Mathematics Department, Princeton University

The Problem of Plateau on a Riemannian Manifold

Author(s): Charles B. Morrey, Jr.
Source: Annals of Mathematics, Vol. 49, No. 4 (Oct., 1948), pp. 807-851
Published by: Mathematics Department, Princeton University
Stable URL: https://www.jstor.org/stable/1969401
Accessed: 23-12-2024 02:18 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Mathematics Department, Princeton University is collaborating with JSTOR to digitize,

preserve and extend access to Annals of Mathematics

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 ANNALS OF MATHEMATICS
Vol. 49, No. 4, October, 1948

THE PROBLEM OF PLATEAU ON A RIEMANNIAN MANIFOLD*

BY CHARLES B. MORREY, JR.

(Received September 10, 1947)

The problem of Plateau or the proof of the existence of a surface of least area
bounded by a given Jordan contour in Euclidean space is one of the classical
problems of the Calculus of Variations. This problem has been studied at least
since the time of Riemann but it was not until 1930 that satisfactory proofs for
a general contour were given independently by Douglas [25]' and Rado [15].
However, before that time, considerable information was gathered and the
problem was solved in many interesting special cases. For the literature of this
period, we refer to the excellent report of Rad6 [38] on the problem of Plateau.
Shortly after the appearance of the solutions of Douglas and Rado, McShane
presented an interesting solution which was perhaps more along the lines of the
straightforward "direct methods of the Calculus of Variations" (see [30]). In
the past seventeen years the problem has been generalized by replacing r by
several contours r1, * * *, rk and by prescribing the topological structure of the
desired surface ([5], [19], [27], [28], 129], [33], 135], [41], [43]), or by replacing the
condition that the surface be bounded by a given contour by the condition
that only parts of the boundary be prescribed, the remainder of the boundary
being merely restricted to lie along certain manifolds (see [18], [22], [23], [24]).
Finally minimal surfaces which do not furnish even relative minima have been
discussed (see [20], [21], [31], [32], [33], [34], [35], [39], [40], [42], [43]).
The present paper generalizes the problem by replacing the underlying
Euclidean space by a Riemannian manifold of considerable generality. The
writer knows of only one paper in which the Plateau problem has been solved in
any space other than Euclidean space, namely the recent paper by A. Lonseth
[7] where the space considered is "hyperbolic space". However, S. Bochner [2]
has studied the "harmonic" functions in a general Riemannian metric (of suffi-
cient differentiability) but has not demonstrated their minimizing property in
general. This work is, of course of great interest as it is still true in a Riemannian
space (of sufficient differentiability) that if a minimal surface is mapped con-
formally on a plane region, then the representing vector is "harmonic" in
Bochner's sense and also that a surface of least area (if of sufficient differenti-
ability) is a minimal surface (in the sense of Differential Geometry).
The present writer does not attempt to generalize all of the preceding results
on the Plateau problem but restricts himself to the case of a surface of the type
of a k-fold connected plane region bounded by k given contours. The space is
a "homogeneously regular" (see Def. 3.1) Riemannian manifold 9) of class C'.
Every compact, regular Riemannian manifold of class C' is seen to be homo-
* A preliminary report of this paper was presented to the Society at the summer meeting
in September 1939.
Numbers in brackets refer to the bibliography at the end of the paper.
807

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
808 CHARLES B. MORREY, JR.
geneously regular; thus this restriction relates to the behavior of 91 at infinity.
In such a space, a vector function x(u, v) of class $2' (see Def. 1.15 and [4] de-
fined on some region B bounded by k circles is shown to exist under hypotheses
identical with those of Courant [5] such that E = G, F = 0 almost everywhere
and x = x(u, v) is a surface of least Lebesgue area bounded by the given contours.
If 9W is sufficiently differentiable, it is then shown that x(u, v) is of class C" (on
the interior B of B), E = G, F = 0 everywhere, and x = x(u, v) is a minimal
surface in the sense of Differential Geometry. The results concerning the
differentiability follow immediately from the writer's recent results concerning
the differentiability of the solutions of minimum problems (see [14]). A simpli-
fication of this work has been suggested in a recent paper by Shiffman [44].
The writer does not attempt to avoid the use of the theory of the conformal
mapping of plane regions and, in fact, uses even more general mappings developed
by him in a recent paper [12]. However, the conformal mapping of the com-
peting surfaces is avoided by the use of e-conformal mapping (see Def. 1.20 and
Theorem 1.2) an idea which was used by Rad6 (see [15]).
It is true that any non-degenerate surface (see [10]) of type k which is of finite
Lebesgue area can be mapped (generalized) conformally on a plane region of
type k and the writer included this proof in an earlier draft. However, this
proof necessitates a rather long topological discussion which is rendered unneces-
sary by the theorem concerning e-conformal mappings. It may be objected
that no purely geometric definition of the Lebesgue area of a surface is given in
terms only of distances in 9W. The recent definition of area given by Busemann
(see [3]) is satisfactory in this respect and coincides in Euclidean space with
the Lebesgue area for surfaces of class C', both being given by the classical
integral. The Lebesgue area of a general surface could then be defined in terms
of surfaces of class C' as it has been in Chapter I, ?8 in terms of surfaces of class
W'. The writer has assigned to a graduate student the problem of modifying
Busemann's definition in order to obtain a definition which yields the Lebesgue
area in Euclidean space in all cases and which will be given by the usual integral
for surfaces of class 3' in a Riemannian manifold of class C'.
The methods used in the proof of the existence theorem are essentially those
used by Courant in [5] although ideas borrowed from other writers are also
used. Chapter I makes precise the elementary topological notions concerning
curves and surfaces (see also [17]) and proves the important theorem concern-
ing e-conformal maps. Finally the Plateau problem itself is solved in Chap-
ter III.
The writer has been unable to establish the existence of a "harmonic" function
taking on given continuous boundary values by finding an equicontinuous
minimizing sequence. Thus, since the functions of class p,, etc. (see [4], [13]),
have proved important in Euclidean space, the writer has defined and discussed
such functions in a Riemannian manifold 9YI in Chapter II and has used them to
obtain the desired result in Chapter III when 9W is homogeneously regular. In
this case it is also possible to prove the result without introducing the more
general functions but the developments are approximately as long as those

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 809
in Chapter II and the final result is not as general as that obtained in Chapter
III in which it is shown that the minimizing function is continuous even when
the more general functions are allowed among the competing functions. That
this is not true for the general manifold 9R is seen by the following example
(which is trivial but indicates the difficulties which may be encountered in deal-
ing with the general manifold): Let 9N be the three dimensional manifold
given by
W = f(x, y, z) 2-' log (2 + y2)
and let r be the curve x2 + y2 = 1, z = 0, w = 0. It is clear that r bounds no
continuous surface in W. On the other hand the surface

w = 2 1 log (2 + y2), Z = 0, 0 < X + Y 1

is easily seen to be of finite area and it is easy to obtain explicit formulas for
mapping this infinite surface conformally on the unit circle in the (u, v) plane so
that x = 0, y = 0 corresponds to u = v = 0. The resulting vector function
of (u, v) is of class 2 and is not equivalent to a continuous function.
If E is a set, E denotes its closure. A region is an open connected set. If G
is a region, G* denotes its boundary. If u[= (u1, U**, N) or (u, v)] or p is a
point, C(u, r) or C(p, r) denotes the open sphere or radius r and center at u or p.
The cell ai _ ui _ bi, i = 1, * * *, N. or a < u < b, c < v _ d, will be denoted
by [a,, ... , aN; bi, ... , bN] or [a, c; b, d], respectively.

CHAPTER I
CURVES AND SURFACES IN A RIEMANNIAN MANIFOLD.
1. In this chapter, we shall set up a technique for dealing with curves and
surfaces in a Riemannian manifold and shall carry over known results concerning
the area and conformal mapping of surfaces to the case where they are embedded
in a Riemannian manifold. We shall see that a Riemannian manifold is a metric
space so that we shall devote part of this chapter to curves and surfaces in a
general metric space 9W. We shall denote points in 9IQ by letters x, y, X, Y, etc.
or by capital letters P, Q, etc., and points in the coordinate sets on which the
curves and surfaces are represented by u, v, i, a, (or pairs of these letters) or by
small letters p, q, etc. The notation x(u, v), for instance, will stand for a vector
function defined on some set in the (u, v) plane with values in 9W. The distance
between points P and Q in 9R will be denoted by I P - Q j .
2. Let 93 be a metric space. We assign the obvious meaning to the continuity
of a vector function with values in W1.
DEFINITION 1.1: Let xl(u) and X2(v) be continuous vector functions defined
for a ? u _ b and c _ v < d. Let T:v = v(u) be a 1-1 continuous transforma-
tion from the interval (a, b) into the interval (c, d) and let x2(u) = x2[v(u)].
We then define
X1, X2 = g.l.b. max I xl(u) - x2(u) I
aourb

for all such transformations T.

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 810 CHARLES B. MORREY, JR.
DEFINITION 1.2: Let x(u) be a continuous vector function defined for
a < u < b. We define 1(x) as the least upper bound of
n

x I X (ui) -x(ui-) I

for all subdivisions a = uo < U, < ... *< u = b.

The following facts concerning these quantities are well known:
LEMMA 1.1: If xi(ui) is a continuous vector function defined for ai ? us ? bI),
i = 1, 2, 3, then (a) II xi, X2 11 > 0, (b) jjX1, X2 11 = 11 X2, X1 11, (C) X1i, XI3 <
jjX, X2jj + 11 X2, X3 . If 1jI X1,X2 = 0, then 1(xi) = l(x2) and XIX1,X3II =
X2,X3j.

3. We now define what we mean by a curve in the sense of Fr6chet:

DEFINITION 1.3: By a curve, we mean a class a of continuous vector functions
x(u) each defined on some interval a < u < b (perhaps a different interval for
each function) such that jI xi , X2 II = 0 for any two members xi(u) andx2(u) of
a and such that if xi e R and I I x1, x2 II = 0, then X2 e a. If C is a curve in this
sense, and x(u) e R, we say that the equation x = x(u) is a parametric representa-
tion of C and write C: x = x(u). If C possesses a representation which is also
1-1, we say that C is an arc.
From Lemma 1.1, it follows that the following definitions have meaning:
DEFINITION 1.4: Let C1 : x = x1(u) and C2 x = X2(v) be curves. We de-
fine the length 1(C,) of C, by 1(C,) = 1(xi). We define the distance II C1, C2 ||
between the curves C1 and C2 by 11 C, , C2 X = II X1 , X2 11. We say that Cn con-
verges to C (C-n C) if I I Cn , C l O.
The following facts concerning curves are known:
LEMMA 1.2: (a) I| Cl , C2 I| = ||C2 X C1 I| X (b) II C1, C2 || > 0 unless C1 = C2,
(C) II ClI C2 11 + || C2 X C3 1> C1 X C3 11 , (d) if Cn -+C, then l(C) < lim infnoo.
(Cn), (e) if Cn -k C and x = x(u), a < u < b is any representation of C, then there
exist representations x = Xn(U), a < u < b, of Cn such that Xn(U) converges uni-
formly to x(u) for a < u _ b.
It is clear that we may consider continuous vector functions x(p) defined on
the circumferences of circles or, indeed, any simple closed curves in the plane
and may define 1(x) and II xI, X2 11 for such vector functions and note that the
statements in Lemma 1.1 hold.
DEFINITION 1.5: By a closed curve C, we mean a family a of continuous vector
functions x(p), each defined on a simple closed curve in the plane (perhaps differ-
ent curves for different functions) such that I I X1 , X2 I = 0 for each two members
xi and x2 of R and such that if xi e R and jI X1, X2 11 = 0, then X2 e R. C is said
to be a simple closed curve or a Jordan curve if it possesses a 1-i continuous para-
metric representation on the circumference of a circle. 1(C) and C| 01, C2
are defined in an analogous way.
It is clear that the statements analogous to those of Lemma 1.2 hold for closed
curves C.

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 811
4. We may carry through an entirely analogous analysis for surfaces.
DEFINITION 1.6: A plane region G is said to be of type k if it is bounded and
its boundary consists of k mutually exclusive simple closed curves.
DEFINITION 1.7: By a surface of type k, we mean a family t of continuous
vector functions x(u, v), each defined on some closed region G of type k such that
,X2I|| = 0 for each two members xi and X2 of a and such that if xi e a and
lxI, x2l e , then' 2Ea. If S is a surface and x(u, v) e 't, we say that
x = x(u, v) is a parametric representation of S. We define II S,, S2 11 for two
surfaces S, and S2 both of type k as before.
It is clear that the analogs of statements (a), (b), (c) and (e) of Lemma 1.2
hold for surfaces of type k.
DEFINITION 1.8: Let r, and r2 be two sets of k closed curves. Let the curves
of ri and r2 be ordered. For each such ordering 0, we define ,u(rF, r2, 0) as
the maximum distance two curves which correspond under 0. We then define
11 ri XF2 II as the smallest of the numbers ,u(r, , r2, 0) for all such orderings 0.
We say that rF, -- r if r and all the rn consist of the same (finite) numbers of
curves and if II r, r i -- 0.
DEFINITION 1.9: Let S: x = x(u, v), (u, v) e G, G of type k. Then the equa-
tion x = x(u, v), (u, v) e G*, defines a system r of k closed curves. We say that
S is bounded by r.
The following lemma is evident.
LEMMA 1.3: If Sn -> S and if r,, and r are the boundaries of S and S, respec-
tively, then rn -- r. If rn-- rF, all consisting of k closed curves, if G is a region of
type k, and if x = x(p), p e G*, is any representation of r, there exist representations
x = xn(p), p e G*, of Fn such that the xn(p) tend uniformly to x(p) on G*. Finally,
suppose that { Sn } is a sequence of surfaces of type k , that Sn is bounded by Fn, that
rnF, - rF and that G is a region of type kc; then there exists a subsequence {S, } of
ISn } and a sequence x = x, (u, v) of parametric representations of the S, on (; such
that the functions x,(u, v) converge uniformly on G* to a function x(u, v) defined on
G* such that x = x(u, v), (u, v) e G*, is a representation of r.

5. We now define what we shall mean by an N-dimensional Riemannian mani-

fold of class C(k)e k > 1:
DEFINITION 1.10: By an N-dimensional Riemannian manifold (without bound-
ary) of class C(k) I > 1, we mean a connected, complete metric space which
contains a finite or denumerable number of open sets 9i with the following
properties:
(1) each point of 9 is in at least one of the Pi,
(2) each 9i is the 1-1 continuous image under a transformation Ti of the open
cell Ro: I xi < 1, j = 1, **, N, the Ti being so related that if P99tM #0
and 9Ji 9,I corresponds to Gi under Ti and to G1 under T1, then the transforma-
tion P' = TY' [Ti(P)] from Gi to GI is of class 0(o) together with its inverse, and
(3) for each i, there exist functions gp)(xl, ... , xN), p, q, = 1, * *, NA, with
9 = q( which are of class C(k-1) in Ro such that

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
812 CHARLES B. MORREY, JR.
(i) the quadratic form g(x) ap is positive definite in t for each x in Ro,
(ii) if r is an arc in 9i which corresponds under Ti to a regular arc C, in Ro,
of class C', then

l(r) - f [gW(x) dxa dx]lI2

(iii) if W1M. 9)I # 0 and corresponds under T, to Gi and under T, to GI and if

x= x'(x) is the transformation in (2), then the functions gp(q(x) and gp )(x)
are connected by the usual condition that

(1.1) g(1)(X) dxta dX'1 = 0 dXa dx.

DEFINITION 1.11: Let T be a 1-1 continuous transformation of a region G in
N-space into an open set r on W. Then T is said to be of class Cel, 1 < k, if
all the transformations T-1 T for which 9i)- r # 0 are of class C11) together with
their inverses. We define the associated functions gp,q(x) uniquely over G (in
case 1 > 1) by the relations analogous to (1.1) using the transformations T- ' T.
Since 9) is connected, any two points xi and x2 of 9) can be joined by a curve.
We shall hereafter assume that I xi -x2 I denotes the greatest lower bound of l(C)
for all such curves C.

6. Now, let x(ui, * , u,) be a continuous vector function defined on a set S

consisting of an open set plus some (or all or none) of its limit points. Let Si
denote the subset of S for which x e Pi ; then each Si is open in S or null. For
(u1, ... , up,) e Si, there are uniquely defined N continuous functions x(ui, * ... up),
j = 1, - , N, such that Ti[x (u), * , x (u)] x(u).
DEFINITION 1.12: The functions x2(u) above are called the components of
x(u) with respect to Ti ; a similar definition may be made if T is any transforma-
tion as in Definition 1.11.
DEFINITION 1.13: The vector function x(u) is said to be of class 0), I < k,
on S if, for each T of class C)l the components of x(u) with respect to T are of
class C&) where they are defined.
DEFINITION 1.14: A vector function x(u), defined for a < u < b, is absolutely
continuous (AC) on [a, b] if its components with respect to any T of class C' are
absolutely continuous on each closed segment of their range of definition. We
define I xI I or I x' I by the expression [9a ,(x)xUx ]'12 wherever this is defined;
it is clear that this is defined almost everywhere on [a, b] and is independent of
T. We say that x(u) is of class $2' on [a, b] if I x. i2 is summable on [a, b] and
x(u) is AC on [a, b].
The following lemma is immediate:
LEMMA 1.4: Let x(u) be AC on [a, b] and let C: x = x(u). Then

l(C) = I(x) = f , I du.

DEFINITION 1.15: A vector function x(u, v) defined on a closed (or open) region

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 813
B (or B) is said to be of class 93"2 on B (or B) if it is continuous on B (or B) and
if the components, with respect to any T of class C', are of class $" where they
are defined. We define I xu xv , E, F, and G by
| I | = [ga,3(X)Xa Xf 2 Xv I = [g((X)g Xv]
E = I xu J2, G = J xv J2, F = 9a(X)XU X
wherever these expressions are defined; it is clear that they are defined almost
everywhere on B and are independent of T. The function x(u, v) is said to be
of class 3'/ on B (or B) if x(u, v) is of class $" on B (or B) and if E and G (and
hence F) are summable over B.
DEFINITION 1.16: If x(u, v) is of class i3' on B (or B), we define

D2(X, B) = ff (E + G) du dv, L(x, B) = ff (EG - F2)112 du dv.

7. We first state two definitions:

DEFINITION 1.17: Let B and D each be sets consisting of an open set together with
some or all of its boundary points. A transformation u' = u'(u, v), v' = v'(u, v)
of B into D is said to be a regular transformation of class C' if it is 1-1 and u'(u, v),
v'(u, v), u(u', v'), and v(u', v') are all of class C' and satisfy uniform Lipschitz
conditions.
We state two well known lemmas:
LEMMA 1.5: Let B and D be regions of type k bounded by circles and let
ul = u'(u, v), v' = v'(u, v) be a 1-1 continuous transformation of B into D. Then
there exists a sequence u' = u (u, v), v = v' (u, v) of regular transformations of
class C' of B into D such that u$ (u, v) and v' (u, v) converge uniformly on B to
u'(u, v) and v'(u, v) respectively.
LEMMA 1.6: Let Bn and B be regions of type k bounded by circles and suppose
that Bn -* B (that is the centers and radii of the bounding circles of Bn tend to those
of the corresponding circles of B*). Then there exists a sequence un = un(u, v),
Vn = vn(u, v) of regular transformations of class C' of B into Bn such that Un(U, v)
and vn(u, v) converge uniformly with their derivatives on B to u and v and their deriva-
tives, respectively.
The following two lemmas are well known:
LEMMA 1.7: Let {Ixn(U, v) } be a sequence of functions of class $12' on a region B
of type k with D2(xn, B) uniformly bounded, and suppose that xn(U, v) tends uni-
formly on B to x(u, v). Then3 x(u, v) is of class 32 on B and

n -*ox -o
D2(x, B) ? lim inf D2(xn, B), L(x, B) < lim inf L(xn, B).4

In fact, if x(u, v) is known to be of class 3' on B, the inequalities above hold even if
the D2(xn, B) are not uniformly bounded.

2 See [4], Def. 4.4, Th. 4.4, and Def. 4.3.

3See [9], ?1-
I See [8].

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
814 CHARLES B MORREY, JR.
LEMMA 1.8: Let x'(u',. v') be of class $2 on a region D (or I)) and u' = u'(u, v),
v' = v'(u, v) be a regular transformation of class C' from a region B (or B) into D
(or D)). Then the vector function x(u, v) = x'[u'(u, v), v'(u, v)] is of class $2P' on B
(or B) and the derivatives of the components of x(u, v) are connected almost everywhere
with those of the components of x'(u', v') at corresponding points by the usual rules
of the Calculus. Thus E, F, and G transform in the usual way and L(x, B) =
L(x', D). This lemma holds also if the map is conformal but not necessarily regular
and in this case it is also true that D2(x', D) = D2(x, B).5
The following useful extension of the lemma above and the theorem on con-
formal mapping has been proved by the author :
LEMMA 1.9: Let B be a region of type k with k = 1, let p, q, and r be distinct
points on B* and let 7r, K, and p be distinct similarly ordered points on C*(O, 1).
Let a(u, v), b(u, v) and c(u, v) be bounded and measurable on B with a > 0, c > 0,
and ac -b2 = 1 on B. Then there exist unique functions t(u, v) and -q(u, v) of
class 932 on B such that t = t(u, v), X = q(u, v) is a 1-1 continuous map of B into
C(O, 1) which carries p, q, and r into 7r, K, and p, respectively and such that t(u, v)
and -q(u, v) satisfy

7JU = (biu + ctv,) 71v = atu + b~t,

almost everywhere on B.
This transformation is such that (i) measurable sets correspond, (ii) functions of
class $3' correspond, that is, if X( , -q) is of class 23 ' on C(O, 1), x(u, v) =
X[{(u, v), 7(u, v)] is of class $2/ on B, and if x(u, v) is of class $2 on B, then
X(%, q) = x[u(%, i), v(%, i)] is of class $2 on C(O, 1), (iii) if x(u, v) and X(, r) are
of class $32 and correspond, their derivatives are connected almost everywhere by the
usual laws of the Calculus.

8. In this section, we define the Lebesgue area of a surface and demonstrate

some of the elementary theorems concerning area.
DEFINITION 1.18: Let x(u, v) be a continuous vector function defined on a
region B of type k. We define Lp(x, B) for p > 0 as the greatest lower bound
of L(y, B) for all y(u, v) of class l3" on B such that I x(u, v) - y(u, v) I < p on S.
We then define L(x, B) = lim,.o L,(x, B).
LEMMA 1.10: Let xn(u, v) and x(u, v) be continuous on a region s of type k and
suppose x (u, v) -- x(u, v) uniformly on B. Then L(x, B) < lim infn_<.o L(x, X B).
In any case there exists a sequence {yn(u, v) of functions of class $3' on B which
converges uniformly to x(u, v) on B and is such that L(yn, B) -* L(x, B).
LEMMA 1.11: Let B and D be regions of type k and let xl(u, v), (u, v) e B, and
X2(S t), (s, t) e D, be continuous vector functions such that x1 xl, x2 = 0. Then
L(x1, B) = L(x2, D).
PROOF: First suppose that there exists a conformal map u = u(s, t), v = v(s, t)
of D into B with inverse s = s(u, v), t = t(u, v), such that X2(S, t) =
I See [13], ?6; the last statement follows from a simple limit process.
6 See [12], ?2, Theorems 3 and 4.

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 815
xi[u(s, t), v(s, t)] and xl(u, v) = X2[s(U, v), t(u, v)]. Let {yin(u, v)} be a se-
quence of functions of class $i' on B which converge uniformly on B to x1(u, v)
and suppose L(yl, X B) -* L(x1, B). Let Y2,n(S, t) = yln[u(s, t), v(s, t)]; then
each Y2,n(5, t) is of class $2' on D with L(y2,,, D) = L(y1, X B) by Lemma 1.8,
and Y2,n(S, t) -* X2(S, t) uniformly on D. Thus L(x2, D) < limnM. L(y2,n , D) =
L(xi, B). In a similar way, it follows that L(x1, B) _ L(x2, D) so that the
equality holds. Thus, using the conformal mapping theorem, we may replace
B and D by regions bounded by circles.
Now, in general (B and D circular) if Xi1 , X2 1 = 0, it follows from Definition
1.1 that there exists a sequence u = fiu(s, t), v = bn(S) t) of 1-1 continuous trans-
formations of D into B such that 22,n(S t) = x1[un(s, t), Vn(s, t)] converges uni-
formly on D to X2(s, t). Now, let Iy1 n(u, v) } be as in the first paragraph. For
each n, let u = un(s, t), v = Vn(s, t) be a regular transformation of class C' from
D into B such that I un(s, t) - in(s, t) 0, I vn(s, t) - bn(s, t) I < n-1; this is possi-
ble by Lemma 1.5. For each n, define y2,n(5, t) = yln[un(s, t), vn(S, t)]; then each
Y2,n(S, t) is of class 3' on D with L(y2,n , D) = L(y1,n, B) and y2,n(8, t)
x2(s, t) uniformly on D. Thus L(x2, D) _ limn- L(y2,nf, D) = L(xi, B).
As above, we may also prove L(xi, B) _ L(x2, D).
The Lemma above justifies the following definition:
DEFINITION 1.19: Let S: x = x(u, v), (u, v) e B of type k. We define the
Lebesgue area L(S) of S by L(S) = L(x, B).
THEOREM 1.1: If Sn S (all of type k), then L(S) _ lim infn 0. L(Sn). If
S: x = x(u, v), (u, v) E B of type k, where x(u, v) is of class "' on D, then L(S) =
L(x, B).
PROOF: The first statement follows from Lemma 1.10 and the second follows
from Lemmas 1.10 and 1.7.

9. We conclude this chapter with a highly interesting and important theorem

concerning the mapping of surfaces. This theorem is similar to a theorem proved
Rad6 in his solution of the Plateau problem [15]. It is well known that
2(EG - F2) < E + G wherever these numbers are defined, the equality holding
if and only if E = G, F = 0. Thus if x(u, v) is of class $3' on B, then 2L(x, B) =
2L(x, B) < D2(x, B). We discuss representations of surfaces which are
called e-conformal and are defined as follows:
DEFINITION 1.20: Let x = x(u, v), (u, v) e R, be a representation of a surface
S. Then x = x(u, v) is said to be an e-conformal representation of S(e > 0)
if x(u, v) is of class 13 ' on D and D2(x, B) < 2L(x, B) + e.
THEOREM 1.2: Let S: x = x(u, v), where x(u, v) is defined and of class $'I3 on B
(of type k). Then for each e > 0, S possesses an e-conformal map on a region Df
of type k bounded by circles.
PROOF: Define

(1.2) E + n-'G, F =F + n-'E

where we have chosen n so large that

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
816 CHARLES B. MORREY, JR.

IL J ( - P2) dudv = fL [EG - F2 + n '(E + G) + n 2EG]1 du dv

< (EG - F2) du dv + e/2 = L(S) + e/2.

Let Cl be the outer boundary of B and let B, be the region bounded by Cl.
At all points of B where E, F, and G are defined and not all zero, we define et(u, v),
b(u, v) and c(u, v) by

(1.3) a = a/H, b = -F/H, = /H, H = (E -_2)

at all other points of B3, we define a = = 1, b = 0. Then a, b, and a are
bounded and measurable on BI with a > 0, a > O, a-b2 = 1. Thus by Lemma
1.9 there exists a unique transformation U = U(u, v), V = V(u, v), with inverse
u = u(U, V), v = v(U, V) of B1 into C(O, 1) in which U(u, v), V(u, v), u(U, V),
and v(U, V) are of class $2' on B, and C(O, 1), measurable sets correspond,
X(U, V) = x[u(U, V), v(U, V)] is of class 3' on C(O, 1), etc., and

(1.4) Vu= -(bUU + UV)) VV = fUU + bUV

almost everywhere. By following this with a conformal transformation, we
may ensure that B is carried into a region D bounded by circles, the resulting
transformation being of the same type (see [12]).
Now
I IX 12 EVv-2FVu Vv + GV

=I XV 12= EU2-2FUu Uv + GU2S

-E UvVv + F(UuVV + VvUUu) -GUuIU

8= 3~~~~~~~~~~~2
(1.5) @e) 92 EG - F2

3 = UuV-Us Vu = dU~L + 2bUu U, + cUV

= &VSU + 2bVUVV + c

S + () daE + 2bF + WG

almost everywhere. Let us define L;, , and 5 almost everywhere in terms of

E, P, and a by the same formulas (above) which define A, Ay, and 5 in terms of
E, F, and G. By reference to the formulas (1.2), (1.3), (1.4) and (1.5), we con-
clude that E, 0, and ($ are summable over D, and that
or + _ _ z t _ of + n-a; t ~~~~~~~~~~~~~~~~~~~2 (

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 817

D2(X, D) _ ff (@ + @) dU dV = 2 fD - m 2)1dU dV

= 2 (f20 -_F2)1 du dv < 2 (EG - F2)1 du dv + e = 2L(S) + e.

Thus x = X(U, V) is a representation of the desired type (it is to be noticed that

e = ( = 0 at almost all points (U, V) corresponding to points (u, v) where
P = P = = 0, i.e. where E = F = G = 0, so that, of course, e = @5, = 0
at these points also).

CHAPTER II

VECTOR FUNCTIONS OF CLASS %3a ,3a ETC.

1. In this chapter, we introduce vector functions x(u, v) defined on plane

regions and having values in a Riemannian manifold 9T which are of class 3a .
The developments are very similar to those in [4] [13], and [14] so that we shall
have to assume that the reader is somewhat familiar with these papers, particu-
larly with [4], [13] and ?6 of Chapter III of [14]. We shall be able to simplify
the discussion considerably by restricting ourselves to plane regions of class C'
(see [13], Def. 7.1) or D' which we now define.
DEFINITION 2.1 A region is said to be of class D' if it is bounded by a finite
number of mutually exclusive simple closed curves, each of which consists of a
finite number of regular arcs of class C' such that if two of them have a (or two)
common end points, they have distinct tangents there.
It is clear that a region is of class C' if and only if it is bounded by a finite
number of mutually exclusive simple closed regular curves of class C' and that a
region is of class D' if and only if it can be subdivided by means of a finite number
of regular arcs of class C' into a finite number of non-overlapping closed regions,
each of which is the image of a closed rectangle under a regular transformation of
class C'.
2. In this section, we consider functions x(u) [u = (ul, .**, UN)] which are
measurable or of class La on a bounded measurable set E.
DEFINITION 2.2: The vector function x(u) is said to be measurable on the
measurable set E if for every open set 0 in 9Y, the set of points u of E for which
x(u) e 0 is measurable.
LEMMA 2.1: Let xl(u) and x2(u) be measurable on a measurable set E. Then
x(u) - x2(u) I is measurable on E. Any continuous function is measurable.
PROOF: For each n, divide 9T up into at most No mutually exclusive Borel
measurable sets Sjn X each of diameter < n 1; it is easy to see how this may be
done by using the neighborhoods Pi. For each j, k, and n let Ej,k,n be the
subset of E on which xl(u) e S,,n and x2(u) e Sk, ". It is easy to see that each
Ej,k,n is measurable and E = Ej,k Ej,k,n for each n, the Ej,k,n being mutually
exclusive. For each j and n, let xjn be some point of Sjm. For each n, let
pn(u) = I Xn- Xk, " I for u e E,,k,n . Then each pn(u) is measurable and pn(u) --

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 818 CHARLES B. MORREY, JR
xj(u) - X2(U) for each u in E, so that I xl(u)-X2('1) I is measurable. The last
statement is obvious.
DEFINITION 2.3: A function x(u) is said to be of class La(a > 1) on a bounded
measurable set E if it is measurable on E and if I x(u) - i a is summable for
every xo in W1.
The following Lemma is immediate, 9)1 being complete:
LEMMA 2.2: Let x(u) be of class La on cach of a finite number of bounded measur-
able sets E1, Ek. Then x(u) is of class La on E + + Ek. If xl(u) and
X2(u) are in La on E, then I x1(u) - X2(U) I' is summable on E. If we identify
equivalent functions in La on E and define the distance

(X1 , X2) = {f 1 X1(U) - X2(U) a duI

then the resulting space is a complete metric space. Finally, if x(u, v)

(u = u,* **, UN,v = v1, .. **, VP) is of class La on a cell a} < ui < bi ci <
vij ? dj, then, for almost all u, x(u, v) is in La as a function of v on [c, d].
PROOF: Of these, the only statement which requires further examination is
the third. This in turn depends on the fact that if each xn(u) is measurable on
E and xn(u) -+ x(u) almost everywhere on E, then x(u) is measurable on E.
To prove this, let Z be the set of measure zero on E where the convergence fails
to hold, let 0 be any open set on 91, let Ok consist of all points PO of 0 such that
all points P with P - Po I _ 17' also belong to 0, let Snk be the set of u e E - Z
where xn(u) e Ok, and let S be the set of u e E - Z where x(u) e 0. Then it is
easily seen that

S = II E jJSn, k
k=1 N-1 n-N

and is therefore measurable, since each Ok is open.

The following Lemma is useful in the sequel:
LEMMA 2.3: Let x(u) be of class La on a bounded measurable set E. Then there
exists a function x(u) defined almost everywhere on E and coinciding almost every-
where on E with x(u) such that

lim [m(e)]fl I x(u)-(uo)du = O

for every uo where X(uo) is defined and every regular family7 of sets e C E about uo .
PROOF: Let {XnI be an everywhere dense sequence of points in 9)1. Since
l x(u) -n I is summable for each n, there exists a set Z C E with m(Z) = 0
such that if uo e E - Z, then

h-O E' Rh e-+u o

(2.1) lim (2h)-N fI x(u) - Xn du = x(uo) - Xn 1, lim [m(e.E)/m(e)] = 1

for every n and every regular family of sets e about uo. It follows easily that,
if uo e E - Z, then (2.1) holds for every x in 9N and hence, in particular for
x(uo). This proves the lemma.
7See [6], pp. 59-61.

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 819
3. In this section, we introduce our functions of class 83a and W . We con-
fine ourselves to the two dimensional case but the theorems generalize immedi-
ately to the N-dimensional case.
DEFINITION 2.4. A cell function (p(R) is said to be normal on a region B if,
whenever a cell R in B is divided into cells R1, * , RN then ep(R) _ Ji'l<(Ri)
We define the variation and the absolute continuity of such cell functions in the
usual way.
The following Lemma is due to Banach [1]:
LEMMA 2.4: Let jo(R) be a non-negative, normal cell function of bounded variation
on a region B. Then the Lebesgue derivatives exists almost everywhere on B and is
snmmable on B, and

Vv(B)> ?fD du dv,

ir,(B) denoting the variation of p over B. The equality holds if and only if p is AC
DEFINITION 2.5: The function x(u, v) is said to be of class ha on the bounded
region B if
(i) it is of class La on B,
(ii) there exist two non-negative, normal, absolutely continuous cell functions
<1(R) and <2(R) on B with (Dvpl)' and (D(P2)a summable on B such that
d

(2.2) (P1(R) = f |x(b, v) - x(a, v) dv,

b

Vo2(R) = L x(u, d) -x(u, c) du, R:a < _ < b c ? v ? d

for almost all cells9 on B.

LEMMA 2.5. Let x(u, v) be measurable on a region B and suppose that, for almost
all uo , x(uo , v) is AC in v on each closed segment of u = uo in B and for almost all
Vo , X(u, vo) is AC in u on each closed segment of v = vo in B. Let Si be the subset
of B on which x(u, v) e Pi . Then almost all points of each Si lie along a set of
lines U = constant such that (i) the totality of these U form a measurable set and
(ii) the part of Si on one of these lines is open on that line; a similar statement holds
with the roles of U and S3 interchanged.
The components x'(u, v) are uniquely defined and measurable on Si and their
partial derivatives exist almost everywhere on Si and are measurable there.
Finally if (uo, vo) is a point of Si at which all the partial derivatives of the xi(u, v)
exist and if (uo , vo) also belongs to some Si, then the partial derivatives of the x'(u, v)
all exist at (uo, vo) and are connected with those of the x'(u, v) by the usual rules of
the Calculus. It follows that the formal expressions for I xu x , I xv l, E, F, and G
are defined almost everywhere on B, are independent of the T; and T, and are measur-
able on B.
PROOF: All the statements in the lemma are immediate except perhaps the

8 See [6], pp. 59-61; naturally the sets must be cells in this case.
9 See [4], Definition 3.1.

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
820 CHARLES B. MORREY, JR.
measurability of the partial derivatives of the x'(u, v). We prove the existence
and measurability of xi u, for instance as follows: Define
-81[( + 8XV i(,V] (8X V) 1E Si, (U + 8X V) e Si
p8(u, v) = [x(u + s, v) - xj(u, v)
lO (U, V) E Si, (U + S, V) e Si
and define 4)h(U, v) as the least upper bound of po8(u, v) for 0 < s < h (h > 0).
It is clear that wp8(u, v) is measurable on Si for each s N 0 and, since x%(u, Q3)
is continuous in u for each value of Q8 above, it follows that for almost all (u, v) in
Si Cbh(U, v) is the least upper bound of (p8(u, v) for all rational s, 0 < s < h,
and is therefore measurable. Since Dtx' = lim-...o+ 4?h(u, v), it follows that
D+xZ is measurable on Si. In a similar way, the other Dini partials of xi
with respect to u are proved measurable on Si. Since x2(u, Q3) is AC in u for
almost all the S3 above, it follows that all the Dini partials coincide almost every-
where on Si.
DEFINITION 2.6: The function x(u, v) is said to be of class 13a on the bounded
region B if it satisfies the conditions of Lemma 2.5, if it is of class La on B, and
if I x. Ia and I xv Ia are summable on B.
4. In this section, we exhibit the relationship between functions of class h
and $1 (a _ 1). Hereafter B shall always denote a bounded region.
THEOREM 2.1: Suppose that x(u, v) is of class Ia on B. Then x(u, v) is of class
La on each segment interior to B which is parallel to an axis. If (p1(R) and s02(R)
are defined by (2.2) for every cell R in B, then (p1 and p2 are non-negative, normal,
continuous, and AC and

Di,= xu l, DjP2 = { x,
almost everywhere.
PROOF: Let [a', c'; b', d'] C B. For almost all U, x(U, v) is continuous in v
and hence in La with respect to v. From the continuity of x(u, v) in u for almost
all v, etc., it follows that x(U, v) is measurable in v for all U, a'< U _ V' and
x(u, V) is measurable in u for all V, c' < V < d'. Thus, from Fubini's theorem
x(U, v) is in La in v for all U and x(u, V) is in La in u for all V and

1i x(b, v) - x(a, v) i dv < , i f u i du dv,

f i x(u, d) - x(u, c) i du < f I xvI du dv

for all [a, c; b, dI C [a', c'; b', d']. Obviously, then, the spi(R) defined for all R
on B by (2, 2) are normal and AC on B.
Now, clearly, if c < e < d

(oi([a, c; b, d]) = poi([a, c; b, e]) + (pi([a, e; b, d]).

Thus we may obtain the variation of (p1 over [a, c; b, d] as the limit

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 821
nd
lim j I x(ui, v) - x(ul, v) I dv
n-0oo 0= c
rb rd
- 1bd I du dv, ui = a + i(b - a)/n.

Thus, from Lemma 2.4, it follows that D<1 = I xu I almost everywhere. In a

similar way, we see that D(P2 =i xv I almost everywhere.
To show the continuity of spj([a, c; b, d]) as a function of the four variables,
we proceed as follows: Suppose [a, c; b, d] C B. Then there is an h > 0 such
that [a', c'; b', d'] C B whenever [a' - a I < h, I - b I < h, I c' - c I < h,
I d' - d I < h. Suppose [a', c'; b', d'] is such a cell and suppose, for definiteness,
that a' < a, b' < b, c' < c, d' < d; the other 15 cases may be handled similarly.

d d'
Then
rd'

i f1([a', c' ; b', d']) - p1([a, c; b, d]) i = i j i x(b', v) -x(a', v) I dv

+J ftx(b, v)-x(a, v) I dv I < x(b', v)-x(a', v) x(b v)l-dx(a, v)v d

+ | tIx(b, v) -x(a, v) |dv -1 x(b, v) -x(a, v) I dv

dr
< I x(b', v) - x(b, v) I dv + f I x(a', v) - x(a, v) I dv

+ I x(b, v) - x(a, v) I dv + f I x(b, v) -x(a, v) I dv = rp,([a', c'; a, d'])

+ pj([b', c'; b, d']) + spj([a, d'; b, d]) + (pl([a, c'; b, c]).

The result follows since spj is AC. The proof for P02 is similar.
THEOREM 2.2: If x(u, v) is of class ,a on B, the equivalent function t(u, v) of
Lemma 2.3 is of class $' on B. If rp and (P2 are any cell functions satisfying the
conditions of definition 2.5 for x, then
rd

(P1([a, c; b, d]) = ( ib, v) - (a, v) I dv,

fb

02([a, c; b, d]) = t i(u, d) - (u, c) I du

for almost all [a, c; b, d] C B. Moreover

D<p= X.ul D(P2=
almost everywhere on B.
PROOF: Since B can be represented as the sum of No overlapping open cells
whose closures are in B, we choose Ro = [a', c'; b', d'] C B and consider cells
interior to Ro. Moreover we shall consider only p2(R) and the variation of

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
822 CHARLES B. MORREY, JR.
x(u, v) is a function of v, the proofs for Se1 and the variation with respect to U1.
being similar.
Since x(u, V) is in La in u on [a', b'] for almost all V on [c', d'] and
since sP2([a, c; b, d]) is given by (2.2) for all [a, c; b, d] C1 R for which (a, c; b, d)
is not in a 4-dimensional set of measure zero; it follows that there is a set ZA of
two dimensional measure zero such that if (c, d) is not in ZA, c' < c < d ? d',
then x(u, c) and x(u, d) and in La in u for a' < u < b' and

(2.3) (P2([a, c; b, d]) I x(u, d) - x(U, c) I du < f Do du dv

for almost all (a, b), a' ! a < b ? b'. Since the last two members are continuous
in (a, b), the second inequality holds for all (a, b), a' < a < b < b', provided
that (c, d) is not in Z1 .
NNow, there is a set Z2 of one dimensional measure zero such that if uo is not
in Z2 , then (uo , v) is defined for almost all v, c' < v < d', and
uo+h r2 'V2
(2.4) lim h-O
(2h)-1 uO-h
D(2 du Il
dv-= D(uo v) dv
uniformly in (v1 , v2) for all (v1 , v2) for which c' ? V1 < V2 _ d'; this is true for
almost all uo on account of Fubini's theorem and the theorem on the convergence
at each point of a sequence of continuous monotone functions to a continuous
monotone function.
Now, suppose that uo is not in Z2, a' < uo <b' and c' < v1 < v2 < d'; let ho
be the smallest of d' - v2, v1 - c', b' - uo, and uo - a'. Now,
clearly x(u, V2 + 712) and x(u, v1 + 771) are measurable in (u, 771, 12). Integrating
(2.3) we obtain
h h uo+h
fd71f d?72 1 x(u, V2 + 72) -X(uiVi + 71) I du
25-h _h u O-h
h h fuo+h '12+q2
< h h - [ 2 Dv dudvf d711dn2.
_h h O-h J+J
From the uniformity of the convergence in (2.4), we obtain
h h (puO+h rV2+712 y2
(2.6) lim (2h)-3 ' | J DVpd dudv d71 d72 = J lDo2 dv.
h-- _ 4-h - O u0-h l+n1 2
Now, suppose that (uto, v1) and x(uO , v2) are both defined. Then

3h h f uo+h -xuv +n ) Id
(2h)-3 f L f I x(is, V2 + 72) 1) du d1 d72
-h I, uO-h
h h fuo+h
- Iz(uov) s)- (o s vi) I _ (2h3) h i J_h
|hhx(u,
u O-h
V2 + 772)-x(u, vi + 711) I

-| X(uo, V2) -Xt(Uo, v) || du dnildn2

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 823
h h u onh
<(2h)-3 L (U, V2 + 712)- (UoV2)1
h h 0o-h

+ I x(u, vi + 11) - (uo, Vi) 1] di? dv6dv2

rUoth vt2+h
- (2h) L J I x(u, v) - (uo, v2) I du dv
u0-h 2-h
u o+h v 1+hz

+ (2h)2 uu0-h
I x(u,
IOv-4
v)-?(uo, vi) I du dv
which tends to zero with h. Thus, by-dividing (2.5) by (2h)3 and letting h --O,
we obtain
V2

JV(o XV2) -(Uo , V1) s f Dqp2dv.

Thus x(uo, v) coincides for almost all v, c' < v < d' with an AC function {(v)
such that

(2.6a) (v2) - t(vl) I _ fD~02dv.

V2

I1

Now, let v be any number, c' < v < d', and choose e > 0. Choose S > 0 so
small that
V2

JDfp2dv < e/4

whenever I v2 - V < 6, c' < vi < d'. Choose vi so that x(uo, vi) = M(vO).
c' < v1 < d', I - v < &. From (2.5) we may choose hi > 0 so small that
the difference there given is numerically < e/4 and also so that
ruo+h vrjrh
(2h) J J x(u, v) - x(uo, v1) I du dv < e/4, 0 < h < h.
u0-h I1-h

Then
uo+h v+h
(2h)2 u0-h
J Jx(u,
i-h
v)- du dv
h h uo+h
= (2hY) L dq L dq L | x(u, v + ) - Z() du
h h u0-h
h h ruo+h
? (2h)3 j d7 d77 f x(u, + 7) - x(u, vI + -1) du
h h 0o-h
uo+h 1'l+h
+ (2h)-2J J x(u, v) - t(vi) I du dv + I j(vi) - ) | < e, 0 < h < hi.
u0-h v1-h

Thus x(uo, v) is defined for all v, c' < v ? d' and is AC in v.

Finally, suppose we define 4' (R) and sop (R) by (2.2), with x replacing x, for

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
824 CHARLES B. MORREY, JR.
all cells R C B, and suppose (uo, vo) is a point of B at which Depj, DA02, Dpi,
and Dqp4' all exist. Since spo(R) = 4p*(R), i = 1, 2, for almost all R, it follows that
Djoi = D4pj at (uo, vo), i = 1, 2. The result follows from Theorem 2.1.
5. The following theorem is an immediate consequence of the definitions and
results of ??3 and 4:
THEOREM 2.3: Let B be a bounded region. Then the following conclusions hold:
(i) if xl(u, v) is of class $3a on B and X2 (u, v) is equivalent to xl(u, v) on B, then
X2(U, v) is of class $J3a on B and X2(U, v) is defined at the same points as is x1(u, v)
and coincides with tl(u, v) at these points,
(ii) if x(u, v) is of class $3 on B, then E, F, and G are summable over B,
(iii) if xi(u, v) and X2(U, v) are both of class $ on B and are equivalent on B,
then E1 = E2, F1 = F2, and G1 = G2 almost everywhere.
DEFINITION 2.7: Let x(u, v) be of class 32 on a region B of class D' and let x(u, v)
be the equivalent function of Lemma 2.2. We then define E, F, and G as
the Lebesgue derivatives of the respective set functions

f lu 12dudV, f("u, xv) du dv, I v 12 du dv.

We define

Da(x, B) = fL (E + G)al2 du dv.

We shall use E, F, and G in the above sense, reserving I X. 12, (xu, xv), and
I XV 12 for the formal expressions for E, F, and G in terms of partial derivatives
in case x is of class $2
The following theorem is immediate:
THEOREM 2.4: (a) If xi(u, v) and X2(U, v) are equivalent functions of class 32
on B, the corresponding functions E , Fj, and Gi exist at exactly the same points
and coincide at these points; (b) if x(u, v) is of class $3 on B, then

E = lxu 12, F = (x", xv), G = I xv 12

almost everywhere on B.
6. In this section, we prove an important theorem on change of variables.
THEOREM 2.5: Let x(u, v) be of class 'J3a on B, let T: u = u(s, t), v = v(s, t) be a
regular transformation of class C' of a region D into B, and let y(s, t) =
x[u(s, t), v(s, t)]. Then y(s, t) is of class %3a on D, g(s, t) = 4[u(s, t), v(s, t)],
and the partial derivatives of a set of components of g are connected with the cor-
responding ones of x by the usual laws of the Calculus almost everywhere.
PROOF: We define

7ri(s, t) = p1[u(s, t), v(s, t)]I u. I + P2[U(8, t), v(s, t)] I V. X P' = I uI, P2 = I I

Tr2(s, t) = pl[u(s, t), v(s, t)] Ut u + p2[u(s, t), v(s, t)] I t I v

Then I 7rl Ia and I 72 la are summable on D. From the definitions of x and y and
from the fact that regular families of sets correspond, it follows that p(s, t) =

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 825
x[u(s, t), v(s, t)], both being defined at the same points. If (p(s, t) is summable,
we define
8+h t+h
SPh(s, t) = (2h)-2 JJ A r) da dT.
s-h t-h
Now, let R: a < s < b, c _ t ' d, be any cell interior to D. We may choose
numbers ho > 0 and Bo > 0 so small that the transformations
{u = u19p0(s, t) = u(s, t) + a[U(S + 0, t) - u(s, t)]/,,

(2.7) v = v1 08 ?(s t) = v(s, t)

(2.7)(U= U t = u(s + , t)
V = V2, a,(S, t) = v(s, t) + a[V(S + 3,B t) - v(s, t)/1
are defined, are regular and of class C', and the functions ui o, and vi , f and the
inverse functions satisfy a uniform Lipschitz condition independent of i3 and a
onthecell[a-hc-h;b+h,d +h]C D, provided0 ? h ? ho, 181 - goI
I a I _ I ,0 1I 3Oa ? 0. It is also clear that Tj~o and T2,0,. (and their inverses)
converge uniformly with their derivatives to T (and its inverse) on [a - ho
c-ho;b+ho,d+ho]as(3O-*0,a--*0. Wedefine
(2.8) 7r,,(S t) = p,[u1 = 0(s, t), V ( t)]. I u(s + I , t)+ - u(s, t) I / I 1,
+ P2[U2,0,B(S, t), v2,S,,(s, t)]. I v(s + ,3, t) - v(s, t) | / I ,B1.

For each i3 and a with I a I < 1I , _ o, IBa _ 0, 7rj,,,(s, t) is in La on [a -h,

c - ho ; b + ho , d + ho] and, for each h, 0 < h < ho ,we conclude that 7r1,ftj,,,h(8 t)
converges uniformly on R to Wlh(S, t) as , (and hence a) -+ 0, all the functions
being continuous in (s, t; ,, a) for (s, t) e R and 0 _ I a I _ I ,B < ,Bo ,13 ? 0.
Now suppose (s, t) e R, (s + 3, t) e R, and 1 ? < 1o, and choose h, 0 < h ? ho.
Then, for almost all (Q, -), I I I < h, I - I _ h, x[u(s + S + 1, t + nq), v(s + t + ,
t + 7n)], x[u(s + S + 1, t + 7n), v(s + {, t + 7n)], and x [u(s + #, t + 1),
v(s +t, t + nq)] are all defined and

x[u(s + + 01 t + -0), v(s + + 13 t + 77)] - t[u(s + $, t + n),

v(s + i, t + i7)] I

< I[u(s + + 0, t + 7), v(s + + , t + )]- [u(s + + 3, t + 7i)

v(s + $, t + n)]

+ L4u(s+ t +13,t+ r)),v(s+ ,t+ 4)]-x5u(s + ,t+ r)),

v(s + $, t + n)]
rU(8+t+ft't+17)
(2.9) < pl[u, v(s + , t + n)] du
U (8+tI t++)

oV(S+t+04,+17)

+ p2[U(S + t+ ) t + y), v] dvI

V (a+t, t+P)

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 826 CHARLES B. MORREY, JR.

= fP{pd1u1,3(s + t, t + y), vI 13,c(s + #, t + j u(s + t + ,3 t + 7i)

-u (s + {, t + ti) / 13! } do-

+
+ {P2[U2,0,, (S + t, t + -q)7 V2,0,S (s+ ,t + 77)] - |v(s + t+ t + s

-v(s+ ,i+ d) 1/1131 do-

= fiv(s + t t + 7)da

referring to (2.7) and (2.8), all the integrals existing as Lebesgue integrals.
Thus, by integrating (2.9), we see that
hh
(2h)2 f I[U(s + + t + 7), v(s + + ,t + 7)]
- T[u(s + (, t + r), v(s + (, t + 77)] | dt drl

(2.10) = (2h)-2
(2.10) A h+ yst:tv
-h -h
(2h)L- L4I( + 013 t +-~7n)
~ ~~~dd
-j(s + ~, t + 7n) I d~ ~

-< f 7r1,$la,h(,s t) da.

Novw, choose e > 0, choose , so small that I 7rl,O,,,h(s, t) - 7rlh (s, t) I < e/2(b - a)
onR,1I31 < B,leta_^y< <b, andlety=so<s1< ... < Sn = beasub-

i=l
division of [-y, 6] such that 13i = si-s < ,, i = 1, * , n, and

7r1, h (S tO (Si - 1) - f 7r1,h(s, t) ds < e/2.

Then
hh
(2h) f2L f i(6 + , t + 7) -y + (, t + 7) I dt d7
hh
nhh
E (2 h -2 h Ig(si + , t + 7i) - (si + - , t + 7) |d d7d
i=l h h1
n $- ~~~~~~~n
i=l 1
< 7 | tsr1,a,7,,h(s5-1 , t) da < Z (Si- t) * (si - si-) + e/2

< lrl,h(S, t) ds + e.

Since e is arbitrary, we conclude that

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 827
h hz
(2h)-2 f f) | Y( + Z, t + 7n) -y(7 + 4, t + 7l) 1 dZ d77
hh

< 7rl,h,(s, t) ds, a < < a < b;

(2.11) h h
(2h)-2 f[i (s + , + X)- s + , X + 7) I dZ d-o
hh

< f72, h(8, t) dt, c < X < , < d.

From this point on, we show that g(s, t) is of class $2 on D exactly as this was
done in the proof of Theorem 2.2.
Now, let us neglect the points where x and g are not defined and let Si and
S' be the sets of points (u, v) and (s, t), respectively, for which t(u, v) and 9(s, t)
are in Mi ; then Si is the transform of S.. As we have seen in Lemma 2.5 all
the points (u, v) of Si except those in a set Z1,j of measure zero are such that there
exists an a (u, v) > 0 such that all the points (u + h, v) and (u, v + h) also belong
to Si if I h I < a(u, v). Thus, choose e > 0. We may first choose a set S1,i C
S. with m(S1i,) > m(Si) - e/3 such that there is an a > 0 such that if (u, v) e
Su,, then all the points (u + h, v) and (u, v + h) with I h I < a belong to Si .
Now the partial derivatives x1X, x , Y X and yt all exist (the latter at the
corresponding points) on a set S2,i C SIi with m(S2,i) = m(S1,i). Now, using
Egoroff's theorem and the possibility of approximating measurable functions by
continuous functions, we may find a set S3,i C S2,i with m(S3,i > m(S2,i) - e/3
on which ?(u, v), x , and x are all continuous. Since these partials all exist
at every point of S3,i, we may (again using Egoroff's theorem) pick a closed
subset F. of S3,i with m(Fj) > m(S3,i) - e/3 > m(Si) -e such that x, xiiu

hh
and x1 X are all continuous on F.;, the partials Y and ,t all exist at each cor-
responding point and the difference quotients.

x, (u + h, v) - x,(u, v) xi(u, v + h) - x(uv)

all tend uniformly to the corresponding derivatives for (u, v) on Fi.

For almost all points (u0, vo) in Fi, the linear metric density of the part of
Fi on the lines u = uo and v = vo and the curves s = so and t to through (uo, vo)
is unity at (uo , vo). Thus also the projection on u = uo or v = vo of the part
of Fi on the curve t = to , say, is also of metric density unity at (uo , vo). Suppose
the projection on v = vo has this property, then the product of this projection
with the points of Fi on v = vo has this property as does the set of points s on
t = to which correspond to points whose projections are in the product set. For
such values of s = so + As, we have
4i (so + As, to) - q (So , to) = X i(uo + AU, vo + AV) - X,(uo, vo)
= [X(UO + Au, vo + AV) - (U0 + AV, vo)] + [Xt(Uo + Au, vo) - (U0 , vo)]

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
828 CHARLES B. MORREY, JR.
= ?,(UO + AUu, Vo) A\V + IE3?V + ,,u(Uo, Vo)A\U + EIAU

= z U , vo)AU + t v(U0,Vo)Av + EIA/U + E2AV, limren = 0, i = 1, 2,

A8-0

it being understood that as -* 0 in the above set. A similar argument for other
derivatives and the arbitrariness of e show that the usual formulas for change of
variables hold almost everywhere on Si and S' . Thus the usual transformation
formulas for E', etc., hold almost everywhere on B'.

7. In this section, we introduce the notion of boundary values of functions of

class M3a on a region B of class D' and carry over some of the results of [13], ?7.
THEOREM 2.6: Let x(u, v) be of class Ma on B. Let C be any arc or simple closed
curve of class D' which is interior to B. Then t(u, v) is defined almost everywhere
along C and is of class La with respect to arc length along C.
PROOF: It is evidently sufficient to consider only the case that C is a regular
arc of class C'. We may map a region A C B which includes C, onto a rectangle
R: I s I < a, I t I < c, a > 0, c > 0, by a regular map u = u(s, t), v = v(s, t) of
class C' so that C corresponds to the segment t = 0, I s I < a. From Theorem

aa aat
2.5, it follows that y(s, t) = x [u(s, t), v(s, t)] is of class l3a, a ndg(s, t) = x [u(s, t),
v(s, t)] on R. Since g (s, t) is AC in s for almost all t and AC in t for almost
all s (along interior segments), we see that 7(s, 0) is defined for almost all s and
is measurable since the AC functions g(s, tn) -y g(s, 0) for almost every s and
a suitable sequence { tn 0. Also

(2.12) f i (s, t) -9(s, 0) la ds < ta1 fft Ii(s 7) la dn, 0 < t < c

which -* 0 as t -O 0. Thus g (s, 0) is in La. The theorem follows by applying

the inverse transformation.
We sum up our results on boundary values in the following theorem:
THEOREM 2.7: Let x(u, v) be of class Ma on a region B of class D' and let x(u, v)
be extended arbitrarily to B. Then (1) t(u, v) is defined almost everywhere (with
respect to arc length) on B*. (2) If uo is not in a set of measure zero then the inter-
section of B with the line u = uo is either null or a finite number of non-abutting
open segments with endpoints on B* and t(uo, v) is AC on each of these closed
segments. (3) If u = u(s, t), v = v(s, t) is a regular transformation of class C'
of a region D into B, then 9(s, t) = t[u(s, t), v(s, t)] for almost all (s, t) along D*,
the two functions being defined at exactly the same points. (4) If B is a cell
[a, c; b, d], then
rd

lim (u, v) - x(a, v) a dv = 0

u --a+c

and similar results hold for the other sides of B.

PROOF: It is well known that if uo is not in a set of measure zero, the inter-
section of u = uo with B is null or is a finite number of non-abutting segments

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 829
with endpoints on B* at each of which B* has a definite tangent which is distinct
from the line u = uo . Let uo be such a number and suppose, for example, that
(uo, vo) is an upper endpoint of a segment of u = uo in B. Since (uo, vo) is
not a corner of B*, we may express the part of B* sufficiently near (uo, vo) in
the form v = f(u), uo - h ? u ? uo + h, where h > 0 and f(u) is of class C'.
If h is chosen small enough, there exists a v1 < f(u) for u - uo I < h such that
the region i\; u - uo I < h, v1 ? v ? f(u) is a subset of S. The transformation

s = u - uo, t = (v - V1)/[f(u) -vi]

maps a on the rectangle R: I s h, 0 < t < 1, and maps a regioni'D i
with i'B = A, on the rectangle R': I s I < h, 0 < t _ 1 + k, this trans-
formation being regular and of class C' on R' if k is small enough. Let
g(s, t) = t[u(s, t), v(s, t)] be the transform of t(u, v) as defined on A; ?(s, t)
is of class 83a on A. Since I yt I' is summable on R, it is summable in t for
almost all s, so that g(s, t) tends to a value which we call y'(s, 1) for almost all s.
Now define y'(s, t) on the whole of R' by

y'(s, t) = 9(s, t), 0 < t < 1; y'(s, t) =9(s, 2-t), 1 < t < 1 + k,
and define x'(u, v) as the transform of y'(s, t). It is evident that y'(s, t) is of
class 8 on R' so that x'(u, v) is of class 83a on A' and that V'(u, v) = x(u, v)
on i. Since any arc v = f(u), I u - uo I _ h' < h of B* is interior to i', the
first two statements follow from Theorem 2.6 and the definition of N(u, v). The
third statement is obvious since regular families of sets correspond and the last
statement follows, using (2.12).
We may now state an important substitution theorem which follows imme-
diately from Theorems 2.6 and 2.7.
THEOREM 2.8: Let x(u, v) be of class 93a on a region B of class D' and let A be
a subregion of B which is of class D'. Let y(u, v) be of class 83a on A and suppose
that g(u, v) = z(u, v) almost everywhere along A*. Let z(u, v) be defined on B by

z(u, v) = y(u, v), (u, v) e A; z(u, v) = x(u, v), (u, v) e B -A

Then z(u, v) is of class 83a on B, 2(u, v) = t(u, v) almost everywhere on B* + A*
and 2(u, v) = 9(u, v) on A*.

8. In this section, we introduce the notion of weak convergence and prove

some important results concerning it.
DEFINITION 2.8: A sequence { xn(u, v) } is said to converge weakly in $a(a > 1)
to x(u, v) on B if
(i) each xn(u, v) and x(u, v) are of class ha on B,
(ii) Da(Xn , B) is uniformly bounded,
(iii) xn(u, v) -> x(u, v) strongly in La on each cell X interior to B.
If a = 1, we require in addition that the set functions Da(Xn X e) be uniformly
AC on B.
THEOREM 2.9: If xn(u, v) converges weakly to x(u, v) [xn(u, v) -> x(u, v)] in

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
830 CHARLES B. MORREY, JR.
813a on B, the convergence holds on any subregion of B and weak convergence in 93a
is invariant under regular transformations of class C'. If B is of class D', it is
also true that xn(U, v) 4 x(u, v) (strongly) in La on the whole of B and tn(U v)
tends (strongly) in La to x(u, v) along any arc or simple closed curve C of class
D' in B.
PROOF: The first statement is obvious. We first prove the second for a
cell B: [a, c; b, d]. Let {xp(u, v)} be any subsequence of {xn(u, v)}. Using the
convergence in La on any cell [&, y; p, 6] interior to B and the diagonal process,
we may extract a subsequence {xq(u, v)} of {xp(u, v)} such that tq(u, v) tends to
N(u, v) in La on any segment u = uo , y < v ? 6, for which a < uo < b, c < y <
a < d and uo is not in a set of measure zero. Now choose e > 0. By referring
to (2.12) (and using the uniform absolute continuity of D(xnX e) if a = 1), it
follows that we may choose a 6 > 0 such that

(2.13) d
d

f i.(Ul , V) - n(U2 X V) la dv < e,

f Lx(ul,v) - x(u2,v) ladv < E, U1 - u1 <5,

rf ra a
independently of n. Thus, let c < y < 8 < d and choose a and f, a < a <3 < b,
so that a - a < 8, b - f < 6, and so that x v) and xn(3, v) tend in L. Ofl
[ey, 6] to t(&, v) and t(j3, v), respectively. Then there exists a q(e) so large that

LJL5 I xq(u, v) - x(uv) Ia du dv < e, L I xq(&, V) - (&, V) ad < e,

(2.14) a
| 2 (3, v) -X v) ladv < e

for all q > q(e). Then, from (2.13) and (2.14), it follows that
f ^ > l/a f ^ 5 u~~~~~~~~~~~~~~1a
fL I q(u v) - (u, v) Iadv | I q(u, V) - q(uov) Iadv}

(2.15) + { I tq(U0, V) - (U0, V) Iadv

l | I(U0 V) - (U, V) lIdV} < 3a11C

a _ u _ l < u _ b, q > q(e)

where uo denotes & or fi according as u < & or u > d. Thus
ba
LbL5I g (U, V) -(U, V) Iadudv < e(1 + 3a 2), q > q
Thus x q(u, v) x(u, v) in La on [a, ey; b, 8], and also Xja, v) and Xq(b, v) tend
strongly in La on [y, 6] to x(a, v) and x(b, v). As above, we may then choose a

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 831
further subsequence t,(u, v) which tends strongly in La to t(u, v) on [a, b] for
almost all v, c < v < d. A repetition of the argument above shows that
xt(u, v) -> x(u, v) on B and also along the top and bottom sides of B. Finally a
further subsequence t8(u, v) may be obtained which also converges to X(u, v)
all around B*. Since {x,(u, v)} was any subsequence of {x,,(u, v) 1, it follows
from a well known theorem on mean convergence that the whole sequence
x"(u, v) -* x(u, v) in La on both B and B*.
Now, if B is any region of class D', we may write B = , where the Bi
are non-overlapping and each is the image of a cell under a regular transforma-
tion of class C'. Since weak convergence is invariant under such transforma-
tions, the conclusions of the first paragraph apply to each Bi, and hence to B.
If C is any arc or simple closed curve of class D' in B, we may find a region A
of class D' in B of which C is part or all of the boundary. The theorem is com-
pletely demonstratee.
THEOREM 2.10: If Xn(U, v) -* x(u, v) in 83a on B, then Da(x, B) < lim infnx
Da(x, , B).
PROOF: From Theorem 2.9, it follows that spi,n(R) -> pi(R), i = 1, 2, for every
cell R with R C B. Now, define

tJ.[Pn (R) = ipm(R) + 2n(R)Ial2/[m(R)Ial

(2.16) #(R) = [<(R) + 02p(R)Ia 2/[m(R)]a-
Now, we recall the inequalities

[(al + a2)2 + (b, + b2)2]* ? (a2 + bi)1 + (a2 + b2

(2.17)
(2.17) 1 (a~(a, + a2)
+ 5_' <*~ ++' -i
= a-a1 a a. > 0, bi _ 0, cS > 0, i= 1, 2

{ [ff fi du dv] +[ff f2du dv] ? ff (f2 + f2 du dv.

Using (2.17), the Holder inequality, and the absolute continuity and normality
of the spi.n(R), we see that if R = R1 + R2 (non-overlapping), then

IP(R) ? {([{p(R,) + zp1(R2) }2 + {<p2(Ri) + sp2(R2) }2 i}a/[m(Ri) + m(R2)]al

? {[pol2(R1) + p2(R1)]? + [ypi(R2) + p 2 (R2)]1Ja/[m(Ri) + m(R2)]1

_ #(R1) + t(R2),

(2.18) #(R) [m(R)Ia - [{[ff Dv du dv] + [ D2 du dv]}

< [f f {(Dv, )2 + (Dp2)2}1 du dv]

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
832 CHARLES B. MORREY, JR.
similar inequalities holding for the ip(R) and for R = R1 + * + Rk , k > 2.
Furthermore it is evident that

D-n= [(D(p1,n)2 + (DP2, n)2]al2, DA = [(D(pi)2 + (DeP2)2Ja12.

From the normality and absolute continuity of A and An, proved in (2.18),
it follows that

VI(B) = Da(xn, B), V; (B) = Da(x, B).

Now, choose e > 0 and choose non-overlapping cells RI l , RN in B such
that

Z a(Ri) > V{(B)-E = Da(x, B)-it

Then, since An(S) -> 4,(R) for each R, we have

IV N
n --x oo n- oo n-oo i=1
lim inf Da (xn , B) = lim inf VP,,,(B) ? lim E An(R) = I1 7p(R) > Da (x, B)-e.

The theorem follows from the arbitrariness of e.

9. In this section, we present a theorem on the uniform boundedness of a

family of functions of class $a on a region B. The theorem follows easily from
the following lemma:
LEMMA 2.6: Let {x(u, v) I be a family of functions each of class h.3a on
R: [a, c; b, d] with Da(x, R) uniformly bounded and let xo be any point of W. Then

(i) if ff i x(u v) xo 1 du dv is uniformly bounded, so is / |(s) - x0 a ds,

(ii) if f i x(u, v) _ 1a du dv is uniformly bounded for some cell - in R,

or if / i (s)-Xo I ds is uniformly bounded for some set e on ft* which is open

on R*, then so is ff I x(u, v) - X0 I du dv.

PROOF: This is proved precisely as is Lemma 8.4 of [13].

THEOREM 2.11: The lemma above holds if R is replaced by any region B of class
D'.
PROOF: We may write B = E~LBi where the Bi are non-overlapping and
each is the image under a regular transformation Tj of class C' of the unit square
Q; if Bi B* #z 0, then B1i B* corresponds under ri to one or more entire sides of
Q. Then statement (i) follows immediately from (i) of Lemma 2.6. If
ff2 x(u, v) Xo 1a du dv is uniformly bounded, then ff xj(ui, vi) -
I0 I aduidvi is uniformly bounded for some rectangle a; in Q, xi(ui, vs) being the
transform of x(u, v) under some ri. Then, by (i) and (ii) of Lemma 2.6, it

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 833

follows that ff i x(u, v) - xo a du dv and also i(s) -o ds are uni-

formly bounded. If Bj*, B,* contains an arc, it follows by using rj and (ii) of
Lemma 2.6, that ff i x(u, v) - xo Ia du dv is uniformly bounded. Proceeding

in this way, we see that ff | x(u, v) - xo Ia du dv is uniformly bounded. If

I i x(s) xo ia ds is uniformly bounded for some set e, open on B*, it follows

that Iixi(si) - xo a ds1 is uniformly bounded for some e, on Q*, ej being open

on Q*. Thus, by (ii) of Lemma 2.6, it follows that ffi xi(ui, v) - xo Ia du,

dv; is uniformly bounded and hence as before that f i x(u, v) - xo Ia du dv

is uniformly bounded.

10. We now prove the following important theorem:

THEOREM 2.12: Let {x (u, v)} be of class 9ta (a > 1) on B with Da(xn , B) and

f l t xn(u, v) - o Ia dudv uniformly bounded. Then there exists a function

x(u, v) of class 3a on B and a subsequence {xp(u, v) I of {xn(u, v) } which converges
weakly in 9,a on B to x(u, v).
PROOF: We first prove the convergence in La for a cell R: [a, c; b, d]. Let M

be a uniform bound for Daw(x.n, R) + f I x,(u, v) - xI adudv. For each n

and each integer k > 0, let Unk be the set of values of u such that xn(U, v) is
AC in v on [c, d] and
d

(2.19) f [ I tn v(u v) a + ( v) - ia] dv ? k.

Then, for each n and k, m(Un,k) _ b- a - M/k. Let Uk = lim SUPn-. Unk =
Ht=iZ#0=n Uj,k; Uk consists of all u which are in infinitely many of the U,,k.
Let U = ZkX=lUk; then clearly m(Uk) > b - a - Mik and m(U) = b - a.
Moreover if u e U, it belongs to U' for some k and hence there exists a subse-
quence {x,(u, v) } of { n(u, v) } for which (2.19) holds the tu, v) then being equi-
continuous in v on [c, d] with I t,(u, v) - xo I uniformly bounded.
For each p, divide [a, b] into 2P equal closed intervals 1,j and choose upj in
I,,j. U. Clearly u,,j lies in one (or both) of the intervals Ip+1,2j-l or Ip+1,2j ; we
shall choose the {up j, so that Up+l,2j-l = up j in the first case or up+1,2j = up
in the second (or both). From the results of the first paragraph, it follows that
we may extract a subsequence {xp ,(u, v)} of {x,(u, v)} such that tp qup j, v)
converges uniformly in v on [c, d] to some continuous function Upj(v), j = 1, ... *
2P; we shall assume also that {Xp+1,q} is a subsequence of Ixp, I for each p so that

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
834 CHARLES B. MORREY, JR.
Xp, ,(u, v) = xp, ,(u, v) for some q' > q, provided p' > p. Finally, we define our
final subsequence {xp(u, v) } by taking xp(u, v) = xp p(u, v). We shall show be-
low that {xp(u, v) } is a Cauchy sequence in La on R and hence converges in La
on R to some function x(u, v) in La on R.
Choose e > 0 and choose pi so large that [(b - a)/2P]a-1 < E/3aM(b - a) for
p 2 p. p Next, choose q, so large that
d

fdl jp ,q(Up j, a) - 1pi,r(Up,j, v) ladv < E/3a(b-a), j= 1, 2' q r > qq.

Let P be the larger of pi and q1 . Then

d

fdVq(Upjj 2 V) -tr('Up1,j V)dv

(2.20) = / zpjq'(Up1j , V) - ,r (Upj y v) a dv < E/'3(b - a),

j = 1 . ,2p; q'r > P, >.... . q'>, r'_r
Moreover, if u e I,, we have (as in (2.12)),
1

f 1 up(, V) - T(utP1,j 2 V) la dv ? I u - upj a-AD2(Xp, R,

(2.21) C
< E/3a(b-a), p= 1, 2,..

d I /a d 51/a
Thus, if a < u ? b and j is chosen so that u e I,, we have (from (2.20) and
(2.21))

{fd( I ZQ(U, V) - (U V) ; adv} < IZ(u V) -Ziq(Upjj, v)ladv}

d V) 11 a 1
(2.22) + {f q(Upjj V) - r(Upij, v) la dv
+ {jd ]Ku, ( ) - i r(Ulpiwj , V) lad < [e/(b - a)l/a.

Integrating the a power of (2.22) with respect to u, we see that

ffI q(U, V) -yxr(U, V) adudv < e, q, r > P.

R

Thus {x,(u, v)} is a Cauchy sequence in La.

Now, suppose B is a general region of class D' and let B = ZEi=1Bi where the
Bi are non-overlapping and each Bi is the image of the unit square Q under a
transformation ri- which is regular and of class C'. Using these 'ri and the first
part of the proof, we see that there exist functions xj(u, v) of class La on each Bi
and a subsequence {xp(u, v)} of lx,(u, v) } which converges in La, to functions
xj(u, v) on Bi, i = 1, * , k and on any rational (and hence any) cell in B. It

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 835
follows easily that these xi(u, v) join up to form one function x(u, v) in La on B
and that x,(u, v) -* x(u, v) in La on B.
Now, by referring back to (2.22), we see also that {xp(u, v) } converges in La
to something on each segment u = uo , c ? v < d. By choosing a further subse-
quence, we may assume also that this convergence holds on segments v = vo,
a < u ? b. Clearly, then it must happen that if uo and vo are not in sets of
measure zero, {tp(uo, v)} -> x(uo, v) in La on any segment c < v < d in B and
{xp(u, vo)J -* x(u, vo) in La on any segment a < u < b in B. Thus, the cell
functions (pi,(R) (i = 1, 2) tend for each R to cell functions sp^(R) which are
given by (2.2) for almost all R in B. Since the D,(xp , B) are uniformly bounded,
it follows that the (pi,(R) are uniformly absolutely continuous and normal so
that the same holds for op1(R) and sp2(R). Moreover, the functions 4Jn(R) of
(2.16) are of uniformly bounded variation and tend to 6(R) for each R. An argu
ment like that at the end of the proof of Theorem 2.10 shows that 46(R) is f
bounded variation from which we conclude, using Lemma 2.4, that

+ a: > V+(B) >f [(D<o1)' + (D2)]] dudv.

B

Thus x(u, v) is of class <a on B and {xp(u, v)I -* x(u, v) in Sala on B.

11. In this section, we prove a theorem concerning the continuity of t(u, v).
THEOREM 2.13: Let x(u, v) be of class 2 on B, let C(po , a) C B (po (uo, vo)),
and suppose that

(2.23) D2[x, C(po, r)] < k2(r), fp7lk(p)dp = K(r) < oo., 0 < r < a.

Then x(uo, vo) is defined and, for almost all 0, N(p, 0) is AC in p for 0 < p _ a
with tO, 0) = x(uo , vo), (p, 0) being polar coordinates with pole at (uo , vo). Thus,
for almost all (u, v) in C(uo , vo ; a) we have
I

I t(u, v) - t(uo , vo) I < r f l[(1 - t)uo + tu, (1 -t)vo + tv]dt,

l(u, v) = [E(u, v) + G(u, v)]", r = [(u -uo)2 + (v - VO)2]"2

and the integral exists as a Legesgue integral.
PROOF: From the change of variable theorem, it follows that 12(r, 0) =
Ir 12 + r-2 | ;e 12 almost everywhere and that
r 2r
(2.24) D2[x, C(po, r)] = 101 | [l T(P 0) I 2 + p2 1 Xe(pI 0)12] dp dO.

Define

(2.25)~~ ~ h r o
r 2 [0
(2.25) ( ) | | ~1/ 2[ rX (p Ia) 1 2 + p-2 1 'TO(pI a) i|']1 dpdO.

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
836 CHARLES B. MORREY, JR
Then, from (2.23), (2.24), and the Schwartz inequality, it follows that

(2.26) h(r) ? (27rr)12k(r).

Then
2 2x- r2
Jr l(r2, 0) - (r, 0)I dO ? I [IX (r,) 12 + r-2 e(r, 0)12112 drd0

(2.27) = f h(r) dr = r2 h(r2) -_rL 2h(ri) + 2' 2 r312h(r)dr

? (27r)"2[k(r2) + K(r2)], 0 < ri < r2 < a

which -O 0 as r2 > 0, 0 < r1 < r2 . Thus there is a function x(O, 0) in L1 on
(0, 27r) such that

(2.28) lim x (r, 0) - t(0,) 0 dO = 0,

r27

on account of Lemma 2.2. Also, by referring to 2.23 and 2.24, it is seen that
there is a sequence {r} -O 0 such that x(r,, 0) is AC in 0 for each n and

(2.29) limr f (rn;n-oo0)O I dO = O.

2rt

Thus x(O, 0) some constant xo for almost all 0. Integrating (2.28), we see that
r2r

r-0
lrm (rr2Yf p| t |(p,0) - xo I dpdo = 0

so that x(uo, vo) is defined and equals xo = x(0, 0) for almost all 0.
Now, for almost all 0, x(r, 0) is of class $2 in r for ro ? r < a for any ro > 0,
and 1(r, 0) = [ (r, ) 1 2 + r-2 te(r, 0)12112 for almost all r. From (2.27), it
follows that

ra r2 a

1010 l(r, 0) drd0

exists as a Lebesgue integral so that l(r, 0) and hence I Jr(r, 0) I is summable in

r for almost all 0. Thus, for almost all 0, we have for all r that
r r ~~~~~~~~~~~~~~~~1
I(r, 0) -x(uo, vo) I < ]tr(Pj 0) I dp < j (p, O)dp = r J l(tr, 0) dt

from which the lemma follows.

DEFINITION 2.9: A function x(u, v) of class $2 on B is said to satisfy a condition
A2[X, M2(a)] (X > 0) on B if

D2[x, C(p, r)] < M2(a) (r/a), 0 < r < a

for all p in B at a distance _ a from B*.

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 837
DEFINITION 2.10: A function x(u, v) defined on B is said to satisfy a condition
B[g, N(a)] (,A > 0) on B if

x(Ul, V1) - X(U2, V2) I _ N(a). (r/a)', 0 < r < a,

r = [(U2 - U1)2 + (V2 - V1)2]2/

for every pair of points (u,, vl) and (u2, v2) in B such that every point of the
segment joining them is in B and at a distance _ a from B*.
It is clear that if x(u, v) satisfies a condition Bgu[, N(a)] on B, it satisfies a uni-
form Holder condition with exponent A on any closed subset of B.
THEOREM 2.14: Let x(u, v) be of class $2 and satisfy a condition A2[2X, M2(a)]
on B. Then x(u, v) is defined at each point of B, is of class $'' on B, and satisfies
a condition B[X, N(a)] on B with

N(a) = 21- X1[7rI2(a)]1/2,

PROOF: That t(u, v) is defined at each point follows from Theorem 2.13.
Thus, let (u1, vl) and (u2, v2) be any two points as in definition 2.10 and rotate
and translate axes so that (u, , vl) is (0, 0) and (u2 , v2) is (a, a), a = -1/2r. Now

( Ix(a, a) - (O,0 ) I _ | (uv) -x O,0) | + (u, v) -x(a, a) ,

From Theorem 2.13, it follows for almost all (u, v) that

11
l ~(u, v) -~(0, 0) l < (u2 +I-. v2)"2 ] l(tu, tv) dt ? 2112a ] l(tu, tv) dt

(2.31) x(u, v) - (a, a) | < 2 /2af l[(1 - t)a + tu, (1 - t)a + tv] dt
l(u, v) = [E(u, v) + G(u, v)]"2.

The remainder of the proof follows from the double integration of (2.30) and
(2.31) exactly as was done in the proof of Theorem 2.1, Chapter II, of [14].

CHAPTER III

SOLUTION OF THE PROBLEM OF PLATEAU.

1. In this chapter, we show under certain conditions that there exists a sur-
face of type k in 9W which is bounded by k given mutually exclusive simple closed
curves in DI and which minimizes the Lebesgue area among all surfaces bounded
by these curves. If DI is sufficiently differentiable, it is shown that the surface
is a minimal surface in the sense of Differential Geometry. The solution is
shown to exist by the method of Courant [5] with the necessary modifications.
We first prove the analogs of Courant's various lemmas and then deduce the
existence theorem from these lemmas.
In order to carry through this program, the writer has found it necessary to
restrict the generality of DI somewhat as indicated by the. following definition:

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
838 CHARLES B. MORREY, JR.
DEFINITION 3.1: A Riemannian manifold 9N on class C' is said to be
homogeneously regular if there exist numbers m and M, independent of xo, with
0 < m < M, such that any point xo in 9N lies in an open set in 9W which can be

N N'
mapped on the unit hypercube R: I xi I < 1 by a map of class C' such that xo
corresponds to the origin and the gi,j(x) satisfy the condition that

(3.1) m E (ti)2 _ 9CF :(X)tafl < E ) ()2

for all (x', * * N, x') on R and all (41 * * N)

It is to be noticed that any compact regular Riemannian manifold has this
property. A great many non-compact manifolds also have this property as for
example do Euclidean N-space and "hyperbolic N-space"'0. We say that W is
a hyperbolic N-space if the whole of 9 can be mapped on the half-plane xN >
o with
N

ds2 = (xN)-2 2 (dxi)2

i-i

It is clear that we may also map 9 on the interior of the N sphere aL(xt)2 <a2
say, taking any point into the origin. When this is done, the metric becomes
- N -12 N
ds2 = 4a2 a2 - E (xi)2DJE (dxi)2.

If a2 > N, this sphere includes R in its interior and we may take m = 4a-2 and
M = 4a2(a2-N)2 in (3.1). We shall assume throughout this chapter that 9
is homogeneously regular and of class C' with associated numbers m and M in defi-
nition 3.1.

2. In this section, we prove our first important lemma and introduce one of
Courant's lemmas [5].
LEMMA 3.1: If t(0) is periodic and AC on [0, 27r] with

(3.2) f 1 48(0) 12 dO < 2m/r

2r

there exists a function x(u, v) of class $'2' on a given circle C(po, a) such that
x(a, 0) = t(O) and
2wr

D2[X, C(po, a)] < (M/m)f 12 d@.

PROOF: Let {(0) satisfy (3.2), let 0 < Oo ? 21r, and map a neighborhood 9 of
9J onto R as in definition (3.1) carrying {(0o) into the origin. Then if P is in
)-91, it is clear that I P - t(0o) I _ m/ 2, since we may join P to {(0o) by a
regular arc of class C' of length I P - {(0) + e and not all of this arc is in 91.

10 See [7].

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 839
Then
2Tr

J I o (0) I dO < [2r 2m/lr]112 < 2m"'

so that I {(O) - t(4o) I < m112 for every 0 and t(O) e91 for every 0. Now
we define x(u, v) as the function of class $32' on C(po, a) whose components
with respect to 9 are harmonic on C(po, a) and coincide with those of t(O) on
C*(po, a). Then

D2[x, C(po, a)] = ff gac(x) (Sx~ + x44) dudv

i U iV n n
C (poa)

< M CL 2 [(Xt)2 + (Xi)2I dudv = Mar ,i: n[a')2 + (b')21}

cor? 2 r N
C ( pea)

i31 nl0
? M Cr n2[(a')2 + (b )2]} = Mf (Ni) do

< (M/m) gJst do = (M/m) I e(O)j2dO

where the as and b' are the Fourier coefficients of Zt(o).
LEMMA 3.2: Let B be of type k, let Ci be one of the boundaries of B, let {xn(U, v)
be a sequence of functions of class $3' on B with D2(xn? B) uniformly bounded, let
I?, be the system of curves x = xR(u, v), (u, v) e B*, and suppose rn-+ r (see def.
1.8) where F consists of k mutually exclusive Jordan curves. Let D be another
fixed region of type l, and map F and rn non-degenerately i.e. so that no arc is
carried into a point (possible unless one of the curves of Fn is a point, i.e. if
xn(u, v) is constant over some C, of B*) on D* by functions x = X(S) and x =
Xn(S) so that Xn(S) -+ X(S) uniformly on D*. For each n, let IN-: S = Sn(s) be
a continuous monotone transformation from B* to D* such that Xn[Sn(s)] = xn(s).
Then either the transformations IN are equicontinuous on C; or there exists a subse-
quence {I u-}, a fixed boundary Ai of D, a sequence {c,j,} of arcs of C; which tend to
a point of C;, and a sequence of arcs {6I ,} of Ai which tend to a point of Ai such
that rv(ci ,) = Ai - bi,. In other words rp, maps a small arc of C; onto practically
the whole of Ai .
PROOF: This is Courant's lemma 6, ?2 [5], and is proved exactly as is that
lemma.

3. In this section, we introduce one of Courant's lemmas [5] and prove the
existence of "harmonic" functions taking on certain boundary values.
LEMMA 3.3: Let x(u, v) be of class $32 on B with D2(x, B) _ K, let po be any point
in the plane, and let 0 < 6 < 1. For each r, let C,. = C*(po, r)-B. Then there
exists a number ?, 6 < f < 6a12 such that either Cr is null or t(r, 0) is AC in 0 on
Cr and

o l( , a) 1 2do _ 2K/Ilog a

so that the oscillation of t(f, 0) over each arc of C; is ? [47rK/ I log a i ]1/2,

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
840 CHARLES B. MORREY, JR.
PROOF: This is Courant's Lemma 5, ?2 [5], and is proved in exactly the same
way using the theory of Chapter II.
DEFINITION 3.2: Let B be of class D' and let x*(s) be of class L2 along B*.
If there is no function x(u, v) of class $2 on B and coinciding with x* on B* (that
is, V(u, v) = x*(u, v) almost everywhere on B*) we define d(x*, B) = + ct;
otherwise we define d(x*, B) as the greatest lower bound of D2(x, B) among all
functions x(u, v) of class $2 on B which coincide (in the above sense) on B* with
x.
LEMMA 3.4: Let B be of class D', let x* be continuous on B* and suppose that
d(x*, B) is finite. Then there exists a function x(u, v) of class $'' on B which coin-
cides with x* on B* and for which D2(x, B) = d(x*, B). Also, x(u, v) satisfies a con-
dition A [X, K] and hence a condition B[X/2, L] on B for some X > 0 which depends
only on m, M, and d(x*, B).
PROOF: Let {xn(u, v) } be a sequence of functions of class $2 on B such that
D2(Xn, B) -> d(x*,B), each coinciding on B* with x*. From Theorems 2.11 and
2.12, it follows that there exists a subsequence {xp(u, v) } of f xn(u, v) } which con-
verges weakly in $2 on B to some function x(u, v) of class $2 on B and, from
Theorem 2.10, it follows that D2(x, B) _ d(x*, B). From Theorem 2.9, it
follows that xp(s) x-> (s) strongly in L2 on B*. Thus x(s) = x*(s) almost every-
where on B* and hence D2(x, B) = d(x*, B). It remains to show that x(u, v)
is continuous on B, and hence of class $'3' on B, and coincides with x* on B* in
the ordinary sense.
Since x(u, v) minimizes D2(x, B) among all functions of class $2 on B which
coincide with it on B*, it follows from Theorem 2.8 that it must have this prop-
erty on every subregion A of class D' in B. Now, from Lemma 3.1, it follows
2r

that if f j (r, 0) 12 dO _ 2m/7r, then

27

d[t, C(po, f)]<(ll/r) I (rf, 0) 12 dO.

Since D2[x, C(po, a)] _ d(x*, B), a being the distance of po from B*, it follows
that
27r

(3.3) D2[x, C(po, r)] < 1 f I x(r, 0) 12 dO

for almost all r, 0 < r < a, where 1 is the larger of the numbers M/m and
7rd(x*, B)/2m. But now, let
r27
(3.4) p(r) = D2[x, C(po, r)] = f f p[ | Xr(p 0)! 2 + p2 1 Xe(p, 0) 2] dpdO,

0 ? r ? a.
Then sp(r) is AC for 0 ? r < a, and, from (3.3) and (3.4), we see that
Sp(r) < lrp'(r), p(a) _ d(x*, B)

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 841
for almost all r, 0 < r ? a. It follows, as in the proof of [14], Chapter III,
Theorem 6.1, that

(3.5) D2[X, C(po 2 r)] _< d(x*, B) * (r/a)" '.

Since Po was arbitrary, x satisfies a condition A[X, K] and hence a condition
B[X/2, L] on B which implies that x is continuous on the interior of B.
To show that x is two dimensionally continuous on B*, we proceed as in the
last part of the proof of Theorem 6.2, Chapter III of [14]. The proof there given
goes through without essential change. However we remark first that the
boundary value function for w(%, v) on C*(O, 1) is the transform of those of
z(x, y) on Be C(po, r); this follows from the analyticity of the map on the bound-
ary, except for the two corners. Since the Dirichlet integral is preserved by
conformal transformations, it follows that w(, v) is harmonic (see def. 3.3 below)
on C(O, 1) and hence satisfies a condition of the form (3.5) with the same value
of 1. This allows us to proceed with the proof as mentioned above.
DEFINITIoN 3.3: Any function x(u, v) of class $? on B such that D2(x, B) =
d(x, B) is called harmonic.

4. In this section we prove a weaker analog of Courant's Lemma 2, ?2 [5]

which, however, is sufficient for our purposes.
LEMMA 3.5: Let x(u, v) be harmonic on B with D2(x, B) = K, let (uo0 vo) = Po e B,
let a be the distance of Po from B*, let xo be any point of 9N, and let e > 0. Then
there exists a 6(e; K, a, xo) > 0 and, for each 6, 0 < 8 < 8 (e; K, a, xo), a function
xa(u, v) of class 3' on X and coinciding on B* with x(u, v), such that D2(xa, B) <
K + e and xa(u, v) xo for (u, v) E C(po, 6) C B.
PROOF: Let x(po) be joined to xo by a regular arc y of class C' of length 1. Then
there exists an open set % on 9)D which contains y which can be mapped by a
regular transformation of class C' on a region H
N-1
H: I(Xi) 2 < W, - h < xN < a + h

so that xo corresponds to the origin, x(po) corresponds to (0,. , 0, 1), y corre-

sponds to the segmentx = 0, i = 1,** N -1, 0-< xN < , and so that the
9a:(x) satisfy

i=31 2=
(3.6) 2' by (ti)2 < gao(X)tja < 2 j (ti)2
for all t and all x e H.
Now, let X = X(K; m, M) be determined as in Lemma 3.4 so that

(3.7) D2[x, C(po, r)] < K(r/a)', 0 ? r < a.

Let 6(K, a, xo) > 0 be so small that x(u, v) e 9 for (u, v) e C(po, 61/2) and C(po,
6 1/2) c B for any 6, 0 < 6 < 6(K, a, xo) [6(K, a, x) depends only on these things

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
842 CHARLES B. MORREY, JR.
since x(u, v) also satisfies a condition B[X/2, L] where L depends only on K and XI.
For each 6, 0 < a < S(K, a, xo), define

(0, 0 < r < 5 0 ? 0 27r

- 2(log )- log (r/) a 5 r < a/2, 0 <0 2Xr.

N 81/2 or N
Then if x'(u, v) are the components of x(u, v) with respect to our mapping,

f | 2 [(rXi)2 + (rX)] dud2 < 2J rf T2 [(x)2 + -2(X )2] d1 d

61/2 2v N
C( po,61/2)

(3.8) + 2 f 4r'1(log f)-2 J E (Xi)2 drdO ? 4D2[x, C(po, 61/2)]

+ 87r[h + (1 + h)2](-log 5)-1 < 4K(5112/a)" + 87r[h2 + (1 + h)21(-log a)-}

using (3.6) and (3.7).

Thus, if we define xa(u, v) = x(u, v) for (u, v) e -C(Po, 6112), and to have the
components rx' for (u, v) e C(po, 6112) we see from (3.6) and (3.8) that

D2(xa, B) < D2(x, B) + D2[xa, C(po, 51/2)] < K +

(3.9) 2 f| | [(TXi)2 + (-rxt)2 dudv _ K + 8K(6112/a)x
C( pO56/2) i1 + 16r[h2 + (1 + h)21(-log b)'
which is < K + e if 5 is sufficiently small.

5. In this section, we formulate our problem and prove a simple lemma con-
cerning lower semicontinuity.
DEFINITION 3.4: Let r be a system of k closed curves and let p > 0. If there
is no surface S' of type k with L(S') finite bounded by a system r' with
11 r, rF ' < p, we define lp(r) = + oo; otherwise we define lp(r) as the greatest
lower bound of L(S') for all surfaces S' of type k bounded by systems rF with
1 F, F' 1 < p. Finally, we define 1(r) = lim,..o+ l(r).
DEFINITION 3.5: Let r and p be as in Def 3.4. If there is no vector function
x'(u, v) of class " on some region B' of type k such that F': x = x(u, v), (u, v) c
B'* is at a distance < p from r, we define d,(r) = + a; otherwise, we define
d,(r) as the greatest lower bound of D2(x', B') for all functions x'(u, v) of class
$2 on some region B' such that 11 rF, r II < p where r' is defined as above.
Finally, we define d(r) = limpdo+ d,(r)
DEFINITION 3.6: If k = 1, we define d*(r) = + cc. Otherwise, we define
d*(r) as the greatest lower bound of DP1 d(rF) (2 < p < k) for all possible sub
divisions of the system r into non-empty, (logically) mutually exclusive sets

In the definitions above, we are not assuming that the curves of r are Jordan
curves or are distinct. We simply consider r as a system of curves (C *,
Ck). By the logically mutually exclusive sets rFt, . , rF in Def. 3.6, we simply

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 843
mean that we divide the integers 1, * *, k into mutually exclusive sets and then
take the corresponding Ci .
LEMMA 3.6: if 1l(r) is finite and r consists of k curves, there exists a sequence of
surfaces Sn of type k such that L(Sn) -> l(r) and rn 1,-+ r, rn being the boundary
of Sn. If d(r) is finite, there exists a sequence {xn(u, v) of functions, each of
class 92 on some region B. of type k such that D2((x. , Bn) - d(r ) and I, r, r,
being the boundary of the surface x = x.(u, v), (u, v) e Bn . Furthermore d(r) =
2l(r) and both are lower semicontinuous with respect to convergence of r. to r.
PROOF: The first two statements are immediately evident from the definitions.
Since for any x(u, v) of class $'I on B, we have D2(x, B) > 2L(x), we see that
d(r) > 21(r). On the other hand, choose surfaces Sn: x = x(u, v), x*(u, v) of
class j12' such that L(S,,) -I1(r) and rn--- r. From Theorem 1.2, it follows that
each Sn may be represented on a region Bn bounded by circles by x = xn(u, v)
in which xn(u, v) is of class $3. on Bn with D2(xn,, Bn) < 2L(Sn) + n '. Thus
d(r) ? 21(r) and the equality holds. To see that l(r) is lower semicontinuous
in r, choose r and p > 0. Letil r, r' 11 < p/2 and choose r" with II r', r"I I <
p/2 so that there is a surface S" bounded by r" with L(S") < 14/2(r1) + p. Then
1f r, r" II< p so that 4(r) < lp/2(17) + P _ l(r') + p. If l(r) is finite, we may
choose p so small that lp(r) - p > 1(r) - e; if 1(r) = + a, we may choose
1,(r) - p > N for any N. This proves the result.
In Section 6, we shall show that d(r) < d*(r). The result above shows that
if we obtain a function x(u, v) defined and of class $32' on some region B of type
k for which D2(x, B) = d(r) < X where S: x = x(u, v) is bounded by r, then
L(S) = l(r) and S certainly is a surface of least area bounded by r and x =
x(u, v) is a "generalized conformal" representation of S on B. Accordingly, we
shall prove (in Section 11) the following existence theorem; Let r be a set of k
(>1) non-intersecting simple closed curves and suppose d(r) is finite and d(r) <
d*(r). Then there exists a region B of type k bounded by circles and a function
x(u, v) harmonic on B such that D2(x, B) = d(r). We shall close the chapter
with a discussion of further differentiability properties of such functions implied
by further differentiability hypotheses on 9W.

6. In this section we prove Courant's Lemma a, ?4 [5] (our Lemma 3.7) and
another lemma which is implied by his other results.
LEMMA 3.7: d(r) ? d*(r).
PROOF: If d*(r) = + o, the lemma is obvious. Now suppose that k > 1,
r = (C1, ,Ck), ' = (Cl, C,Cp) andr7" = (C+l,* Ck) 1 _ p < k,
and suppose that d(r') + d(r") < Oo. Now, let B' and B" be regions of type
p and k - p, respectively, each of which contains the origin, and let xf (u, v) and
x','(U, v) be of class $' on B' and B", respectively, such that D2(x , B') < d(17) +
e/4, D2 (C'' , B") < d(17") + e/4, It r', r1' < (, 1 r", 17" 11 < e where r' and
r1f. are the boundaries of the surfaces S': x X (u, v) and S'': X = X. (u, v). We
may assume x and x',' to be harmonic. Let xo be any point of W). From Lemma
3.5, we may choose a S > 0, and functions y'(u, v) and y" (it, v) of class $' on B'

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
844 CHARLES B. MORREY, JR.
and B" which coincide on (B')* and (B")* with x' and x", which equal xo on
C(0, 6) C B'* B", and such that D2(y , B') < D2(x'e, B') + E/4 < d(r') + E/2
and D2(y't', B") < D(x"', B") + E/4 < d(r",) + E/2. Now, let B' = B' -
C(0, 6) and let B"' be the transform of B" - C(0, ) by an inversion in C*(O, 6),
and let B = B' + B"' + C*(0, 6). Then B is of type k. On B, define xt(u, v) =
y'(u, v) on Da and to be equal to the transform of y'"' on Bf" . Then x.(u, v) is of
class W3t on X with II rF, r II < e and

df(T) ? D2(x , B) = , B') + D2(y'"', B") < d(rI) + d(r") + e.

Since this is true for every e > 0, it follows that d(r) < d(r') + d(1"). By in-
duction, it then follows that if d*(r) is finite, so is d(r) and d(r) < d*(r).
LEMMA 3.8: Letr = (Cl, ,Ck), k _ 1, and let Ci be the point xo. Then if
k = 1, d(r) = 0. If k > 1 and r' =. (cl) * * *,ci-1) ci+l, .. ** Ck), then dwr)
=d(r).11
PROOF: The first statement is obvious. From Lemma 3.7, it follows immed-
iately that d(r) _ d(wT) + d(Ci) = d(nT).
To prove the equality, it is sufficient to prove d(r') < d(r). If d(r) = + oo,
this is obvious. Thus, suppose d(r) finite. Let {xn(u, v)} be a sequence of
functions of class ?J3 ' on regions Bn of type k such that rn-- r (rn: x = Xn(p);
p eB*) and D2(xn , Bn) -*d(r). Since rnT- r and Ci = xo , it follows that there
is a boundary bi n of Bn on which I xn(u, v) - I < en, limn-- e1n = 0; using con-
formal mappings, we may assume that bin is always an inner boundary.
Now, let a neighborhood 9' of x0 be mapped on the unit cube I x9 I < 1 by a
map of class C' so that xo corresponds to the origin and the ga9 (x') satisfy (3.1).
By means of a further homogeneous linear transformation from the x' to the x

N N
space, we may find a subneighborhood 9 of V' which may be mapped by a map
of class C' onto a sphere ELZ (xj)2 < R2 so that xo corresponds to the origin,

j=1 j1
ga0(0) = 5aO (the Kronecker 6), and such that

m(r) II (Zj)2 _ ga8(X)aZ < M(r) Zj ()2, 2-1

? m(r) < M(r) < 2, lim m(r) = lim M(r) = 1,

r -0O r-0
for all x e C(0, r) and all Z, 0 ? r < R. Now, let {rn} and {Rn} be a sequence
of numbers -* 0 such that rn/Rn -* 0 and for all n > n1, rn < R. < R and
[4n(u, v)]2 < rn for (u, v) e bi,n. For each such n, we define a transforma-
tion y' = y'(x) of C(0, R) into itself by

,ziX Rn < r < R

yn'(x) = {Rn(r - rn)x'/r(Rn - rn), rn ? r < Rn r = [ (3)2]
0 < r < rn

11 For a similar procedure, see 111], Theorem 1, as corrected in [16].

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 845
Then, we observe that yn(Xi) - yn(X2) I< R? Ilx - X2 | -(R- rn) for xl, X2 e
C(0, R).
Then, for each n > n1, we define yn(u, v) = xn(u, v) for (u, v) on T3n when
xn(u, v) is not in 91. For x (u, v) e91, we define yn(u, v) as the function whose
components with respect to 9 are y' [x'(u, v), * , x'(u, v)]. Then we see that
each yn(U, v) is of class 43 on , that yn(U, v) = xo on bin X and that

D2(YnX Bn) = D2(Yn X dn) + D2(Yn X Bn- dn) _ D2(Yn, Bn- dn)
N

+ M(Rn) f f n 2 [(Ytu)2 + (ynv)2] dudv _ D2(Yn , Bn- dn)

ijl
dn

+ [R2M(Rn)/(Rn - rn)2] f fi [(x ,U)2 + (xi ,)2] dudV ? D2(xn, Bn - dn)

d,

+ [RnM(Rn)/(Rn - rn)2m (Rn)]D2 (Xn , dn)

where dn denotes the open set on Bn for which xn(u, v) e 9 and Yn(U, v) $ xn(u, v).
Thus, we see that D2(Yn , Bn) -+ d(r). Now, let Bn$ = Bn + bin + the interior
of bn ; then Bn is of type k - 1. Moreover if we extend Yn(U, v) from Rn to
BR by defining Yn = xo on the interior of bin we see that rF -+ rn and
D2(yn, B') -+ d(r), r': y = Yn(U, v), (u, v) e (B')*. Thus d(F') ? d(r) as was
to be proved.

7. Now, let r be a set of k mutually exclusive Jordan curves C1, ** , Ck, let
D be a fixed region of type k bounded by circles di, - -*, dk of which di is the
outer boundary and is the unit circle. Let t = t (q) be a topological mapping of
D* on r so that di corresponds to Ci . Let {x,1(u, v) } be a sequence of functions
of class $32' on regions Rn of type k such that D2(xn, Bn) -+ d(r) and rn --+ r,
rn: x = Xn(u, v), (u, v) e B n. Since rn -- r, it is possible to label the curves of
rn = Cln, - - *, Ck n so that Cin -* Ci, i = 1, - - *, k. Using conformal maps,
we may assume that each Bn is bounded by circles b1, , - - *, bk n of which b1,n is
the outer boundary and is the unit circle and bin is mapped on Cin , i = 1, -* *,
k. Since Cin -+ Ci, it is clear that none of the Cin is a point if n is sufficiently
large. Thus we may map the Con (n > n1) on di by non-degenerate maps Q =
{n(q) so that (n(q) --+ (q) uniformly on D*. For each n > ni, let q = rn(p) be
a continuous monotone map of B* on D* such that in[r (p)] X=Xn(p), p 6 B* . Let
a, f3, and y be three equally spaced points on the unit circle. By mapping the
unit circle into itself by conformal mappings, we may assume that a = rn (a),
13 = n(/3) by = rn('y) for each n > ni .
Before reducing our problem further, we state Courant's Lemma c, ?4 [51 which
may be proved exactly as he has proved it.
LEMMA 3.9: IWith the Bn normalized as above, each pair of the circles bi,n and
ban, whose radii remain above some positive bound independently of n, remain at
a positive distance from one another independently of n.

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
846 CHARLES B. MORREY, JR.
Now, for each i and n, let bi n = C*(pi,, , ri n). Since the pi,n and ri,, are uni-
formly bounded, we may, by extracting a subsequence (still called In}), assume
that the pin -t pi and ri - ri ; clearly Pin = Pi = (0, 0) and ri n = ri = 1.
Let us relabel the bi n , di, Ci5,n and Ci (leaving i = 1 fixed) so that none of the
ri for i = 1, 1 < k are zero and that ri = O for i > 1. From the lemma
above, it follows that the circles bi = C*(pi, ri), i = 1, , * * 1, bound a region B
of type 1. Using the transformations of Lemma 1.5 of the region B into the
regions Bn of type 1 bounded by the circles bi nX i = 1, * * 1, we may assume
that the circles bin X i = 1,*, 1, are the fixed circles bi, i = 1, *, lwhich
bound B. With the Bn completely normalized as above, the following cases of
"degeneracy" may occur.
CASE I-I < k and one of the pi with i > 1 is interior to B;
CASE II-1 < k and one of the pi with i > 1 is on a bj with 1 < j < 1 along
which the xn(u, v) are (one dimensionally) equicontinuous (the x, are equicon-
tinuous along bi on account of Lemma 3.2 and the three point condition).
CASE III-1 ? k and the x, fail to be equicontinuous along some bi , 1 < i < 1.
We shall consider these cases in the next three sections and shall prove the exis-
tence theorem in Section 11. In ??8-11, we shall retain the notations and nor-
malizations of this section.

8. In this section, we consider Case I.

LEMMA 3.10: Suppose Case I holds. Then d(r) = d*(r).
PROOF: Let Po = pi. Since Po e B, there is a number a > 0 such that po is at
a distance _ 2a from all the circles bi, * , b1 and all the pj for which pj F po.
Choose e > 0 and let K be a uniform bound for D2(xn , Bn). For all n > ni,
the circles bjn for which I pj - Po I _ 2a remain at a distance > a from po.
Now, let a be the smallest of the numbers

a, e,e
2 -2kr/2m -4kM/me

Then for n > n2 _ ni, it is also true that all the circles b j n for which pi = Po are
interior to the circle C(po , 6). Since a < a2, the annular ring C(po, (112) -
C(po, () is interior to B, for all n > n2. Thus, on account of Lemma 3.3, we
may pick an fn, a < n <_ 112 , on which x,(fr, 0) is AC in 0 with
2r

f I Xn1e(1 X, ) 1 2d6 < 2K/I log a I _ 2m/wr, me/2M, n > n2.

From Lemma 3.1, it follows that there exists a function yn(u, v) of class 3 on
C(po, Tn) which coincides on C*(po, fn) with Xn(fn, 0) and for which

D2[Yn, C(O) r,)] < e/2, n > n2 -

We let Zn be the function defined over the exterior of C(O, Tn) as the transform of
yn by an inversion in C*(po, f")._
Now for each n,vwe define B' = Bn - Bn-C(po, Br()B - B' + C(po, =B),
B" as the transform of Bn.C(po, f') under an inversion in C*(pi,n, ri,), and

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 847
n2) as the transform of Bn C1(po, fXn) + [I - C(po fr)] under this inversion,
H denoting the whole plane. Then, for each n > n2, B") is of type k1 = - k2
and B L2) is of type k2 where k2 is the number of indices j for which pj = po.
For n > n2 , we define x(1) on B(1) to coincide with Xn on A' and with Yn on
C(pO, fn), and define x(2 on Bn to coincide with the transform of Xn on Bfn' and
with the transform of zn on - B'n$. Then x(1) and x() are of class $ '2'and

D2(X'1) B(')) = D2(xn B') + D2[Yn X C(O, fn)] < D2(xn I B ) + e/2,

D2(X(2), B(2D) 2(Xn X Bn - B) + D2[yn, C(O, fn)] < D2(xnX Bn - B') + e/2,

since the Dirichlet integral is preserved by conformal mappings. Let r(l) denote
the curves C, of r for which I pj -po ] 0 and let r1(2) be those C, of r for
which pj = po. For each n > n2, let rpti) be the curves x = x(i)(p), p e B(,
i = 1, 2. Then rib) + r 2 = rnp, rib) -rl, and r 2 - r(2). Also, we have
D2(Xn , Bn) = D2(xn I B ) + D2(xn X Bn - B) > D2(X(1), B(1)) + D2(XA2), B(2)) - e.

Our lemma follows by letting n - o and using the arbitrariness of e, the defini-
tions of d(ri(1)') and d(r(2)), and Lemma 3.7.

9. In this section, we consider Case II.

LEMMA 3.11: If Case II occurs, d(r) = d*(r).
PROOF: Let pi e bj, 1 < j _ 1, i > 1, and choose bo so small that C*(pi, 5o)
intersects B* only at two points of bj and C(p*, bo) contains no other pX except
those which coincide with pi. For n > ni, all the bx,n for pA = pi are interior
to C(p*, bo) and the remaining circles (except bj) are outside C(pi, Qo). From
Lemma 3.3, it follows that we may choose a sequence An -- 0 such that for n > n2,
A n < bo , all the circles bxn with px = pi are interior to C(pi, Xn) and all the re-

maining circles are outside C(p, X n), and ' I Xnes 12 ds -+ 0. Since the
C*( Pi,80)Bn
xn(u, v) are equicontinuous along bj , we may, by choosing a subsequence, assume
that they converge uniformly on bj to a representation of Cj . For each n, let
r$l) be the curves of rn corresponding to the circles outside C(pi, Xn) without
Cjn, plus the curve C',n defined by

x = xn(u, v), (u, v) e b =n [bj -bj bC(p, , An)] + C*(pi Xn) -nX

let rF2) be the curves of rn corresponding to the circles bxn with px = pi, let On
be the curve x = xn(u, v), (u, v) e bn = bj*C(p XA) + C*(p X n and let
r(3) = r + aYn. Let r(l) be the curve Cj plus the CA for which px $ pi , let r2)
be the curves CA for which px = Pi, let y be the point of Cj corresponding to pi
under the limiting map of Cj on bj , and let r'3' = r(2) + y. Then r1 ( + r(2) =
r r'l') rl), r 2() r( 2), 2r3' r(3). Now

D2(xn Bn) = D2[xn Bn- Bn- C(pi X n)] + D2[xn X C(pi 1 An) ) Bn].

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 848 CHARLES B. MORREY, JR.
Letting n s0, we obtain

d(r) _ lim inf {fDAx [X Bn - Bn * C(pi XA)] + D2[XnX C(pi Xn) ) Bn]}
n -400

> d(r('1) + d(r(3)) = d(r (') + d(r'2)) > d* (r)

remembering Lemma 3.8. Then d(r) = d*(r) by Lemma 3.7.

10. In this section, we consider Case III.

LEMMA 3.12: If Case III occurs, d(r) = d*(r).
PROOF: We suppose that the xn(u, v) fail to be equicontinuous on one of the
bj, 1 < j _ 1 (< k). Now, choose three equally spaced points afj, Oj3, and es
on d, (see ?7) and choose points ain, Ojn3, and yj,n on bj such that r(aGj,n) =
a;, rn(Oj,.) = 3i and r(,yj,n) = yj. Then, from Lemma 3.2 if we choose a
subsequence (still called {In }) and if a,, Oj3, and -yj avoid the point of di referred
to in that lemma, we see that ajn, Oin , -yjn all tend to some point po on bj.
Now, map Bn conformally on another region B' by a map an which consists of a
fixed linear transformation of the exterior of bj into the interior of the unit circle
* which carries a?,n,, in, and 7yn into a',n, Gf,, and y',n on * followed by a
(variable) conformal transformation of z into itself which carries a4,n , 0,n
and 'Yn into a, A , and y, respectively on *, and let x' (u, v) be the transform of
xn(u, v) under an . Then the x' (u, v) are equicontinuous along A* on account
of the 3 point condition and Lemma 3.2. On the other hand, if we choose a
subsequence, we see that all the circles corresponding under an to the other
bj, 1 < j < 1 all tend toward some point of ZV. Thus we have transformed
Case III into Case II since the xn(u, v) were originally equicontinuous along bi
and hence j $ 1. Then d(r) = d*(r) in this case also.

11. In this section, we prove our existence theorem. We observe first that
d(r) < d*(r) automatically if k = 1 and d(r) is finite, on account of Def. 3.6.
Thus our theorem covers all cases.
THEOREM 3.1: Let r be a set of k(> 1) non-intersecting simple closed curves and
suppose d(r) is finite and d(r) < d*(r). Then there exists a region B of type k
bounded by circles and a function x(u, v) harmonic on B such that D2(x, B) = d(r).
If S: x = x(u, v), (u, v) e B, then L(S) = l(r) so that S is surface of least area
bounded by r and x = x(u, v) is a generalized conformal map of S.
PROOF: Let xn(u, v), (u, v) e Bn , be chosen so that D2(xn , Bn) -d d(r) and rn
r, rn: x = Xn(u, v), (u, v) e B . Let the Bn be normalized as in ?7 and let B be
the region mentioned in that lemma. Since d(r) < d*(r), none of the cases of
degeneracy can occur on account of Lemmas 3.10, 3.11, and 3.12, so that all the
Bn coincide with B and the xn(u, v) are equicontinuous along B*. Thus (remem-
bering Theorems 2.9, 2.10, and 2.11, and 2.12), a subsequence (still called Xn)
can be chosen which converges uniformly along B* to a function {(p) which yields
a representation x = t(p) of r on B* and which converges weakly in $2 on B
to some function x(u, v) with D2(x, B) < d(r) which has t(p) as its boundary
value function. A verbatim repetition of the last three paragraphs of the proof

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 849
of Lemma 3.3 show that t(u, v) is of class $'1' on B. Since it is obvious that
d(a, B) _ d(r), it follows that t(u, v) must be harmonic on B and D2(x, B) = d(r).
Since d(r) = 21(r) and D2(x, B) = d(r), it follows that if S: x = t(u, v), then
L(S) = l(r) so that S is a minimizing surface. Since D2(x, B) = 2L(x, B), it
follows that x = x(u, v) is a generalized conformal map of S on B (i.e. E = G,
F = 0 almost everywhere).

12. In this section, we discuss the differentiability properties of harmonic

functions which are implied by imposing further hypotheses of differentiability
on 9W.
DEFINITION 3.7: An ordinary function 4p(x) = 4o(x', * , x ) (or 4p(u), etc.)
is said to be of class C~') on an open set B if it is of class C(8) on B and if the
derivatives of the first n orders each satisfy a uniform Holder condition with
exponent y on each bounded closed subset of B, 0 < 'y < 1.
DEFINITION 3.8: A Riemannian manifold 9) is said to be of class C(f)
(O <ry < 1) if in Definition 1.10, we require all the transformations P' =
TYj'[Tj(P)] and their inverses to be of class C(7) and if we require the g,,q(x) to
be of class C(f-') .
THEOREM 3.2: Let 9) be of class C"' and be homogeneously regular. Then the
harmonic functions of ?3 are of class C' on the interior of the regions of definition
for any y, 0 < -y < 1. If 9) is of class C n _ 3, 0 < y < 1, then the har-
monic functions are of class C ') on the interiors of their regions of definition. If
) is analytic the harmonic functions are analytic.
PROOF: Let x(u, v) be harmonic on B. Then x(u, v) is continuous on B,
is of class C) on the interior of B for some ju > 0, and satisfies a condition
A [2,, kc] on B. Suppose 9) is of class C"' and choose a point po e B and sup-
pose x(po) e Y1i . Then, for (u, v) e C(po , ro) C B, ro > 0, x(u, v) e9Y. Let Pi

N N
be mapped on R by Ti and let the go(x) be the functions of Def. 1.10, (3);
they are of class C" in x. Let R': I x' I < 1 - a where a is so small that
x(u, v) e mia , the corresponding part of wi . Then we have

j=1 i=1
(3.1) m Si: ({j)2 < g.,(X)taO < M A (4j)2

for all i, all x in R', and some numbers m and M, 0 < m < M, and the ga(x)
are of class C" and have uniformly bounded first and second derivatives.
Since x(u, v) is harmonic, its components x2(u, v) minimize

(3.10) f f gao(x)(xux + xaXg) du dv

C(poro)

among all vector functions of class '132 on C(po , ro) which coincide with {x(u, v) I
on C*(po, ro) and are sufficiently near x(u, v). This fact, together with the
A[2 , k] and B[;, L] conditions satisfied by x on B are sufficient (as one sees
by reading the proofs) to ensure that the conclusions of [14], Chapter VII,
Theorems 1.1 and 2.1 hold for x'(u, v) [= x2(u, v)], the f(u, v, x, p, q) being

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
850 CHARLES B. MORREY, JR.
f(u, v, x, p, q) = f(x, p, q) = g a(X)(paPP + q qv) (pi = ,u = xv)
which is obviously of class C" in all its arguments. Since the ga(x) are of
class C", it is easy to see that the hypotheses on f of Theorem 3.1 of [14] Chap-
ter VII are satisfied for any y, 0 < y < 1, and what we know of the minimizing
property of x and its A and B conditions is sufficient to replace the hypothesis

that {xI(Z, v) minimizes ff f(x, p, q) du dv among all {xi} of class 2

on C(po, Q). Obviously {xi} is of class C' on each closed circle C(po , 1) with
&l < do. Thus {x'} is of class C' on C(po, 8o) for any a, 0 < - < 1. The
remaining statements follow from [14], Chapter VII, Theorem 3.2.
UNIVERSITY OF CALIFORNIA
BIBLIOGRAPHY
(The writer does not pretend that this bibliography is complete.)
[I] S. BANACH, Sur une classe de functions d'ensemble, Fund. Math., vol. 6 (1924), pp. 170-
188.
[2] S. BOCHNER, Harmiionic surfaces in Riemann metric, Trans., Amer. Math. Soc., vol. 47
(1940), pp. 146-154.
[3] PI. BUSEMANN, Intrinsic area, Ann. of Math., vol. 48, no. 2 (April 1947), pp. 234-267.
[4] J. W. CALKIN, Functions of several variables and absolute continuity I, Duke Math. J.,
vol. 6 (1940), pp. 170-186.
[5] 1R. COURANT, Plateau's problem and Dirichlet's principle, Ann. of Math., ser. 2, vol. 38
(1937), pp. 679-724.
[6] C. DE LA VALLEE POUSSIN, Integrales de Lebesgue, fonctions d'ensemble, classes de Baire,
Paris, Gauthier-Villars, 1916.
[7] A. LONSETH, The problem of Plateau in hyperbolic space, Amer. J. Math., vol. 64 (1942),
pp. 229-259.
[8] E. J. MCSHANE, Integrals over surfaces in parametric form, Ann. of Math., ser. 2, vol. 34
(1933), pp. 815-838.
[9] C. B. MORREY, JR., A class of representations of manifolds I, Amer. J. Math., vol. 55
(1933), pp. 683-707.
[10] ----, An analytic characterization of surfaces of finite Lebesgue area I, Amer. J.
Math., vol. 57 (1935), pp. 692-702.
[11] ----, The same, part II, Amer. J. Math., vol.58 (1936), pp. 313-322.
[12] --, On the solutions of quasi-linear elliptic partial differential equations, Trans.,
Amer. Math. Soc., vol. 43 (1938), pp. 126-166.
[13]----, Functions of several variables and absolute continuity II, Duke Math. J., vol. 6
(1940), pp. 187-215.
[14] ---, Multiple integral problems in the calculus of variations and related topics, Uni-
versity of California Publications in Mathematics, new series, vol. 1, no. 1, (1943),
pp. 1-130.
[15] T. RAD6, On the problem of least area and the problem of Plateau, Math. Zeit., vol. 32
(1930), pp. 763-796.
[16] -- , AND P. REICHELDERFER, On a stretching process for surfaces, Amer. Jour.
Math., vol. 61 (1939), pp. 645-650.
[17] J. W. T. YOUNGS, The topological theory of Frechet surfaces, Ann. of Math., ser. 2, vol. 45
(1944), pp. 753-785.
FURTHER LITERATURE ON THE PROBLEM OF PLATEAU
[18] R. COURANT, The existence of a minimal surface of least area bounded by prescribed
Jordan arcs and prescribed surfaces, Proc., Nat. Acad. Sci., vol. 24 (1938), pp.
97-101.

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms
 THE PROBLEM OF PLATEAU 851
[19] The existence of minimal surfaces of given topological structure ulnder prescribed
boundary conditions, Acta Math., vol. 72 (1940), pp. 51-98. This work presents
a complete bibliography of this subject up to the approximate date of publication.
[201 -, Critical points and unstable minimal surfaces, Proc., Nat. Acad. Sci., Vol. 27
(Jan. 1941), pp. 51-57.
[211 -- , On the first variation of the Dirichlet-Douglas integral and on the method of
gradients, Proc., Nat. Acad. Sci., vol. 27 (May 1941), pp. 242-248.
[221 , On a generalized form of Plateau's problem., Trans., Amer. Math. Soc., vol. 50,
no. 1 (July 1941), pp. 40-47.
[231 , AND N. DAVIDS, Minimal surfaces spanning closed manifolds, Proc., Nat. Acad.
Sci., vol. 26, (1940), pp. 194-199.
[24] N. DAVIDS, Minimal surfaces spanning closed manifolds and having prescribed topological
position, Amer. J. of Math., vol.64 (1942), pp. 348-362.
[251 J. DOUGLAS, Solution of the problem of Plateau, Trans., Amer. Math. Soc., vol. 33 (1931),
pp. 263-321.
[26] , The problem of Plateau, Bull., Amer. Math. Soc., vol. 39 (1933), pp.227-251.
[271 , Minimal surfaces of higher topological structure, Ann. of Math., vol. 40 (1939),
pp. 205-298. This paper contains a detailed bibliography of Douglas' previous
papers.
[28] , The most general form of the problem of Plateau, Amer. J. of Math., vol. 61
(1939), pp. 590 -608.
[291 , The higher topological form of the problem of Plateau, Annali, Scuola Normale
Superiore di Pisa, ser. 2, vol. 8 (1939), pp. 195-218.
[301 E. J. MCSHANE, Parametrizations of saddle surfaces with applications to the problem of
Plateau, Trans., Amer. Math. Soc., vol. 35 (1934), pp. 716-733.
[31] M. MORSE, The first variation in minimal surface theory, Duke Math. J., vol. 6 (1940),
pp. 263-289.
[32] M. MORSE, AND C. B. TOMPKINS, Existence of minimal surfaces of general critical type,
Ann. of Math., ser. 2, vol. 40 (1939), pp. 443-472.
[33] U- , nstable minimal surfaces of higher topological types, Proc., Nat. Acad. Sci.,
vol. 26 (1940), pp. 713-716.
[34] - , Minimal surfaces not of minimum type by a new mode of approximation, Ann.
of Math., ser. 2, vol. 42 (1941), pp. 62-72.
[35] , Unstable minimal surfaces of higher topological structure, Duke Math. J., vol. 8
(1941), pp. 350-375.
[36] I. F. RITTER, Solution of Schwartz' problem concerning minimal surfaces, Tucuman
Univ., Revista A., vol. 1 (1940), pp. 49-62.
[37] T. RAD6, An iterative process in the problem of Plateau, Trans., Amer. Math. Soc., vol. 35
(1933), pp. 869-887.
[381 , The problem of Plateau, Ergebnisse der Mathematik und Ihrer Grenzgebiete
(Berlin, Julius Springer) vol. 2 (1933). This work presents a complete biblio-
graphy of this subject up to the approximate date of publication.
[39] M. SHIFFMAN, The Plateau problem for minimal surfaces which are relative minima,
Ann. of Math., ser. 2, vol. 39 (1938), pp. 309-315.
[40]- , The Plateau problem for non-relative minima, Ann. of Math., ser. 2, vol. 40
(1939), pp. 834-854.
[41] , The Plateau problem for minimal surfaces of arbitrary topological structure,
Amer. J. of Math., vol.61 (1939), pp.853-882.
[42] , Unstable minimal surfaces with any rectifiable boundary, Proc., Nat. Acad.
Sci., vol. 28 (1942), pp. 103-108.
[43] -- , Unstable minimal surfaces with several boundaries, Ann. of Math., ser. 2, vol. 43
(1942), pp. 197-222.
[44] , Differentiability and analyticity of solutions of double integral variational prob-
lems, Ann. of Math., vol. 48, no. 2 (April 1947), pp. 274-284.

This content downloaded from 140.112.25.41 on Mon, 23 Dec 2024 02:18:36 UTC
All use subject to https://about.jstor.org/terms