The Banach-Tarski Paradox

Zaichen Lu October 2009

Introduction
The Banach-Tarski Paradox1 is generally quoted as stating: It is possible to partition a solid ball in R3 into nitely many disjoint parts and rearrange them through rigid motions to form two identical copies of the original. This is sometimes rather fancifully described as cutting up a pea and reassembling the pieces into a ball the size of the sun, which follows as a simple corollary to the original theorem. This result, published in 1924 by Stefan Banach and Alfred Tarski2 , goes some way into showing the necessity of adopting restricted concepts of measure, such as Lebesgues construction, and also shows our conception of sets not to be wholly intuitive. It has been described by some mathematicians as the most surprising result of theoretical mathematics 3 . size in some sense. Someone with a background in mathematics, who is meeting the paradox for the The paradox plays on the idea of a proper subset B A, but where B and A are of the same

rst time, might, after a little thought, regard this merely as an odd consequence of the uncountability of R (and Rn ) 4 . However, the whole story is much more interesting, as the paradox calls on rigid transformations, and only holds in Rn where n 3. The analogous theorem in R2 would require the weaker condition of partitioning into countable pieces. Furthermore, BTP requires the Axiom of Choice (AC), and cannot be proven nor disproven in ZF. Initially, the Banach-Tarski paradox caused a stir amongst mathematicians, some of whom cited it as reason to reject the Axiom of Choice, upon which the paradox depended, from the list of basic axioms of set theory. The following is the one common formulation of the axiom: Axiom Of Choice (AC): Let A be a family of non-empty sets. Then there exists a function f : A SA S, such that f (x) x, x A. The function f is referred to as the choice function, and the range f(A) is called the choice set. In a bid to eliminate such highly counter-intuitive results, many mathematicians also proposed weaker forms of AC, ones which were not strong enough to imply BTP, but still allowed for many of the same positive results as Choice. However, many important results, such as the Tychono Theorem5 in general topology, still required AC in full generality, and the appearance of AC in many dierent branches of mathematics, lead to its common acceptance amongst most mathematicians.

also referred to as BTP. Stefan; Tarski, Alfred (1924). Sur la dcomposition des ensembles de points en parties respectivement congruentes Fundamenta Mathematicae 6: 244277. 3 Jan Mycielski in his Foreword to Wagons The Banach-Tarski Paradox. 4 The bizarre nature of innity has been investigated since ancient times, with seemingly paradoxical consequences. Examples include Zenos Paradox, Hilberts Hotel, Cantors theory of cardinality, and of course BTP. 5 The product of any collection of compact topological spaces, is itself compact.
2 Banach,

1 Hereafter

Background
In this section, I shall present some of the denitions and theorems I shall be using in the proof of the Banach-Tarski Paradox. As mentioned in the introduction, the crux of the paradox is to generate two sets of identical size as the original. We shall be using the geometric concept of equidecomposability to represent this idea more technically. Denition 1. Two sets A, B Rn are congruent, denoted A B, if there is some isometry = n n f : R R such that f (A) = B. Denition 2. Let X be a set, and let G be a group. G is said to act on X if to every g G, g(h(x)) = (gh)(x) and 1(x) = x, where 1 denotes the identity element in G. there corresponds a bijection, also denoted g : X X, such that g, h G and x X, we have Over the course of this paper, I shall be using the following notational conventions for groups acting on Rn : En - The group of all isometries. SOn - the special orthogonal group. In R3 this consists of the rotations around the origin. Denition 3. Let G be a group acting on a set X and suppose A, B X. We dene A and B to be subsets. That is, A=
i=1 n n

G-equidecomposable, if A and B can be partitioned into nite disjoint unions of pairwise congruent Ai , B=
i=1

Bi

with Ai Aj = Bi Bj = whenever i = j, and that 1 i n, g1 , . . . , gn G such that Bi = gi (Ai ). We denote this A G B. Proposition 1. The relation G is an equivalence relation. Proof. Reexivity and symmetry are trivial, and transitivity doesnt require much more eort. that there exists partitions A1 , . . . , Am A, B1 , . . . , Bm B and rigid motions g1 , . . . , gm G, such such that Cj = hj (Bj ). each i, and similarly Transitivity: Suppose A, B, C Rn , such that A G B and B G C. By denition, this means

that Bi = gi (Ai ). There are also partitions 1 , . . . , n B and C1 , . . . , Cn C and h1 , . . . , hn G Let Yi,j = Bi j . Clearly the collection of Yi,j also partitions B. Furthermore,
i Yi,j = j for each j. j

Yi,j = Bi for

Zi,j = (h g)(Xi,j ) for each i, j.

1 Let Xi,j = gi (Yi,j ), and Zi,j = hj (Yi,j ). Then, {Xi,j } partitions A, and {Zi,j } partitions C, and

Denition 4. Let G be a group acting on a set X, and suppose E X. We dene E to be Gparadoxical if E is G-equidecomposable with two disjoint copies of itself. That is, if for some A, B E, s.t. A B = , and A G E G B

With this denition, we can understand BTP in the more general setting of paradoxical sets. Clearly, it is of particular interest when G is En or SOn , since only rigid motions are then allowed. Over the course of this paper we shall investigate certain particular cases when X is some Euclidean space, and G a subgroup of En . We can now phrase the Banach-Tarski Paradox using this terminology: BTP: S 2 is SO3 -paradoxical. Denition 5. Let M be a set. The free group F with generating set M is the group of all nite reduced words using letters from {, 1 : M } under composition. A word is reduced if equivalent. The identity of F will be denoted 1, and called the empty word: clearly 1 = 1 = 1, M all adjacent pairs of letters of the form 1 or 1 are removed. Thus, no two reduced words are

Figure 1: Cayley graph of free group on two generators. The center represents the empty word, each vertex representing a unique reduced word, and moving along a branch represents multiplying by one of 1 , 1 . 3

Theorem 1. A free group F on two generators (i.e of rank 2) is F-paradoxical, where F acts on itself by left multiplication. Proof. Suppose , are the two generators of F. We denote the set of words in F whose rst letter is as W (), where is one of 1 , 1 . We can then partition F into ve sets: F = {1} W () W ( 1 ) W ( ) W ( 1 )

However, note that if we pre-multiply an element in W ( 1 ) by , we obtain a word beginning with 1 . . . where is one of 1 , 1 , which upon reduction, gives a word beginning with . That is: W ( 1 ) = {1} W ( 1 ) W ( ) We therefore obtain that: W () (W 1 ) = F = W ( ) W ( 1 ) = F \ W (). W ( 1 ).

This result is important to our dealing with BTP, which is based on the realisation of the free group on two generators as a group of isometries in Rn , and also uses the idea of transferring a paradox from a group to a set upon which it acts. For the following theorem, we dene a subset S of a group G to be independent if S is a generating to the empty word. Here, we shall be considering rotations in R3 , such that no non-trivial sequence of compositions results in the identity rotation. Proposition 2. There exists two independent rotations, and , about axes through the origin in R3 . Hence SOn has a free subgroup of rank 2, for n 3. Proof. It is clear to see that, for example, if and are two rotations of the same irrational angle (in degrees), but through perpendicular axes, then they are independent. Consequently, and generate a free group of rank 2, which by Theorem 1, was shown to be paradoxical. It is also useful to talk of groups having the property of being paradoxical, where the group acts on itself by left multiplication. To transfer the paradoxical quality from a group to a set upon which it acts, we need the extra condition that the group acts without nontrivial xed points, that is, no non-identity element of the group xes a point of the set. Theorem 2 (AC). Let G be a paradoxical group acting on a set X without nontrivial xed points. Then X is G-paradoxical. Proof. Recall that a group is paradoxical if it is equidecomposable with two disjoint copies of itself. and transformations which witness that G is paradoxical. Suppose therefore, that A1 , . . . , Am , B1 , . . . , Bn G, and g1 , . . . , gm , h1 , . . . , hn G are subsets Let p X. We dene the orbit of p under G by o(p) := {g(p) : g G}, i.e. it is the set of set for a free group H G. This means that no non-trivial word composed of elements of S corresponds

all points that can be reached from p by action from G. Note that as G acts without nontrivial xed 4

points, any two orbits are either identical or disjoint. Hence, X partitions into a disjoint union of orbits. Formally, there exists a choice set M containing one element from each G-orbit in X. Then {g(M ) : g G} partitions X, because G acts without nontrivial xed points. Let A = {g(M ) : g Ai } and Bj = {g(M ) : g Bj }. Then, by the paradoxical decomposition i
m i=1 gi (Ai )

of G, we have that

=G=

n j=1

hj (Bj ). Therefore,
n

gi (A ) = X = i
i=1 j=1

hj (Bj )

Corollary. Let G be a group. Suppose H is a paradoxical subgroup of G. Then G is paradoxical. Hence, any group with a free subgroup of rank 2 is paradoxical. Proof. A subgroup H of a group G acts by left multiplication on the whole group, without nontrivial xed points. By Theorem 3, G is H -paradoxical. Then, G is G-paradoxical. An immediate problem we encounter is that groups often have non-trivial xed points. For example, if we consider the group of rotations about the origin in R2 , we can see that all rotations leave the origin as xed. Hausdors Paradox presents a precursor to BTP, where we restrict attention to exclude nontrivial xed points.

The Banach-Tarski Paradox

be the set of all such rotations. Let M be a choice set for the equivalence classes given by , i.e. M by AC). Since Q is countable, the set of rotations R is also countable, hence can be ennumerated thus: Let Mn = rn (M ), the set obtained after rotating M by some rn R. Clearly {Mn : n N} partitions S 1 . Furthermore, for each i, j N, we have that Mi Mj by rotation, since for each r R, = the reverse rotation r1 R. R := {rn : n N}

consists of one point from each equivalence class, as picked out by some choice function (which exists

We now note that the Naturals are equinumerous with the positive even integers, and also with

the positive odd integers. That is, they can be placed in 1-1 correspondence. We can therefore, rotate each of the sets in {Mn : n is even} to obtain the entire of S 1 . Similarly, we can rotate each of the sets in the remaining {Mn : n is odd} to obtain S 1 . This theorem gives us a taste for the style of proof required in the Banach-Tarski Paradox, and provides us with an example of a paradoxical set which only requires rigid motions. In fact, the proof can be streamlined by using Theorem 3. In 1914, Felix Hausdor proved that there is no nitely additive, rotation-invariant measure on (S 2 ), all subsets of a sphere, such that congruent pieces have equal measure. To do so, he constructed the following paradox: Theorem 4 (Hausdor Paradox, AC). There is a countable subset D S 2 such that S 2 \ D is SO3 -paradoxical. Proof. Suppose and are two independent rotations about axes through the origin in R3 . Let F be

the free group generated by the two rotations. We have done most of the hard work with Theorem 2 and its corollary, and so we would like to use it here. However, it cannot be applied without modication, as we require a paradoxical group that acts without nontrivial xed points, and clearly each nonidentity word xes two points in S 2 , namely the intersection of the axis of rotation with S 2 . We therefore need to remove the troublesome points. countable. Now, F acts on S 2 \D without nontrivial xed points, and so theorem 2 can be applied. 6 Let D be the set of all points which are xed by some g F. As F is countable, so D is also

This is close to our goal, but to get to BTP we still have to account for the countable subset D, and to shift from S 2 to the unit ball. Firstly, a theorem to show that as D is countable, S 2 is still paradoxical. Theorem 5. S 2 and S 2 \ D are SO3 -equidecomposable. We only require 2 pieces for this decomposition. To better appreciate how this proof works, it is useful to rst consider a similar example: Proposition 3. S 1 \ {x} and S 1 are SO2 -equidecomposable for any point x S 1 . Proof. Identify R2 with C in the usual way. Suppose that for some , x = ei . Let A = {ei+n : n Z} the set of all points reached from x by anti-clockwise rotation through a positive whole number of radians. By irrationality of 2, all the points in A are unique. Let B = S 1 \ A, so then S 1 = B A. Leaving B xed, rotate A by 1 radian anti-clockwise, to obtain that A A\{x}, and that S 1 S 1 \{x}. Note that this proof is a variation of the famous Hilberts Hotel. The crux of the argument for Theorem 5, then, is to nd some rotation of S 2 , such that a similar way to the above example to obtain the desired result. {D, (D), 2 (D), . . . , n (D), . . .} are all pairwise disjoint. We can then decompose and rotate in

Proof of Theorem 5. Let l be a line through the origin that doesnt intersect D; it will act as our axis of rotation. Such a line exists because D is countable. Then consider the set A of all angles , such that for some x D and n Z \ {0}, the rotation of x around l by n gives another point in D. by our careful choice of , we have that for any 0 m < n, m (D) n (D) = . Let =
n=0

A is countable, so we can choose some angle A, with corresponding rotation about l. Then, / n (D). Then:

S 2 = (S 2 \ ) () (S 2 \ ) = S 2 \ D

Theorem 6. Suppose G acts on X, and E, E X such that E G E . If E is G-paradoxical, then so is E. Proof. We recall that equidecomposability is an equivalence relation (refer back to Proposition 1). By use the fact that A B = and E G E to see that there exists disjoint subsets C, D E such that A G C and B G D. denition, E is G-paradoxical i E contains disjoint subsets A, B such that A G E and B G E. We

transitivity, C G E G D.

By simple application of the transitivity of G we have that A G E and B G E . Again, by

Theorem 7 (Banach-Tarski Paradox, AC). S 2 is SO3 -paradoxical. Consequently, any solid ball in R3 is E3 -paradoxical. S 2 \ D is SO3 -paradoxical. By Theorem 5 we have that S 2 SO3 S 2 \ D. Applying Theorem 6, we have that S 2 is SO3 -paradoxical. Next, we consider the unit ball B in R3 . By setting up a correspondence between a point P S 2 Proof. The Hausdor Paradox (Thm. 4) showed that there exists a countable subset D S 2 for which

with all the points directly below it along a radius, i.e. identifying every point P on the surface of the sphere with the set of points {P : 0 < 1}, and translating the radial points in relation to the rotation of P , we have that B \ {0} is G3 -paradoxical. It then suces to show that B B \ {0}.

We can use the same idea as in Proposition 3, by simply considering, say, a circle C, through the origin, with radius 1 and center ( 1 , 0, 0). Since 0 C, by applying Proposition 3 we obtain that 2 2 C C \ {0}. As C B, we can write B = C (B \ C) = B \ {0}

(C \ {0}) (B \ C)

Therefore, by Theorem 6, B is E3 -paradoxical. Since E3 includes translation, and nothing in the proof has required a limit on the size of the ball, we have that any ball in R3 is paradoxical. Corollary. R3 is E3 -paradoxical. Proof. Simply extend the correspondence by setting up a radial correspondence between S 2 and R3 \{0} as opposed to just B \ {0}. Everything else then follows in a similar fashion to the above.

Bibliography
[1] Capinski, Marek and Peter E. Kopp (2005). Measure, Integral and Probability. 2nd Edition. Springer. [2] Wagon, Stan (1985). The Banach-Tarski Paradox. Cambridge: Cambridge University Press.