Você está na página 1de 5

Conways Tribones Theorem

Rich Schwartz
April 19, 2016

References

These notes give an exposition of Conways Tribones Theorem. Here are


some references for further reading.
[CL] J. H. Conway and J. C. Lagarias, Tilings with polyominoes and combinatorial group theory, J. Combinatorial Theory, Ser A. 53 (1990): pp 183206
[P] J. Propp, A Pedestrian Approach to a Method of Conway, or, A Tale
of Two Cities, Mathematics Magazine, 1997.
[T] W. Thurston, Conways Tiling Groups, American Mathematical Monthly
1990.

The Result

Let Tn be a triangular array of dots in the plane. Figure 1 shows T5 . Figure


1 also shows how we have crossed out some subsets of 3-in-a-row, which we
call tribones. Conways Theorem is that it is impossible to write Tn as a
disjoint union of tribones. That is, we cant keep crossing off tribones and
get to an empty board.

Figure 1: The coarse hyperbolic plane


Tn nas n(n + 1)/2 dots, and this number is divisible by 3 if and only if n
or n + 1 is divisible by 3. So, in one out of three cases, Conways Theorem
has a trivial proof. One can use a vertex coloring argument to rule out some
other cases. Label all the dots in the kth row by k mod 3. Each tribone
involves vertices whose labels sum to a number divisible by 3. So, if the total
point sum is not divisible by 3 then Conways Theorem holds. This happens,
e.g. for T6 . However, this will not rule out T9 . So, the labeling trick does
not always work.

Transplantation

Now we start down the road to a proof. The proof is based on the fact that
there are 2 canonical edge colorings of the hexagonal lattice using 1, 2, 3. The
first coloring has the property that two edges get the same label if and only if
they are parallel. The second coloring is a little more complicated. We first
3-color the hexagonal faces so that adjacent faces have different labels. We
then color each edge by the label which is not either of adjacent face labels.
Figure 2 shows parts of the two edge colorings.

Figure 2: Two edge colorings of the hexagonal graph


There is a basic operation defined on walks in the hexagonal graph. Draw
the path in the first copy of the hexagonal graph, read off the labels, and then
draw a new path in the second graph following the labels. For instance, if our
path is a closed hexagon then it corresponds (say) to the path 123123. When
we draw this in the second copy of the graph, we get a zigzag. Were we to
take the path 231231 the transplant would also be a zigzag. Curiously, if we
interchange the roles of the two edge labelings, the transplantation process
works the same way.
Say that a tile is a union of 3 hexagons stacked in a row. There are 3
possible positions for a tile, up to a translation. The word corresponding to
the boundary of a tile has the form 1(323232)1(323232) up to permutation
of the letters and cyclic relabeling. Figure 3 shows the transplant of a path
going around the boundary of a tile. Ive separated out the central part of
the path to show it more clearly.

Figure 3: The transplant of a tile boundary.


There are several things to observe:
3

The transplant is closed.


The transplant encloses 2 hexagons having the same face label, and it
travels different directions around each one.
We define a path to be strongly closed if it is a closed loop and if it
surrounds each kind of hexagon 0 net times. To put it another way, we take
each kind of hexagon labeled k and we assign an integer depending on how
many times the loop winds around the hexagon. the sign tells whether we
have gone an excess of clockwise or an excess of counterclockwise. We sum
up all these numbers and get a single number n(k). What we are asking is
that n(1) = n(2) = n(3) = 0.
Let A denote the transplant of A. We define a loop A to be null if A is
strongly closed. For instance, the loop representing a tile boundary is null.
Here are some basic facts about null loops.
If A is null then any cyclic retracing of A is also null. What we mean is
that if we just start tracing A from a different starting point, then the
transplant is again strongly closed. The point is that the transplant of
the cyclic permutation is just a cyclic permutation of the transplant.
If A and B are null then so is the concatenation AB. The point here
is that (AB) = A B .
If C is any path then CC 1 is null. Here C 1 is just the same path
traced in the opposite direction. The point here is that (C 1 ) =
(C )1 .
A is null if and only if A(CC 1 ) is null. Here C is any path. This
follows from the previous results.
Now we can prove the main technical result.
Lemma 3.1 Suppose A is a closed polygon which bounds a union of nonoverlapping tiles, then A is null.
Proof: The proof goes by induction on the area of the region bounded
by A. Choose some tile which touches the boundary of A and has the
property that still has a polygonal boundary. For instance, choose
to be a tile having a point with the largest y-coordinate in it. Let B be the
4

boundary of . Let C be the boundary of . Let D be the path at the


interface of and . Figure 4 shows how A DD1 is the concatenation of A
and B. Here A is some cyclic relabeling of A. Now B is null because it is a
tile boundary. C is null by induction. Hence BC is null. Hence A DD1 is
null. Hence A is null. Hence A is null. This completes the proof.

Figure 3: The transplant of a tile boundary.

The End of the Proof

Well argue by contradiction. Suppose that some Tn is a union of tribones.


We think of the dots of Tn as the centers of hexagons in the hexagonal tiling.
But then the union of all these hexagons is covered by non-overlapping tiles.
But then the boundary of this union is null. The boundary curve has the
form (12)n (23)n (31)n . The transplant of this curve winds essentially n times
around each of 3 hexagons. The numbers associated to this path, even if it
is closed, are essentially (n/3, n/3, n/3) rather than (0, 0, 0). In other words,
the transplant is far from being strongly closed, This is a contradiction. That
completes the proof.

Você também pode gostar