The Banach-Tarski Paradox Author(s): Karl Stromberg Reviewed work(s): Source: The American Mathematical Monthly, Vol.

86, No. 3 (Mar., 1979), pp. 151-161 Published by: Mathematical Association of America Stable URL: http://www.jstor.org/stable/2321514 . Accessed: 07/01/2012 01:12
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

THE BANACH-TARSKI PARADOX

KARL STROMBERG

In thisexposition clarify meaning and provethefollowing we the of "paradoxical"theorem which was setforth Stefan by Banach and Alfred Tarskiin 1924[1].We wereinspired do this to by a recent paperof A. M. Bruckner JackCeder[2],wherethistheorem, and amongothers, is brought into theirinteresting discussionof the phenomenon nonmeasurable of sets. We are grateful Professor B. Burckelfor calling this paper to our attention. to R. We warmly recommend to thereader.It is our intention it hereto present strictly a elementary accountof thisremarkable thatwillbe accessibleto readers fact withvery little mathematical background. We do presume little a matrix theory theelements real analysis. first and of We statethemain and theorem thengiveprecise We definitions before launching intoitsproof. mayas welladmit in advance thatits proofdependson Zermelo'sAxiom of Choice, whichis used in a very obviousway in the proofof TheoremC below (the set C selectedthereis notspecified a in finitely constructable way). subsets R3 having BANAcH-TARsKI THEOREM.If X and Y are bounded of nonempty interiors, then there exista natural number andpartitions 1 < j < n} and { Yj: 1 < j < n} ofX and Y, n {Xj: respectively n pieceseach),suchthatXj is congruent Yjforall j. (into to the Looselyspeaking, theorem thatifX and Y are anytwoobjectsin space thatare each says in to smallenough be contained some(perhaps to very large)ball and each largeenough contain of number pieces and some (perhapsverysmall)ball, thenone can divideX into some finite falseifwe Y. thenreassemble to them (usingonlyrigid motions) form Thisseemsto be patently withthe "real" submitto the foolishpracticeof confusing "ideal" objects of geometry the ball can does seemto be folly claimthata billiard to objectsof theworldaroundus. It certainly of statue Banach. be choppedintopieceswhich thenbe putback together form life-size to a can is the We, of course, make no such claim. Even in the world of mathematics, theorem astonishing, true. but
DEFINITIONS. For x = (x1, x3) in R3 we define x2, thenorm x to be thenumber of Ix= (X2+ x2 + X3)1/2. closedball of radius > 0 centered a E R3 is theset {x E R3: Ix- aI r}. A subset The r at X of R3 is bounded it is contained somesuchball, and X has nonvoid if in if interior, it contains somesuchball. An orthogonal matrix a squarematrix is with realentries whosetranspose also is its inverse product (its withits transpose theidentity is matrix). a rotation shall mean a By we 3 x 3 orthogonal matrix whosedeterminant equal to 1. We also regard p is sucha p as a mapping of R3 onto R3 by writing p(x) forthe vectorobtainedby multiplyingby the colunmn p vector x X : P(X) =Y = (Y1,Y2,Y3) where = (x1,X2, X3), 3 P1 P12 P13 p=

P21 P31

P22 P32

P23JX

P33

Yi

j=1

pijx

fori = 1,2,3. A rigid motion Euclidean (or transformation) mapping ofR3ontoR3having is a r the form = p(x) + a forx ER3 where is a fixed rotation a ER3 is fixed. denotethe3 X 3 and We r(x) p
Karl Stromberg received Ph.D. at theUniversity Washington 1958underthedirection Edwin his of in of Hewitt, with whom is thecoauthor Real andAbstract he of Analysis (Springer-Verlag, He served the 1965). on faculty theUniversity Oregon of of 1960-68 and has beenProfessor Mathematics KansasStateUniversity of at since 1968.He hasalsospent single years thevisiting on faculty four at other universities: (1958-59), Yale Chicago (1959-60), Uppsala(Sweden) (1966-67), York(England) (1974-75).His mainresearch interests in Analysis, are bothclassical abstract. is theauthor theforthcoming Classical and He of book RealAnalysis (Prindle, Weber, and Schmidt, 1979).-Editors 151

152

KARL STROMBERG

[March

matrix t. Two subsets and Y of R3 are said to be congruent we write identity by X and X Y if thereexistssome rigidmotionr for whichr(X)= Y. (Here, as usual, r(X) denotesthe set { r(x): x E X }.) By a partition a setX we meana family setswhoseunionis X and anytwo of of members whichare either of or identical disjoint. Thus,to say that{Xj: 1 < j < n} is a partition of X inton subsets meansthat X=X1uX2u* ** U Xn and Xin Xj=4 ifi j. It is allowedthatsomeor all Xj be void. The geometrical of is significance our purelyalgebraicdefinition a rotation perhaps of clarified thenextproposition. by
PROPOSITION.

Letp be a rotation. wehavethe Then following.

(i) The imagep ofanylineis a line:p(b+ tc)= p(b) + tp(c)forall b,c ER3 and tER. are (ii) Inner products preserved p; ifx,x' ER3, p(x) =y and p(x') =y', then by
3
i=I

3 3 j=l Xi Xi,>

EYi Y"=

(iii) Distances preserved p: ifx ER3, then are by Ip(x)l= lxi. there a p in R3 is (iv) If p /t, then setA = {x eR3: p,(x)= x} is a linethrough origin: the the suchthat = {tp: tER} and IpI= 1. We call A theaxis of p. A (v) If q is any pointofR3 having two the properties p in (iv),then =p orq = -p. We call of q p and -p the poles of p. Proof.Assertion is obviousand (iii) follows (i) from by taking = x. To prove(v) notice (ii) x' thatif { tp:tER) = { tq: tER} and Iql = IPi= 1,thenq = tpforsomet and t2= t2Ip12 =ItpI2= Iqlj2 =I so t is 1 or- 1. = To prove usethe that is orthogonal jpPk ijk= 1 or0 according = k orj#kJ (ii), fact p asj

to write

Y=

y kXk

Xjk

To prove(iv) we need a modest amount matrix of theory realanalysis. and The characteristic polynomial = det(p X) of p is a cubic polynomial f(A) havingreal coefficients the Interso mediateValue Theoremassuresthatit has at least one real root. Let X1,X2,X3 its three be A1 roots(counting (complex) multiplicity) where is thelargest root.Thenf(X)= (AXA)(X2 real X)(X3-X)so (*) X1X2X3f(O)=detp= 1. If Xk is a real root,thenthesystem equations of
3
]=1

(Pij-AkLij) = Xj

(i-=1,2,3)

has a real solution x1,x2,x3 (not all 0) so there an x ER3, JXl is #0, such thatp(x)=Xkx from = which(iii) yieldsIXk1 1, and so Ak=1 or - 1. If X2is not real,thenA3 is its complex conjugate 1 SO = and (*) becomesXI1X212=X1 1. If A2is real,thenso is X3,all three rootsare 1 or - 1,and (*) showsthatXI= 1 and A2 =X3. Since XI= 1, we can takek= 1 in theabove system finda to

1979]

THE BANACH-TARSKI PARADOX

153

vectorp ER3 withIpI=1 suchthatp(p)=p. Then tpEA forall tER. Ourjob is to see thatthere for is u:/ztp all t ER. Choose a nonzero are no other vectors A. Assumethatthere a u EA with in u, to vector thatis perpendicular theplane containingp, and 0; thatis, vjpj=f vjuj = 0. Since v to from thatp(v) is also perpendicular thisplane and thence (ii) p(p) =p and p(u) = u, it follows as from(iii) that p(v)= v or - v. Any vectorx ER3 can be written x = ap + fu + yv for havep(v)=v. appropriate/,yER and,by(i), p(x) = ap +,fu + yp(v). Sincep'4' t, we cannot a, Therefore =v. The matrix p(v)
a=

PI

P2

U1
U2

VI
V2

P3

U3

V3

product independent) thematrix and (becausep, u, and v are linearly has nonzero determinant
PI

pa=

P2
p3

Us
U2 U3

-V1
-V2 -V3

satisfies - deta = det(pa)= (detp)(deta) = deta completes proofof (iv). s the so deta= 0. This contradiction whichare of considerable interest themselves in We nowproveseveral theorems lemmas and of as well as beingvitalstepping three these,of which stonestowardour maingoal. The first in TheoremC is the real key to our story, were set forth Felix Hausdorff 1914 [4, pp. by the 469-472].We consider tworotations

-1/2
A,= and V /2 0 (-cos O 0 sin9

-XV/2
-1/2 0 0 0 sin 9 0 cos9

0
0 1

4=1

4, R3 where9 is a fixedreal number, be chosenlater.(Geometrically, rotates by 1200 about to the z-axis and 4 rotatesR3 by 180? about the line in the xz-plane whose equation is 42 4, x cos I 9 = z sinI 9.) One checksthatthematrix iS thesame as thematrix exceptthat\/ is replacedby - VX and that
A3=02=l (1)

as that where is theidentity matrix. Now let G denotethesetof all matrices can be obtained a l each of whichis 4 or 4. Becauseof (1), it is clear of of factors, product a finite number (matrix) thatG is a groupunder matrix p multiplication p,a E G, then - I, pa E G) and thateach plt in (if G can 1e expressed at leastone wayas a product in (2) P=OIO2 0,n n one where > 1,each aj is 4 or 4 or 42, and if 1 < j <n, thenexactly of aj and aj I is 4. We call the suchexpressions words theletters 4,,and 4,2.For example, expression in 4, reduced 4,4,24,4,2 is nota reducedwordbecause of thetwoadjacent4's, but it is equal to thereducedword4,, of thant,4, Thus each element G other products thesamematrix). are ("equal" meansthatthese in forms 4, and 4,2 can be expressed at leastone of thefour a = 4,Pi,OOP24, ... 4%Pm+4, /= 4,4P I4P2 **4,Pm,
= y44PI4141P2... * 4,Pm4,,

a3 =

4,PI4,4,P24,... *

nPm

(3)

154

KARL STROMBERG

[March

m where > 1 and each exponent is 1 or 2 (for8,m> 1). These are thereducedwordshaving pj on wordsthat morethanone letter. Depending our choiceof9, it mayhappenthattworeduced are appearto be different actually equal; i.e.,whenmultiplied theyequal thesame matrix. out, = For example, we choose9= if = =, one checksthat414 442 and 41+++ t. However, do have we theorem whichis thekeyto ourlaterresults. thefollowing remarkable then THEOREM [4]. If cos0 is a transcendental A. thanl has number, each element G other of in word theletters 41, 4,2. Thatis, if as exactly expression a reduced one 04, and
(i)
ala2**an

=P1P2...

Pm

where each side of thisequationis a reducedword,thenm= n and aj = pj for1 6j < n. Proof.We need onlyshowthatno reducedwordis equal tot, forthenif(i) held truewith n as smallas possibleit wouldfollow thatn= 1 and Pi = al. We first showthatif a is as in (3), thena& t. We have a = am I ... a2a1 whereeach a is am 4, or 424. That is, each a is one of thetwomatrices either

1Cos
a= + %2

+ \/2
-

I
+ _S

sin9
sin9

a V 3

cos9

sin9 0 cos9 One checksby induction m thatifK= (0,0,1), thenamam . aj(K)=(sin9Pm-1(cos9), VJ on l-I with rational Rm(cos9)) where the P, Q, and R are certainpolynomials sin9Qm-1(cos9), their are and their are coefficients, subscripts their degrees, leadingcoefficients
-

1(3)

?2

()2

In respectively. fact, simplecomputations showthat

PO(X) Pm(x) =

2(2)

1
I

31Q(x)=-2
xPmi (X)?

,RI(x)=x,
Qm- l(X)Rm(x)

QM(x) = + 2 JXPm(x) Rm+I(x)=(

+ 2 Qm- l(x)-

Rm(x)

This done, we see thatsince cos9 is a root of no polynomial withrationalcoefficients, is it impossible thata(K) = K (else Rm(cos 9)-I =0) and so a t. = Now we see that no /8as in (3) can equal t, for otherwisea = 4,34 4t4= 42 = t. Similarly,if
y = t, then 8 = 4y4 = t, so it remains only to rule out the possibilitythat 8 = t. Assumethat8 = t where8 is as in (3) and m is thesmallest natural number whichthisis for 42 true.Of coursem> 1. Ifp1=pm then4,P +P. is either or 4A=40so
= t= 41P3418d1PI
* 044dP2... 4JPI +Pm

-X2)Pmi(X)+XRm(X).

/3 is a reducedwordof theform whichis impossible. Thuspi +pm=3. In case m>3, we have

l = 041PmS41P = 4JP24... I(A
(A41Pm-I

whichis again of theform contrary theminimality m. Therefore = 2 or 3. But m= 2 to of m 8, = yields t 4P2 SOPI = ) while m =3 yields t= 4Aw03 '4= 41P2and these resultsare ridiculous. We 83IP concludethat8 = l is impossible. U We hereby chooseand fixany9 suchthatcos9 is transcendental. courseall butcountably Of 9 manyreal numbers have thisproperty. (Incidentally, follows it from Generalized the Linde-

1979]

THE BANACH-TARSKI PARADOX

155

9 mannTheorem thatany nonzeroalgebraic willdo; e.g.,9= 1.) If an element e G is expressed itsuniqueway as a reducedwordas in (2), we call n the p in length p and we say thatal is the of with We write = n and letter p or thatp begins of first al. l(p) l(t)= 0. As usual,by a partition a setX, we meana pairwise of of of disjoint family subsets X whose unionis X.

{ THEOREM B. There existsa partition GI,G2, of G intothree subsets suchthat nonvoid G3} for each p in G we have
(i) pEGI-)pEG2U G3, (ii) pE Gl<p E G2, (iii) p E G 4I2pE G3. (Note,forexample, 4. that4p need not beginwith If p= 040, then4p = 4i beginswithA.) Proof.Assigntheelements G inductively of Put to as according their lengths follows.

IE GI, EG2,4E G2,42EG3.

(4)

Supposethatn > 1 is someinteger suchthateach a E G with < n has been assigned exactly to l(a) one of GI, G2,and G3.We nowassignall elements length + 1. If 1(a)= n and a beginswith 4 of n or i2, put

4o xG2ifa EGI,
)aE GI ifcEG2U G3.

If 1(a) = n and a beginswith4),put

1 Oa E Gj+ ifa E G., 42a E Gj+2 if a E G

(6) (7)

forj=1, 2, 3 where G4= G1 and G5= G2. By inductionour partition now formed. is The assignment any element length can be easilydetermined n steps.For example, of of n in if p+p2+42, thenl(p)= 7 and we notesuccessively, withthelast letter, p= beginning that p2E G3, )2 E G1,42A2 E G3,4 42(2 4E G
E pqp20p2
G2, (p4/qp2qp2 pE G2. G1, E

One easilychecksthattheelements length of twosatisfy

41.E G3, c G1, and therefore (i)-(iii) hold if l(p) < 1 (forexample, that both sides of equivalence are false (i) unlessp = t). For an inductive proof (i)-(iii), supposethatn > 1 is someinteger thatthese of and three equivalences known holdforall p E G having <n. Now letp E G withl(p) = n be are to l(p) given. CASE., 1. Suppose that p beginswith+. Then (6) and (7), witha = p, imply(ii) and (iii), respectively. Sincefp has length - 1, our induction n hypothesis yields pa GI#O(op)=pE G2UG3 <=bpE GI'?:cpaG2u G3 and so (i) also holdsforp. CASE 2. Supposethatp begins -24 with Then (i) follows from with = p. We have4p= (5) a 41. wherel(a) = n- 1 and a beginswith4, so (7) and (6) yield
{ q)4/'042,/20

whichproves(ii) and (iii) forp.

4p= 42ae G2<>aUe G3'p = 4a e GI'Ap

= a E G3

156

KARLSTROMBERG

[March

from Herewe have4p = a (5). with As in Case 2, (i) follows CASE3. Supposethatp begins 4,2. n has length -1 and beginswith0. So again (6) and (7) yield
4p= aE
G2.p=

=p2a

Gl<.?a

G2-?.42p-4'aE

G3

proving and (iii) in thisfinalcase. U (ii)

+ X3 = 1)

unit S a {P, exists partition SI,S2,S3} ofthe sphere {x ER3: IX12= C. THEOREM There
suchthat into foursubsets (i) P is countable,
(iii) 41(S1)= S2,

X2 + X2

(ii) O(S1)=

S2 U S3,

(iV) 42(S1)= S3.

Since G is countableand each Proof.Let P= {p ES: p(p) =p forsome p E G withpl=/t}. For we of leavesjust twopoints S fixed (thepolesofitsaxis ofrotation) see that(i) obtains. each x E S\P, let G(x) = {p(x): p E G). Each such G(x) is a subsetof S\P (ifp(x) E P forsome any two such sets G(x) and G(y) are eitherdisjointor identical(if tE G(x) n G(y), say p(x) = t= a(y), and z E G(x), say z = T(x), thenz = T(x) = Tp - '(t) = Tp - 'a(y) E G(y); whence, = of the G(x) n G(y)#7fr?G(x) G(y)). Therefore, family sets6 = { G(x): xE S \P) is a partition so of each member F and denotethesetof points one of S \P. Next,chooseexactly pointfrom chosenby C. The set C has theproperties: CcS\P, cl 7=c2 in C= G(cl)n G(C2)=4,, somec E C E S \P=x E G(c) for x becausex E G(c).c
E

p, then ap(x)

p(x) for some a -- so p - lap(x)

x, p - lap t, and x E P), x E G(x)[x = t(x)], and 7=

(a)
(b)

(c)

G(x) forall x,c E S \P. Now define

Sj= Gj(C)=

{p(c):p E Gc EC}

G2, forj = 1,2,3 whereG1, G3 are as in TheoremB. Using (a) and the factthatG(x) c S\P if xES\P, we see thatSjCS\P foreachj. The factthat G=GIuG2uG3 and (c) implythat for If j]#i in {1,2,3}, then SjnS1=4, (otherwise, xESjnSi, we have S\P=S1uS2uS3. x = p(c) = a(c2) forsomecl,c2E C,p E Gj,a E G,so (b) yieldscl = c2= c, say,and hencea c'p(c) = c whilec 7 P from to {P, which - = t and p = a contrary Gjn G =4). Therefore S1,S2,S3} iS a 'p a partition S. of B we Finally, apply(i)-(iii) of Theorem to write
O(SI1)= {p(c):pEGI,cEC}= iP(51)= {4'p(c):pEGI,cEC}
+2(Sl)

{T(c):TEG2U

G3,cEC}=S2US3, =S2,
=

= {T(c):TEG2,cEC}
T={r(c)C: T E G3,c E C)

={2p(c)

which proves(ii)-(iv). U The following lemma and its use in deducingTheoremsD and E fromTheoremC are of contributions W. Sierpiniski [6]). (see set X there existsa countable Q and a rotation of subset S, then LEMMA. P is anycountable If suchthatP c Q c S and o(Q)= Q \P. for selectan axis of rotation Wthat Proof.The idea of the proofis verysimple.We first of containsno point of P, then we use the countability P x P x N to select one of the P for uncountable supplyof anglesof rotation w thatmake w satisfy nWn(P)p forall n> 1, and finally put we
00

: p E GI,c E C)

S3

Q=PU

n=1

U &n(P)

(8)

1979]

THE BANACH-TARSKI PARADOX

157

We now givedetails. Amongall vectors = (vI, v2, in S having = 0, there onlycountably v V3) are V3 v manyforwhich or -v is in P. Selectanyv = (vI,v2, E S suchthatneither nor -v is in P. Writing = (1, 0) 0) v u 0, and
VI
a=-V2 V2 VI

?
?

0 we see thata is a rotation, = u, and the set a(P) containsneither nor - u. For real u a(v) numbers consider rotations t, the (1
Tt=

0 cost -sint
0 sint cost

thatleave u fixed. For each triple (x,y,n) withx,ye a(P) and nEN, it follows the easilyfrom factthatx2 + x2 > 0 thatthere existeither n 0 exactly or exactly valuesof t in [0,24r forwhich according xl =YI or xl #,Yy. as Since there onlycountably in are Ttn(x)=y manysuchtriples all, there onlycountably are manyt forwhichtheequality
00

a(P)n

n=1

U Tna(P)=4 w Tt. Now define =

n=1

(9) e - 'Ta and define as Q from whichwe U w(n(p))=

fails.Fix anyteR forwhich obtainsand write = (9) T have Pnw(Q)=p. But Q=Puw(Q)

in (8). Since Tr=an'n forall n, (9) yieldsa(Pnw(Q))=a(Pn

so theproof finished. * is

THEOREMD. Thereexistsa partition Tj: 1 <j < 10) of theunitsphere intoten (disjoint) { S subsetsand a corresponding {pj: 1<j<10) set such that {pj(Tj): 1 <j6) of rotations is a partition S into subsets {pj(Tj): 7 ]j < 10) is a partition S into of six and subsets. of four Moreover, we can take T7,T8,and Tgto all be rotates SI and take TI, T2,T3,and T10to all be countable. of

Proof.We continue previous our notation and define

Ul =4 (S2), U2== O(S2),

U3= 4A2(S2),

= VI= 4(S3), V2= 4'(S3), V3 420(S3). By TheoremC it is clear that { Uj,Vj) is a partition Sj for 1,2,3 and thatthesesix sets of j= alongwithP form partition S intosevensubsets. a of Now let T7= U1, T10=P, U2, T87= Ts9=U3,
P7=

20 P8 =0!,p2p9=

P4',plo=i

and checkthatpIo(T,0)=P and pj(Tj)=Sj-6 forj=7,8,9 so that{pj(Tj)):7j< 10) is indeeda partition S. We shallnow divideS \(T7U T8u Tg T10)= V1UV2U V3 intosix pieces.Let Q of U and w be as in thepreceding Lemmaand define T7=P8(SIn Q),T2=P9(S2n Q),T3=p7(S3n Q) T5 Q). T4=p8(SI\Q), = p9(S2\ Q), T6= p7(S3\
Plainly,

{ TI,T4) partitions P8(SI)= { T2, partitions T5) pg(S2)=

{ T3,T6) partitions p7(S3)=

VI,
V2,

and thuswe see that{ Tj: 1 <j < 10) partitions Next define S.

V3,

158

KARL STROMBERG P4 = P8 ,P5 - P9 I P6 = P7

[March
P+3

and p

for = 1,2,3. Evidently, j (j= 1, 2,3) pj+3(Tj+3)= Sj\Q setsis S \Q. Finally, have we and, sinceP c Q, theunionof thesethree

(j= 1,2,3) p1(7))=1-p+3(T7))=&P-'(Sjn Q) so thesethree and their unioniso -1(Q\P)= Q. U setsare disjoint If, fora subsetT of S, we writeT'={tx:x ET,O<t< 1), thenS'={y ER3:0<IyI 61) is thepunctured obtained ball from solidunitball B = { y ElR3:IyI 1) by removing origin the the I 0 = (0,0,0), and it is clearthatthefirst sentence Theorem remains of D trueifwe replaceS by S' and Tj by TJ' in We throughout. use thisobservation thenextproof.
( THEOREM There E. exists partitionBk: 1 S k 6 40) oftheclosedunit a ball B into subsets forty and a corresponding {rk: 1 < k 6 40) ofrigid set motions that{rk(Bk). 1 6k 6 24) partitions such B intotwenty-four subsets and { rk(Bk): 6 k 6 40) partitions intosixteen 25 B subsets.

Proof.Applytheabove Lemmato thecase thatP is thesingleton { u) where = (1,0,0)E set u S to obtaina countable Q withu E Q c S and a rotation such thatpo(Q) = Q \{u). Next set po let N1= {2(q-u): q E Q) and define rigidmotion by the ro
ro(x) = po(x + 2u) u.

Plainly the vector 0 is in N1 and rO(N1)=N1\{O). Writing N2=B\Nl, sl=rO, s2=t, and B Mh=sh(Nh) for h=1 and 2, we see that (N,,N2) partitions and {M,,M2) partitions S' =B\{O). We complete proofby combining the thesepartitions rigidmotions and withthe { of S' and the rotations partition Tj': 1 1 following {pj: 1 1j? 0) as in the remark i10) Theorem D. Notice that, eachj(l S j < 10),thefamily Tj'n pj-I(M): I 6 i 6 2) partitions and that for TJ' in turn{Mhn Tj'npj-I(Mi): 1 ?h<2) partitions y(M) for i=l andi=2. Thus{Mh n TJ' Tj'npjsets of and theforty subsets n pj- I(M;): 1< h 6 2, 1 < i 6 2, 1 6 j 10) is a partition S' intoforty
Bhy =S*-'[

Mhn Tj'npj- I(Mi)]

form partition B whileforeach fixed thefour a of sets j npj(Mh n Tj') (10) (1 S h ? 2, 1 6 i < 2) form partition pj(Tj').We nowinvoke a of Theorem to see thatthefamilies D
pjsh(Bh j) = Mi

{Pjsh(Bhy): 1 6h < 2,1 ?i<2,1

?j16)

{Pjsh(Bhy): 1 6 h < 2, 1 < i 6 2,7 <j < 10) are each a partition S' while, fixed (10) showsthattherespective of for i, families twelve of and eightsets are each a partition M, whichwe can in turnmap to partitions Ni via s7 '. of of Therefore, writing = s' lpjsh,we infer that rh {rhj(Bhj).

1 6hS2,1
1

SiS2,1

??< 6)

and
{rh,(Bh,):

h < 2, 1 6i 6 2,77 < 10) Sj

are partitions B intotwenty-four and sixteen of sets sets,respectively. Finally, relabeltheforty setsBhi and theforty rigid motions withsinglesubscripts = 1,2,... ,40. m k rhii
DEFINITION. We shall say thattwo subsets and Y of R3 are piecewise congruent X and we write X- Y if,forsome naturalnumber thereexista partition n, {Xj: 1 < j6n) of X into n

1979]

THE BANACH-TARSKI PARADOX

159

such that{f1(Xj):1 j 6n) is a subsetsand a corresponding {ff: 1 j < n) of rigidmotions set of to partition Y. In case X is piecewise congruent a subsetof Y, we shallwrite XC5Y. of Our nexttheorem givessomesimpleproperties therelations defined. just
THEOREM

F. For subsets Y, and Z ofR3 we have X,

(i) (ii) (iii) (iv) (v) (vi) (vii)

X'X, X Y==>Y'X, X- Y and Y-Z=>X-~Z, X- Y=?'XC Y, X-< Y and Y-<Z=>X-5Z, X c Y=>Xc Y, XC5Y and Y-<X=>X- Y.

Proof.Since Y c Y, (iv) is banal. Sincetis a rigid n= motion, and (vi) are obvious(with 1). (i) Assertion follows (ii) from factthatinverses rigidmotions rigidmotions. the of are of To prove (v), suppose that {Xj: 1 j<n) and {Yi: 1 i<m) are partitions X and Y, 1 respectively, that {ff: <j n) and { gi: 1 i < m) are sets of rigidmotionssuch that and of is a partition some Y0ocY and { gi(Y):Ii<m) is a partition some of {(f(Xj):1Ij<n) of Z0ocZ. Then one readilychecksthatthe mn setsA. =Xjnfj-'(Y,) forma partition X (for unionis Xj) and,forfixed the and their are disjoint i, fixedj,them setsAlj,A2j*..., Amj pairwise n setsf(AU)= Yinfj(Xj)(1<j <n) form partition Yin YOso { g1f1(Au): 6i <6m, <j <n) is of a 1 1 a pairwise disjoint family whoseunionis somesubsetZ1 of Z. Each composite mapping is a gj) rigidmotion, we have X-Z1 and henceX-<Z. This proves(v). The same argument so proves (iii) by takingY0= Y and Z = Z. be To prove(vii),supposethatX- Y0 and Y-X0 whereY0c Y and X0CX. Let thenotation as in the preceding withX = Z and X0= Z0. We prove thatX-, Y by copyinga paragraph well-known proofof the Schroder-Bernstein Theorem.Firstdefine on X and g on Y by f f(x) =fj(x) ifx E Xj and g(y) = gi(y) ifyE Yi. For E c X, define E'c X by E'= X\g[ Y\f(E)]. Plainly, (12) EcFcX= >E'cF' Let 6D ={E: E CX,E cE'). Notice thatf E 6D.Let D= U6D be theunionof all the setsthat belongto 6D. For each EE6D we have E'CD' by (12) so ECD'. Thus DCD' and so (12) yieldsD'C(D')'; hence,D'C6D, D'ccD, and D'=D. Put E=D in (11) to obtain D =X\g[ Y\f(D)],X\D =g[ Y\f(D)] X\D cXO. Now define, 1 <j<n and 1 <i<m, Clearly, for A-=D n Xj,An+i=g[Yi\f(D)],hj=fj, and hn+i=gi-. It followsthat {A1,...,An) partition {An+l,-..,An+m} D, partitions X\D, {h1(Al),....hn(An)) partitions f(D), and {hn+1(An+1), .. .,hn+m(An+m)} partitions Y\f(D). Therefore Y. m XRecall thata closedball in RI is anysetoftheform = {x CR3: Ix - aI <} where cR3 and A a E > 0 are given. Recall also thata translate a setA cR3 is anysetoftheform + b = {x + b: x of A EA } where eR3 is given. b A, then
THEOREM

(11)

G. If A cR3 is a closed ball and ifA 1,A2,... Anare a finitenumberof translatesof A- U A1.
j=1

160

KARLSTROMBERG

[March

E Proof.Wemay suppose A = {x ER3: IxI <E} for that some > 0. Choose a ER3 for any which 2e and letA'=A+a={yER3:ly-a We use Theorem to showthat E I<e}. laI> A-(AuA'). For So let theBk and rk be as in thattheorem. any setD cR3 and any 8 > 0, let SD ={x: x E motions by D }. We consider partition the {EBk: 1 < k < 40) ofA. Definerigid Sk
Sk(X)=Erk( sk(x) = erk(

x)

ifI <kS24,
if25 <k < 40.

x) +a

motion and (Note thatifr is a rigid (r(x) = p(x) + b where is a rotation) s(x) = Er( p s , then is a rigidmotionbecause s(x) = p(x) + eb.) From TheoremE we see that {sk(eBk): 1 k 24} A partitions {sk(eBk): 25< k < 40} partitions and so,since nA' = +. {sk(eBk): 1< k < 40} A, A', A partitions U A'. This provesthat A (A U A'). is We now provethetheorem induction n. The theorem obviousifn= 1. Supposethat by on and n > 1 is such thatA is piecewisecongruent the union of any n- 1 of its translates to letA1,,...Anby anyn ofitstranslates. hypothesis [A1U ... U An- j and it is obviousthat By A ... to (by An\[A1U U An-1] is congruent translation) a subsetofA' so we have

Aiu ...

A -<AuA'-A.

But clearly -A 1u ... u A so Theorem yields -A 1u U. UA. A F A We now statetheBanach-Tarski Theorem again and thenproveit.
THEOREM

then nonvoid H. If X and Y are bounded subsets R3 having of interiors, X- Y.

and Proof.Choose interior pointsa and b of X and Y, respectively, thenchoose E > 0 such A thatA = {x eR3: IxI<E} satisfies + a c X and A + b c Y. Since X is bounded,thereexista X. finite number of have,using A,,..., An of translates A whoseunion contains We therefore Theorem G, of Y-A. Another F so it follows from TheoremF thatX-A. Similarly application Theorem givesX- Y. U
REMARKS. 1. The number thatappearsin Theorem is notthesmallest E 40 possible.In fact, R. M. Robinsonshowedin 1947[5] thatthere a partition B intofivesets(one of thema is of whichcan be reassembled rigidmotions form singleton) by to twodisjoint closed balls of unit radius.Moreover, J. Dekkerand J. deGrootproved[3] thatthesefivesetscan be chosenso T. thateach is bothconnected and locallyconnected. 2. It follows from TheoremC that,since Lebesguemeasure 3 on R3 is rotation invariant, none of thethree S;' = { tx: x Sk,0 < t < 1), 1< k ? 3, can be Lebesguemeasurable. sets 3. A poor analogueof Theorem can be explicitly C constructed Axiomof Choice) in the (no plane as follows. Fix any transcendental complexnumber with ci= 1 (plenty theseexist, c of sincethere onlycountably are manyz withIzI= 1 thatfail to be transcendental; can take we c = e'). Now letX be theset of all complex numbers theform of

A5<Xc(Alu ... uA -A

z= E ajc j=O

where and ao,a,,.. ., an are nonnegative n integers. Each z E X has a uniquesuchexpression. Let X. XO be theset of thosez forwhich = 0 and letX, =X\Xo. Then {Xo,X1} partitions Define ao the rotation of the plane by p(z) = cz and definethe translation by T(z) = z +1. Then p T p(X) = Xo and T(X) = X1 so thesetsXOand X1 are each congruent X. The reasonthatthis to analogueis "poor"is twofold: is bothcountable unbounded. X and

1979]

SEX DIFFERENCES

IN MATHEMATICS

161

References congruentes, en de des respectivement Sur 1. S. Banach& A. Tarski, la decomposition ensembles points parties Fund.Math.6 (1924)244-277. 23 Mat.Tidskr. (1975)59-68. Nordisk On measure, Lebesgue & 2. A. M. Bruckner J.Ceder, improving Fund.Math.43 (1956),185-194. of Decompositions a sphere, & 3. T. J.Dekker J.de Groot, 1914. Leipzig, der Grundzuge Mengenlehre, 4. F. Hausdorff, 246-260. Fund.Math.34 (1947), of On 5. R. M. Robertson, thedecomposition spheres, NewYork,1953. of Chelsea, Congruence sets, 6. W. Sierpinski, KS DEPARTMENTOF MATHEMATICS,KANSAS STATE UNIVERSITY, MANHATTAN, 66506.

SEX DIFFERENCES IN MATHEMATICS: HOW NOT TO DEAL WITH THEM

EDITH H. LUCHINS

Even casual observation this distinguished of assemblagereveals sex differences among mathematicians. There are both male and femalemathematicians! This may seem to be a vehement way of expressing obvious.But it seemsto be not at all obviousto thosewho the portray history mathematics. case in pointis an important the of A collection portraits of and biographies mathematicians of throughout ages on a wall map entitled the "Men of Modem Mathematics" Thereis a womanamongthem, [7]. Emmy Noether. absentare other But women who,despite enormous obstacles, contributed significantly mathematics, SophiaGermain to e.g., and SonyaKovalevsky. a similar In vein,a well-known otherwise and excellent textbook the on history mathematics no womenlistedin thename index-and seemingly mentioned of has not in the text-not even EmmyNoether, although her father, Max Noether,is listed [3]. Still another well-known on thehistory mathematics text of referred Hypatiaofthefourth to century as the first woman mathematician be mentioned the history mathematics-but to in of it referred no otherwomen,at least not in its first to threeeditions, even as recently 1969; as however, there a brief is reference EmmyNoether themostrecent to in edition thetext[5]. of These are illustrations waysin whichnotto deal withsex differences mathematics. of in Do not ignoreor overlook hide theachievements one sex. Let us findout moreabout these or of achievements makethem and knownto our colleagues, students thegeneral our and public. can True, famouswomenmathematicians throughout history be countedon one's fingers. But whenmathematics wereasked to name such women,theyusuallydid not reach students women mathematicians was eventhefirst whentherequest namefamous to finger. example, For level made of 26 mathematics majorsin a junior-senior algebraclass,24 did notlistany names. In contrast, 22 when theywere then asked to name threeto fivefamousmathematicians, to It students an answered, listing averageof four(male) mathematicians. is important increase in theawareness thecontributions womenmathematicians thepast (cf.[4],[20]). of of Nor shouldwe belittle women'scontributions. a recent the At conference womenin the on history mathematics, oftheparticipants of one remarked on thewholeshewas disappointed that
EdithH. Luchins graduate did workat New York University KurtFriedrichs thelate Richard with and Courant. obtained Ph.D.in Banach She her of where she algebras Bertram with Yood from UniversityOregon, the was research associate under fellowship theAmerican a of to Association University of Women. Before going Rensselaer Polytechnic Institute, where is Professor Mathematical she of at Sciences, heldfaculty she positions in Brooklyn Collegeand the University Miami.She is concerned of aboutwhythere not morewomen are mathematics. interests applications mathematics philosophy psychology reflected eleven Her in of in to and are booksand monographs somefifty and articles, many which wishes point of (she to out)arecollaborative efforts with husband, her Professor Abraham Luchins SUNY-Albany.-Editors S. of