Author(s): A. S. Besicovitch
Source: The American Mathematical Monthly, Vol. 70, No. 7 (Aug. - Sep., 1963), pp. 697-706
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/2312249 .
Accessed: 18/07/2013 03:38
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
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.
Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to
The American Mathematical Monthly.
http://www.jstor.org
697
698
[September
A three-cornered
hypocycloidinscribedin a circleof diameter3/2 also belongs
to the class (Fig. 3). For it is well known that the tangentline at any point M
of the hypocycloidmeets the hypocycloidat two other points A and B distant 1 fromeach other. Thus if we let one end of the segment describe the
hypocycloidwhile keeping the segmenttouchingthe hypocycloid,we have the
otherend of the segmentalso movingon the hypocycloidand so the wholeof the
segment remains all the time withinthe hypocycloid.
B
ALf
FIG. 1
1
FIG. 2
The area of the hypocycloidis 7r/8=.39, that is exactly half of the area of a
circle of diameter 1. It was conjecturedthat the hypocycloidwas the figureof
minimumarea. The problemaroused great interest.In 1925 G. D. Birkhoff,
one
of the greatestmathematiciansof his time, writingabout unsolved problemsin
his book The Origin,Natureand Influenceof Relativity,firstmentionsthe fourcolor problem,then adds: "Of like intriguingsimplicityis the question raised
a few years ago by the Japanese mathematician Kakeya."
My solution of the Kakeya problem was published in the Mathematische
in 1928 and was reportedto the American Mathematical Society by
Zeitschrift
J. D. Tamarkin.
The figurearrivedat in solving my problem,togetherwith the "joins suggested by a Hungarian mathematician J. Pal, representmy solution of the
Kakeya Problem.
My solutionshows that the hypocycloidconjectureis false,and that in fact,
thereare figuresof arbitrarilysmall area which permita unit segmentto change
its directionby 360? while moving continuouslywithinthem.
The plan of the solution is this. We take a square of side 2 (see Fig. 4) and
divide it into fourcongruentrighttrianglesby joining the centerto the vertices.
The hypotenuseof each triangleis divided into a large numbern of equal parts.
Joiningeach point of division to the centerof the square, we have 4n "elementary" triangles,each of height1.
We enumerate the elementarytriangles in the order in which they come
as we move around the boundary of the square in the counterclockwisedirection, startingat the vertex A. The directionsof the various segmentswhich
1963]
699
FIG. 3
B
FIG. 4
B
FIG. 4 (a)
join the vertex of each elementarytriangle to every point of its base have a
range of 3600. The same will remain true if we give arbitraryparallel translations to the elementarytriangles(see Fig. 4a). As we shall show,parallel translations can be given to these elementarytriangleswhich achieve such a degree
of overlappingthat the totalarea coveredby thetrianglesin theirnew positionis
as small as we please.
Now ifwe place an end-pointof the unit segmentsuccessivelyat the vertices
01, 02, - * -
ofthefirsttriangle,
thesecondone,and so on,intheirpositionafter
700
[September
one the segmentwould not remain withinthe area of the figure.We eliminate
this difficulty
by means of Pal's joins, as follows:
L
EG
FIG. 5
ILMNI - I GHKI
8- I GH
The figureconsistingof the lines GL, DL and of the triangleLMN will be called
the join. We see that the area of the join is less than E/8times the area of an
elementarytriangle.Connectingevery pair of consecutiveelementarytriangles
we shall get 4njoins oftotalarea less thanE/8timesthearea ofthewholesquare,i.e.
<e/2. The join added to the trianglesDEF, GHI permitsthe unit segmentto
come fromthe triangleDEE to GHI remainingalways on the area of the triangles or of the join. For, fromthe position of the segmenton the side DF we
let the segmentslide down along the line DL until its lower end-pointreaches
L, then rotate about L until it reaches the side LN and then slide up until its
top end reaches G, that is, gets in the second triangle.Thus the problemis reduced to findingparallel translationsof elementarytrianglessuch thatthe area
coveredby thembe small.
0
B
FIG. 6
1963]
701
order, which has the advantage of being still simplerin details and easier to
visualize. It is essentiallythis versionthat we shall adopt here.
We consider the coordinate plane and an integerp>2. We constructthe
isosceles right triangleA=OAB of our original square with its hypotenuseof
length2 on the x-axis. The base AB of A is divided into n= 2X-2equal parts and
n elementarytriangleswith vertex 0 are constructed.Figure 6 shows the case
p-5, n=25-2=8.
4
3
2
0
A1
3
T1
TI
T2
4
T,,
4
T1
t:A2
4
73
4
T4
5
T1
T9
6
3
.
fi
T4
fi
T5
S
TS'
5
8
A5
A4
FIG. 7
We next draw the lines y-k/IP, for k= 1, 2, * * *, p (Fig. 7) and call the
line y = k/pthe line of level k, or simply levelk. We then constructp isosceles
right triangles A1, A2, * *, A
Ap each with hypotenuse on the x-axis, and the
opposite verticeson the 1, 2, . . . , p levels respectively.Note that A.= A. For
, p
each k, k =3,
p, the base of Ak is divided into 21-2 equal parts and the
elementarytrianglesare constructedon the subintervalsof each base. Notice
that Ak+1 is divided into twice as many elementarytrianglesas Ak; A2 is not
divided into elementarytriangles,A3 is divided into two, A4 into four,A5 into
eight,and so on (see Fig. 7).
We shall now assign labels to the elementarytrianglesin each Ak. These will
A . . .
be labeled fromleft to right as rik , k*,
The superscriptk
T*,
tk-2.
*,
shows that Tj is part ofAk and the subscriptj says that Tr is thej-th elementary
trianglein Ak countingfromleftto right.A2is not divided into elementarytriangles. We shall say it coincides with the elementarytriangler'; A3 has r3 and
T3as elementarytriangles;A4 has r4,
T4, and r4;
1 r4,
4 and
r2asA
2, 13, an
n so
onon.
Note a simple relationshipbetween the elementarytrianglesof Ak and of
Ak+l
0. -
0\
~
:_
NL
0
NL
MRNLS
R N
FIG. 8
Let us start with 2= LON (Fig. 8). Bisect it by the median OM into two
trianglesOLM and ONM and expand them to similartrianglesPLR and QNS
to the level 3. We shall call this operation the bisection and expansion. The
702
[September
resultof this operationis a pair of trianglescongruentto the pair r' and r' ofA3.
Similarlyis definedthe operation of bisectionand expansion of the trianglesrj
for any k> 2: rj is bisected into two trianglesby the median fromits vertex,
-andeach of the trianglesis expanded to the next level. The operation transformsrj into parallel translates of r2+t1and r21 (Fig. 10), and applied to the
set of all trianglesTi, or to any set of their parallel translates,transformsthe
set into a set of parallel translates of elementarytriangles of Ak+1. Figure 9
representsa particularcase of k = 3.
T3
3a
T2
r1
T4
3
FIG. 9
k+1
_~~~~~~~~~~~~~~~~~~~~T'
~~
k+1
k+1
tl~~~~~+1k+I
FIG. 10
The part of Ak (or of any elementarytriangle rj' of Ak) which lies between
levels k - 1 and k will be called the "top end" of Ak (or of rj). Notice that the
top end of Ak is congruentto A1and that the sum of the areas of the top ends
of all elementarytrianglesof Ak is equal to the area of the top of Ak,that is to
IA1, (see Fig. 11).
k+1
k
~~~-l'-
..
FIG. 11
Now let us look at the change in area when bisection and expansion are
applied to a triangle.Consider an elementarytriangle j = LMN with vertex N
at level k (see Fig. 12). Let NP be the median of LM1N, bisecting it into the
two subtrianglesLPN and MPN. If we expand LPN upwards and to the right
to the level k+1, we get a similartriangleLRQ. If we expand MPN upward to
the leftto level k+1, we get a similar triangle MTS.
1963]
THE
KAKEYA
703
PROBLEM
k+1
NS
k+1
kS//
k-
k-1
L~~~~~~~~
L
TP
XL
FIG. 12
The two trianglesLRQ and MTS togethercover the triangleLMN and the
two "end pieces" SNV and QNU.
We have
= |QNUj
ISNVI
INUVI
so that the two overlapping trianglescover an area equal to the area of the
originaltriangleLAIN plus twice the area of the top end of LMN.
Observethatfor constructing
the end pieces to LMN, one merelyhas to know
thetop end of thetriangleLMN. The rule is this: producethesides LN and MN
to thelevelk+ 1 and join theend-pointsto thepointsof thesides on thelevelk- 1.
If we start with a complete set of elementarytriangles of Ak or of their
parallel translatesand apply bisectionand expansion to each of these triangles,
we shall arrive at a set of parallel translates of all elementarytrianglesof Ak+1
and introducean increase in area at most equal to twice the sum of the areas
of the top ends of these triangles.(The "at most" is necessaryhere because of
possible overlapping.) The sum of the areas of the top ends of a completeset of
elementarytrianglesis equal to the area of A1,so that the total increase in area
is at most21A1.
p2
704
[September
Taking p> 16/e we shall get the area <e/8. With similar translationsfor the
other 3n elementarytriangleswe shall get the total area covered by the translates <e/2. Adding 4n -1 joins of total area <e/2 we shall get a figureof area
<e on which the unit segmentcan turn round through3600,which represents
a solultionof the problem.
A2=22
(a)
2;3
A3
(b)
(c)
FIG. 13
(a)
2 A4
(c)
(b)
FIG. 14
1963]
705
of constructionis confinedonly to the two top strips and there is no need for
tracing fullyall elementarytriangles.Figure 16 representsthe process.
The figure1, of overlapping triangleshas been called by Schoenberg the
Perron tree; we shall call it the Perron-Schoenbergtree. The processof its constructioncan best be described as the "growthof the Perron-Schoenbergtree"
(see Fig. 16).
24
(b)
(a)
(C)
FIG. 15
22
24
23
26
FIG. 16
706
[September
Besicovitch [1] poses the question,what is the solutionof the Kakeya problem ifwe restrictourselvesto simplyconnecteddomains. It is known,forexample, that it is possible to turn a unit segment through3600 within the threecorneredhypocycloidof area wr/8.Insofar as I am aware, no simplyconnected
domain of lesserarea is knownto satisfythe Kakeya criterion.Here we demonstrate the existenceof a class of domains, regular star polygons,which satisfy
the Kakeya criterionand forwhichthereexistareas exceedingir/8by arbitrarily
little. Is wr/8
then the minimumarea forsimplyconnecteddomains, or at least,
forstar-likedomains? If so, we have here a curious instance of a domain functional which is minimizedin two entirelydifferent
ways.
We constructa sequence of star polygonsin the followingmanner. First, a
regular (2n+1)-gon is inscribedin a circle of radius r<1/2. Each side of the
(2n + 1)-gon is then used as the base of an exteriorisoscelestrianglewith vertex
at unit distance fromthe diametricallyopposite vertexof the (2n+1)-gon. The
domain of interestis the union of the inscribedpolygonwiththe 2n+ 1 triangles
attached to the sides. In Fig. 1 we depict the case n =3.
Let us denote the center of the circle by 0 and label the vertices of the
isosceles triangles by V0, V1, * *, V2., V2f+, * * *, in counter-clockwise order,
whereifi _j (mod 2n + 1) then V-= V,. The diametricallyopposite vertexof the