Approximate Self-Assembly of the Sierpinski Triangle

The Tile Assembly Model is a Turing universal model that Winfree introduced in order to study the nanoscale self-assembly of complex (typically aperiodic) DNA crystals. Winfree exhibited a self-assembly that tiles the first quadrant of the Cartesian plane with specially labeled tiles appearing at exactly the positions of points in the Sierpinski triangle. More recently, Lathrop, Lutz, and Summers proved that the Sierpinski triangle cannot self-assemble in the “strict” sense in which tiles are not allowed to appear at positions outside the target structure. Here we investigate the strict self-assembly of sets that approximate the Sierpinski triangle. We show that every set that does strictly self-assemble disagrees with the Sierpinski triangle on a set with fractal dimension at least that of the Sierpinski triangle (roughly 1.585), and that no subset of the Sierpinski triangle with fractal dimension greater than 1 strictly self-assembles. We show that our bounds are tight, even when restricted to supersets of the Sierpinski triangle, by presenting a strict self-assembly that adds communication fibers to the fractal structure without disturbing it. To verify this strict self-assembly we develop a generalization of the local determinism method of Soloveichik and Winfree.

💡 Research Summary

The paper investigates the limits of strictly self‑assembling the Sierpinski triangle within the Tile Assembly Model (TAM), a framework introduced by Winfree to study DNA‑based nanoscale construction. Winfree’s original construction tiles the first quadrant so that tiles appear exactly at the points of the Sierpinski triangle, but later work by Lathrop, Lutz, and Summers proved that a “strict” self‑assembly—one that forbids any tile outside the target shape—is impossible for the exact triangle.

Motivated by this impossibility, the authors ask how closely a set can approximate the triangle while still being strictly self‑assembling, and whether any non‑trivial subset of the triangle can be assembled under the strict rule. To answer these questions they introduce the Hausdorff (fractal) dimension as a quantitative measure of the size of the disagreement between the assembled set and the ideal fractal.

The first major result is a lower‑bound theorem: for any set A that strictly self‑assembles, the symmetric difference A Δ S with the Sierpinski triangle S must have Hausdorff dimension at least dim_H(S) ≈ 1.585. In other words, any strictly self‑assembled approximation must differ from the true triangle on a set that is itself as “large” (in fractal terms) as the triangle. Consequently, any subset B ⊆ S whose Hausdorff dimension exceeds 1 cannot be strictly self‑assembled. This establishes a sharp negative bound: only subsets of dimension ≤ 1 (essentially linear pieces) could possibly be assembled without extraneous tiles.

The second contribution is an upper‑bound construction that shows the lower bound is tight. The authors augment the Sierpinski triangle with a system of “communication fibers”—thin strips of tiles that run between the recursive copies of the triangle. These fibers carry information needed to enforce deterministic growth while never altering the underlying fractal pattern. The resulting superset still has Hausdorff dimension dim_H(S) and can be strictly self‑assembled.

To verify the correctness of this construction the authors extend the local determinism method of Soloveichik and Winfree. The original method guarantees that each tile’s placement is uniquely forced by its neighbours, ensuring error‑free growth. However, the added fibers introduce new adjacency relations that the classic criteria do not cover. The paper therefore defines an “extended local determinism” condition: (a) fiber tiles are uniquely determined by the surrounding triangle tiles, and (b) triangle tiles remain uniquely determined regardless of the presence of fibers. By proving that the augmented system satisfies these extended conditions, the authors formally certify the strict self‑assembly of their construction.

Overall, the work delivers a two‑fold insight. First, it quantifies the unavoidable “noise” in any strict self‑assembly of a fractal: the disagreement set must be at least as complex as the fractal itself, ruling out strict assembly of any high‑dimensional subset. Second, it demonstrates that by allowing a controlled, dimension‑preserving superset—implemented via communication fibers—strict self‑assembly becomes feasible. This bridges a gap between theoretical impossibility and practical design, suggesting that error‑correcting or signaling structures can be embedded in DNA tile systems without compromising the target fractal geometry.

The paper concludes with several directions for future research: extending the dimension‑based lower bound to other classic fractals (Cantor set, Koch curve), designing more efficient auxiliary structures that replace communication fibers, and experimentally validating the proposed constructions with actual DNA tiles. By linking fractal geometry, computational self‑assembly, and nanotechnological implementation, the study advances our understanding of what complex patterns can be reliably built at the molecular scale.