# Non-empty, disjoint subsets of $\Bbb R^2,$ both isometric to their union?

This question was posed to me by a friend who tells me the answer is yes, but I cannot see why.

Does there exist two nonempty, disjoint sets $A, B \subset \mathbb{R}^2$ such that $A$ and $B$ are both isometric to their union?

I cannot for the life of me come up with a way of constructing both sets - I've noticed that the measure of both has to be $0$, but that's about it.

• The measure could also be $\infty$. Or else they could be non-measurable. – user357151 Jul 15 '18 at 15:37
• Perhaps A and B are of the same range and 'content' but of opposite sign. – poetasis Jul 15 '18 at 15:39

Let the group with presentation $\langle x,y \mid x^2=e \rangle$ (that is, the free product $C_2 * \mathbb Z$) act on the plane by mapping $x$ to a rotation by $180^\circ$ about $(1,0)$ and $y$ to a rotation about the origin by some angle $\theta$ whose cosine is transcendental.

(Originally I claimed that this action is faithful, but actually it isn't -- the group element $y^{-2}xyxyxy^{-2}xyxyx$ has the identity action no matter what $\theta$ is. Fortunately less can do for this particular purpose).

Let $K$ be the set of group elements that can be written on the form $$y^{n_m}xy^{n_{m-1}}x\cdots xy^{n_1} x y^{n_0}$$ where $m\ge 0$, and each $n_j\ge 0$ except that $n_0$ may be negative. This set is not a subgroup, but it has two useful properties:

1. $K$ is the disjoint union of $\{yk\mid k\in K\}$ and $\{xyk\mid k\in K\}$.

2. Different elements of $K$ map the point $p_0 = (3,0)$ to different points. (This can be seen by going to the complex plane where $y$ is multiplication by $e^{i\theta}$ and $x$ is the map $z\mapsto 2-z$. Then each element of $K$ maps $3$ to a different Laurent polynomial in $e^{i\theta}$ with integer coefficients, and those all have distinct values because $e^{i\theta}$ is transcendental. Phew!)

Now set $$A = \{ykp_0 \mid k\in K\} \qquad B = \{xykp_0 \mid k\in K \}$$ Then $A$, $B$, and $A\cup B$ are related by rigid motions of the plane, namely: $$y^{-1}A = (xy)^{-1}B = \{kp_0 \mid k\in K \} = A \cup B$$ so they are isometric.

This construction is inspired by the initial step of the proof for the Banach-Tarski paradox. It doesn't need the axiom of choice because it doesn't need to select a representative of each orbit, because it is not required that $A\cup B$ is an entire pre-existing shape.

It is natural to ask how $A$ looks -- but unfortunately it can't really be seen: it is dense in $\mathbb R^2$.

• Thank you, amazing answer! Really basic question, I know, but is the group you described abelian? Also, how do you know $A$ is dense in the plane? – NMister Jul 16 '18 at 19:51
• @NMister: No, it is quite non-abelian -- it needs to be; otherwise $yxp_0 \in A$ and $xyp_0 \in B$ would be the same point and the construction wouldn't work. Nontrivial free products are never abelian. – Henning Makholm Jul 16 '18 at 20:27
• @NMister: A is dense in the plane because the image of $(3,0)$ under iterations of $y$ is dense in a circle around the origin with radius $3$. Doing an $x$ now moves the circle away from the origin -- so now the distances from the origin are dense in the interval $[1,5]$. Then doing more $y$s will smear the moved circle out to points that lie densely in an annulus with radii from $1$ to $5$, and now the next $x$ will move that annulus again ... as we have more and more $x$s in the word we can get arbitrarily far from the origin. – Henning Makholm Jul 16 '18 at 20:33
• Thanks! One more question - How do we know we can't cancel the $e^{i\theta}$s? – NMister Jul 19 '18 at 17:17
• @NMister: The Laurent polynomials must be different for different $k\in K$ because (a) the cofficients are always integers, (b) the ones that are not $0$ or $\pm 3$ encode $x$s in the word $k$, and the distances between them count the number of $y$s between successive $x$s, (c) exactly one of the coefficients is odd, and its position relative to rightmost "not $0$ or $\pm 3$" encodes the power of $y$ after the last $x$ (d) the exponent corresponding to the odd coefficient is the total power of $y$s in $k$. – Henning Makholm Jul 19 '18 at 18:03