I had initially begun the pursuit of learning how to achieve this paradox as a result of a

conversation I had on the subject while I was an undergraduate student. A professor was
offering a reading course to me on the basics of measure theory, and, predictably, the
concession that there exist non-measurable sets was presented as a necessary evil to the
theory. From there, sparked by previous conversations about the Axiom of Choice, the
dialog went something like this:

Professor : As a matter of fact it turns out there is a theorem, which requires the
Axiom of Choice, which states that a pea can be cut into finitely many pieces,
rearranged, then glued back together to have a ball the size of the Earth.

Me : Wow, how is that possible?

Professor : Well the sets used are really bizarre.

Me : What do they look like?

Professor : That’s outside the scope of this course.

His last statement, repeated in several of my analysis books prior to jumping ship on the
explanation, would ring in my ears for two years until I finally had a good excuse, and
time, to study the paradox. Interestingly enough, while the sets must indeed be bizarre,
the proof I will cover reveals little insight to their form. Instead, and possibly more
interestingly, the emphasis is placed on how the pieces are moved around! The goal of
this presentation is to describe exactly how the pieces are moved around.
 We start with a definition.

(1) Definition: Let G be a group acting on a set Χ and suppose E ⊆ Χ . E is G -

paradoxical if for some positive integers m, n there are pair-wise disjoint subsets
n
A1 , L , An , B1 L Bm of E and g1 , L , g n , h1 L hm ∈ G such that E = U g i (Ai ) and
i =1

m
E = U hi (Bi ) .
i =1

Figure 1

 n   m 
It is interesting to notice that  U (Ai ) U  U (Bi ) ⊆ Χ , not necessarily all of Χ .
 i =1   i =1 
The next theorem provides an example to illustrate this concept.

(2) Theorem : A free group F of rank 2 is F - paradoxical, where F acts on itself by

left multiplication.

Proof : Suppose that F is generated by α and β . The words in F are derived from the
alphabet consisting of two letters, α and β , and each word must therefore begin with

one of these letters, with the exception of the empty word denoted by {1}. Now, define

w (α ) to be all words in F which start (on the left) with α and define w (β ) , wα−1 , ( )
 ( )
and w β −1 similarly. Thus we have, using our new notation, another representation for

( )
F . That is, F = {1} U w (α ) U w α−1 U w(β ) U w β −1 . ( )

Figure 2

( ( )) ( ( ))
Since {1}(w(α)) U α w α−1 = F and {1}(w(β )) U β w β −1 = F , we are done. To verify,

( ) ( ) ( ( )) as
suppose ξ ∈ F \ w (α ) , then α −1ξ ∈ w α−1 and therefore ξ = α α −1ξ ∈ α w α −1
desired. QED

Next we establish a link between the abstract notion of a paradoxical free group
and something tied more closely to the spatial world. First we need to recall a definition
from algebra.
Definition : Let G be a group and let Χ be a set. The G -orbit of x ∈ Χ is the set
Ox = {gx | g ∈ G} .

Note that each x ∈ Χ is in some orbit, for if 1 is the identity in G , then

x = 1x ∈ {gx | g ∈ G}.

 nπ 
Ex. Let G =  rθ : θ = , n ∈ {0,1, K ,7} where rθ is
 4 
counterclockwise rotation through the angle θ . Let
Χ = S 1 , then each orbit consists of 8 points. The
illustration to the right illustrates the orbit O(0,1) . Note

that O(1, 0 ) , O(−1, 0 ) , and O(0, −1) represent the same orbit.

Figure 3
 Now, suppose we have a group G that operates on a set Χ and that G is G -
paradoxical. What does the paradoxical nature of G have to do with the paradoxical
nature of Χ ? We will show that if G is G - paradoxical then Χ is G - paradoxical.
Since G is paradoxical, we know that there exist pair-wise disjoint subsets
A1 ,K , An , B1 , K, Bm of G as well as elements of G g1 , K , g n , h1 , K , hm which satisfy
the definition. In order to show the set Χ is paradoxical, we use the Axiom of Choice
and orbits to establish a link to the group. By AC there must exist a set, call it Θ which
contains exactly one element from each G -orbit in Χ . The set {g (Θ) | g ∈ G} certainly

covers Χ since Θ contains one x ∈ Χ from each orbit left multiplied by each gα ∈ G ,
thus regenerating all orbits. If we further assume that G acts on Χ without any
(nontrivial) fixed points, then {g (Θ) | g ∈ G} forms a pair-wise disjoint partition of Χ .

That is {g (Θ) | g ∈ G} would then be a pair-wise disjoint family, which covers Χ . To see
this assume there exist x, y ∈ Θ and g1 , g 2 ∈ G so that g1 x = g 2 y ∈ {g (Θ) | g ∈ G} . Then

notice that this implies that x = g1−1 (g 2 y ) . Thus, x is another element in the orbit of y
which contradicts x, y ∈ Θ since we only chose one element from each orbit. The
previous contradiction will fail if it turns out that x = y . However, this implies that

g1 x = g 2 x and so the group element g1−1 g 2 ∈ G fixes x , contradicting the lack of

nontrivial fixed points.
Next, let Ai* = U {g (Θ) | g ∈ Ai } and B *j = U {g (Θ) | g ∈ B j } and notice that since

{Ai }U {B j } are pair-wise disjoint subsets of G , {Ai* }U {B *j } are pair-wise disjoint subsets

of Χ . Finally, for the coup de grâs, recall that since G is paradoxical we have
G = U g i ( Ai ) =U h j (B j ) and, therefore, by using the associativity of G and the fact that
i j

{g (Θ) | g ∈ G} forms a partition, we can deduce ( ) ( )

Χ = U g i Ai* =U h j B *j . Indeed,
i j

U g (A ) = U g (U {g (Θ) | g ∈ A }) which using associativity and the paradoxical nature

i
*
i i i
i i

of G gives us {g (Θ) | g ∈ G} a partition of Χ . This gives us the following theorem.

Theorem : If G is paradoxical and acts on Χ without nontrivial fixed points, then Χ is
G - paradoxical.

Thus we have the immediate corollary,

(3) Corollary : Χ is F - paradoxical whenever F , a free group of rank 2, acts on Χ

with nontrivial fixed points.

To proceed, we identify a free group in ℜ 3 with two generators. One of the

features of the Banach-Tarski paradox is that the way the pieces are moved around is via
isometries. This limits the generators of a rank 2 free group to two possible candidates,
rotations and reflections. It will be shown in the next statement that rotations provide the
necessary group structure.

(4) Theorem : SO 3 has a free subgroup of rank 2.

The proof of this statement is provided by Stan Wagon and is given below. The
idea behind the proof is to show, inductively, that any non-identity reduced word in the
alphabet of rotations around two coordinate axis will not generate the identity matrix. The
way this is done is by showing that the general form of any such word acting on a
standard basis element can not possibly give back that basis element, thus eliminating the
possibility of the word generating the identity matrix.

Proof : Let ϕ and ψ be counterclockwise rotations around the z-axis and x-axis,

respectively, each through the angle cos−1 ( 13 ) . Then ϕ ± and ψ ± are represented by
matrices as follows:

 13 m 2 2  1 0 0 
 3
  
ϕ ± =  ± 2 32 1
3 0 ψ = 0
± 1
3
m 2 32 
 0 1  0 ± 2 2 

1
 0 3 3 
We wish to show that no nontrivial word in ϕ ± , ψ ± equals the identity. Since
conjugation by ϕ does not affect whether or not a word acts as the identity, we may

restrict ourselves to words ending (on the right) in ϕ ± . Hence to get a contradiction,
assume that ω is such a word and ω equals the identity.
( )
We claim that ω(1,0,0 ) has the form a, b 2 , c 3k where a , b, c are integers and
b is not divisible by 3. Note, 0 is divisible by any integer. This implies that
ω(1,0,0 ) ≠ (1,0,0) which is the required contradiction. The claim is proved by induction

(
on the length of ω. If ω has length one, then ω = φ ±1 and ω(1,0,0) = 1, ± 2 2 , 0 3 . )
( )
Suppose then that ω = φ ±1ω' or ω = ψ ±1ω' where ω' (1,0,0 ) = a ' , b' 2 , c' 3k −1 . A single

( )
application of the matrices above shows that ω(1,0,0 ) = a , b 2 , c 3k where a = a 'm4b' ,
b = b' ±2a ' , and c = 3c' , or a = 3a ' , b = b'm2c ' , and c = c'±4b' according as ω begins
with φ ±1 or ψ ±1 . It follows that a , b, c are always integers.
It remains only to show that b never becomes divisible by 3. Four cases arise
according as ω equals ϕ ±1ψ ±1v , ψ ±1ϕ±1v , ϕ ±1ϕ ±1v , or ψ ±1ψ ±1v where, possibly, v is the
empty word. In the first two cases, using the notation and equations of the previous
paragraph, b = b'm2c ' where 3 divides c' or b = b' ±2a ' where 3 divides a ' . Thus if b' is
not divisible by 3, neither is b . For the other two cases, let a" , b" , c" be the integers
arising in v (1,0,0 ) . Then in either case, b = 2b'−9b" . For instance, in the third case,

b = b'±2a ' = b'±2(a" m4b") = b'+b"±2a" −9b" = 2b'−9b" ; an essentially identical proof
works in the fourth case. Thus if b' is not divisible by 3, neither is b, completing the
proof. QED

At this point we would like to combine the results (3) and (4) so that we could
produce a paradoxical subset of ℜ 3 using the group of rotations. Unfortunately, a
problem arises when trying to apply these results to a particularly fundamental object,
S2.
 Define the group of rotations from (4) to
be ℑ . Then notice that any element from
ℑ will fix exactly two elements of S 2
where the axis of rotation intersects the
sphere. Thus we may not use (3), yet. Note
however that the group ℑ is countable
since it consists of words containing only
finitely many syllables, thus so is the set of
Figure 4 points of S 2 that it fixes, call it Ω .
Then the set S 2 \ Ω still has uncountably many elements and ℑ acts on it without
nontrivial fixed points. We may want to be careful and be sure that if p ∈ S 2 \ Ω then

g ( p ) ∈ S 2 \ Ω as well for all g ∈ ℑ . So, proceed by contradiction. That is, assume that

p ∈ S 2 \ Ω and g ( p ) ∉ S 2 \ Ω , then g ( p ) ∈ Ω . So that means there is an element f ∈ ℑ

so that f [g ( p )] = g ( p ) . However, if this were true we would have the following,

( )
f [g ( p )] = g ( p ) ⇒ g −1 fg ( p ) = p and so g −1 fg fixes p contradicting p ∈ S 2 \ Ω . We
have worked out the following result, which is the first real taste of the Banach-Tarski
paradox.

(5) Theorem : (Hausdorff Paradox) (AC) There is a countable subset Ω of S 2 such

that S 2 \ Ω is SO 3 - paradoxical.

We next need to know what we mean when we say the set A looks like the set B.
The version for “looks like” we will use in this presentation is known as
equidecomposable.
(6) Definition : Suppose G acts on Χ and A, B ⊆ Χ . Then A and B are G -

equidecomposable ( A ~ G B ) if A and B can each be partitioned into the same finite

number of respectively G - congruent pieces. Formally,

n n
A = U Ai , B = U Bi
i =1 i =1

Ai I A j = ∅ = Bi I B j if i < j ≤ n , and there are g1 , K, g n ∈ G such that, for each i ≤ n

g i ( Ai ) = Bi .

Figure 5

There is another nice property about the above definition.

Proposition : ~ G defines an equivalence relation.

Proof : Suppose G acts on Χ and A, B , C ⊆ Χ .

i) Since {1}∈ G we immediately have {1}A = A and so A ~ G A .

ii) Assume that A ~ G B . Then we know there exist {Ai }in=1 and {Bi }ni=1 which

are each pair-wise disjoint partitions for A and B , respectively, so that

for each i ≤ n g i ( Ai ) = Bi according to the definition of ~ G . Then since

G is a group it necessarily contains an inverse for each of its elements and

so we also have, Ai = gi−1 (Bi ) and thus B ~ G A .

iii) Assume that A ~ G B and B ~ G C . Then we have pair-wise disjoint

{}
families {Ai }in=1 , {Bi }ni=1 , Bˆ j =1 , and {C j }mj=1 as well as group elements
m
g1 , K , g n , h1 , K hm ∈ G so that for each i ≤ n , g i ( Ai ) = Bi and for each

( )
j ≤ m , h j Bˆ j = C j . Now further partition A into at most mxn pair-wise

disjoint pieces using the following scheme,

(
Aˆ1 = g1−1 B1 I Bˆ1 , )
Aˆ 2 = g1−1 (B 2 I Bˆ ),
1

M
( )
Aˆn = g1−1 B n I Bˆ1 ,

Aˆn +1 = g (B I Bˆ ),
−1
2 1 2

Aˆn +2 = g (B I Bˆ ) ,
−1
2 2 2

M
(
Aˆnm−1 = g n−1 Bn −1 I Bˆ m , )
Aˆ nm = g n−1 (B n I Bˆ m .)
Next, define the following maps,
k1 = g1 ,

k2 = g2 ,

M
kn = gn ,

k n+1 = g1 ,

k n+2 = g 2 ,

M
k nm−1 = g n −1 ,

k nm = g n .
Finally, notice that for all i = 1, K, n , and α = 0, K , m − 1

( ( )) ( ( (
hα kαn +i Aˆαn +i = hα kαn +i g i−1 Bi I Bˆα )))
= h (g (g (B I Bˆ )))
α i i
−1
i α

= h ((B I Bˆ )) ⊆ C ,
α i α α
 U h (k (Aˆ )) = C
n
and that α αn + i αn +i α . See figure 6 for a visual interpretation.
i =1

Figure 6

Therefore, it follows that A ~ G B and B ~ G C ⇒ A ~ G C . Combining i), ii),

and iii) we have shown that ~ G is an equivalence relation.

Using this notion of equidecomposability, and the fact that it defines an
equivalence relation, we can now redefine our notion of paradoxical in a more useful
way.

Definition : ( G - paradoxical II) E is G - paradoxical II if E contains disjoint sets A

and B such that A ~ G E and B ~ G E .

For an illustration, see Figure 1.

Lemma: E is G-paradoxical II iff E is G-paradoxical.

Proof : If E is G - paradoxical then there exist pair-wise disjoint subsets
n
A1 , K, An , B1 , K, Bm of E and g1 , K, g n , h1 ,K , hm ∈ G such that E = U gi ( Ai ) and
i =1

m
E = U hi (Bi ) . We are to show there are disjoint sets A and B such that A ~ G E and
i =1

B ~G E .
n m
Let A = U Ai and B = U Bi . Then notice g i ( Ai ) ∈ E so define Ei = g i ( Ai ) . Is it
i =1 i =1

true that Ei I E j = ∅ for all i < j ≤ n ? If not there are i ≠ j so that Ei I E j ≠ ∅ and,

hence, elements x ∈ Ai and y ∈ A j so that g i (x ) = g j ( y ) . However, we need not be

redundant, that is we only need either x ∈ Ai or y ∈ A j not both. Remove either x from

Ai or y from A j . Continuing in this manner it is possible to restrict the sets Ai so that

C i = g i (Ai ) is a partition of E . Then, by construction, we have A ~ G E . The case to

show B ~ G E is identical.

Now assume there exist disjoint sets A and B subsets of E such that A ~ G E

and B ~ G E . We are to show that E is G - paradoxical. This is immediate. Since

n
A ~ G E we know there exist A1 , K , An ∈ A such that A = U Ai , Ai I A j = ∅ for all
i =1

n
i < j ≤ n , and there exist g1 , K, g n ∈ G such that g i ( Ai ) = E i , where UE i = E and
i =1

Ei I E j = ∅ for all i < j ≤ n . Similarly for B ~ G E . Finally since A and B are disjoint

we are done. Therefore our new definition is equivalent to the former. QED

If E and E ′ are equidecomposable then there is a way to break them up into the
same number of pieces and have maps which take one piece from E directly onto a piece
of E ′ . So then, if E or E ′ also happens to be paradoxical then there should be some way
to refine the equidecomposable maps so that pieces used to show E ~ E ′ disjointly cover
the pieces which represent the paradoxical nature. Thus, using the obvious composition
of maps, we see that both sets would necessarily be paradoxical. The proof of the next
theorem is very similar to the proof of the transitivity of the equidecomposable relation
~G .

(7) Theorem : Suppose G acts on Χ and E , E ′ are G - equidecomposable subsets of

Χ . If E is G - paradoxical, so is E ′ .

The proof of this statement is similar to the proof of the transitivity of the ~ G

equivalence relation and will be omitted.

We now define a partial ordering of the ~ G classes. First we introduce some new
notation.

Notation : Suppose a set A is equidecomposable to a subset of another set B with

respect to some group G . Then we write, A p B .

To show that p defines a partial order, we must show that p is reflexive,

antisymmetric, and transitive. Fortunately two of these properties have already been
shown. The fact that p is reflexive follows from the fact that any set is
equidecomposable to itself via the identity map and any partition. The transitivity follows
from the previous proof of transitivity for the relation ~ G . The very desirable property of

antisymmetry is not quite so obvious and is a result credited to Banach, Schröder, and
Bernstein.

(8) Theorem : (Banach – Schröder – Bernstein) Suppose G acts on Χ and A, B ⊆ Χ .

If A p B and B p A , then A ~ G B . Thus p is a partial ordering of the ~ G classes in

℘(Χ ) .

The proof of this theorem hinges on the use of two lemmata.

Lemma 1 : If A ~ G B then there is a bijection g : A → B such that C ~ G g (C )
whenever C ⊆ A .

Lemma 2 : If A1 I A2 = ∅ = B1 I B2 , and if A1 ~G B1 and A2 ~ G B2 , then

( A1 U A2 ) ~ G (B1 U B2 ) .

Proof (Lemma 1) : Using the maps that witness A ~ G B we can easily define a piece-
wise map which takes A onto B . The fact that this map is 1-1 follows from the fact that
each element contributing to the piece-wise map is an invertible group element acting on
Χ which contains both A and B . So, formally, if Ai , Bi , and {g i }ni=1 ∈ G are the sets

 g1 , ∀ a ∈ A1
 g , ∀a ∈ A

and maps which witness A ~ G B then, g :=  2 2
is the desired bijection. It is
 M
 g n , ∀ a ∈ An

then an immediate consequence that C ~ G g (C ) whenever C ⊆ A . QED

Proof (Lemma 2) : This is almost immediate. Let A1i , B1i , and {g i }ni=1 be the sets and maps

which witness A1 ~ G B1 and let A2i , B2i , and {hi }im=1 be the sets and maps which witness

 Ai , i = 1, K , n
A2 ~ G B2 . Then since, A1 I A2 = ∅ = B1 I B2 define C i =  i −n 1 and
 A2 , i = n + 1, K , n + m
 g i , i = 1,K , n
also define li =  . Thus we have constructed the sets and maps
 hi −n , i = n + 1, K , n + m
which witness ( A1 U A2 ) ~ G (B1 U B2 ) . QED

Proof (Banach – Schröder – Bernstein) : Since A p B we know there exists a set B1 ⊆ B

so that A ~ G B1 and because B p A there exists a set A1 ⊆ A so that B ~ G A1 . Thus by

lemma 1 we know there exist bijections f : A → B1 and g : A1 → B so that if C A ⊆ A

 1 1
( )
and C A ⊆ A1 then C A ~G f (C A ) and C A ~ G g C A . Now, define C 0 = A \ A1 and
1

∞
inductively define C n = g −1 f (C n −1 ) and let C = U C n .
n= 0

Claim : g ( A \ C ) = B \ f (C )

Pf : It should be noted that the above map g ( A \ C ) is indeed acting only on its

domain. That is, the set C o = A \ A1 ⊆ C , therefore ( A \ C ) ⊆ A1 and so g ( A \ C )

is well defined. Notice that C = g −1 f (C ) ⇒ g (C ) = f (C ) . Then, since g is the

bijection constructed using the maps witnessing B ~ G A1 we have

g ( A \ C ) = B \ f (C ) .

Therefore, we have from lemma 1 ( A \ C ) ~ G (B \ f (C )) and f (C ) ~ G C . So, using

lemma 2 we finally have C U ( A \ C ) ~ G f (C ) U (B \ f (C )) ⇒ A ~ G B as desired. QED

We are now in a position to get an improvement on Hausdorff’s paradox. Recall

that we were able to show that S 2 \ Ω is SO 3 paradoxical. The question now is, how

much does S 2 \ Ω look like S 2 ? Well, if we use the notion of equidecomposability to

define “looks like” then the answer is they look the same! We will construct two
different representations for S 2 and S 2 \ Ω , respectively, which are equidecomposable.
To do this we show that a certain rotation will, in a sense, absorb the fixed points. Indeed,
suppose ρ is a rotation so that ρ(Ω ), ρ 2 (Ω ), K is a pair-wise disjoint sequence of sets.
∞
( )
Then, if we define Ω = U ρ n (Ω ) , we have S 2 = Ω ∪ S 2 \ Ω . Next, notice that this
n =0

( )
isn’t much different from S 2 \ Ω . Indeed, S 2 \ Ω = ρ(Ω ) ∪ S 2 \ Ω since the only copy

( )
of Ω in the set Ω ∪ S 2 \ Ω is removed when Ω is left multiplied by ρ . Therefore, S 2

is decomposed into two distinct pieces and S 2 \ Ω is decomposed into two distinct pieces
with maps ( ρ and {1} ) which take one decomposition directly onto the other we have

S 2 ~ S 2 \ Ω . All that remains is to show that such a rotation exists.

(9) Theorem : If Ω is a countable subset of S 2 , then S 2 and S 2 \ Ω are SO 3 -

equidecomposable.

Proof: Since Ω is a countable subset there must be a line l which passes through the
origin and does not intersect Ω . We now show exactly what the rotation can’t be, then
show that there are choices remaining to choose from. Since Ω is countable, give it the
representation Ω = {xn }n∞=1 and for each x n ∈ Ω define Φ n as the set of all angles θ so

that when xn is rotated about the line l through θ radians it falls back into Ω . Since Ω
∞
is countable, Φ n is countable and therefore UΦ n is also countable. Thus let ρ be any
n=1

∞
angle so that ρ ∉ U Φ n . Then we immediately have ρ n (Ω ) I Ω = ∅ by construction
n=1

and so it follows that ρ n (Ω ) I ρm (Ω ) = ∅ whenever 0 ≤ m < n . This follows because

ρ n (Ω ) I ρ m (Ω ) = ρ n−m (Ω ) I ρ(Ω ) = ∅ . QED

The following is an interesting corollary to this result which, when combined with
previous results, yields the strong form of the Banach – Tarski Paradox.

(10) Corollary : (Banach – Tarski Paradox weak form) (AC) S 2

is SO 3 - paradoxical,

as is any sphere centered at the origin. Moreover, any solid ball in ℜ 3 is G3 - paradoxical

where G3 is the group of all isometries on ℜ 3 , and ℜ 3 is itself paradoxical.

Proof: The previous theorem (9) shows that S 2 contains two equidecomposable subsets,
S 2 and S 2 \ Ω , one of which, S 2 \ Ω , is paradoxical according to the Hausdorff Paradox
(5). Thus, by theorem (7) we immediately have S 2 is paradoxical. Notice further that
none of the arguments use the radius of the sphere in any way. So, then, any sphere
centered at the origin in ℜ 3 is paradoxical.
 To verify the statement that any solid ball in ℜ3 is G3 - paradoxical first note that

since G3 contains all translations it suffices to assume that we are dealing with a ball
centered at the origin. Observe that the unit ball take away the origin deformation retracts

. Indeed, the unit ball minus the origin ( B (0,1) \ 0 ) can use
x
onto S 2 via the map x a
x

the same decomposition as the sphere. Visualize this as the laying S 2 \ Ω on top of
B(0,1) and removing the lines in the ball which connect points in Ω to the origin. Then
we have a continuum of shells which are all paradoxical by the same isometries, so we
may do it all at once so to speak. So, all we need to show is that the origin can somehow
be absorbed. To do this use the same trick involved in the last proof to absorb the
countable set Ω . Let p = (0,0, 12 ) , and let ρ be a rotation about an axis which passes
through p and misses the origin. Then further notice that when the origin is rotated about
this axis it will never go outside of the ball. So we have,
{ } { }
B (0,1) \ (0,0,0) U ρ ρ n (0,0,0) | n = 0,1,2, K ~ ρ n (0,0,0) | n = 0,1,2, K U B (0,1) similar to

the previous proof. Since ℜ 3 \ (0,0,0) also deformation retracts onto S 2 using the radial

contraction map, we also have ℜ 3 is paradoxical. QED

Finally, using the weak form of the paradox we can say something even a little
more surprising.

(11) Theorem : (Banach – Tarski Paradox strong form) (AC) If A and B are any two

bounded subsets of ℜ 3 , each having nonempty interior, then A and B are

equidecomposable.

Proof: Since A and B are arbitrary bounded sets we only need to show A p B since
showing B p A would be the same. Then using result (8) we will be done.
 Since A and B are bounded with
nonempty interior we can encase A in a
solid ball K and let L be a solid ball
contained in B . Without loss of generality
assume that K has larger volume than L .
Since both are bounded, there exists an
integer n so that K can be covered by n
Figure 7
copies of L .

Now suppose S is a set of n disjoint

copies of L . Use the result of theorem (7)
to construct n paradoxical subsets of L .
Next use the weak form of the Banach –
Tarski Paradox to generate n copies of L
using these subsets and then use
Figure 8
translations to move the copies to obtain
S p L.
Since we know that we can cover the set K with n copies of L we have K p S
via the identity map or translations alone. Thus, A ⊆ K p S p L ⊆ B , and so A p B and
we are done. QED

To see that the above usage of the weak form of the Banach – Tarski Paradox
along with Theorem (7) will generate a cover of the set S recall the following facts. L is
paradoxical and therefore contatins two sets which are each equidecomposable to L , thus
by Theorem (7) each of these sets are paradoxical since L is paradoxical. One of these
sets can be rotated via the predefined isometries to generate a copy of L and then
translated over part of S . Then, since the other set is paradoxical, it contatins two subsets
which are equidecomposable to the whole set, and by transitivity they each are
equidecomposable to L . They are also each paradoxical by Theorem (7). Take one of
these two, generate a copy of L and translate over S and use the other to generate two
more paradoxical sets. Repeat this procedure as many times as necessary.
 There was only one source used for this paper. All of the above ma terial was an
elaboration on the first three chapters of a marvelous book writtten by Stan Wagon. The
specifics are as follows:

Wagon, S. (1985). The Banach-Tarski Paradox. Cambridge: Cambridge University

Press.