Maths Olympiad Prep

Library / /459 of 462

Combinatorics Difficulty 7.8 National Olympiad, round 2 Prove it Ireland

Let nn be a positive integer. An nn-level honeycomb is a plane region covered with regular hexagons of side-length 11 connected along edges, such that the centres of the boundary hexagons are lined up along a regular hexagon of side-length n3n\sqrt{3}. The diagram shows a 22-level honeycomb from which the central hexagon has been removed.

Figure 1

A trex is a sequence of 33 hexagons with collinear centres such that the middle hexagon shares an edge with each of its neighbours in the trex.

An nn-level honeycomb from which the central size-11 hexagon has been removed is to be completely covered by trexes without any overlaps. Find all values of nn for which this is possible.

Solution

We will first prove that coverings can be found if n=3kn = 3k and if n=3k1n = 3k - 1. When k=1k = 1 these are the cases n=2n = 2 and n=3n = 3 for which the diagrams below show a covering by trexes. In these diagrams, the unit hexagons are represented by their centres.

Figure 2

To extend this for all k1k \ge 1, we observe that, for any n1n \ge 1, the region obtained by removing an nn-level honeycomb from an (n+3)(n+3)-level honeycomb can be covered by trexes without overlap, see left diagram below. The statement follows now from the cases n=2n = 2 and n=3n = 3.

Alternatively, we may cover the large hexagon formed by the centres of the unit hexagons by 33 congruent parallelograms, each covering n(n+1)n(n+1) centres, as shown in the diagram on the right below.

Figure 3

When the number of hexagon centres along one side of such a parallelogram is divisible by 33, we can cover it by trexes. More specifically:

* if n=3kn = 3k then each parallelogram of size 3k(3k+1)3k(3k+1) can be covered by k(3k+1)k(3k+1) trexes;
* if n=3k+2n = 3k + 2 then each parallelogram of size (3k+2)(3k+3)(3k+2)(3k+3) can be covered by (k+1)(3k+2)(k+1)(3k+2) trexes.

Figure 4

Next we prove that no coverings exist if n=3k+1n = 3k+1. We present three different proofs. In all of them we label the centres of the hexagons by 00, 11, 22 with 11 in the centre of the honeycomb and so that two hexagons with the same label never share an edge. It follows that every trex contains each label 00, 11, 22 exactly once, hence the honeycomb (with central hexagon removed) can only be covered with trexes when it contains the same number of 00-s, 11-s and 22-s.

We now prove that this is not the case when n=3k+1n = 3k + 1.

Proof 1. We note that the labelling is preserved by rotation around the centre with angle 120120^\circ. Hence by splitting the honeycomb minus the centre into three parallelograms of sides n=3k+1n = 3k + 1 and n+1=3k+2n + 1 = 3k + 2 as above, we note that the numbers of 00-s, 11-s and 22-s in each parallelogram must be one third of the total numbers of 00-s, 11-s and 22-s in the honeycomb minus the centre. However, the number of hexagon centres in such a parallelogram is (3k+1)(3k+2)(3k + 1)(3k + 2) which is not a multiple of 33 and so cannot contain the same number of 00-s, 11-s and 22-s. Hence the numbers of 00-s, 11-s and 22-s in the honeycomb minus the centre cannot be equal to each other.

Proof 2. As we have seen above, the region obtained by removing an nn-level honeycomb from an (n+3)(n+3)-level honeycomb can be covered by trexes without overlap. This implies that the numbers of 00-s, 11-s and 22-s in an (n+3)(n+3)-level honeycomb (with centre removed) coincide if and only if they do so in an nn-level honeycomb (with centre removed).
Because in case n=1n = 1 there are three hexagons labelled 00 but no hexagon labelled 11, the numbers of 00-s and 11-s in a (3k+1)(3k+1)-level honeycomb (with centre removed) do not coincide.

Proof 3. For each a{0,1,2}a \in \{0, 1, 2\} and each nn, let hn(a)h_n(a) denote the number of hexagons labelled aa in HnH_n, the nn-level honeycomb (including the central hexagon). We wish to find hn(a)h0(a)h_n(a) - h_0(a). Let
cn(a)=hn(a)hn1(a)=bn(a)+vn(a) c_n(a) = h_n(a) - h_{n-1}(a) = b_n(a) + v_n(a)
denote the number of labels aa on the collar HnHn1H_n - H_{n-1}, where vn(a)v_n(a) counts the labels at the 66 corners and bn(a)b_n(a) counts the remaining vertices.

Figure 5

We can compare HnHn1H_n - H_{n-1} with Hn2Hn3H_{n-2} - H_{n-3}, as the labels at the outer border (except at the corner) repeat the labels from the inner border. Hence the following recurrences for n3n \ge 3:
bn(a)=bn2(a)+2vn2(a)and so b_n(a) = b_{n-2}(a) + 2v_{n-2}(a) \quad \text{and so}
cn(a)=bn(a)+vn(a)=cn2(a)+vn2(a)+vn(a) c_n(a) = b_n(a) + v_n(a) = c_{n-2}(a) + v_{n-2}(a) + v_n(a)
while v3(1)=6v_3(1) = 6 and for n4n \ge 4 we have vn(a)=vn3(a)v_n(a) = v_{n-3}(a). Also using hn(a)=hn1(a)+cn(a)h_n(a) = h_{n-1}(a) + c_n(a) we can fill in the following table.

nvn(0)v_n(0)vn(1)v_n(1)vn(2)v_n(2)cn(0)c_n(0)cn(1)c_n(1)cn(2)c_n(2)hn(0)h_n(0)hn(1)1h_n(1) - 1hn(2)h_n(2)
0010000000
1303303303
2303363666
3060666121212
4303969211821
53039129303030
6060121212424242

Applying the recurrence relation 33 times for n7n \ge 7:
cn(a)=cn2(a)+vn(a)+vn2(a)=cn4(a)+vn(a)+2vn2(a)+vn4(a)=cn6(a)+vn(a)+2vn2(a)+2vn4(a)+vn6=cn6(a)+12 \begin{align*} c_n(a) &= c_{n-2}(a) + v_n(a) + v_{n-2}(a) \\ &= c_{n-4}(a) + v_n(a) + 2v_{n-2}(a) + v_{n-4}(a) \\ &= c_{n-6}(a) + v_n(a) + 2v_{n-2}(a) + 2v_{n-4}(a) + v_{n-6} \\ &= c_{n-6}(a) + 12 \end{align*}
as vn6(a)=vn(a)v_{n-6}(a) = v_n(a) and vn4(a)=vn1(a)v_{n-4}(a) = v_{n-1}(a) and vn(a)+vn1(a)+vn2(a)=6v_n(a) + v_{n-1}(a) + v_{n-2}(a) = 6
for all a{0,1,2}a \in \{0, 1, 2\}. Hence by induction we can prove
hn(0)=hn(1)1=hn(2)for all n=3k and 3k+2,hn(0)=hn(1)+2=hn(2)for all n=3k+1. \begin{align*} h_n(0) &= h_n(1) - 1 = h_n(2) && \text{for all } n = 3k \text{ and } 3k + 2, \\ h_n(0) &= h_n(1) + 2 = h_n(2) && \text{for all } n = 3k + 1. \end{align*}

Hence after removing the central hexagon, we have shown that the nn-level honeycomb cannot be covered by trexes if n=3k+1n = 3k+1, since at least 33 hexagons labelled 11 would remain uncovered.

Final answer:

All nn except those congruent to 11 modulo 33 (i.e., all nn such that n≢1(mod3)n \not\equiv 1 \pmod{3}) can be covered by trexes as described.

Want a route through all this instead of an archive? The track puts 2,000 problems in a working order, from AMC 10 level to the IMO shortlist.

Source: MathNet, licensed CC-BY-4.0. Statement reproduced verbatim; metadata (topic, difficulty) added by this project.