BOUNDARY FUNCTIONS

KACZYNSKI, THEODORE JOHN

ProQuest Dissertations and Theses; 1967;
1967; ProQuest

This disiertation has been ._. ------

microfilmed exactly as received 67-17,?90
_- KACZYNSKI, Theodore John, 1942-
c BOUNDARY FUNCTIONS.
..

The University of Michigan, Ph.D., 1967

.
v.p-_.q.—.

Mathematics
.n

University Microfilme, Inc., Ann Arbor. Michigan

 BOUNDARY'FUNCTIONS

by
Theodore John Kaczynski

A dissertation submitted in partial fulfillment

of the requirements for the degree of
Doctor of Philosophy in the
University of Michigan

1967

Doctoral‘Committee:

Professor Allen L. Shields

Assistant Professor Peter L. Duren
Associate Professor Donald J. Livingstone
Professor Maxwell 0. Reade '
Professor Chia-Shun fiih
 BOUNDARY FUNCI‘IONS

m Theodore John Ksceynski '

Let 8 denote the set of all points in the Euclidean plane

hsving positive y-coortunste. and let x denote the x-exis. If

p is e point of x. then by on me; p we mean s simple are

Y . having one endpoint at p, such that \K-gp} C__' a. Let r

be e function mapping 8 into the Riemnn sphere. m s

bouncing function for f we mean s function. (9 defined on s

set E E X such thet for each p e: E there exists en src Y

st p for which

'13:}, fin) = tM12»)-

56 y

The so; 9;, W W _o_f_ f is the largest set on

which I: boundary function for 1‘ can be defined: in other words.

it is the set of all points p E X such that there exists an

ere at p along which I approaches a limit. A theorem of J. E.

McMillan states thet if f is s continuous function napping H

into the Risnsnn sphere. then the set of curvilinear convergence

of f is of type F55. . In the first of the two chapters of

this dissertstion we give s more direct proof of this result than

McMillsn's. end we prove. conversely. that if A is s set of

type __ 0.5 in X. then there exists e bounded continuous

complex-valued function in it having A as its set at curvi-

uncu- conversant». Noxt. we provc that a boundary function for
of
a contimoua function can always bc made into a function

Bairo claaa 1 by changing its valuco on a countable act of

potato. Coma-och. wc show that 1: (P 1: a function mapping a

m. E 9 x 1m tho 31am sphcrc. audit (9 canbcnado

a
into a function of Rain class 1 by changing in van” on

mum. not. thou than exists a conunuou. moon in 3

having (P on a boundary muon.,(1Ma ia a slight germ-c.1-

1aaucn of a thcom of 34t and Human.) In tho accord

ohaptcr w pron that. a boundary {motion for a function of
It
Bah-c class 5 Z 1 in a 1: of Bah-c class at moot EM.
:-
follow. troll on. that a bcmdary function for a Enrol-moan
a
ahlc function to always Bord-numb]... but u. chow that
not
boundary function for a ubcognc-ncaaurablc function mod

In booms-loam}... Tho dissertation ccnoludcc an: a not

of problm rue-firing to bc solved.
 TABLE OF CONTENTS

Page

LIST OF ILLUSTRATIONS..... ....... ........ iii

' INTRODUCTION ......................................... . ..... .. 1

Chapter

I. BOUNDARY FUNCTIONS FOR CONTINUOUS FUNCTIONS.......... 15

II. BOUNDARY FUNCTIONS FOR DISCONTINUOUS FUNCTIONS ....... 57

SOME UNSOLVED PROBLEMS .............. ..; ............ ... ....... 73

REFERENCES ................................................ ... 75

ii
 LIST .OF ILLUSTRATIONS

F_igure P_age

1. OIOOILOOIDIIOI I00...ccoooououo-oouuluuotltOInon-o... ‘27

2o Trap (J, a, e) IIIIII onocacao-conaooouoouoilo.uuto... 30

3- Tri (J, e)I.I..0030.0IOOOIIOIIIIIIOCIIIOOOOII... can

uuuuuuu OODIOIIIIIOIOIOIIIDI..0000... 62

iii
 INTRODUCTION

1. Pre liminary Remarks

Let H denote the upper half-plane, and let X denote its

frontier, the x-axis. If xex, then by an £333 x we mean a simple

arc y--with one endpoint at x such that y -' {x} E H‘. Suppose that f

is a function mapping H into some metric space Y.. If B is any subset

of X, we will say that a function gnfi -> Y is a boundafl function for

f if, and only if, for each REE there exists an arcYat x such that

lim f(z) '= 90:) .

z + x
Z 6 Y

The study of boundary functions in this degree of generality was

initiated by Bagemihl and Pirani-an {2] . A function defined in H may

have more than one boundary function defined on a_ given set E E X,

but it follows from a famous theorem of Bagemihl [1] that any two

such boundary functions differ on at most a countable set of points.

If f is defined in H, then the set 9}: curvilinear convergence

of f is the Set of all points xéx such that there exists some are

at, 2: along which f approaches a limit. Evidently, this is the

largest set on which a boundary function for f can be defined.

J. E. McMillan [10] discovered that the set of curvilinear convergence

of a continuous function is always of type F05, and in this paper we

show that every set of type F06 in X is the set of curvilinear

 convergence of some continuous function. Nent, we show that if «p is

a function defined on a subset E of X, then «9 is a boundary function

for some continuous function if and only if (f can be made into a

function of the first Baire class by changing its values on at most

a countable set of points. (This solves a problem of Bagemihl and

Piranian [2, Problem_l].) we then consider functions that are not

assumed to be continuous, and we prove that a boundary function for

a function of Baire class E :_1 is of Baire class at most 5 + 1 (thus

proving another conjecture of Bagemihl and Piranian [2]). It follows

from this that a boundary function for a Borel-measurable function

h l

is always Borel-measurable, and in the last section we show that a

I

boundary function for a Lebesgue-measurable function need not be

Lebesgue-measurable.

Most of the results appearing here have already been published

([8] and [9]). At the time I published these papers I did not expect

' to have to make use of this material for a dissertation.

2. Notation

R will denote the field of real numbers.

R? will denote n-dimensional Euclidean space.

Points in Rn will be written in the form (x1...., 2;) rather

than ($1,..., an) (to avoid confusion with open intervals of real

numbers in the case n = 2).

If véRn, then IvI denotes the length of the vector v.

82 denotes. {v6R5: [VI =; 1}. 82 will be referred to as the

Riemann sphere.
 .Let.

H = fi{(5c,y)e'R2 :y> o}

H ='{<$c.y>eR2 =-,1;>y>0}
'{<x,0) :xeR} .

>4
ll
§1='{<L%) :xéR}.

We consider X as being identical with R. Thus, for example,

<x,0) i {5:50) means x :x', and for p, q 6 X, the notations

[p,q], [p,q), etc. refer to the obvious intervals on X.

If E is a subset of a topological space, then E'denotes the,

closure of E,'E* denotes the interior of E, and E' denotes the

complement of E. Of course, if E is a subset of X, then‘E* means the

interior of E relative to X, not relative to the whole plane. In

Section 7, we often denote two line segments by s and 5'. Since the

prime notation is never used for complementation in that section,

there is no danger of confusing s' with the complement of s.

If f is a function defined in a subset of R2, then f(x,y)

. means f((x,y)). Thus we write Hz) for 26R2 and f(x,y) for x,yER

interchangeably. '

3. Baire Functions

In this section we review the main facts concerning Borel

sets and Baire functions, and we prove some results that will be

needed later.

If C is any family of sets, let C6 be the family of all sets

that can be written as‘a countable intersection of members of C, and

let Co be the family of all sets that can be written as a countable

union of members of C.

Suppose M is a metrizable topological space. Let P1 (M) be

the family of‘alil Open subsets of M and let Q:l (M) be the family of

all closed subsets of M. If E is an ordinal number greater than 1,

let

P504) = .“OcUQcMn
n<€

QgCM) = (U P"(M))
' n<€

For any F, E€Q€(M) (:2) E' 6 P€(M).

For any subset L of M, E 6 PE’(L) (respectively Qg (L)) if and

only if there exists a set D € PE(M) (respectively Qg (M)) such that

E = p n L.
Pg(M) and QE(M) are closed under finite unions and finite

intersections. P§(M) is closed under countable unions and Q£(Mj is

closed under countable intersections.

If mg, then PnCM) u Q“(M) E P€(M)n QgCM).

Let F°(M) be the class of all PC sets of M, and let 65““) be

the class of all G6 sets of M.

PZCM) = F004) and cM) ' = 6604).

Let Y be a metric space. For any family C of subsets of M

we will say that a function f : M + Y is of 'clfi (C) if and only if

f-1( U)€C for every open set u E Y'.

The following definition of the Baire classes is somewhat

different from the classical definition, but it seems more: convenient

 for ourpurposes. A function f : M -r Y is said to be of Baire class

0(M,_ Y)_ if and only if it is continuous. If E, is an ordinal number

. greater than or equal to 1, then f is said to be of'Baire class

€(M, Y)_if and only .if there enists a sequence of functions {fn}n=1
mapping M into Y, fn being of Baire class "1104:”. for some “n < E, ’

such that fn -> f pointwise. ‘

If f : M + Y is of Baire class £(M, Y)_ and if L is a subset

of M, then fIL is of Baire class €(L, Y). I

If K is a metric space, if g : K -> M is continuous, and if '

f : M +_ Y is of Baire class 5(M, Y) _,‘then the composite function f9 g

is of Baire class £(K, Y) _.

If Y is separable and if f : M -> Y is of Baire class E(M, Y),

then f is of class (Pg+l(M)) .[4, page 294].

_If Y is separable and arcwise connected, if E + l, and if

f : M -r Y is of class .(P§+1(M)), then f is oftBaire class ECM, Y). [4].

For any a, if f : M -> R is ,of class (pmcm), then f is of

Baire cllass, €(M, R) [6].

If L€QE+1(M) and f:L + R is of Baire class E(L, R), then f

can be extended to a function f : M -> R of Baire class £(M, R) [6].

We say that a function f : M -> R is Eel measurable if, and

only if, for'every open set UE-R, f'1(U) is a-member of the o-ring

of subsets of M generated by the open sets.

If f : M + R is of some Baire class. £(M, R), then f is Borel-

measurable, and, conversely, if f : M + R is Borel-measurable, then

f is of Baire class E(M, R) for some countable ordinal number 5

-[7, page 294] .

 The proofs of Lemmas 1 through 6 are based on standard

techniques in thestudyof Baire functions.

Lemma 1. .Let M be a metric space, and let E and F be two Fa sets in

M. Then there eicist two disjoint Fa sets A and B S M such that

B—FEA and .F-EEB.

Proof. Let E = $31 En and F = $31 Fn, where En and Fn are closed.

Then

En, Fn é F°(M)n 66w).

It is easy to check that FG(M) [\G5CM) is an algebra (i.e., is closed

' I

under complementation, finite unions, and finite intersections). We

inductively define a sequence of pairs of sets (An, En) as follows,

Let

A1=E1, Bl=F1nAi.

For n > 1, let

' . n-l n
An = En“ ._ BjI , an = !
ao AJ..
3- . J-1.
By induction, An’ Bné FO(M) n G 6 (M) '. Let

' A = DA
n=l n
, B = °°
n=1
B
n
.

Then A and B are F0 sets. Notice that

n-l n
B. E F and U A. E E .,
5:1 3 j=1 J
from which it follows that
and
n V
3n = acjgl'AJJ' 2 aa'

Therefore

AEU (EnnF') = E - F
n=1

and

82U(Fn/\E') F-E.

It only remains to show that A n B = ¢~ Suppose x 6. An 13. Choose

2., mwith iceAz and icéBm. If R. > m, then 2. >1, so that

. ~2-1
A2 = ‘Ezng BJ! 9131'“.
I

Hence ice B]; --a contradiction. On the other hand, if 2, 1 m, then

m
Bm _— agDIAj'I 9A2v ,

so that nEAi --another contradiction. We conclude that An B = g).-

If E is a subset of a space M, we let XE denote the charac-

teristic function of E.

Lemma 2. Let L be a. subset of a metric space M, and suppose that

E e FO(L) n G6 (L) . Then there exists a sequence {fn}n:1,of continuous

real-valued functions on M such that fn -> XE pointwise on L.

Proof. Both E and L - E are in FUEL), so there eicist sets E1,—

F1 6 FO(M) such that

5:331a and ,, L-Er—F 1 nL.

By Lemma 1, there exist A, B e FOCM) such that A A B = d and

El-FIEA,F1-E1§ B. Wehave
 i

E=AnL and. L-BABnL.‘

co . a: . .

Write A = q An, B = 1 En, where An, Bn are closed and Ans; Amp
n: n:
Bn E Bn+1 for each n. By Urysohn'sLemina there exists a continuous
function fn : M -r [0,1] such that

fn (x) 1 when x 6 An

fn(x) 0 when x e Bn .
._,.

{fn}n=l is the desued sequence.-

Lemma 3. Let L be a subset of a metric space M, f : L -> R a function

of class (F°(L)) that takes only finitely many different values.

'l'hen there exists a sequence. {fn}n:1 of continuous real-valued

functions on M such that fn -> f pointWise on L.
a.

Proof. From Banach's Hilfssatz 3 [4], we see that there exist real

numbers a1,..., a and sets

#4.
E1, .',. .', Eq 5 FU(L) n 66(L)

such that

If we choose for each j a sequence {fn'J }n:l of continuous real-valued:

functions on M such that' fnJ -> XE pointwise on L, and if we set

‘.n u

q j
fn —'J':1ajfn '
then. {fn}n:l is the desired sequence.-

Lemma- 4. Let L be a metric space, f a bounded real-valuedfunction

on L of Baire class '1(L, R). Then there exists a sequence. {fn}n=1

of real-valued functions on L converging uniformly to f, such that

each fn is of class (FOCLD and takes only finitely many different

values .

Proof. f is of class (FO(L)) and the range of f is totally bounded,

so an obvious modification of the proof of Banach's Hilfssatz '4 [4]

' gives the desired result.-

~-~.-

Lemma 5. Let M be a metric space, L a subSet 'of M, f : L + R a

I

function of Baire class 1(L, R). Then there erists a sequence. {fn}n:1

of continuous real—valued functions on M such that fn + f pointwise

on L.
....

Proof. We first prove the lemma under the assumption that f is

bounded. For any bounded real-valued ftmction h, let

“h“ = sup.{ Ih(r)| : x6 domain of h} .

ByLemma 4 we can choose, for each n, a function gn : L + R of class

(FO(L)) such that gn takes only finitely many different values and

ugn - in -3}; Let

h1=_g1, hnfign-_gn_1‘forn>1.

Then, for n > 1,

’ ..1 ..1...1'
Ilhn||= "gn'f*f'gn-1ll 13+ F<2n72'

Each hn is of class. (FOCLD and takes only finitely many different

values, so by Lemma 3 we can choose .(for each_n)_ a sequence. {hnj}j:1

of continuous functions on M such that hn:j ,-'> hn pointwise on L.

3
Set
 10

kax) = _- \\ hn“ .if tm 1 — H hn“

knc) = “hnll if 1111333111111.“
knjcio = hnjcic) if - u hn“ < hnjm < “nu“ .

Then knj is continuous, knj 3r hn pointwise on L, and " knju' .5. “hnu
_ < _ 1 . Therefore, if we set
2n-2

f. = 2 kj
3 n=1 n ’

then the series converges uniformly and fj is continuous on M. We

claim that fj + f pointwise on L. Take any x 6 L and any a > 0.

i 1 <- 3'- e. For each n, choose j.(n) so

Choose m large
. . _ . so that 2m—2
enough 3

that
. . ' I1, j . ' 1 i
J :J(n) z) ._..n (x) — hn(x)| <___2n+1 3,

Let i0 = max: {j (1),..., j(m)}. Then j 1 io implies that

.. m . m . . m .

f.(x) - f(x)| : f.(x) - z k J-(x) + z k 3m - 2 h (x)

I 3 I 3 n=1. “- I In=1 “ n=1. “ I
m . .
+ | z hn(x) - f(x)|
n=1

°° . m . ,

2 “kg“ + 11:21 lknJCX) -hncx)l + llgm - ill

z n=m+1

1 m 1 e ..1 é_
12m-2+(n§1;m‘33*;fi‘33 ' 8'
I

Thus fj (it) 7' fix) for each x e L, and the lemma is proved for bounded
3
f.
I

If f is not bounded, let

 11

_ n) =-. arctan f(l{) (x e L). .

Then - g _<_ g(ic) < ;- for every ké‘ L, and g is of Baire class 1(L, R),

so there enists a sequence' -{gn}n:1 of continuous functions on M

converging to_ g pointwise on L. Set

' 1
hnu) = '12r'+% 1f '%(x)< “3”}?
' _ 7T 1 ' 1r 1
hnu) ' 7—5 If .gn(x)2'f'fi

hum = gn(x) 1f -12'.+}_1<gn(x)<;-fi1

Then hn is continuous on M, — 321- < hn(i<) < 721, and hn f g pointwise on

L. Let fn(x) = tan hn(ic) . Then fn is continuous on M and fn -> f

pointwise on LI _ ’

Lemma 6. If L is a subset of a metric space M and f : L + Rm is a

function, then the following are equivalent.

(1) ‘ f- is of Baire class 1(L, Rm).

(iii f is of class (FOCLD.

(iii) There eitists a sequence. {fn}n:1 of continuous functions

mapping M into RIn such that fn + f pointwise. on L.

This lemma is an easy consequence of Lemma 5.

Definition. Let q be any point of R:5 lying inside the bounded open

domain determined by $25 By the q-Erojection of R3 -' {q} onto 82 we

mean the function Pq defined as follows. If a is any point of

R3 -. {(1}, let 2. be the unique ray, having its endpoint at q, that

passes through a, and let Pq(a) be the intersection point of 2, with

82.‘ Pq is a continuous mapping of R3“. {q} onto 82 that flies every

point of S2 .
 12.

Theorem 1. Let'. L be an arbitrary subset of R2. Then a function

f : L + $2 is of-Baire' class 1(L, 82) if and only if it is .of class

(FocL-n .
Proof. Assume that f : L -> $2 is of class (FO(L)). 829 R3,, so by

Lemma 6 there exists a sequence. {fn}1:1 of continuous functions. mapping

R2 into R5 such that fn + f pointwise on L. Let

An _ -
fn -1-({veR 3..IvI -L-
— 2})
_ -1~ 3, . l-
Bn-fn ({vER.l.12})
—1- 5. -. 1 '
. cn _-’fn ({veR .IvI 5-2-1).

Let fno = nA . According to [5, Lemma 2.9, page 299], fno can be
n
extended to a continuons’l—ftmction gn : R2 —> {v 6 R3 : |v| = 2%}.

Define hn : R2 + R3 -. {0} by setting .._t_s,_. _ l

hn(x) = _ gnCX) - if. xéBn

ha) = fn(x) if xecn.

Since 311’ Cn are closed, hn is continuous, and it is easy to verify

that hncic) -> £3) for each x e L. Let kn : R‘ -> s2 be the composite
function P0 9 hn' Then kn is continuous, and for each x E, L,

kncx) + P°(f(x)) = f(x). Thus f is of Baire class 1(L, 52).-

Definition. Let M and Y be metric spaces. Then a function f : M + Y .

is said to be of‘honorary Baire'class 2(M, Y). if and only if there

efists a countable set NQM and a function g : M + Y of Baire .class'

1(M, Y)_ such that flit) =. g(x) for every he M - N.

Theorem 2. Let L be an arbitrary subset of R2 and let Y be either

the real line, a— finite-dimensional Euclidean space, or 82. Then a

 15.

function f : L i Y is .ofhonorary Baire class 2(L, Y)..if and only if .

there enists a countable 'set N 9 L such that f'L-N is of class

(FocL - N)).
Proof. .Suppose that f : L + Y is of honorary Baire class 2(L,_ Y) _.

Then there exists. g : L '*. Y of. Baire class 1(L, Y)_ and a comtable

set N ._ L such that flL-N ' glL-N'

C = But glL-N 15 of class (FOCL
I l . — ND.

Conversely, suppose that flL-N is of class (FOCL - N)), where

N is countable. We must show that f is of honorary Baire class

111
2(L, Y)_. First consider the case where Y .= R . Write

fCX) = (flu), fzcic), £1,100).

Then-fiIL-N is of class (FO(L — M) (i=1,..., m), and it follows that

filL-N is of Baire class 1(L - N,’R). Since L - N 666(L), we can

entend fiIL-N to a function. -gi : L + R of Baire class 1(L, R). If we

set g(x) = (gICIx),...,. gm(x)) , then g is of Baire class 1(L, Rm)

and n) s 1562:) for is L -‘N, so we have the desired result.

Now consider the case where Y .= 82. Since 52 E R3, there

exists, as we have just shown, a function g : L + R3 of Baire class

1(L, R5) such that. gun ,~= £05) for all ice L - N. Then an - $2 is
countable, so there exists some point q in the bounded open domain

determined by_S2 such that qé g(L) . Let h be the composite function

Pq o. g. Then h maps L. into 82, and for each rel; - N,

MK) = Pq,(gCX)) = .Pq(f(X)) = f(X)-

If U Q 82 is open, then _

h -1 (U) =,.g -1 (Pq —1 cuneFocm,

 14

so h is of class (Fa (L)) . By Theorem 1, 'h is of‘Baire class

1(L, 52); so we have the desired resiflt.‘

 CHAPTER I

' BOUNDARY FUNCTIONS-FOR CONTINUOUS.FUNCTIONS

If r is a positive number andif yo is a point of a metric

space Y having metric p , then

S(r, yo) ‘ denotes '{yé Y _: p(y’ yo) < 1.};

We will repeatedly make use of Theorem 11.8 on page 119 in

[11] without making eicplicit reference to it. This theorem states

that if D is a Jordan domain in R2 or in 'R20%}, ifv is the

frontier of D, and if a is a cross—cut in D whose endpoints divide Y

into arcs Y1 and 72, then D-a has two components, and the frontiers

of these components are respectively a u Y1 and a u Y2- (The term

cross-cut is defined on page 118 in [11].)

4_.__rl_)omain of the Boundary Ft'mction ‘

Definition. ,If f is a function mapping into a metric space Y, then

the set '9f curvilinear convergence of f is defined to be

' ix EX : there eitists an arc y at x and there exists y 6 Y .

' such that

lim f(z) = 'y}.

z+x
zvey

J. E. McMillan [10] proved that forsuitable spaces Y,_ the set

of curvilinear convergence of a continuous function is always of type

Fag. We give a more direct proof _of this result than'McMillan's.

(This proof can be modified to_ give a more general result; see [9].)

.15
 go’s-Hug“...
16

.__i.. s‘. .__.- .

An interval of X will be-Called‘nOndegene‘rate ,if and only if

it‘ contains" more than one point. — I

i A!
Suppose y is a cross-cut of 1-1. If V is the bounded component

of H - y, let L(y) = a. Then L(y) = _[c, d], where c and d are the

endpoints of y and c < d. Suppose 9 is a domain contained in H. Let

I‘ denote the family of all cross-cuts y of H for which y n H E 9, and

let

1m) = .U c)*.
yéI‘

Let accm) denote the set of all points on X that are accessible by

arcs in (2.

Lemma 7. Assume that ace (9) is nonempty. Let a be the infimum of

ace (9) and let b be the supi-enrum of acc'm) . Then

1(9) = (a, b) .

Proof. Suppose x 61(9) . Let y be a cross-cut of H such that

Y n H Sn and ice L(Y)*. L(Y) =,[c, [d], where c and d are the end- .

points of Y and c < d. It is evident that c and d are in accm), so

a i c < x < d :b, and ice (a, b). Conversely; suppose 5c? 6 (a, b).

Then there exist points c', d' 6 accm) with c' < ic' < d'. Since 9

is arcwise connected, it is easy to show that there eitists a" cross-

cut 7' of H, with y' n H 99, having c', d' as its endpoints. But A

then x' e (c', d') = mm", so x' 5 1(9)} I

Lemma 8. If 91 and 92 are domains contained in H, and if

(1) 162]) n' accinli and 1(92) n accinzi

 '17

'are.not disjoint, then 91 and 522 are.not disjoint.

Proof. .We assume that $21 and 522 are disjoint and derive a contradic-

tion.- Let a be a point in both‘ Of the. two sets (1) . Let'yi be a

cross-cut of H, with fin H E oi, such that a e Lfiil)’: (i = l, 2). Let

Ui and Vi be the components ,of H - yi, whereiVi is the bounded component.

Observe that Yln H and 72 AH are disjoint.

Suppose Y1“ H 9V2 and 720 H 9V1. Then, since yln H S? ,

U1 has a point in common with V2. But, since U1 is unbounded, U1

caImot be 'contained in V2, so U1 must'have a point in common with

-Y2 n H. This contradicts the assumption that Y2 n H 9V1, so we

conclude that either 71 n H $ V2 or “Y2 n H g V1. Hence, either

yl n H EUZ or Y2 n H Eul. By symmetry, we may assume that

72 an 5 U1.
92 does not meet yl, and 92 does meet U1 (because yz n H S

01 n 522) , so 92 QUI. Since a G, 555E237, there exists a point

b e L(y1)* such that b e accCszz). But then. b e 5, <.-'-. ul, and this is
it
impossible because the frontier of U1 is disjoint from L(yl) . I

Theorem 3 (J. E. McMillan). Let Y be a complete separable .metric

space and let f : H —> Y be a continuous function. Then the selt‘of

curvilinear convergence of f is of type Fe 5

——._

Proof. Let'{pk}k:"1 be a countable dense subset' of Y.' Let' {Q(n, m) }m:l

be a counting of all sets of the form

' ' l. ' 1

{(x,y.) :Q<y<fi and r<t<r+fi}

where r is a rationalnumber. .Let' {U(n, m, R, an“; be a counting

(with repetitions allowed) of the components of

 '18

, .
Flcsci? pkn nqcn', m).
(We Consider ¢ to be a component of 45.) Let

Acn, m, k,' 9.) =-.-acc[U(n,' m, k; 2)].

Set' .

B = n U U {’11o m, k, a—Acn. m; k, 2).

‘n=1 m=1 k=1_ 2:

Since I(U(n, m, k, 3)) is open in X it is of type I; . It follows that

B --i-s of type F06 . Let C denote the set of curvilinear convergence

of f. I claim that B EC; l‘ake any b 6 B. For each n, choose m[n],

k[n], g,[n] with l

(2) b é MUCH. m[n], k[n], mm “Mn, m[n], a], Hill)

(11 = 1, 2, 3,...).

For convenience, set Un = U(n, m[n], k[n], 2,[n]). By (2). and Lemma 8,

Un and Un+1 have some point zn in common. For each n, we. can choose

an arc Yn E Un+1 with one endpoint at 2n and the other at and. Then

Yn Q Q(n+1, m[n+1]) . Also,

b 6 ATn+1. m[n+1], k[n+1]. “5+3? SEUn+1 EQCIHI, m[n+1] .

and therefore each point of Yn has distance less than n—i—l- from b.

5725" 0 as n + oo;_hence, if we set Y =‘ {b} UU Yn’ then Y’is an arc

' n=1
with one endpoint at b.

Since "11 and ”n+1 have a point in common,

-1 -1
f (“z—n? pk[n])) and f-1 (ac-2m,
‘ ‘ l .
pklmun

have ac'ommon point, and hence

I
 19

cfi'f Pk[n-])- and 5(2'11+1,’ Pk[n-I_-'l])'

have a common point. Therefore, if p is the metric on Y, _then

'.‘.1 .. 1 .'.1

, pcpk['n]’ Pk[n+'1]).i—n'+ 'n+l‘< n-l ’

2 2 - . 2 .

and therefore
r - r ..
C . ) < z c . ) < z 1 ,< 1
p Pk[n]’ PRU-1.1.1.] _ i=1 p Pk[n+'i.-1]’ Pk[n+i] i=1 2n+1-2 zn-Z '

Thus {pk [11]} is a Cauchy sequence and must; converge to some point

p 6 Y._ Since

'y CU Cf1(S(——
n - n+1- 2n+1’ Pk[n+1])) and

Pk[n] ; Pa

ilim, f(z) = p. It is possible that y is not a simple arc, but

H.

:21 ' according to [12] we can replace y by a simple arc

W157. Thus b e C, and.we have shown that B EC.

Suppose c E G. Let Yo be an arc at c such that f approaches

a limit p' along Yo . Take any 11. Choose k with p' E S(—, pk} .

' Cheese m so that c is in the interior of qin, m n X. Then Yo has

a subarc yo' , with one endpoint at c, such that

v , - ' {c} c_ Q(n, m) n :3-1 cscz—n,

1
pkn.
Hence, for some 2., c €acc[U(n, m, k, 2.)] = Mn, m, k, 9.). This

shows that

cgfio 9H
0 A(n,m,k,2.).
n=1m=1_

It is easy to deduce from Lema 7 that the set

 20

A(n, m; k, z) _ I(U(n; m, k, 2)) = .[lTTTTTTITTT"

Mn”. m; 16.32) - mum"; m; k; mnAcn; m; k; m

contains at most two points. It follows by a routine argument that

n U Mn; m; k, z) -n U '[Icucn', m, k, 2)) nAcn. m. k, m

n' m,k,2,
n m, ,2.

is countable . ' Since

0 mU [1(UCn, m,k 233nm ssc

0U A01, "bk ‘1'),
n m,,k2;

C - B is countable, and therefore C is of type F06“ .

Next we will show that the foregoing theorem is as strong as

possible, in this sense: if A is any set of type F05 contained in X,

then there exists a bounded Continuous complex-valued function f

defined in H such that A is the set of curvilinear convergence of f.

The proof is unfortunatelyquite long.

Definition. Let E1 and E2 be two sets on the real line. A point p

on the real line will be called a splitting'point for E1 'and E2 if

either

xi :p for all x1e E1 and p _<__x2 for all x26 E2

or xzipforallJcZEE2 and p_<_x forallxlefil.

1

.We will say that two sets E1 and E2 split, or that E1 splits with E2,

if and only if there efists a splitting point for E1 and E2.

Lemma 9. Let E .be an F0 set in R. Then there is a sequence. {En}n:l

of sets such that

 ‘21‘

(i) En is bounded and closed,

(ii) if n =|= m, then either'Er-1 andlE.m are disjoint or En and

Em split ‘ h '
(iii) ' E = U En.
n=1

Proof. We can write E = g An where An 15 closed, An E An+1 for all

n, and A1 = «p. '

Observe that if I is any open interval, then there eitists a

countable family. {Jn}n:1 of bounded closed intervals such that

n 4= m :§ JH and Jm split, and I = nL-Jl Jn' Since any open set of

real numbers is a countable disjoint union of open intervals, it

follows that for any open U there exists a countable family {In}n-

of bounded closed intervals such that n + m I? In and Im split, and

U = U In.
n=1_ _
For each n, let {13%; be a family of bounded closed inter-

vals such that j + k 2} I? and Ik split, and An =U 121L612

, 2,.uo}o
f =I{A1}U{IrjlnAn+1:n=1,2,..-5j=1

Thenf is a countable family of bounded closed sets, and

C

9
>

E nA')
I

-C
51>
ll

H
H

.4.
:5

8
8

5
C
>

D
u. :5
D

I.
9
C

V
A
>

:l>
V

H
II

+
”I;
H
H

J
5

ll
:1
7"

, 1
Let F1 and F2 be any two distinct members of ‘F . If either F1 or F2

is A1 = ,¢, then P and F2 are automatically disjoint. If neither

1
F1 nor F2 is 'Al’ then we can write
 . 22‘

'.F1"_II-1(1)nA n(2).
t" n(.1)+1'.and l=2" f- -' I 3(2)" An_(2)'+1
If nu). <. n(2); then nu) + 1.531(2); .so
F2 —_ . Ij(2)
IDCZ) n An(2)+1C AnCZ) E An(l)+l’ and th-refore
A F1 - and F2 are'

disjoint. A similar argument shows that if n(2) < n(l), then P and
1
F2 are disjoint. Thus, if F1 arid F2 are not disjoint, then n(l) =

n(2). and we have

F1 ' Ij(1)"““n+1 and F2 ' I5(2)"An+1’

_ n -

where n = n(l) n(2) . I But then j(1) In

J'(2), so and In
5(1) 3(2)
split, and therefore F1 and F2 split. So we have shown that any two

distinct members of F either split or are disjoint.

If 1" has infinitely many distinct members, let .E

l’ E 2’ E 3'”
be a counting of g: . If f: has only finitely many distinct members,

let E1,..., Em be the members’of F and let Ek = 4; for k > m. In

either case,’ {En }11:1' is the desired sequence. I

If F is a closed subset of the real line, then by a comple-

mentary interval of F we mean a component of F'. (If F = R, then 4,

is considered to be a complementary interval of F.)

Definition. By a; special family we mean a family F of subsets of R

such that

(3) 5- is nonempty

(4) . each member of F is bounded and closed

(5) there exists a sequence {Fn }n°°'1. of members of F such that

.everymember of F is equal to Some Fn, and the following condition

is satisfied.

”(5a) If m > n, then either Fm is contained in one of the complementary

 .23.’

intervals .of Fn, orelse Fm splits witt.

Lemma 10. If E is an F6 set in R, then there emists a special family

fisuch that E =Uf.

Proof. By Lemma 9 we' can choose a sequence {En}n:1 .of bounded closed

sets such that if n + m then En and Em either split or are disjoint,

andE = (2 En.
11:

Let 111 = _1 andlet P1 = E1. Now suppose that n1, n2,..., nS

are chosen and F1, F2,..., Fn are chosen so that
s
(1) l=n1 <n2 < <ns

(ii) Pi is closed and bounded (i = 1,. .., ns)

(iii) if ns‘ 1 r > t Z 1, then either Fr is contained in one of

__ _

the complementary intervals of Ft’ or else Fr’ splits with Ft

(iv) if 1 i 1 ins’ then there eiists j €{1,..., s} such

that Pi E E. ,
JD 5

(V) L2
1:, F1 = -U
i=1.
Ei'

We construct F. ,..., F as follows. Let-"PJ- be the family of

“5+1 ns+1
complementary intervals of the bounded closed set

n s
UFi = .U Bi.
i=1. i=1 .
We assert that Es+1 meets at most finitely many members of Jr. If this

assertion is false, then there emists an infinite sequence {In}n:1 of

members of :9 such that n =I= m implies Inn Im = g», and there emists

(for each m) a pomt xm 6 Im A Es +1. {xm}m=1_ 15 a bounded sequence,

- O l l 'l m 0

and n i: m implies that xn + 9cm. From this it follows that {xm}m=1

has either a strictly increasing or a strictly decreasing convergent
 24-

subsequence. We will assume'that-Ixmm) }k:1 is a strictly increasing

convergent subsequenCe; the reasoning is Similar in the case ,of a

strictly decreasing convergent subsequence. Say IMk) = .(ak, bk) .

Then a.k < mc)‘ < bk’ so since xmfk)‘ < xm(k+1) < bk+l and

'xm(k)¢ Imum-1) .we “.1“ have xm(kj i ak+1 < xm(k+1) . Therefore, if
we let » I
sq: lim
= .k-m xm(k)’
then '2? = iii} ak alsoé Moreover, for k 1 2, ak is a finite real
number, so that aké U 131' Therefore there exists u 6 {1,..., 5}
i=1
such that akE Eu for infinitely many values of k. Consequently
x 6 Eu. But Since xm(k) e Es+1’ x €Es+l also. But then x E Eu n E91,

so that Eu and Es+1 must split and Y must be a splitting point for

Eu and Es+1' Since infinitely many ak lie in Eu’ Eu contains pomts '

that are less than 35; and Es+1 also contains points less than E;

therefore Eu and Es+1 cannot split, and we have a contradiction. This

proves the assertion. Let

«5' = {¢}u {I n as“ : 16:9— and I mas“ 4: ¢}.

Let 115+1 equal ns plus the number of members of ,8. Let F“5+1“ . . ,
Fn 1 be all the members of ,8. We must show that conditions (1)
5+
through (v) are still satisfied when 5 is replaced by‘s+l. Conditions
(1), (ii), and (iv) are obvious. The Verification of (iii) is divided

into three parts. Suppose ns+i _>_ r > t i l.

Case I. Assume that ns‘ 1 r > t 1 l. In this casewealready know

that either Fr is contained in one of the complementary intervals of

Ft or else F.1- splits w1th Ft‘

 25- -

Case 11.. Assume that ns+i :r > ns‘ 1.1? 1 I. There exists v €.{1_,...,s}
' c_ I
such.that Ft “.1311" Either‘Ev and Esflare disJomt or they split.
4 . I a u . ' .

Case Ila. Ass‘ume.Ev and Es+1 are disjoinp; Either-Fr = ¢ (in which

case Fr' 15' certainly'contained in a complementary interval of Ft) or}

else Fr =} 1: and F1; = In Es+1 for some 16:9. Let J be the: smallest
closed interval containing Fr' Then J E? and J1“; I E (H Bi)" so
* _ . . _
that 'J does not meet Ev. The endpoints of J lie in Fr 5554.1: so
neither endpoint of J lies in Ev' So J does not meet Ev and therefore '

J does not meet Ft; from which it follows that Fr '15 contained in a

complementary interval of Ft' '

‘
,Case IIb. -
Assume '
that BV and Es+1 split. '
Since Ft c
_ Ev and Fr C
_ E-5+1

it follows that Ft and Fr split.

Case III. Assume. that ns+l :r > t > ns. If either F1‘ or Ft 15 qt,

it is clear that Fr is contained in a complementary interval of Ft‘ ‘

Otherwise, there eicist I1, 1269- such that- I1 A 12 = 51> and

’Fr =_Ilr\Es+1 and Ft =12nE

Since I1 and I2 evidently split, FI. and Ft must Split.

Thus condition (iii) is verified.

As for (v), it is clear that

I s C “5+1
E E. _ . ' P __.E
s+1 H 1 J= s+1 J s+1’

so that
5+1 5 s
_
1-].
E1 = (U E1) U (Es+1' - U E1) -
=1 I 1:].
 26.

's+l n n n
» ,s+ 1' 5+ . '
HenCe UEi
i=1.
=. _' (61.11.)
j=1, J
uc| b p.)‘ = - IQWF
j=ns+1' J 3"- J

... Thus we have shoWn that we can construct sequences. {11.J}j:1?

'{Fk:}k_1 in such a way that conditions (i) through. (v) are satisfied

for. every value of 5. If we 'set’F= {Fk : =1, 2, ...}, it is easy

to verify that? is a Special family and that E = UfFI. I

Definition. If f1 and 5-72 are two families ,of sets, let

fiAs = {Fln F2 : F16 971 and es}.

.Lemma 11. Iffandfare two special families, then T’Af’is a

special family .

Plroof. Conditions ,(3) and (4) in the definition of a special family

are clearly satisfied, so we just have to verify (5) .

Arrange all pairs of positive integers in a sequence

according to the scheme shown in Figure 1. Let (aCk), bfk)) be the

kth term of the..sequence1 (k = 1, 2, . ..). Observe that k < 1 if and

1The reader may find it amusing to derive the following

formulas for (a(k), b(k)). For real t, let [[t]] denote the largest
integer that is strictly less than t. Then

V81k+ + 1 2 '1’8k+1 +11

6100 = 2([[-——--
-——-.-7]] + [[————]]) - k + 1

%([['/8'—_]k+1 ] + — %—( 1)[['8k+1]])([[/“T‘]8k+1 ] + .1-%(—1)[[”8E+1]]) -k +1

+1] + 3)([[VT']8 111+ 1) - k + 1

émnm if [[léfim is odd

.- %([[V8E+1]] +‘2)[[V8E+1I]-- k 5|- 1 ' if [[1’8E+1|]- is' even,

-and
 I

.23

onlyif either a(k) + b(k)o< a0.) + b0.) or-else 'a(k) + b(k) = a0.) +

b0.) and b(k)' < bob). Thus k < 2'implies.that either a(k), < aUL) or

. b(k) < bCl). I I

Let {Fn}n- be a sequence ,ofelements off}7 such that .every

member of {F is equal to some Fn and Such that condition .(Sa) in the

definition of a. special family is satisfied. Let-'{F } : 'be a

similar sequence forfF. Set

_1 is a sequence in T’Aflé such that .every member of {FA j?

Then {Fk}k:

is equal to some Fk' We must show that7 condition (5a) is satisfied.

Suppose that 1b > k. Two-cases occur.

Case I. a(k) < am).

Note that FkE Fa(k) and F2, Fm E1ther Fa) is contained in

one of the complementary, intervals of Fa(k) (in which case F2 15

contained in a complementary interval of Fk')’ or else Fan) and Fa(k)

split (in whichrcase F”, and Fk split).

Case 11. bCk) < b(2.).

In this 'case a similar argument shows that either F2115 contained in

1:00 = ,2([['—-——111‘*§* 1.11 - [tL—-81‘*§ * 1111) + k

-(W8,E+111 + — - %(-1)11"11*111)([[/§E+_111 - — - -1-c-11”’81*1“) +k

'*—§c11'/‘I<‘“18 +11+ mums +1.1 — 1) + k if [[ms +1] is odd

‘1 -%[[18E+1]]([[18E+1]] " 2) '1‘ k if [USE-I'll] is even.

 29

a.¢omplem'entary.interva1of Fk'or Finand‘lgk split... Thus condition

(5a) is~satisfied', andfF/f is a special'family.I

Lemma 12. Let E1, E2 be two F0. sets in R such that E19 B2,. and

suppose that?"1 andj-‘r2 are Special families such that 1?.1 = U971 and

2=Ufi. ThenE1=U(fi/\’FZ)-

The proof is obvious.

Neict .we introduce some notation.

Let J be a nonempty interval on X with endpoints a, b (a_ i b) .

' By Trap (J, a, 6) (where 6 €"(0; 1) and e > 0) we mean the interior

of the trapezoid shown in Figure 2. That is,

Trap(J, 5, e) =‘{ <1c,y) :0<y<e,a+yctn6<1c<b~yctn6}.

For 66 (0; 1), let Tri (J, o) be the closed triangular area shown in

Figure 3. That is,

Tri 0,6) ='{ <1,» :yloanda+yctneirib-yctne}.

If 5‘06- X, a > 0, and e 6 (0, '12:), let SCXO, 6, 9) denote the open
Stolz angle‘shown in Figure 4. That is,

S(i0,8.9)= {<X.y) =0<y<e,

-xo+yctn(1T-e)<x<xo+yctn6}

If K is a closed set on a real line, let J(K) be the smallest

closed interval containing K. If K is bounded, closed, and nonempty,

’e>9,and0<s<a<12'-,thenwe. define

m, e, a; s) = Trap (JCK). a, a) - U Tri (1', 3);

I63-

where eQ denotes the set of complementary intervals of K.

Figure 2.~-Trap(J,£ , 9)
 31

Figu re 3.--Tr1(J, 9)
32

x axis

o
 33

We . state without proof the following .readily .verifiable facts .

(6) B(K, e, a, '8) is an open subset of H.

(7) S(e, a, e) is an open subset of H.

(8) If K1 and K2 split, then for any 51, e2, ., s,

BCKI: £1: a: B) and B(K2: 32: a, B)

are disjoint.

up]:
(9) Suppose that KIE K, e > $1 > 0, and 0 < B < 61 < :31 < a. $
Then

B(K1, 61, (x1, 81) nH§B(K, e, a, B).

(10) Suppose K 1 is contained in one of the complementary intervals

of K, and suppose e, (.1..- B are given. Then there exists 6 > 0

such that for every n 1 6,

c, e, a, B) and cl, n, a, B)
are disjoint. -

(11) ' Suppose that a < e é gand x0 1}: IJ(K)*. Then, for any a, 61’

5.
c, e, a, a) and 3(k0, e1, e)
are disjoint .

(12', Suppose that x0 e K n JCK)* and [B < a < e < %. Let a be. given.

Then $39.3"??? '55. > 0 such that for every n f 6,

S(xo, n,"e) nl—I §B(K, e, a, B).

x

(13) Suppose that e < e' and 6'. < 9. Then

STE; 6, :ei nI-l 586:0, e', 6').

 .34 .

(14) Suppose x0 é Kand e, .a, B, 6 are given. . .Then there exists

6 > 0 such thatforevery n g 6,

scxo, 1.1.9) and ' MK, a, 6,6)

are disj.oint..

(15) If 5:0 4: 5:1 and e, e are given, then there exists 6 > 9 such
that for every n i 6,

80:0, 6, e) and S(x1, n, e)

are disjoint.

(16) ——‘_TB(1<, e, a; s n n.
(17) s 5: , 'e, e nx =' {5:0}.

Definition. If? is a special family, let?"2 .be the set of‘all

members of T" that have two or more points.

Definition. Let F be a special family, let E be the-set of allend-

points .of intervals J (F) where F 6 T, F + e, and suppose that
0 .< B < at < 6 £321. By a pa_i_1_'g_f_'.sp'ecia1 a, B; 6 functions forwe
mean a-pair (a, 6), where e and 6 are positivereal-valued functions;

the domain of e is E, the domain of 6 is 1‘2; and

(18) for each n > 9,, there exist at most finitely many F6152 such
that 6(F) 1 n;

(19) for each n > 9, there erist at most finitely many e e B .such

that e(e) in:

(20) ,if e, e' 6. E and e + e", then

S(e; 5(a), '6) - and S(e"; 2(e'), e)

 . .35

' 'are disjoint;

(21) ,if F, K 61:2- and F =1 K; then

Bcp; SCF); a; e) and - scx; §(K), ¢;:e)
' are disjoint;

(22). ‘ if e €13 and F€f2,'then

S(e, e(e), e) and a 30:, up); one)
are disjoint .

Lemma 15. Let’f-be a special family and suppose that 0 < 'B < a < e < %.

Then there exists a pair of special a, B, 9 functions forf.

Proof. _1 be a sequence of members of f of the type referred

Let {Fn}n:

to in condition (5) in the definition of a special family. Let

{F2(n) =' {F6152 : F = Fk for some k _<_n}

E = set of all endpoints of intervals J (F) for

Fe£F+¢
E(n) =' ie e E : e is an endpoint of J(Fk) for

‘ some k in for which Fk + 4)}.

If J(F1) has one endpoint e, set e(e) = 1. If J(F1) has two

endpoints e1, e2, then by (15) we can choose 6(61] 1 1 ande'(e2) _<_ 1

so that S(e1, E(el), 9) and S(e2, E(eé), 6) are disjoint. :If F16132,

set 6(F1) = . In this case, J(F1) has two endpoints el and e.2 and

(by (11)) B(F1, 6(F1), a, B), S(e1, e(e1), e) and S(e2,- a(e2), 6) are

all disjoint.

-.Now suppose that e(e) and SEE) have been defined for all
 .36

e EE(n) and all P 6T2(n)=in.s'uch a-waythat .

(i) if‘e, efi €E(n) and e + e", then S(e', ace), :_9) and .

S(e!,e(e'); '6) are disjoint;

(ii) if F, Kefzcn’) and P +_l(_, then B(F, aw), (1)3) and

B(K,.6(K)l, a, a) are disjoint; ‘
(iii) if e eE(n) and Peg-'20:), then S(e, e(e), e) and
B(F; 6(F), (Sc, 3) are disjoint; .

(iv) if e e E(n) and k i n is the least integer for which

e e E(k), then e(e) 111?;

(v) if F éfiFz (n) and k _<_ n is the least integer for which

F efzck), then 6(F) :11? '

We must extend the definitions of a and 5 to E(n+1) and

9:2 (ml) in such a way that conditions (1) through (v) are still
satisfied when n is replaced by n+1.

' . 2
Case I. If 'l gp orvif l = Bk for some k in, thenf (n+1) = .

fzcn) and E(n+1) E(n), so that nothing is required to be done.

Case 11. If Fn+1 consists of a single point e and if e E Fk for some

k _<_n, then (since Fn+1 and Fk must split in this case) e is an

endpointof J(Fk), so that _againfz(n+1) =fz(n) and E(n+1) = E(n),

and nothing is required to be done.

Case III. Suppose that. Fn+1 consists of a single point co and that

for each k _<_n, eo ~$ F‘x‘ By (14), (15), and the fact that E(n) and
fzm) are finite, we can choose E(eo) E (0, 571-) so that S(eo, E(eO), e)

is disjoint from S(e, e(e), e) and from B(F, 6(F), a, B) for each

e 6 E(n) and-each F efin) . The construction is then finished for

 .' 757

E (n-p-l) and ‘ F'(n'+1) '.

Case" IV. Suppose that Fn+1 contains atleast two points andathat, for

each k in, Fk + Enid" For each k _‘<_n, either Fn+lsp1its with Fk’ or
else Fn+1 is contained in a complementary interval 61’ Pk"- Since T2 (n)

is finite, (8) and (10) show that we can choose 6(Fn+1) E(O, n+—-—1—) so 1

that Ba+l’ 6(Fn+1), a, B) is disjoint from E(F, 5(F), a, B) for each '

F 6 (F: (n) .

Say e e E(n). Then e is an endpoint of J (Fk) for some k i n,

so (since Fn+1 either splits with Fk or is contained in a complementary

intervalof Pk) e ¢J(Fn+i)*. By (11), B(Fn+1, 6(Fn+1)_, a, B) and

S(e, E(e), e) are disjoint. h

Let e0, e0' be the endpoints ,of J (Fn+1)‘

Case IVa. 'eo, eo' eE(n). 4

In this case the construction is already finished.

CaselVb. eoe E(n) and eo' i E(n).

If e0' 6 FR for some k in, then Fn+1 splits with Pk, so that

eb' must be and endpoint of J (Fk) --which contradicts the assumption

that eo' $ E(n). Hence, for each R in, e0' 4; Fk' By (14), (15),

and the fact'th'at E(n) andfl=2 (n) are finite, we can choose

e(e° ') E (0, n—1_+1) so that S(eo ', e(eo ‘), e) is dilsjoint from

S(e, e(e), e) and from B(F, 6(F), a, B) for each e6 E(n) and each

F€f2(n). By (11), S(eo ', e(eo '), e) and E(Fm-l’ 6(Fn+1), a, B) are

disjoint. Thus the construction is finished for E(n+l) andfz (n+1).

I

Case 1%. e0 4; E(n) and e0! 6 E(n).

This case is essentially the same as Case._t.
 . '38

Case-.IVd. e0 45130:) and e0 .' $1301) .

If e0 6 FR .for some k _<_ n, then Fn+i—3plits with Pk, so

eo' is an endpoint of J (Fk); acontradiction. Thus e045 PR for k i n,

and similarly eo ' t F. for k < n Therefore, by (14) and(15),. we can

choose e(e o) and t-:(eo ') € (0, n+——1-1—) so that S(eo , e(eo), 6) and

S(eo' , e(eo'), 6) are'disjoint and each of S(e:, e(eo), e) and

S(eo.', e(eo‘), 6) is disjoint fromevery S(e, e(e), e) (e €E(n)) and.

from every 30:, am, a, a) (FefZCnD. By (11), S(eo, a(eo), e) and

S(e0' , e(e°'), 6) are each disjoint from B(Fn+l’ 6(Fn+1), a, B), so the

construction is finished for E(n+1) and {[52 (n+1) .

Wei have shown that we can inductively define s(e) forevery

e'é E and 6(F) for every F 69:2 in such a wayflthat (1) through (v)

are satisfied for every value of 11. Conditions (20), (21) and (22) in

the definition of a pair of a, B, 6 special functions are thus auto-

matically satisfied by (a, 6). We must verify that ('18) and (19*) are

also satisfied.

SuppoSe (19) is false. Then there exists n > O and there

exists an infinitesequence' {ek}k:1 of distinct members of B such that

e(ek) _>_ n for every k. Let m(k) be the least integer for which ek is

an endpoint of J (Fm(k)) . Each J (Fm) has at most two endpoints, so,

since the ek are all distinct, there exists (for given 111) at most two

values of k for which m(k) = m. Therefore there exist infinitely

many distinct integers among m(l), m(2), m(3), Consequently

there exists j with-H13)— < n. But, by (iv), e(ej) i-filtm—< Tl, a .

contradiction. So (19) must be true. A similar argument shows that

(18) is true. I
 .59

.LemI'na-l4... .Let‘Fbe aispecial family, 0 s B <,.a_< a Q %, and.1et E be

the set ofall endpoints of intervals J (F) .for F e ’1’. .Suppose (e, 6)

is a pair of special on; B; efunctions forf. Ifcl; Isi‘are two real-

valued functions'having domains B and respectively, and_if ‘

0 < 81(8) :e(e) for all e 5E, and

~

0 < 61(F) ism for all FefiFz,

. then (:1; 61) is a pair of special a, B, 6 functions for {15.

Proof. This follows from the fact that

S(xo, e', 6) 9.80:0, e", e) ‘

and B“, 3': 0‘, B) EBCK, 5"; 0‘, 3)

whenever s' _<_ e". I

Theorem 4. Let A be any set .of typeFOs in X. Then there exists a

_bounded continuous compleic-valued function f defined in H such that A

is theset of curvilinear convergence of f. '

Q

Proof. We can write A = Q An, where each An 15 of type Fe and

An+1§An for' every n. For each n, let';l be a special family with

L117:1 = An. 'Let

5&1 =31
flu-1 =~fi1Afi1+1 ”for n > 1.

By Lemmas 11 and 12, together with mathematical induction,-{F;'1 is a

special family and UT; = An. 'Moreover, every member of T;+1 is a

subset of some member off-:1. ,

Let' {Bn}n:1 be a strictly ascending sequence in (0, 1)

converging to g—.
 '40

.Let1.{un}n:lzbe a strictly.descendingsequence in (lg-V1741)

.
-. . “I. '

convergrngto -§-.
31'
Let. {_e'n}n:1 be a strictly ascending sequence in (3?"—

converging to $1.

Let En be the set of all endpoints of intervals J (F) for

A

Fe {Fir - . . I

Let (e(1,2), 6(1,’-)) be any pair of special a1, 81, 61

functions for 321

Now suppose that for eaCh k _<_ n we have chosen a pair of

special ck, Bk, 6k functions (e(k, .)’ 6(k, -)) forfF‘l’< in such a way

that ' “ '

. . " '2 -
(1) whenever I _<_k in - l, e 6 Ek+l’ F effk , and

e e F nJ(F)*, then
sce, eck+1, e), eaHEBcF, «sac, F), ak, Bk);
(iii)... whenever I 5-.k in 1, e 6 Ek+1’ and e 6 Bk, then
-
SCe. e(k+1, e). 61M) nHQSCe, 6(k. e), 6k);

(iii) whenever I _<_ k < n - 1, K5 k+1’ Pei-12 , andK_F, then

Bus, 6(k+1,1<), am, Bk 1) nHQBCF sck F), ak. Bk)

Then we construct (e(n+1,-), 6(n-I-1, ')) as follows. Let

(c, 6) be any pair of special an+l’ Bn+1’e2n+1 functions forfFl'1+1 If

e €En+1 - En t,hen for some unique Pei-:12 , e 6 F nJ(F)* , so by (12)

we can choose £(e) > 0 such that n _<_ gCe) implies

Ste, n, enfli nHEB(F, 6(n, F), an, en),

We set t(n+1, e) = min" {e(e), £(e) }. 0n the other hand, ,if e 6 arm n En,
 -41

then'weset e(n+l', e) =‘ min. {8(6): .lf-EUI. é)}-'

,If Fefifl, then there erists: a unique K 6?: WithF 9K.

Set' ..

6(n+1; F) = ,min’{5(1=); %—6(n, -K)}.'

By Lemma 14, (e'(n+l',.‘.)-, 6(n-_I-1,.-‘)) is a-pair of special “n+1’

Bn+l’ en+l functions for fin-1" and by (13) and (9), conditions .(i),

(ii), and (iii) are still satisfied when n is replaced by’n+1. Thus

we can inductively construct a pair (e(n,'), 6(n,-)) of special an,

en, en functions for’f’n in such a way that conditions (i), (ii) and

(iii) are satisfiedfor every n.

Let

Un = [ e\t,}
6 En
S(e, eCn, 6). en )] \J'
) .
\_il B(F, 6(n, F).a 11 , an]
PST?!

Then Un is open. For fired n, all the various sets S(e, aCn, e), an)

(e 6 En) and B(F, 6(n, F), on, 'Bn) (Rafi) are open and pairwise dis-

' joint, so thatevery component of Un is contained in one of the sets

“2
S(e, 6(n, e), an) (e6 En) or B(F, 6(n, F), en, fin) (FEfl‘n). It

therefore follows from (16) and (17) that if 9 is any component of Un’

then

(23) if n x 9. An.

From the fact that (a(n,-), 6(n,o)) is a pair of special on,

an, en functions tori-”h together with conditions as)- and (19), it

follows that
 - .42

[=U —"“"‘T‘—T(e e(n,e nH] u

eeEn

A BiF, 661, F5 , Bn i nH].

[Féf2- an

Consequently, conditions (i), (ii), (iii), together with the fact

that I
1: " . l
e e En+1 - 1511:) e 6 FA J(F) for some Fefi,

Un+1n, H C
show that "' _..Un for every n.

By Urysohn's Lemma, there exists a continuous function

.gn : H -> [0, 1] such that

.gn(z) =11 forzefl-Un

, _l
and Igfi(z) = J tor z 6 Un+l n H. -

Let g(z)'=n21—-—2ngn(z). Then 0 < g(z) _<__ 1, and the series converges

, uniformly, so1 g2 is continuous on H.

If. z 6 H - Un’ then 2 6H - UI“ for every mlln, so that

=, gn(2) =. gn+1(z)' =. gn+2(z)- = . .. , and' hence

m .

' 1 ‘ 1
(24) - g(z)> 2 —=__
m-n 2m 2n 1 a __
(z€H-U).
11

Also, if z €Un+1’ then 2 €U1, U2, ..., Un+1’ so that

l
o = , g1(z) =, 32(2) = =, gnu), and
3 1—m
(25) gm. 5 m=n+l = [1—11 (mum).
2 2

We assert that

(26) for each x0 e A, g(z) -> o as z '-> 5:0 with z e saga, 3%).
Take any naturalnumber n. Since x0 é An+1 = U511, either
' ‘ * A2 .
x06 EM1 or else xo 6 F n J(F) for some Fefnfl. In the first
 43,

case, .set' n = s(n+1‘, x0). In thesecond case, (12) shows .that we 'can

'. choose n > 9 so that

SCXO, n. 3—) QB (F,U6(n+1, F), «n+1, gnu),

Suppose (x, y). E 3(990, 1,. $1) and y < n. Then, in the first'case,

(3:, y) e S(xo, nil) 956:0, e(n+1, 5:0), °n+1) guml, and in the
second case, I

(5:, y) 6 86:0, n,':;’—1)§B(F, 6(n+1, F), «n+1, sn+1)gun+1. So, by

(25). I

(On y) 6 SCXO. 1, g1) and y < nit? (x, y) 6 Un+1

=> 01gCX. y) i .1. nl

N
l
This proves (26).

Let x0 be a point in X and Y any are at 2:0. ,Suppose, g(z) + 0 ‘

as z + x0 along Y. Then Y has a subarc y' with-ohe'endpoint at 5:0

. -1 . 1 '. .1 . .

such that Y l _ {x0} Eg (( _ __

n’ _
2n)). By (24), y' _ {x0} C.
-Un.
2
Therefbre, by (25), x0 e An. Since n was arbitrary, x06 £31 An = A.

Thus ,

(27) if there eicists an arc Y at xo' s'u'ehitthat- g(z) + 0 as 2

approaches :0 along Y, then x0 6 A.

Now . define

. fez, y) = _.gcx, y) sin % + m, y) (<Sc, y> 6-H).

i 5:06 A, then, by (26), f(z) -> o as z + 5:0 with z escxo, 1, 3’81).
Thus every point of A is in the set of curvilinear convergence of f.

Conversely, suppose x0 is any point of theset of curvilinear

convergence of f. Let Ybe an are at 2:0 such that f approaches the

limit c + di along y. Then g approaches the limit (1 along y. If d

 44

is different from zero , then gcfx, y) sin-g;- (the real part of f) cannot

approach any limit along .Y -- a contradiction. , Therefore g approaches

the limit 0 along y, and, by (27), x0 6 A. Therefore A is the set of

curvilinear convergence of f. I

5. Boundary Functions for Continuous Functions

LemmalS. Let E be a metric space, Y a separable metric space.

I
Suppose that q: E +IY is a function having the following property.

;For every open set U SY there exists an Fa set L E E and a countable

set N SE such that

—1 -i
@(U)§LE‘P(U)UN-

. Then there exists a countable set M EB such that ‘PIE-M is of class

(Po. (E - MD-

Proof. Let Bbe a countablebase for Y.. For each 368, let L(B) E E

be an F0. set and let N(B) E E be a countable set such that

191(3) E LCB)C ‘91?) u N03)

Let M = U N(B). Then M is countable. Let E0 = E - M and let

Bee
0 = splfio. We show that I90 15 of class (F<7 (BOD .

Let W be any open subset. of Y.. If p e W, there exists r > 0

such that S(r, p) C; W. Choose 3618 so that p 6 BE S(%— r, p). Then

'FESU, P) EW. It follows that '

W = B = | -§, h— --
'BLcljcm 7 Bg'cjlcm
where Q(W)= {368: BCW} . Therefore '

-1 . -.-1
‘9 cm =E n on =_E n “1(3)
° ° (F seam)
 45

‘5:a L(B)
Beacw)
EB n [Hams]
° Beam) ‘9
U [élcfilu M1
E Eon Baum
= Eon U (1(3)
Beam)
= Eonqilm) = ‘9: (W).
Consequently $3107) = E0 n B GCQCW) .L(B) , so (Palm) is of class

(FO(EO)) . I
Theorem 5. Let Y be a separable metric space and let f : H -> Y be a

continuous function. Suppose that E S X and that '(p: E + Y is a

boundary function for- f. Then there exists a countable set M E B

such that (PIE—M 18 of class (Rom - M)) .

, .
‘Proof. Let U be any'open subset of Y, and let W = (m'. Let

En - {x E X : there exists an arc y at x, having one

endpoint on Xn, such that y -' {x} Ef'1(U)} -
K

' {x e X :‘there exists an arc 'y at x such that

II

y -' {x} Ef'1(W)}.

0b serve that

61w) 9 U En,
n=1
and <p'1(W) 9 K.
For the time being, let n be a fixed natural number. For each

x € K we can choose an arc yi at x such that.

vi -‘ {-x} E Hn n f—1(W) .
 ' .g.-_;....
'46

‘. All- A...“
Since an arc.at x is by.definition a simple arc, YX -'{x} is a

A
connected set and hence must be contained within one.nonempty component

of Hn ‘n f'1(W) .' Let Ux denOte this ,COmponent (for each x e K),

.Let T be the set of all points of K that are two-sided limit

points of fig. We claim that if x, y e T, then u + y implies

Ux n Uy = (b. If U5: n Uy + p, then (since "2: and Uy are two components

of the-same set) UK and Uy are equal._ Let p be the endpoint of Yx

lying in UK and let q be the endpoint of Yy lying in Uy = Ux' We can

join p and q by an arc y lying in Ux' Putting Yx’ Yy and Y together,

we obtain an arc a with one endpoint at x and the other at y, such

that a -' ht, 'y} Q U*. According to [12] we can choose a simple are

(1‘ E on having one endpoint at v and the other at y. Of course,

a' -' {x, 'y} E n Hn n f_1(W) . Let I be the open interval in X with

endpoints at x and y, and let J = X - T1 Let B be the bounded

component of H - a' and let A be the other c0mponent. Since Xn is

unbounded and does not meet o', Xn E A.

Because x is a two-sided limit point of En, we'canchoose a

point w 6 I n En' Let B be an arc at w, having one endpoint on Xn,

such that 3 -'.{w} E f_1‘(U). Then 8 does not meet a‘ (because

u' -' {x, y} Erlcm and f'1(W) n f'1(U) = ¢), and therefore (since
EL '7' {w} contains a point of Xn E A) 13 - {w} SA. It follows that

w E‘KL This, however, is.a contradiction, because the frontier of A

(relative to the finite plane) is u' L! J. We conclude that, for x,y e T,

n+yin1p1iesU$cnUy=¢. .

An open set in the plane has only countably many components,

so it follows that T must be countable. Let S be the set of. all

 47‘

points._of in that are.not two-sided limit.points .of fn' .We.know that

S is .countable, so
I

Ktfi E _ [K n (in - 3)] u [aS]

n
IT u [K nfS] .

is countable .'
00' on . '

Let N: Kn E = U (K APT). Then N is countable, and,

n=1 n n=1 n
since 611W) 9 K, . ~ ‘
-1 c m C m _
(9 (U)- En n=1 En _EnL=J1 En

= EnKn f)u((E-K)n f)
'( gléjln n=1 n

E (E ’n N) u (E - K) g (E n ns—u-(E - <p‘1(W))

= (E n N) u vp'lcfi).
l.
- co

Thus ‘P-ICU) E E n q En Q03 n N) U {1(5) , and the desired result

11:
follows from Lemma 15‘. I

Corollary. Let Y be either the Riemann sphere, the real line, or a

finite-dimensional Euclidean Space. If f : H + Y is a continuous

function, if E Q X, and if (p: E -> Y is a boundary function for f,

then (9 is‘of honorary Baire class 2 (E, Y) ..

Neatt we show that the foregoing corollary is as strong as

possible in the sense that if E is any subset of X and (p is ‘a function

of honorary Baire class 2 mapping E into a suitable space, then there

exists a continuous function in H having «9 as a boundary function. A

proof of this result-ma at least for real- or vector-valued functions «I-

was outlined by Bagemihl and Piranian [2, Theorem 8], in the case
 48

where E = X. Although the construction given here is carried out much

more explicitly than the construction given by Bagemihl and Piranian,

my treatment differs from theirs in only two aspects that are of any

significance. First of all, the proof of the theorem for arbitrary

subsets E of X depends on Lemma 6 of the Introduction. Secondly,

[Bagemihl and Piranian say in the last line of their proof that there

is "no difficulty now in extending f continuously to the whole of D

in such a manner that ¢ is a boundary function for f." While this

appears to be all right for real- or vector-valued functions, it is

not clear why the extension should be so easy for functions taking

values on the Riemann sphere. Theorem 7 of the present paper shows,

however, that the result can be obtained for functions taking values

on the sphere once it is known for vector-valued functions.

The following miniature closed graph theorem will be a

convenience.

Lemma 16. Suppose that M is a metric space and that u : M + R is a

function having the following properties:

(i) if {pn} is'a convergent sequence of points of M, then

'{u(pn)} converges neither to + w nor to - m'.

(ii) if. {p11}; M, p e M, and y e R, and if p11 3 p and

uCPn) Ky. then u(P) = y-

Then u is continuous.

Proof. Suppose that {pn} is a sequence of points in M converging to

a point p E M. Using (1) it is easy to show that {u(pn)} is a bounded

sequence. Suppose that {u(pn)} does not converge to u(p). Then there

exists a subsequence {u(Pn(kl)} that converges to a real number

y + u(p). This, however, contradicts (ii). We conclude that

 49

- 11(P'n) :MpLI

Lemma 17. Let h : R + R be a strictly increasing function such that

MN is neither bounded above norbounded below. Then there exists a

*
continuous weakly increasing function 'h : R -> R such that 'h*(h(x)) = x

fior every x E R.

Proof. Let Z = h(R). Observe that h"1 z z -> R is strictly increasing.

For any x€ R, the set (-°°-, x] A Z is nonempty. Also, h-1((-°°, x] n Z)

is bounded above, because if we choose y 6 Z with x i y, then

h’1((-~, ,1] A Z) is bounded above by h'lcy). -

We claim that for every x E R

-1 '_ -1
(27) sup h (,(-oo, x] n Z) = sup h ((-oo, 2:) n Z).

If x t Z, the equation is trivial. Suppose x e 2. Then

y < h'lcx) 2:) (hcy) < x and My) 6 2),

so that hcc-e, h‘1($c)))_C_(-o, 5:) n 2. Hence
“'1 C '1
(-w. h (X))_h ((-N. X) n Z),

so that sup h‘lcc-oo, Son 2) :h‘lfic) = sup h'1((-m, ,x].r\ 2). The
opposite inequality is trivial, so (27) is established.

We also claim that

. —1 ' -1
{28) inf .h ((x, +00) n Z) = sup h ((-w, .x] n Z).

Obviously, inf h_1((x, +oo)nZ) : sup h_1((-°°, x] A Z). Take any

y > sup’h-lfl-oo, x] n 2). .If My). 5 x, then My) 6 (-«o, .x] n z, and
so yen-110‘», x] n Z)-- a contradiction. Thus h(y) > x and
My) 6 (it, +°°) n Z. Therefore y 6 h'1((ic; +w) n Z), and so

inf'h-1((x, +oo) n Z), _<_y. In view of the choiceof y, this implies.

 . 50

that .
. ‘1 .. "_1 _ .
Inf h (CX. +°°) n Z). i sup h (.(~°°,_,x] n 2,),

and (28) is established.

Define

* ' -1 4
h (x) = sup h ((-°°, x] n Z).
*
It is clear that h* is weakly increasing and thath (h(x)) = x for
' * I
everyreal x.. The continuity of ‘h can easily be deduced from the

equat ions

sup "h*cc-~, no) 'h*(x)

inf h*((x, m.) :h"(x) ,
which are established as follows:

sup -1 .
sup 'h*cc-°°, .50) .y<X sup h ((-m’ Y] n Z)

= sup Elm-8750"}? 2)
= sup h'1((-°°, fx] A Z)

513(k)
inf 'h*((k; «on :2: sup h'1((-~.fy] A Z)-
inf
y>x inf h'1((y,- m) n 2)
II

inf h'1((ic; m) A Z)
.sup h'lcc-w, 2:] n 2) ,
’h*(x). I
Theorem 6'. Let E be any subset of X and let <9: E + Rq be any

functionof honorary Baire class 2(E, Rq) . Then there exists a

wontinuous function f : H + Rq such that (p is a boundary function for

f.
 51

Proof. Let 4; : E + Rq be a function of Baire class 103, Rq) and N a

countable subset of B such that 4.90:) = Mac) for every x e E - N. Let

' {sn}n:l (with n + m implying sn =I: Sm) be a countable dense subset of

X that includes every integer and every point of N. Let

tn = _1 1f sn is an integer

_ 1 . " . .
tn — E if sn 15 not an integer.

Define

h(x) = Z tn if x > 0
.0<$ <x-
—n

.. h(x) = -.-z tn ifi:_<_0.

.xjsnio .

Then h is a strictly increasing function from R into R, and MR? is

. *
bounded neither above nor below. Let h be the function described in

Lemma 17.

Suppose that o. < y. < 1. Then (for fined 50

u-h*(—-—Lx-}(,1-h)u) _""

is a strictly increasing continuous function of u that approaches

+00 as u -> +oo and ~09 as u + ~00. Consequently there exists precisely

one number u(x, y) that satisfies the equation

(29) u(ic, y) _-h*(x_:_£.1_'yg’uifl) = '0.

I claim that u(ic, y) is a continuous function on

H1 =’ {(x, y) :ic, y 6 R and 0. < y <'1}. Suppose'{(xn, yn)} 9 H1

and (Kn, yn) + (x, y) e H1. If uCXn, yn) + +oo, then

_xn - (l-ygaixn, yfi) + m

9
Yn n

and hence .. . . . _ '.

xn-e (l-yn)u(xn.-yn)
ucxn, yn) - h*(t y 3 3 «a,
n
 , .52

which. contradiCts. (29)... Thus qn , yn) cannot approach +°°. . A similar,

argument shows that u(xn’ yn) cannot approach -°°. . Now assume that

nun, yno)->u €R Then, by (29),

x:.--(1- gJq , yn)

0=1iwmlucxfl,y)-h*(“ j,“
n
, * _X -. {Ia-flue
= -uo ' h (—-—-}7—-—) ,

so u0 = u(x, y). By Lemma 16, u is continuous.

From Lemma 6, there exists a sequence. -n-—-1{gn} 00 of continuous

functions mapping X into Rq such that gn(x) + Mac) for each x E E.

n
For n > 2, define .

focx; y) = maul) - rugncucsc, m + (cw) - m(n+1).)gn+1(u(x, y))

when mi), EH. I

Then f0 is continuous on H2 n H. By the Tietze extension theorem,

we can assume that f0 is defined and.continuous on all of H. Let

__ inf
rn ' 'x>s n Mac)

sup
-
12'n ' x<sn h(x)

vn =;-V(sn) - Msn) if sné N ,

vn = .o if 511+” N.

If x and y are real numbers, define xV y = maidx, -y}. For (it, y) 6H,

set

An(x, Y) =
. ..... 1.’ ' _ ”sh-X ‘
[(1 - ny) v 0].[(1.- I,——--—| rn + ”n - i2sn + 2 | )V,0]vn.
='3||'-‘

Then An is continuous in H. Observe that An(x, y)= 0 when y_>_

 55

Using this -fact,. it iszeasy to show that, if we' set

W . _

f = f + 2 A. ,
0 n=1 n

then f is defined and continuous in H. We now show that (p is a

boundary function for f.

Let p be any point of “E. The line

(30) x =.(h(p) ~p)y+P

passes through (p, 0), and the part of it that lies in El is an arc

at p. We will show that f approaches Mp) along this line. If we~

substitute (h(p) - p)y + p for x in the expression for Ana, y), we

I I

obt ain

(31) - Anck, y) =
[(1 - ny) v 0][(1-'-1;---}—£—|:c1'1 + an +‘2(% - 1) (Sn - p)
n n

— 2h(p)| )v‘o] vn.

If p _<_ sn, then h(p). 5. 2n, and one can verify directly that. (31)

vanishes. If p > sn; then h(p) 3_ rn, and again one can verify directly

that.(31) vanishes. Thus An(x, y) vanishes along that part of the line

(50) lying in H.

Solving (30) for h(p) , we find that, along the given line,

h(p) = L'LLiiilR ,

and .hence p = -h*(h(p)) ='h*('l‘;>;£ifl£). Therefore, if 0 < y < 1,

__ ' ' . . . .' ' 1, 'l
p — u(x, y). So, 1f (x, y) satisfles (30), n :2, andn—H—iy. ii"

then

focx; y) = .(yncn+1) - n)gn(p) + ((n+1) — YnCn+1))gn+1(P)-

 . 54

Sincethe coefficientsof gn(p).anc1.‘gn+1(p) in the above eacpression

(ladd‘up to 1 and since both coefficientslie in [0;.‘1],. fo(x; y) lies

on the' 1ine.s_egment joining gn(p) to. gn+1(p); and it follows that

. fo(ic, y) apprOaches Mp) as <5", y) approaches p along the line (30)..

Since each An vanishes on the part of this line lying in H, f(x, y)

alsoapproaches Mp) along the line.

Let sm be any point of N. We show that f approaches (9(sm)

along the part of the line

1‘.. 2.
(32) — 'M
x —_( 2 _. sm)y+sm.

that lies in H. Again, we first consider the value of An along the

. given line. Substituting the value of x given by (32) into the expres-

sion for An, we obtain

(55) Anew)
[(1-ny)V 0][(1_ - rain |rn - rm + 2n - 2m .+ 2% - 1)(sn.- sI) V 01v“.

If sIn < sn, then 2’m < rm ; 2n < rn, and one can verify directly that

(33) vanishes. If sn < s m’ then 2'n < rr'1 _<_ zm < rm, and again one can

verify that (33) vanishes. Thus, for n + m, An(x, y) = 0 when (x, y)

lies on the line (32) and in H.

If we take n = m in (33), we obtain

Am(x, y) = [(1 - my) V Olvmg

Therefore .Ac, y) approaches vm = ¢p(sm) - Msm) along the. given line.

Take any (x, y) 6 H1 satisfying. (32), and take any a and b

satisfying

(34) a<s <b.

 » 55

.r. 1-.2,
Then. h(a) < "m _1n_2____< rm .< h(b), so that .

(h(a) - sm)y + sm < x < (h(b) - 5111)}? + sm; from which. we deduce that

‘ . 'x - (14305“-1
h(a). s. - <, h(b).

Since ‘h* . is weakly increasing,

* . * x ¥.'(I'-.'y)sm *
a = h (h(a)). :‘h (——y———-—). 5h (tD = .13-

Since a and b were taken to be:any two numbers satisfying (34), we

conclude that

.* x - (1-y)sm
sm = h (——;——-——),
' whence it follows that u(x, y) = sm. Thus

. f0 (x, y) = cyn(n+1) - n.)gn(sm) + (gnu) - yncn+1)_)gn+lcsm)

1 < y, 53‘
when (x, y) lies on the given line andn— Consequently
fOCX, y) approaches Msm) along the line (32). So f approaches Msm) 4-

('(sm) -- Msm) = (p(sm), and the theorem is proved. I

Theorem 7. Let E be any subset of X and let (p: E + S2 be any

function'of honorary Baire class 2(E, 82). Then there exists a

continuous function f : H + $2 suchrthat qv is a boundary function for

f.

Proof. The proof of this theorem is very similar to that of Theorem 1.

Since 82 €115, there exists, by Theoremlé, a continuous function

_ g : H +‘R3 having q: as a boundary function. Let

K = _ g'1.({v€'R3 : |v| =% })

L = .. g-1'({v a Rs : |v| 3% })

F =__g'1({veR3 : |v|_<_%}).
 .56

.Let_'.go =_-gIK-. H is.homeomorphic.to.R2,.so_by [5, .Lemina 2.9,.p. 29.91,- ..

_ g6 canbe extended to. acontinuous function

- 5 . , 3'1
'g1.:H+{véR :.Ivl =.-2-}.

.Define f1 : H + R3 -‘ {0} ,bysetting “

f1(z) = f g(z) if z e L

f1(z) = . g1(z). if z E P.

Then, since F and L are closed, fl is continuous on 1-1. It is easy to

verify that (p is a boundary function for £1. Let Po : R3 -' {0} + 52

be the 0—projectipn onto 82 (see page 11), and let f be the composite
. . . 2 .
= «p
function Po 9 f1. ’ Then 1" maps H continuously into S , and Poo q:

is a boundary function for f. I ' '

 CHAPTER II

BOUNDARY FUNCTIONS FOR DISCONTINUOUS FUNCTIONS

6. Boundary Functions for Baire Functions

It is not known whether the set of curvilinear convergence

of a Borel-measurable function defined in H is necessarily a Borel

set; The answer is not known even fpr functions of Baire class 1.

However, a theorem on boundary functions that is similar to the

corre5ponding result for.continuous functions in H can be proved for

functions of Baire class 5 in H.

Definition. If A and B are two sets, we will call A and B equivalent

and write A = B if and only if A - B‘and B --A are both countable.

It is easy to check that = is an equivalence relation.

Lemma 18. If A = B, then S - A = S - E for any set S. If A.n = En

for all n in some countable set N, then

\IAn= \IBnandl \An='flEn.

n E N n E N n<e N n e N

The proof of this lemma is routine.

Definition. An interval of real numbers will be called nondegenerate

if it contains more than one point.

Lemma 19. Any union of nondegenerate intervals is equivalent to an

all .a... .

Open set.

57
 S8

Proof. . .Let <9.be any-family ,ofnondegenerate. intervals. ,It.will

* ..
.sufficeto prove that U I - U 'I .is.countable.. .We: can write
16.3) . Ieé} .

U.“ U
16

where. {Jn} is a countable family of disjoint open intervals. If

'k
x 6 I ) I - U 'I ,
° lee) 16.9
. l *

then x0 is an endpoint of ID for some Ioé (9‘. For some 11, I0 Q Jn’

so that x 63. Butic ¢J , so 2: is an endpoint ofJ . Thus

0 n o n o -n
I - g 1* is contained in the set offlall endpoints of the
16.3. I {9- "
various Jn’ and the lemma is proved. I

Lemma 20. Let h be a weakly increasing real-valued function on a

nonempty set E QR. Suppose that Ix - Mic) I g 1 for every x 6 E.

Then h can be extended to a weakly increasing real-valued function h1

on R.

Proof. Let e = inf B (e may be -°°). For each x e (e, +°°), set

h1(x) = _sup h((-°°, jic] n E).

Since |t - h(t) I j 1 for each t e E,

t ec-oo»; x] n E ::>h(t) $5: + 1,

so h1 is finite-valued. If e = -o° we are done. If e > T°°s then

xe E implies h(x) _>_;c - 1 _>__e - 1, so h is bounded below. For

2: € (-oo, ~e] set

h1(x) = inf 11(5).

It is easy to verify that hl‘ has therequired properties. I

 59

Lemma’Zl. Let Y be a metric space, f : R i Y a function of Baire class

. ECR, Y), and suppose that h : R + R is weakly increasing. Then there

exists a countable set N .C_R such that the composite function

f'o t-N is of Baire class €(R — N, Y).

Proof. Let N be the set of discontinuities of h. By a well-knoWn

theorem, N must be countable. But then t-N is continuous, so that

f o (hIR_N) = (f . h)|R_N is of Baire class :02 — N, Y)‘.‘

Lemma 22. Let Y be a separable arcwise connected metric space, E any

metric space, and let (p: E -> Y be a function having the following

property. For every open set U E Y there exists a set T E PE+l (B) such

that (film) ET (iii-ICU). Then, if F, _>_2, (p is of Baire class 503, Y).

Proof. The proof is similar to that of Lemma 15. Let E be a countable

base for Y,. and suppose that W is any open subset of Y._ Let

(12m) = mete; UEW}.

The argument in the proof of Lemma 15 shows that

w = U U = U U.
i
U 6 (10V) U E (DEW)
€+1
For each U68, let T(U) 6 P (E) be chosen so that

«9401) Q Tang. 61(3). Then

“lcm = U6
U
€(RCW)
(940139 U W)
U 6 (MW)
C -1 . -1
— (FF (W).
ué’acm ‘9 ‘9
Thus Ip'1(W) = U T(U), and since Pad-103) is closed under countable
ueacm
unions, (P'1(W)€ PEHCE). Therefore ‘9 is of Baire class 505, Y). I
 60

TheOrem' 8. Let Ybe a separable arcwise connected metric space, .

f : H + Y _a function of Baire.c1ass’§(H; Y)_where E; _>_ 1, _E. a subset of

x; and I '(p: E -§ Y a boundary function for 15.. Then \pis of'Baire class

g + 1(E,,Y):.

Proof. Let U be any open subset of Y and let _V = Y 7 11'. Set

A = (p (U) a B =.‘P_1(V)
C =IAUB. I

Observe that A n B = ¢- For each x e C, choose an arc vs: at x such

that

lim
v z+x f(2) = $00
2e

§x§{z : Iz - x]. :1}

7x {ix} 9 f'lcm if x e A

yx -"{x} g f-ICV) if 5: e B.
Notice that if x e A and y e B, then yxn yy = 4:.

We will say that y)“ meets Yy in in provided that Yx and Yy

have subarcs yx' and yy' respectively such that x’e yx' 9 En,

Y 6 Yy' gin, and Yx'n Yy' +¢. Let

La =- {x6 A 2 (Vn)(3y) (ye C, y + x, and 'yy meets 7* in T-I-n)}
W“ . VMr—T.“

Lb =. .{x6 B : (Vn),(3y) (y e C, y + x, and 'yy meets -yx in 3-11)} -

Ma =- {xe A : (3n)(yx meets ho yy (with y + x) in in);
Mb =' {X'E B : (3 11) (7x meets no Yy (with y + x) in‘fi'n) }.
L = La” Lb
M = MaUMb'
 61

ObserVe that La,‘ Lb, Ma’ Mb are-pairwise disjoint ,Aand that

A=LaUMaandB=LbIUMb5 .

For each x e M, 1et'n(x) be apositive integer such that Y5:

meets no Yy (with y + x)'1n Hn (x) . Then n i n(x) implies that Yx

meets no yy 1n Hn' Let

Kn =' {x6 C : Yx meets Xn, and, if x 6M, 11 _>_n(x)}.

on
Then KnE Kn+1 for each n, and C = g Kn.

We next show that for each positive integer n and each x 6 La

there exists a nondegenerate closed interval 1: such that

x 6 1:9. La u (X I‘- Kn) . By the definition of La’ there exists

. y E C (y :I: x) such that yy meets-vx in 'I-Tn. Let I: be the closed

interval having‘its endpoints at x and y. Let t be any point of 1:.

We must prove that t e La L) (X — Kn). If t $181, we are done. So

assume t 6 Kn. Then Yt meets Xn, and hence it is clear‘from Figure 5

that yt must meet either Yx or Yy in in. (This argument can be

rigorized by means of Theorem 11.8 on p. 119 in [11].) But, if t 6 M,

then (because t e Kn) n _>_ n(t) , so that this situation, is impossible.

Therefore t 4-: M. Now x 6 La Q A, 50, since Yx intersects Yy’ y 42 3.

Hence y e C - B = A. Similarly, since Yt intersects vi or yy,

t 6 C - B =hA. Thus t6 A — M= La’ and we have shown that

IgeLau (x — Kn).
n
Let Wn - U Ix' . For each n,
xeLa
-
c - .
Lagwnn G _[La u (x Kn)] no,

and therefore
 6?

Xn

Figure 5.
 .63

Lagcqwnn‘c
n:

_' {1101 [La u (x.- Kn'ill'n c ’

'_[LaU (x —k}1'1<n)] n c
n=>

(Lan mum-911%) = ‘Lauo = .La.

on

It follows that La = (fl Wn) n C. By Lemma 19, each Wn is equivalent

' n=1
to an open set, so there exists a (‘3 set Ga EX such that
l

Lag GanC.
-q. '

A similar argument shows that there exists a Gd set Gb E X such that

Lb = (3b n C.

Next we study the properties of Ma‘ -In doing this, it is

convenient to define a function 1r : R2 -> R by setting 1r(x, y) = x.

If x E M nKn, then, starting at x and proceeding along Yx’ let pn(x)

be the first point of Kn that is reached. Define h: : M n Kn -> R

by setting h:(x) = 1r(pn(x)). If x, x' e M n Kn and x < x', then, since

Yx cannot meet Yx' in in" it is evident that pn(x) must lie to the

left of pn(x') ; that is, mango). < “(Pnb‘m' Thus h: is a strictly

increasing function on M n Kn. Moreover, -

|x - 'h:(x)| i 1 because yxg {z :»|z .- x| i 1}.

So, by Lemma 20, hi: can be extended to a weakly increasing function

hn:X->R. Let

,gncx) = ,fchncic); %) (x e R).

 64

Hit; -11T).is.a.fm1ction (ofx) ‘of Baireclass £(Xa Y)_, so, .by Lemma 21.,

there-existsa countable‘set'Nn X.Such that. gnIX-N is .of Baire Class

. ca . ~ . n
gcx - Nn, Y)_. Let N = U Nn' The!) gnl is of>Ba1rec1ass

ECM " N: Y)-

For is M n Kn) gncéc) =, £(h3(5c),%) = f(pn(it)).' “-If 5: am; then

for all sufficiently large n, ice M n Kn, so ~

111:2 gnCi) = 11:: fCPnCil) = 9(x).

Thus gnlM ; QIM, so gnlM-N :‘PIM-N’ hence 'PlM-N 15 of Baire class

E + 1(M - N, Y) . It follows that there exists D 6 Pgi‘2 (X) such that
. _1 .
A n(M - N) = -(‘PlM—N) (U) = n n(M — N).

Obviously A n M = D n M. Now,

L = -LaUI‘h = (Gan C) u (Gbn C) = (Gav Gb)n o,

50
.

a AnM-apnM: DACC-L)
z

II

’Dntc - ((Gauabmcn
-D‘n [X - (Gau Gb)]n C.

Ga and Gb are GG.’ so X - (GaU Gb) is Fa , and hence”

2 5+2
x - (GauGb)€ p (X)§_P (X).

Therefore Ma = F n c, where F e P‘5+2 (X). Now, Ga 6 65m = (220:), and

since 2; :1, Q2(X) £2. P£+2(X), so Gau F €'P€+2(X). But
A =‘LauMa=(GanC)u(FnC) = (Gav F)nC,

so A 2 Sn C, where S e 1.95""2 (X). Since every gauntable set is F0, it

is now-easy to show that

 .65

A = The

for some T e P§+2(X) . From the definition of C it follows that

TEX - B. Thus we have

= 94(6).
‘P-lfll) = AQTnEEB-B = E-‘P'lm
g+2
T n E E'P (E), so Lemma 22 shows that «p is of Baire class 5 + 1(E, Y)...

Corollary. Let Y be a separable arcwise-connected metric space,

f : H + Y a Borel-measurable function, E a subset of X, and «p: E + Y .

a boundary function for f. Then £9 is Borel—measurable.

Proof. f is of some Baire class gU—l, Y) _, hence (p is of Baire class

g + 105, .Y), hence (p is Borel-measurable. I

This corollary raises the question of whether a boundary

function for a Lebesgue-measurable function is necessarily Lebesgue-

measurable, which we answer in the next section.

7. Boundary Functions for Lebesgue-Measurable Functions

Suppose that a0, b0, a1,'b1 are extended real numbers, and

that a0 :b0, a: 1 b1' To make the formalism more convenient we let

(-°°) .- (-°°) = 0 and (+00) - (+4») _= 0. In other respects we adhere to

the usual conventions regarding arithmetic operations that involve

-o° or +00. Let

T(ao, b0; a1, bl) ='{ (x, y) :0 1y :1 and

(a1 - aob' + a0 1x 1 (b1 - bob' + b0}.

A set of this form will be called a closed trapezoid. We also

consider ([5 to be a closed trapezoid. A set S will be called a

trapezoid if there eitists a closed trapezoid T such that T1 Q S ET,

I
 66 .

where T1 denotes the interior of '1' relative .toTi'llt .Ever'y trapezoid is

Lebesgue-measurable, though not .nec'eSSarily Borel-mea'surab 1e .'

If s, s' are disjoint line segments having endpoints (a0, 0) ,

(a1, 1) , and (ao', 0) , (al', 1) respectively, where ai. fiai'

(i = O, 1), then let
__

T(s, s') = T(s', s) = T(ao,-l,ao ' 3 a1, al').

If s = _s', then we 1et;T(s, s") = T(s', s) = s. 'In what follows we

will use the symbol Xo as an alternative designation for the x-axis

X. This will enable us to make statements about xi (1 = 0, 1) (where

X1 denotes, as before; { (x, l) : x 6 R}) .

We omit the proofs of the following two routine lemmas.

Lemma 23-; Let theline segments 51', 52}, 53, 54 each have one endpoint

on x0 and the other on x1, and assume that i + j implies that either

Si n sj = st: or 5.1 = sj. If T(sl, 52) n T(SS’ 54) 4f ¢, then

T(sl, 53) Q T(sl, 52) u MS?” 54).

.Lemma 24. Letx be any set of line segments, each of which has one

endpoint on X0 and the other on X1, and no two of which intersect.

Then K ) T(s, s‘) is a trapezoid.

s,s'€.$ 2 .
Let m denote two-dimensional Lebesgue measure in R . If E

is a measurable subset of some line in R2, let m£(E) denote the linear

Lebesgue measure of B. Let-me and m: denote two-dimensional exterior

measure and linear exterior measure, respectively; i.e. , for any

E 9R2;
meCE) = inf. ImCU): E EU and U is open};
 .67
- l

and if E is.a subsetjof a line L, then'

‘ m:(E) = inf"{m2(il) : E E U £3 L and U is open' relative to L};-

Theorem 9. Let L be any set of line segments, each of which haspone

endpoint on X0 and the other on X1, and no two of which intersect._

Let S = Uat . Then

I
_' -
1 .ns
me(b), - -2- 0 X0) + méCS n X1)) .

Proof.” We may assume that oC is nonempty. Let a be any positive

number. Choose an open set U E R2 such that S 9U and

m(b).'_gme(5) + e.

Let Ei = S n xi (1 = 0, 1). Choose sets Gi E xi that are open relative

to Xi such that Ei C
.. 61 and

'm“(G)%m “(E)+e
i —; e i
(1:01)
’ '

Let V be the union of all lines L ERZ such that L meets both Go and

G1. It is easy to show that V is an open set. Furthermore, S EV

and V nXi = (3i (i = O, 1). Now let W = Un‘,’. Then

Wis open, SGWEU, and ‘ '—

f EiQWAXiEui
"' (1'= 0,1).
If s, s'eL , define s a 5‘ if and only if T(s, s') S- w.
It is easy to verify by means of Lemma 23 that} _ is an equivalence

relation. Let I' be the set of a11 equivalenceclasses. We prove that

I‘ is countable.

If s E .C , we let <ai(s), i) be the endpoint of s on xi

(1 = 9, 1). Then I
’

s = '{(x, y) E R2 _:_0 iylil and x = .(a1(s) - a°(s))y + ao(s)}.

l
 68

Since '5 is‘c’ompact andcontainedinwrthere is no difficulty in

showingthat there exists 6.5'- > Osuch that

'{(x,Y)€R2 :Ogyil and

(a1(s) - a°(5))y + ao(s) - (is. 45: < (alts) - a°(s))y + ao(s) + 65}

9 w'.
Let Ji(s) = _(ai(s) - as, ai(s) +635) Pi = 0, 1). A sketch will

rapidly convince the reader that if s, s‘ 6 «C , Jo(s) n Jo(s') + 9b,

and J1(s) nJ1(s') + h, then T(sl, 5') SW, so that s a 5'. Thus
(JOCS) x cs)) n .cJocs') x cs')) + «p 2 s" a s'.
For each ‘C e I', choose 5(C) E C and let

Q(C) = J°(s(cn x J1me».

Then C1 + C2 :qml) n Q(C2) = g2. Since each Q(C) is a nonempty
open subsetof R2, this implies that I‘ is countable.

If C EI‘, let

TCC) = U T(s,s').
s, s'ec
By Lemma 24, TCC) is a trapezoid. Also,

(35) c g T(C) S w.

Suppose that C1, C2 E I‘ and C1 + C2. We claim that

T(C1) (\T(C2) = q). Assume that T(C1)nT(C2) =|= 4,. Then there exist

$1, 51' e C1 and 52, 52‘ €C2 such that T(sl, 51‘) n TCSZ, 52') + (b.

By Lemma 22., ' k

T(sl, 52) g T(sl, 51')” T(sz, .52..) 9 W,

so that 51‘: 5‘2; a contradiction. Therefore T(C1) n T(Cz) = 95.

 69

.Let'.Ki(C) = me) n x1.L 91-. 9,.1).. ThenKi(C) is an interval

and

(3.6) E. CUK.(C)CWnX. CG. (i=0,1).

CEI‘

Furthermore, 01+ C2 implies that K.1(Cl) nK.1(C2)= . Using the

formula for the area of a trapezoid, we find that

%[m2(UKO(C)) +Im2'CU KICCD]

cer C6?
2 hmhx (CD + m”1m (cm
C€.I‘2

=_ z m(T(C)) = m(U T(C)).

cer
Let or. = ;[mH(UK(C))+rm(CUK1(C))]
cer

= m( Ucn , ' ' ‘

cer
According to (35), s C U TCC) c w 90, so that
car

(37) mew) f. a 2mm). 5me(S) + a.

By (36),

(38) % (mime) + m‘e‘wln _<_ a' :71 (Mac) + mzccln

1 2, 2, _
, :2— (me(Eo) + me(E1)) + e.

Since a is arbitrary, inequalities (37) and (38) imply that

. meCS) 2
1 (meCEo)
_- ~2- 2,
+ me(E1)). I

One wonders to what extent a result resembling the foregoing

theorem might be obtainable without the hypothesis that no two of the

line segments intersect. '1‘Hé”following eitample is relevant to this

question. .Let Mo be a residual set of measure zero inIKo and let M1

 '70

be a.residua1.set ofmeasure.‘zero in'xll .Let (5:0; yo). .be.any point

of H1. we claim that there isia line.‘segment-passing through (x0, yo‘)

. that has one endpoint in M6 and the' other in M1. For 6 6 (0,.1r), let

‘ F1(e)‘ ((1 - yo) ctn e + x0; 1) and

Face) (x0 - yo ctnve, 0).

Then Fi is a homeomorphism of (0, 1r) onto Xi, so FSICMO) and Fi1(Ml)

are both; residual sets in (0',.1r). Choose on e F;1(Mo) nFi1(Ml). Let

L be" the line whose equation is

x = x°+ (y-yo) ctna.

Then L passes through the points (x6, yo) , Feta) and Flat), so that
L nfi'l is the desired line segment. Let A: be the set of all line

I segments having one endpoint in M0 and the other in M1' Let S = LL];

Then S n X0 and S n X1 both have measure zero: but, as we have just

I . .
shown, 1-11 9. S, so that S has infinite measure. See Problem 5 at the

end of this paper.

Lemma 25. For every 6 >_(_) there exists a strictly increasing real-

valued function h on R such that h(R) has measure zero, and, for every

real x, Ix - h(x) I _<_s. .1

l '9'”. .
Proof. For each integer n, let In = [ne, (n + 1):]. Then k] In = R.
n—-°°
.
There exists a strictly increasing function f : [0, l] + [0, .1] .such

that m2’(f([0,1]))= 0. For example, such a function may be defined

as follows. Any number in [0,. 1) may be written in the form

. . ha

-. .va1a2a3. . .an. . . (binary decimal) ,

where the decimal does not end in an infinite unbroken string of 1's.
I
 71

.Set.,

f(fa1a2a3...an...) = . 'b1b2b3....bn... . (ternary dec1ma1),‘

wherebi=pifai=oandbi=2ifai=1 oC

.Set HI) _1. Then 1" maps [0, [1] into the Cantor ternary set, so

m&(f([0, 11)) = 0. It is easily shown that f is strictly increasing.

For each n, it is easy to obtain from f a function fn : In -> In

such that in is strictly increasing and m2'(fn(In)) = 0-. Set

hm =‘ fnfic) for x e(ne, (n + 1);].

There is no difficulty improving that h has the required properties. I

Theorem 10. There exists an indexed family. {Yx}x E x of simple arcs

such that

(i) for each x e X, Y5: is an are at x

(ii) x + y ==>wr,'c mry =.¢

(iii) \ ’ '72: is a set of measure zero.
x 6 X -

Proof. For each natural number 11,- let' hn : R 4 R be a strictly

increasing function such that hnfli) has measure zero and, for every x,

Ix - hn(x) I _<_ i. For every are R, let sn(x) be the. line segment

joining the point (hn(x) , %> to the point (hn+160 , 137). Since

hn'gicm) < hn(x2) ::>x;r 4 x2 :b hn+1(x1). < hmlficz),

we see' that x1 4: i2 implies snficl) n sndcz) = g). Let

Sn = Usnfic) : xe'R}. Then

Snn xn §{{k; %) : xe hn(R)}

and Sn n xn+1 91.65%) : x e hn+1(R)},
2. ‘ _ 2 _ 0. - . easy to deduce from
so In (Sn n Xn) — m (Snn xn+1) - It 15
 . 72.

Theorem. 9 that I

, . _ '1---1 1 2 .z- . _
me(Sn) ' I; ' ET) 7101145110 xn) + mewnn xn-fln " -o'

For x E X, let Yx =' '{ic}ULJ1'sn(x). Since (hndc); El?) 3 it, vi is

n= ,
an are at X.

mc v'x) imam + meaJ Sn)

xeX

' :meCX) + 2 me(Sn) = ‘0’

n=1

so K J y is a set of measure zero. I

xeX x

Corollary. Let (pbe an arbitrary function mapping X into any topologi-

ca'l space Y having an element called 0. Then there 'ekists a function

f : H + Y such that f('z) '= 0 almost everywhere and \p is a boundary

function for f . 1

Proof. If. {Yx}x ex=is the family of arcs described in Theorem 10, let

f(z) 0 if z is in no 75c

f(z) ‘9“) if z eyk.

Then f is the desired function. I

Corollary. There exists a real-valued Lebesgue-measurable function f

defined in H having a nonmeasurable boundary function defined on X.

 SOME UNSOLVED PROBLEMS

1. If A is an arbitrary set of type Fad“ in X, does there necessarily _

exist areal-valued continuous function f defined in H having A as'

its set of curvilinear convergence? If q; is an arbitrary real-valued

function of honorary Baire class 2 on A does there exist a continuous

real—valued function f defined in H having, A as its set ,of curvilinear

convergence and ‘9 as a boundary function? -----

. 2. (First proposed by J. E. McMillan. [10]). If A is any set of type

F05 in X and if (p is any function of honorary Baire class 2(A, 32),

does there necessarily exist a continuous function f : H + 82 having

A as its set of curvilinear convergence and \9 as a boundary function?

3. If f is a real-valued Borel-measurable function defined in H, is

the set of curvilinear convergence of f necessarily a Borel set? What

if f is assumed to be of Baire class 1?

4. Let s =' {(x, y, z) e R3 : z > 0}. If f is a function defined '

in S, we define the set of curvilinear convergence of f in the obvious -

w'ay. If f is continuous, is its set of curvilinear convergence

necessarily a_Bore1 set? Is it necessarily of type F06?

5. Let .c be a set of line segments each having one endpoint on X0

and the other on X1, and let S = IU£. Assume that S is a Borel set.
If m‘Q'(S n_xo) and m2'(_S n X1) are known, what lower bound can be. given

.73 ‘-
 '74.

for MS) ? . , Asolution .to. this "problem might .be ‘helpfulinattackipg a

problem'of-B'ag‘emihl', Piranian, aniYoptng [3, Problem'.'1].-

 I

REFERENCES

1. F. Bagemihl, Curvilinear cluster.sets of arbitrary functions,

Proc. Nat. Acad. Sci. U. S. A., i (1955) 379—382.

F. Bagemihl and G. Piranian, Boundary functions for functions

defined in a disk, Michigan Math. J., 8 (1961) 201-207.

F. Bagemihl, G. Piranian, and G. S. Young, Intersections of cluster

sets, Bul. Inst. Politehn. Iasi, S (9) (1959) 29- 34.

S. Banach, Uber analytisch darstellbare Operationen in. abstrakten

.Raumen, Fund. Math., -17 (1931) 283- 295.

S. Eilenberg and N. Steenrod, Foundations of. algebraic topology,

Princeton University Press, Pr1nceton,New.J6rsey, 1951.

F. Hausdorff, fiber halbstetige Funktionen und deren Verallgemeinerung,

Math. Z. ,5 (1919) 292—309.

F. Hausdorff, Set Theory, Second edition, Chelsea Publishing

Company, .New YorE, N. Y.',1962.

T. J. Kaczynski, Boundary functions for functions defined in a disk,

J. Math. Mech., 14 (1965) 589— 612.

T. J. Kaczynski, On. a boundary property of continuous functions,

Michigan Math. J., 1§_(1966) 313-320.

10. J. E. McMillan, Boundary properties of functions continuous in a

disc, Michigan Math. J., ‘13 (1966) 299-312.

11. M. H. A. Newman, E1ements of the topoiogy of plane sets of points,

Cambridge University Preess, 1964.

12. H. Tietze, fiber stetige Kurven, Jordansche Kurvenbogen’ und

. geschlossene Jordansche Kurven, Math. Z. ,5 (1919) 284- 291.
I