Monomer-dimer tatami tilings of square regions

We prove that the number of monomer-dimer tilings of an $n\times n$ square grid, with $m<n$ monomers in which no four tiles meet at any point is $m2^m+(m+1)2^{m+1}$, when $m$ and $n$ have the same parity. In addition, we present a new proof of the result that there are $n2^{n-1}$ such tilings with $n$ monomers, which divides the tilings into $n$ classes of size $2^{n-1}$. The sum of these tilings over all monomer counts has the closed form $2^{n-1}(3n-4)+2$ and, curiously, this is equal to the sum of the squares of all parts in all compositions of $n$. We also describe two algorithms and a Gray code ordering for generating the $n2^{n-1}$ tilings with $n$ monomers, which are both based on our new proof.

💡 Research Summary

The paper studies tatami tilings of an n × n square grid using monomers (1 × 1 squares) and dimers (1 × 2 or 2 × 1 rectangles) under the “tatami condition”: no four tiles may meet at a single point. The authors first introduce the T‑diagram, a structural representation that classifies four possible local features (loner, vee, bidimer, vortex) and the rays they emit to the boundary. This diagram captures the entire tiling once a single feature is fixed, because the rays uniquely determine the placement of all other tiles. The remaining undetermined part consists of independent “diagonals”, each a chain of alternating monomers and dimers that can be flipped in place without violating the tatami condition.

Key lemmas establish that for any tiling with fewer than n monomers (m < n) there is exactly one non‑trivial feature—a bidimer or a vortex—and that all other features are loners. Moreover, each diagonal can be flipped independently, and the set of possible flips is in one‑to‑one correspondence with binary strings of length equal to the number of diagonals. The distance yₓ from the feature to the nearest boundary determines both the total number of monomers and the number of diagonals:

• If the feature is a bidimer, then m = n − 2yₓ and the number of diagonals is n − 2yₓ − 2 (or n − 2yₓ − 1 when the feature lies on the main diagonal).
• If the feature is a vortex, then m = n − yₓ + 1 with the same diagonal count.

The feature can occupy any of the 4(n − 2k) positions at distance k = yₓ from the boundary. For each position the number of admissible diagonal flips is 2^{d}, where d is the diagonal count given above. Summing over all positions yields the closed formula

 T(n,m) = m·2^{m} + (m + 1)·2^{m+1},

valid when n and m have the same parity and m < n. When m = n there is no non‑trivial feature; all tiles belong to diagonals, giving T(n,n) = n·2^{n‑1}. The authors provide a new, symmetric proof that the n·2^{n‑1} tilings split into n classes of size 2^{n‑1}, answering a question raised by Knuth.

Summing T(n,m) over all admissible m gives

 ∑_{m} T(n,m) = 2^{n‑1}(3n − 4) + 2,

which coincides with the sum of the squares of the parts in all compositions of n, revealing a surprising link between tatami tilings and integer composition statistics.

The paper also presents two algorithms for generating the n·2^{n‑1} tilings with n monomers. The first treats each diagonal flip as a bit in a binary string and enumerates all 2^{n‑1} strings directly, achieving O(1) time per tiling. The second orders the tilings using a Gray code based on the “diagonal flip” operation: tilings are grouped into n classes, each class is traversed by a reflected binary Gray code on the corresponding bit strings, and adjacent tilings differ by exactly one diagonal flip. This yields constant‑time generation with minimal change between successive outputs and uses only O(n) auxiliary space.

Overall, the work delivers (1) an exact enumeration formula for monomer‑dimer tatami tilings with any admissible number of monomers, (2) a symmetric class decomposition for the maximal‑monomer case, (3) a novel combinatorial identity linking tilings to composition squares, and (4) efficient generation methods. These contributions deepen the combinatorial understanding of tatami tilings and provide practical tools for applications in algorithmic generation, statistical physics models, and educational demonstrations.