k ~ ,Concerning this expansion, see [3], p. 50; [4], pp. 98-103 a n d [5], Chapter X I I I . (The roots can be divided into cyclic systems, a n d the n u m b e r n can be chosen as t h e p r o d u c t of the n u m b e r of roots in each system.)
W e m a y assume t h a t ~l/n is real for ~ > 0, which gives the expressions
~ ~ - ~ J . k [ x / 1/n~i ) ~ k = 1, 2, . . . , m.
j = l
Clearly, we need only consider those series in which Cj.g are real for all ?'. (If there are no such series, t h e n there is n o t h i n g left to prove.) Therefore, let us suppose t h a t
~]k(x),
k = 1,2 ...M,
are real for 0 < x < 0 , a n d t h a t ~k(x) are complex (not real) for k > M a n d 0 < x < 0 . W e k n o w f r o m t h e first p a r t of t h e proof t h a t
Fx(r,/(r),
0)= Fx(8,/(s), 0)=0.H e n c e t h e points (r,/(r)) a n d
(s,/(s))
m u s t lie on the curvesy =~k(x)
if r a n d s are sufficiently close to x = 0.ARKIV FOll MATEMATIK. B d 6 n r 22 5) W e shall n o w s t u d y in some d e t a i l t h e f u n c t i o n s
~k(x) -= ~ ~].k
( x l / n ) ~ = g k ( X l / n ) 9j = l
H e r e t h e f u n c t i o n s
g~(z)
a r e a n a l y t i c for ]z[ < 8. F u r t h e r , e v e r yg~(z)
is r e a l if z is real.D i f f e r e n t i a t i o n gives
. _ _ - 1 - - 1
,(X)=gk(xl/n ) 1 "X n
forO < x < 8 n.
n
P u t t =
x ~/n,
w h i c h g i v e s~k (x) = g; (0 [ ~ = ~ ( t ) .
T h e f u n c t i o n ~ ( z ) is a n a l y t i c in 0 < I z] < 8 a n d t h e s i n g u l a r i t y a t z = 0 is e i t h e r re- m o v a b l e or a pole. W e h a v e
l i m ~ (x) = l i m F~(t),
x --.-~ + 0 t ' - - > + O
a n d t h i s l i m i t m u s t be real. I t can b e + o% - 0% f i n i t e b u t ~ 0 a n d 0. Since ]'(0) = 0 , we can e x c l u d e f r o m c o n s i d e r a t i o n t h o s e v a l u e s of/c for which we do
not
h a v e ~ ( 0 ) = 0 . T h e r e f o r e we c a n a s s u m e t h a t ~ ( z ) is a n a l y t i c for l zl < ~ .A c o n s e q u e n c e of t h i s is t h a t t h e r e is a 8 1 > 0 such t h a t ~vk(z ) 4=0 for 0 < ]z] <81, unless ~vk(z)~0. F r o m t h e r e l a t i o n ~ ( x ) =~k(xl/~), we can infer a c o r r e s p o n d i n g r e s u l t for ~ ( x ) .
N o w let us c o m p a r e t h e f u n c t i o n s ~ ( x ) =gk(t) a n d ~ l ( x ) - g l ( t ) . T h e f u n c t i o n s
gk(z)
a n d
gz(z)
a r e a n a l y t i c a n d n o t i d e n t i c a l . H e n c e t h e r e m u s t e x i s t a 82 > 0 such t h a t9k(z)#gl(z)
for 0 < l z ] <82.T h i s gives a c o r r e s p o n d i n g r e s u l t for ~k(x) a n d ~z(x).
W e h a v e f o u n d t h a t ~/k(x)=gk(x
1/'~)
a n d ~ ( x ) = ( v k ( x l / ~ ) , w h e r egk(z)
a n d ~k(z) a r e a n a l y t i c for Iz[ < 8 . L e t us w r i t exl/n=t,
as a b o v e . This gives~'(x, ~(x), o)= ~'(t ~, g~(t),
o ) = u ( t ) a n dF(x, ~k(x), ~'k(x) )= F(t ~, gk(t), qJk(t) )=v(t).
Clearly, t h e f u n c t i o n s
u(z)
a n dv(z)
are a n a l y t i c in a n e i g h b o u r h o o d of z = 0 . H e n c e t h e r e is a 8~ > 0 such t h a t e a c h of t h e f u n c t i o n sF(x,
~k(x), 0) a n dF(x, ~(x), ~'k(x))
is e i t h e r increasing, c o n s t a n t or d e c r e a s i n g on t h e whole i n t e r v a l 0 ~< x ~< 83.
W e h a v e e x c l u d e d t h o s e f u n c t i o n s
~k(x),
for w h i c h we d i d n o t h a v e ~]~(0) = 0 . A f t e r a r e n u m b e r i n g , we h a v e ~k(x), k = 1, 2 . . . P , left.I f we collect o u r results, t h e n we see t h a t t h e r e e x i s t n u m b e r s ~ > 0 a n d % such t h a t : a)
~k(X):4:~l(X)
if 0 < X ~ : r a n dlc:4:l.
This m e a n s t h a t one of t h e f u n c t i o n s ~k(x) is s m a l l e s t a n d one is g r e a t e s t on t h e whole i n t e r v a l 0 < x ~< ~.b) ~]k(x) C C 1 on 0 ~<x ~ cr a n d ~k(0) =~/~(0) = 0 . I f ~k(x) ~: 0, t h e n ~ ( x ) 4=0 for 0 < x ~< ~.
c) I f we h a v e p , q, r a n d s as in 1) a n d 0 < s < r ~ < c ~ , t h e n
](r)=~k(r )
for s o m ek<~p
a n d
](s)=~(s)
for s o m e k 1 ~<P.423
c. xRo~ssorr
Minimization problems. I I
d) E v e r y function
F(x, ~k(x),
0) is either c o n s t a n t or strictly m o n o t o n i c on 0~<x~<~. The same is true forF(x,
~k(x), ~ ( x ) ) (1<~k<~P).
6) W e are interested only in those functions ~k(x) for which one of t h e relations u n d e r c) above really occurs on t h e interval 0 < x ~< ~. I f we exclude t h e others, t h e n we get (after a renumbering) t h e functions ~k(x), k = 1, 2 . . . . , N, left. This does n o t affect t h e validity of t h e s u m m a r y a)-d).
I f N = 1 a n d ~ l ( x ) ~ 0 , t h e n we get a contradiction at once a n d there is n o t h i n g left t o prove.
I f this is n o t t h e case, t h e n t h e greatest of t h e functions ~ ( x ) is > 0 (for 0 < x ~< r162 o r t h e smallest is < 0.
L e t us choose t h e first case a n d let ~N(X) be the function in question. T h e n
~N(X) > 0
a n d ~]N(X) > 0 on 0<x~<r162I t follows from our choice of
~N(X)
t h a t there is an interval (P0, %) such t h a t / ( % )=
~N(ro).
W e have t h u s/(qo)--/(ro)=~N(ro);
/'(qo) = 0 a n d ~N(X) > 0 .H e n c e
/ ( x ) > ~ ( x )
for q0 - ~ < x < % . B u t we also h a v e / ( P o ) =/(So)<~N(So) (owing t o our choice of ~ ( x ) ) a n d/'(po)-O.
H e n c e/ ( x ) < ~ ( x )
forp o < x < p o §
Conse- q u e n t l y there m u s t be a ~, p 0 < ~ < q 0 , such t h a t / ( ~ ) = ~ ( ~ ) a n d / ' ( ~ ) > ~ N ( ~ ) > 0 .N o w assume first t h a t
F(x,
~?N(X), 0) is c o n s t a n t or decreasing. T h e n we haveF(~, ~N(~), O) >~ F(r o, ~]N(rO), O) >~ F(q o, ](qo), 0).
(See t h e first p a r t of the proof.) This gives
F(8, /(8), /'(8))
> F(q0,/(qo),0),
which is a contradiction.Assume t h e n t h a t
F(x, ~N(X),
0) is increasing. A n obvious consequence of this is t h a tF(x, ~N(X), ~N(X))
is increasing. L e t us n o w consider t h e minimization problem between t h e points x l = y l = 0 a n d x2=~, Y2=/(~)=~N(~). Since/(x)
is an absolute minimal, we h a v eM o = H ( / ) >~F(~,/(~), 1'(~)).
:But
F(x, ~TN, ~7"N)
is increasing, which givesM o <~ H(~tN ) = F(~, ~IN(~), ~'N(~)) <~ H(/) = M o.
T h u s
H(~tN ) = M o
a n d~N(X)
is also a minimizing function.Choose a function r C 1 on 0 ~<x ~<~ such t h a t ~ ( 0 ) = r a n d r 0. F o r m t h e function
g(x)-~N(x)+),~b(x),
where A > 0 is a parameter. W e h a v e.F(x, g(x), g'(x) ) = F(x, ~N(x), V]'N(X) ) § ~.(a(x)r d-b(x)r ) § R(x, 2.).
H e r e
a(x) = Fy(x, ~N(x), ~7'N(X))
a n db(x) = F~(x,
~N(x), ~v(x)) a n d [R(x, 4) 1 <~ C~ 2
where C is i n d e p e n d e n t of x a n d 2, if 0 < ~ < 1.Now, recalling the fact t h a t
F(x, ~TN, ~'N)
takes its m a x i m u m o n l y at x = ~ a n d t h a t F~(~, ~N(~), ~TN(~)) > 0 , it is easy to verify t h a tH(g)
< H ( ~ ) , if A is small enough.This gives
H(g) < M o
which is a contradiction. This completes the proof.ARKIV FOR MATEMATIK. B d 6 n r 2 2 2. This section will t r e a t b r i e f l y t h e q u e s t i o n of u n i q u e n e s s for a.s. m i n i m a l s . I t will be s h o w n b y e x a m p l e s t h a t t h e r e m a y be s e v e r a l a.s. m i n i m a l s for a g i v e n mini- m i z a t i o n p r o b l e m . A few sufficient c o n d i t i o n s for u n i q u e n e s s will also be given.
Choose F(x, y, z) = (1 + x ~ + y2)2 § z2, 2
x 1 = - 2 , yl=O, x 2 = 2 a n d y ~ = 0 .
I t follows f r o m T h e o r e m 2.1 t h a t t h e r e is a n a.s. m i n i m a l / ( x ) . Clearly, g(x) = - / ( - x ) is also a n a.s. m i n i m a l . Now, if t h e r e is a u n i q u e a.s. m i n i m a l , t h e n we g e t - ] ( - x ) ](x), i.e. /(0) = 0. C o n s e q u e n t l y M o >~ F ( 0 , 0, 0) = 2.
P u t h ( x ) = 1 2 - x for x~>0,
[2+
x for x ~ 0 . C l e a r l y H(h) = 2 / 9 § 1 < 2.T h e c o n t r a d i c t i o n shows t h a t t h e r e is no u n i q u e a.s. m i n i m a l . (This s i t u a t i o n c a n b e d e s c r i b e d thus: e v e r y m i n i m i z i n g f u n c t i o n m u s t e v a d e t h e m a x i m u m a t x = y = 0 , w h i c h , owing t o t h e s y m m e t r y , l e a d s t o n o n - u n i q u e n e s s . )
P u t F(x, y, z ) = y 2 + z 2 a n d let /o(X) be d e f i n e d as in E x a m p l e 3 in [1]. W e k n o w t h a t all f u n c t i o n s p/o(x+q)(p, q a r e c o n s t a n t s ) a r e a.s. m i n i m a l s . I t is e v i d e n t t h a t t h e r e a r e i n f i n i t e l y m a n y a.s. m i n i m a l s if Y2 = - Y l 44=0 a n d x 2 - x 1 >~r.
I t follows f r o m t h i s e x a m p l e t h a t t h e r e m a y b e s e v e r a l a.s. m i n i m a l s e v e n if 2'(x, y, z) is c o n v e x in y a n d z. I n t h e calculus of v a r i a t i o n s , t h i s c o n d i t i o n l e a d s t o u n i q u e n e s s (for all b o u n d a r y values). I t follows f r o m t h i s e x a m p l e t h a t t h e c o n d i t i o n (Yl - C) (Y2 - C) ~> 0 in T h e o r e m 3.3 c a n n o t b e o m i t t e d .
These e x a m p l e s s h o w t h a t e x t r a c o n d i t i o n s on F(x, y, z) o r on t h e b o u n d a r y v a l u e s m u s t be a d d e d to o u r u s u a l ones in o r d e r to secure uniqueness. Two w a y s to do t h i s a r e s h o w n b y t h e following t h e o r e m s :
T h e o r e m 3.2. Assume that F(x,y,z) satisfies the conditions o/ Theorem 2.1, with o)(x, y ) ~ 0 . Then, as we know ]rom that theorem, there is at least one a.s. m i n i m a l / o r
every choice o / ( x D Yl) and (x2, y~).
Assume now that
~Y(x, y, O)
0 for all x and y.
~x
Then there is a V•IQVE a.s. minimal/or every choice o/(xl, Yl) and (x2, Y2).
Proo/. W e a s s u m e t h a t Fx(x,y,O ) > 0 . W e m a y also a s s u m e t h a t Yl <Y2 (the case yl>y2 is a n a l o g o u s a n d t h e case Yl=Y2 is t r i v i a l ) . L e t /(x) b e a n a.s. m i n i m a l . I f ]'(x) > 0 for x 1 ~ x <~x2, t h e n t h e a s s e r t i o n follows f r o m t h e T h e o r e m s 6 a n d 9 in [1].
I f / ' ( x ) =0 for some x, t h e n i t follows f r o m p a r t 1 of t h e p r o o f of T h e o r e m 3.1 t h a t t h e r e is a n u m b e r ~, x 1 < ~ ~x2, such t h a t
> 0 for xl<~x<~, /'(x) is = 0 for ~ < x ~ < x ~ .
425
a. ARO~SSO~, Minimization problems. I I
I f g(x) is a different a.s. minimal, then the same reasoning holds for g(x). Let ~' correspond to g(x). Assume t h a t ~' <~. This gives g'(xl) >]'(xl) and
B u t we also have
M1: F(Xl,
Yl, /' (xl) ) < F(xl, Yl, g' (x) ) = M e.M 1 = F ( ~ , Y2, 0) > F(~', Ye, 0) = M e.
The contradiction proves the theorem.
Theorem 3.3. Assume that F(x,y,z) satis/ies these conditions:
F(x,y,z) is analytic, as in Theorem 3.1;
2 A and 3;
the condition concerning existence and boundedness o/minimizing/unctions, which was given in Theorem 2.1;
t > : i / y > C '
~F(x,y,Z) is = i/ y = C ,
@ [ < 0 i/ y < C , (C is a constant);
F(x, ~Yl + ( 1 - 2 ) y~, 4z 1 + (1-4)z2) ~<~F(x, y~, Zl) + ( 1 - 4 ) F(x, y~, z~)
/or all x, Yl, Y2, zl and z e and/or all 4 such that 0 < 4 < 1 , equality holds i/ and only i/
Yl =Y~ and z I =z2.
We consider the minimization problem between (x 1, Yl) and (x 2, Ye). We assume that the boundary values satis/y the inequality
(Yl - C) (Ye - C) >~ O.
Then there is a VNIQV]~ a.s. minimal between (x 1, Yl) and (xe, Ye).
The proof is omitted, since it is rather laborious and since the result is not used in this paper.
3. I n the previous section, we discussed the uniqueness question for a.s. minimals.
We shall now briefly consider the same question for minimizing functions. The following theorem shows t h a t those functions F(x,y,z), for which every minimization problem has a unique solution, constitute a v e r y "small" class.
Theorem 3.4. Let F(x, y, z) satis/y the/ollowing conditions/or X 1 <~x <~X2 and all y,z:
F(x, y, z) E Ce;
2 A and 3;
the condition concerning existence and boundedness o/ minimizing /unctions which was given in Theorem 2.1.
We consider the minimization problem between (x~, y)and (xe, Y2) (X1 <~xl <x2 <<-X2) 9 Then there is a VNTQV~ minimizing /unction /or every choice o/ xl, Yl, x2 and Y2 i/ and only i/
426
ARKIV FOR MATEMATIK. B d 6 n r 2 2
_F(x, y, O) =-constant.
Proof.
1) S u p p o s e t h a tF(x,
y, 0) - c o n s t a n t . Consider t h e m i n i m i z a t i o n p r o b l e m b e t w e e n (xl, Yl) a n d (x2, Y2). W e k n o w t h a t t h e r e is a n a.s. m i n i m a l / ( x ) a n d we k n o w t h a t / ( x )E C 1.
If ] ' ( x ) 4 0
for x 1 ~<x ~<x~ t h e n i t follows f r o m T h e o r e m 8 a n d T h e o r e m 6 in [1] t h a t](x)
is a u n i q u e m i n i m i z i n g f u n c t i o n .I f t h e r e is a n ~ such t h a t / ' ( r 1 6 2 = 0 , t h e n
] ' ( x ) - 0 .
I n o r d e r to see t h a t , s u p p o s e t h a t/'(fl) # 0
for s o m e fi > ~. :Put= sup {x Ix < ~ , / ' ( x ) = o}.
T h e n we g e t
F(/~, / @ , / ' @ ) > P ( ~ , 1 @ , 0) = F ( ~ , / ( r ) , 0).
B u t this c o n t r a d i c t s t h e r e l a t i o n
F ( x , / , / ' )
= c o n s t a n t w h i c h h o l d s o n e v e r y i n t e r v a l w h e r e]'(x) #0.
So we h a v e
](x)-yl(=y2).
I fg(x)
is a n a d m i s s i b l e f u n c t i o n a n d g ' ( ~ ) # 0 for s o m e~, t h e n we g e t
H(g)
>~ F ( $ , g(~), g'($)) > F(~, g(~), 0) = H ( / ) , a n d we h a v e p r o v e d one h a l f of t h e t h e o r e m .2) S u p p o s e n o w t h a t t h e r e is a u n i q u e m i n i m i z i n g f u n c t i o n for e v e r y choice of xl, x2, Yl a n d Y2.
I f F=(Xl, Yl, 0) =4=0, t h e n we choose Y2 =Yl a n d x 2 such t h a t Fz(X , Yl, 0) # 0 for
xl<~x<<.x ~.
Clearly,/ ( x ) - Y l
is a m i n i m i z i n g f u n c t i o n , b u t n o t t h e o n l y one. T h i s p r o v e s t h a tFx(x , y, O) - O.
I f
Fy(Xl,
Yx, 0) # 0 , t h e n we chooseYz=Yl
a n d x 2 such t h a tFy(x,
Yl, 0) # 0 for x 1 ~<x ~<x z. W e shall p r o v e t h a t / ( x )=Yl
is n o t t h e o n l y m i n i m i z i n g f u n c t i o n .I t is no r e s t r i c t i o n to a s s u m e t h a t x 1 =Yl = 0 a n d Fy(0, 0, 0 ) < 0. Consider t h e func- t i o n y = ax2 w h e r e ~ > 0 is a c o n s t a n t . T a y l o r s f o r m u l a gives
F ( x , ~x 2, 2~x) =
F(x, O, O) + ax~F~(x, O, O) + l(~2xaF~(x, O. ax2, O.
2ax) + 2. ax~ .2~XF~z(...) + 4~x~F~z(...)).I f we consider a s u i t a b l e n e i g h b o u r h o o d of t h e o r i g i n a n d a s s u m e t h a t 0 < ~ < 1, t h e n F ~ ( x , 0 , 0 ) < - K < 0 , a n d t h e m o d u l u s of t h e e x p r e s s i o n in b r a c k e t s is n o t g r e a t e r t h a n K 1 9 ~2. x 2, w h e r e K a n d K 1 a r e i n d e p e n d e n t of x a n d ~.
This m e a n s t h a t
F(x, ax e,
2ax) ~<F(x,
0, 0) - K . ~x 2 + K 1 ~2x2 =F(x,
0, 0) -ax2(K -
~K1).I f we choose ~ so s m a l l t h a t K - ~ K 1 > 0, t h e n we h a v e
F(x, ax2,
2~r ~<F(x,
0, 0) for 0 ~ x < ~ , for s o m e 8 > 0 .A s i m i l a r c o n s t r u c t i o n c a n be c a r r i e d t h r o u g h for @2, Y2), a n d t h e r e s t of t h e p r o o f is obvious.
Therefore, if F ( x , y, 0) is n o t c o n s t a n t , t h e n we m u s t i m p o s e c o n d i t i o n s on t h e b o u n d a r y v a l u e s in o r d e r t o secure u n i q u e n e s s . T h e following t h e o r e m i l l u s t r a t e s t h i s p o s s i b i l i t y .
427
G. ARO~SSOrr Minimization problems. I I
T h e o r e m 3.5. Assume that F(x, y, z) satisfies these conditions:
F(x, y, z) E C1;
2 A and 3;
the condition concerning existence and boundedness o/ minimizing /unctions which was given in Theorem 2.1;
either ~F(x, y, O) >~ 0 / o r all (x, y) or < 0 / o r all (x, y).
~x
Let us denote the admissible linear/unction by l(x). Finally, we assume that m i n F(x, l(x), l'(x)) > m a x F(x, y, 0),
X I ~ X ~ X ~ ( X , y ) e K
where K is the set I xl ~ x <~ x 2
[
Yl <~ Y <~ Y2 (Yl >t Y >1 Y~)"Then there is a unique minimizing ]unction/(x). Moreover, ]'(x) # 0 /or x 1 ~x<~x2.
Proo/. W e k n o w t h a t t h e r e is a n a.s. m i n i m a l / ( x ) EC 1. I f / ' ( x ) = 0 for some x, t h e n i t follows f r o m t h e T h e o r e m s 9 a n d 10 (with a t r i v i a l change) in [1] t h a t
M 0 ~ m a x F(x, y, 0).
(X,y) E K
B u t i t was p r o v e d in L e m m a 5 in [1] t h a t M0>~minxl<x<x ~ F(x, l(x), l'(x)). So t h e as- s u m p t i o n t h a t ]'(x)~0 for s o m e x l e a d s t o a c o n t r a d i c t i o n . F i n a l l y , i t follows f r o m T h e o r e m 6 in [1] t h a t ](x) is a u n i q u e m i n i m i z i n g f u n c t i o n .
Remark. I t is e a s y t o f i n d n e w r e s u l t s of t h e s a m e t y p e b y using n e w e s t i m a t e s . I t s h o u l d be m e n t i o n e d h e r e t h a t t h e i n e q u a l i t y
M 0~> inf F(x,g(x),g'(x))
x1~x<~x2
h o l d s for e v e r y a d m i s s i b l e m o n o t o n i c f u n c t i o n g(x)EC 1 ( c o m p a r e L e m m a 5 in [1]).
This gives
M 0 >~ sup [ inf F(x, g(x), g'(x))],
g(x) xl<~x~x~
w h e r e t h e s u p r e m u m is t a k e n o v e r all such f u n c t i o n s .
I t is also clear t h a t t h e c o n d i t i o n on F x in t h e t h e o r e m c a n b e r e p l a c e d b y t h ~ c o n d i t i o n in T h e o r e m 1.1 (in t h i s case a s s u m e d t o h o l d for all y).
Chapter 4. Comments and examples
I n t h i s c h a p t e r , we shall i l l u s t r a t e some of t h e t h e o r e m s b y m e a n s of e x a m p l e s . Example 1. S u p p o s e t h a t
F(x, y, z) = G(x, y) + A y 2 + By + z 2,
w h e r e G(x,y) is c o n t i n u o u s a n d b o u n d e d f r o m b e l o w a n d A , B are c o n s t a n t s .
ARKIV FOR MATEMATIK, B d 6 n r 2 2 Is there a minimizing function? W e shall a p p l y T h e o r e m 1.2'. F o r m a l calculation gives
l O(x, y , M ) = § (M - G ( x , y) - A y e - B y ) ~/2, 9 (x, y, M) = - ( M - G(x, y) - A y 2 - By) 1/2.
I f :r is defined for y
~Yl,
t h e n we see f r o m the expression for O(x,y, M) t h a t a(y,M) =O(y), when y--> + ~ . H e n c e t h e integral 11 in T h e o r e m 1.2' c a n n o t exist finite. I t follows in the same w a y t h a t none of t h e integrals Ie, /3 a n d 14 can exist finite.So we can conclude t h a t E ( M ) is b o u n d e d for all M. Consequently there is a mini- mizing f u n c t i o n for e v e r y choice of xl, x2, Yl a n d Y2-
Example 2. W e h a v e n o t studied t h e properties of minimizing functions in general, we h a v e o n l y studied the special cases of a.s. minimals a n d unique minimizing func- tions. The reason for this is clear f r o m t h e following example:
Choose F(x, y, y') = x + y,2, xi = Yl = 0, x 2 = 1 a n d Y2 = 0. This gives M o >~ F(x~, y~, O) = 1, a n d if g(x)~-0, t h e n H(g)= 1.
Thus we h a v e M 0 = 1. I t follows t h a t i f / ( x ) is an admissible function, t h e n it is a minimizing f u n c t i o n if a n d only if I/'(x) l ~ V i - x a.e.
Clearly, the fact t h a t / ( x ) is a minimizing function implies v e r y little a b o u t / ( x ) . This m o t i v a t e s the i n t r o d u c t i o n of a.s. minimals.
On the other hand, suppose t h a t we h a v e a minimizing function g(x) which belongs to C 1. T h e n the local v a r i a t i o n m e t h o d , which was used in the end of t h e proof of T h e o r e m 3.1, can be applied to g(x). This m e t h o d works also in t h e general case of variable ~o(x,y). F o r instance, it can be used to derive t h e following result: if
Fz(X, g(x), g'(x)).o
for all x, t h e n F(x, g(x), g'(x)) = M o for all x. (This is m e n t i o n e d in [6] with a s k e t c h of a proof.) This result leads us once again to the differential equation (dF/dx)" Fz = 0 for a.s. minimals.
Example 3. W e h a v e p r o v e d in Chapter 2 t h a t if F(x, y, z) satisfies certain condi- tions, t h e n there is an a.s. minimal for e v e r y choice of the b o u n d a r y values, a n d we h a v e also p r o v e d t h a t every a.s. minimal belongs to C 1.
Consequently, there is a minimizing/unction in C 1.
However, there need n o t exist a minimizing function in C 2, as can be seen f r o m the following example:
P u t F(x, y, y') =y,2_cos2x, xl =Yl =0, x 2 =n7~ a n d Y2 = 2 n . W e assert t h a t
/(x)= [cos t l dt
is the only minimizing function. Clearly,/'(x) >~0 a n d F(x, /(x), /'(x)) - 0 . If H(g) <~0, t h e n it follows t h a t g'(x)~/'(x) a.c. This implies t h a t g(x)~/(x), which proves o u r assertion.
Finally, it is easy to verify t h a t / " ( x ) has a j u m p at x =ze/2 § k.~, k = 0,1,2 ... n - 1.
4 2 9