Hitomezashi Loops: Theory & Applications

• Hitomezashi loops are closed cycles formed from alternating horizontal and vertical stitches on a square lattice, defined by bi-infinite binary sequences.
• They exhibit rigid arithmetic invariants, including odd width and height, with length and area satisfying specific congruences, often analyzed via Dyck-path encodings.
• Advanced methods such as slicing, excursion induction, and symbolic duality extend their study to cylindrical and toroidal settings, revealing deeper combinatorial and topological insights.

Hitomezashi loops are the closed connected components, or equivalently cycles, arising in hitomezashi patterns: alternating arrangements of horizontal and vertical stitches on a lattice determined by binary data assigned to rows and columns. In the classical square-grid setting, they form one of the central mathematical objects extracted from traditional Japanese embroidery, and they satisfy unexpectedly rigid parity and congruence laws. Subsequent work has recast them in graph-theoretic, combinatorial, probabilistic, topological, and symbolic terms, and has extended the notion to cylindrical and toroidal settings where winding number and homology become intrinsic (Defant et al., 2022).

1. Classical square-grid definition

The standard ambient graph is the square lattice graph $\Cloth_{\mathbb Z}$ on vertex set $\mathbb Z\times \mathbb Z$, where each $(i,j)$ is adjacent to its four lattice neighbors. Given two bi-infinite binary sequences $\epsilon,\eta\in\{0,1\}^{\mathbb Z}$, the associated hitomezashi pattern is the subgraph with horizontal edges

$\{\{(i,j),(i+1,j)\}: i\equiv \eta_j \pmod 2\}$

and vertical edges

$\{\{(i,j),(i,j+1)\}: j\equiv \epsilon_i \pmod 2\}.$

Thus, on each horizontal or vertical grid line, only every other unit edge is present, with the parity determined by the corresponding row or column label.

In the infinite square-grid model, every lattice point is incident to exactly two stitches, one vertical and one horizontal. Consequently there are no branch points, and the stitched graph decomposes into strands. A hitomezashi path is a path made from stitches, a strand is a maximal hitomezashi path, and every strand is either a loop or a bi-infinite path. A hitomezashi loop is therefore a cycle in the pattern; equivalently, in the geometric formulation used elsewhere in the literature, it is a bounded connected component of the union of the stitches. These strands partition the plane into regions, and the bounded regions are bounded by loops (Defant et al., 2022).

The standard loop parameters are $\length(L)$, the number of stitches in a loop $L$; $\area(L)$, the Euclidean area of its interior; and $\width(L)$ and $\mathbb Z\times \mathbb Z$0, the horizontal and vertical spans. Closely related terminology uses longitude and latitude of a stitch, west/east-extremal vertical stitches, north/south-extremal horizontal stitches, and extremal latitudes or longitudes. This geometric vocabulary is central in the extremal and inductive arguments developed later (Defant et al., 2022).

The same loop objects appear in Pete’s work on corner percolation. That identification is structurally important: it places hitomezashi loops at the intersection of lattice graph theory, planar topology, Dyck-path combinatorics, and probabilistic models of random tilings and regions (Defant et al., 2022).

2. Arithmetic and geometric invariants

The first layer of structure is parity. A basic lemma states that on an oriented hitomezashi path, all horizontal stitches at a fixed latitude point in the same direction, and all vertical stitches at a fixed longitude point in the same direction. From this, vertical and horizontal cross-sections of a loop acquire odd spacing properties. In particular, Pete’s theorem implies that every hitomezashi loop has odd width and odd height: $\mathbb Z\times \mathbb Z$1

The principal arithmetic invariants proved for planar loops are sharper: $\mathbb Z\times \mathbb Z$2 These congruences are exact. They strengthen the observation that lattice-loop lengths are even by showing that hitomezashi loops exclude the residue classes $\mathbb Z\times \mathbb Z$3, and they force the enclosed area into the single residue class $\mathbb Z\times \mathbb Z$4 (Defant et al., 2022).

The width and height statement is explained combinatorially by Pete’s bijection between loops modulo translation and pairs of Dyck paths of the same height. If $\mathbb Z\times \mathbb Z$5 has width $\mathbb Z\times \mathbb Z$6 and height $\mathbb Z\times \mathbb Z$7, then the associated Dyck paths have semilengths

$\mathbb Z\times \mathbb Z$8

so $\mathbb Z\times \mathbb Z$9 and $(i,j)$0 must be odd integers. The length and area congruences are subtler and require additional inductive or slicing arguments rather than Dyck-path semilength alone (Defant et al., 2022).

These constraints are sharp. For odd $(i,j)$1, the rug-shaped loop $(i,j)$2 satisfies

$(i,j)$3

$(i,j)$4

$(i,j)$5

These examples show that all widths, heights, lengths, and areas not forbidden by the congruence theorems actually occur. In this sense, odd width and height are the only parity restrictions on dimensions, and $(i,j)$6 and $(i,j)$7 are the only congruence restrictions on perimeter-length and area (Defant et al., 2022).

3. Combinatorial encodings and proof methods

One major proof architecture is Pete’s corner-percolation encoding, presented in the hitomezashi literature as a bijection from loops modulo translation to pairs of Dyck paths of the same height. For a clockwise-oriented loop $(i,j)$8, one reads longitudes intersecting $(i,j)$9 from left to right and latitudes intersecting $\epsilon,\eta\in\{0,1\}^{\mathbb Z}$0 from bottom to top, classifying them by the directions of the vertical or horizontal stitches. After deleting first and last steps, one obtains paths $\epsilon,\eta\in\{0,1\}^{\mathbb Z}$1 and $\epsilon,\eta\in\{0,1\}^{\mathbb Z}$2. This bijection is the source of the odd-width and odd-height theorem and serves as the combinatorial benchmark for later generalizations (Defant et al., 2022).

A second proof architecture is the slicing theory developed for planar loops. Vertical slicing $\epsilon,\eta\in\{0,1\}^{\mathbb Z}$3 and horizontal slicing $\epsilon,\eta\in\{0,1\}^{\mathbb Z}$4 are defined when adjacent grid lines have matching labels, so that a deleted strip can be closed without breaking the alternating structure. To analyze slicing combinatorially, the theory introduces local moves, splits, splices, and square deletions, together with pseudo-hitomezashi paths and the agreeable/anti-agreeable dichotomy. Outdent longitudes are especially tractable: slicing there decomposes a loop into smaller non-nested loops. The length proof then proceeds by induction on width through an exact identity,

$\epsilon,\eta\in\{0,1\}^{\mathbb Z}$5

where $\epsilon,\eta\in\{0,1\}^{\mathbb Z}$6 are the intertwined pre-slice loops, $\epsilon,\eta\in\{0,1\}^{\mathbb Z}$7 are the post-slice loops, $\epsilon,\eta\in\{0,1\}^{\mathbb Z}$8 is the number of local moves, and $\epsilon,\eta\in\{0,1\}^{\mathbb Z}$9 is the number of horizontal stitches at the slicing longitude (Defant et al., 2022).

Ren and Zhang later replaced this loop-based induction by a shorter self-contained proof built around a new object, the $\{\{(i,j),(i+1,j)\}: i\equiv \eta_j \pmod 2\}$0-excursion. An $\{\{(i,j),(i+1,j)\}: i\equiv \eta_j \pmod 2\}$1-excursion is a hitomezashi path with at least three vertices whose start and end vertices lie on the vertical line $\{\{(i,j),(i+1,j)\}: i\equiv \eta_j \pmod 2\}$2, while all other vertices lie in the half-plane $\{\{(i,j),(i+1,j)\}: i\equiv \eta_j \pmod 2\}$3. Their key strengthening is the congruence

$\{\{(i,j),(i+1,j)\}: i\equiv \eta_j \pmod 2\}$4

for an $\{\{(i,j),(i+1,j)\}: i\equiv \eta_j \pmod 2\}$5-excursion from $\{\{(i,j),(i+1,j)\}: i\equiv \eta_j \pmod 2\}$6 to $\{\{(i,j),(i+1,j)\}: i\equiv \eta_j \pmod 2\}$7. The loop theorem then follows by cutting a loop along its leftmost vertical line and decomposing it into excursions farther to the right. The proof is driven by three ingredients: parity of starting vertices controls edge direction, all edges of a path on a fixed vertical line have the same direction, and a planar noncrossing lemma forces a rigid order on the $\{\{(i,j),(i+1,j)\}: i\equiv \eta_j \pmod 2\}$8-coordinates encountered along the cut line (Ren et al., 2023).

Taken together, these methods show that hitomezashi loop theory admits at least three complementary languages: Dyck-path encoding, slicing with local moves, and excursion induction. The recurrence of parity, monotonicity, and planar noninterleaving across all three methods suggests that the congruence laws are manifestations of a common lattice-topological mechanism rather than artifacts of a particular proof strategy.

4. Extremal geometry and long-stitch rigidity

For ordinary square-grid loops of prescribed odd width $\{\{(i,j),(i+1,j)\}: i\equiv \eta_j \pmod 2\}$9 and odd height $\{\{(i,j),(i,j+1)\}: j\equiv \epsilon_i \pmod 2\}.$0, the literature gives sharp extremal results. The minimum possible length satisfies

$\{\{(i,j),(i,j+1)\}: j\equiv \epsilon_i \pmod 2\}.$1

If $\{\{(i,j),(i,j+1)\}: j\equiv \epsilon_i \pmod 2\}.$2 (respectively $\{\{(i,j),(i,j+1)\}: j\equiv \epsilon_i \pmod 2\}.$3), equality holds if and only if each horizontal (respectively vertical) stitch in $\{\{(i,j),(i,j+1)\}: j\equiv \epsilon_i \pmod 2\}.$4 has the same longitude (respectively latitude) as exactly one other stitch in $\{\{(i,j),(i,j+1)\}: j\equiv \epsilon_i \pmod 2\}.$5. Loops achieving equality correspond to Dyck paths of semilength $\{\{(i,j),(i,j+1)\}: j\equiv \epsilon_i \pmod 2\}.$6 (Defant et al., 2022).

The minimum possible area satisfies

$\{\{(i,j),(i,j+1)\}: j\equiv \epsilon_i \pmod 2\}.$7

Equality holds if and only if either $\{\{(i,j),(i,j+1)\}: j\equiv \epsilon_i \pmod 2\}.$8 is a cross or $\{\{(i,j),(i,j+1)\}: j\equiv \epsilon_i \pmod 2\}.$9. Since every loop with $\length(L)$0 is a rug, this identifies the minimum-area loops completely. At the opposite extreme, every loop satisfies

$\length(L)$1

with equality if and only if $\length(L)$2 is a rug. Rugs are therefore exactly the maximum-area loops for fixed width and height (Defant et al., 2022).

For length maximization, the non-rug bound is

$\length(L)$3

with equality if and only if $\length(L)$4 is a horizontal or vertical comb or, when $\length(L)$5, a wand. The standard named families have explicit formulas: $\length(L)$6

$\length(L)$7

$\length(L)$8

$\length(L)$9

and a wand of width $L$0 and height $L$1 has length $L$2 (Defant et al., 2022).

The same paper also studies the long-stitch generalization. In an $L$3-hitomezashi pattern, horizontal stitches are $L$4-over-$L$5-under and vertical stitches are $L$6-over-$L$7-under, with the necessary existence condition

$L$8

Here the rigidity is dramatic: the only possible finite loops are 4-stitch rectangles. In the generic case $L$9 and $\area(L)$0, every pattern consists either entirely of rectangles or entirely of zig-zags of a single type. In the nongeneric case $\area(L)$1 or $\area(L)$2, rectangles may coexist with horizontal or vertical accordions, but the only loops remain rectangles. The classical unit-stitch model is thus rich in finite-loop shapes, whereas the long-stitch square-grid model collapses finite loops to a single combinatorial type (Defant et al., 2022).

5. Cylindrical and toroidal loops

On the torus and cylinder, the parity-based planar definition is replaced by an oriented one. For $\area(L)$3 and $\area(L)$4, the toroidal pattern $\area(L)$5 orients every horizontal edge in row $\area(L)$6 according to $\area(L)$7 and every vertical edge in column $\area(L)$8 according to $\area(L)$9. A toroidal hitomezashi loop is then a directed circuit whose edges alternate between horizontal and vertical. If $\width(L)$0 and $\width(L)$1 are the signed horizontal and vertical displacements along the loop, its homology class is

$\width(L)$2

Writing

$\width(L)$3

and similarly for $\width(L)$4, one obtains exact structural information. When $\width(L)$5 and $\width(L)$6, every nontrivial loop has the same primitive homology class

$\width(L)$7

and the number of nontrivial loops is exactly $\width(L)$8. Every trivial toroidal loop has length $\width(L)$9, while a nontrivial loop of homology class $\mathbb Z\times \mathbb Z$00 satisfies

$\mathbb Z\times \mathbb Z$01

In the symmetric case $\mathbb Z\times \mathbb Z$02, nontrivial loops have homology class $\mathbb Z\times \mathbb Z$03 if $\mathbb Z\times \mathbb Z$04, homology class $\mathbb Z\times \mathbb Z$05 if $\mathbb Z\times \mathbb Z$06, and do not occur if $\mathbb Z\times \mathbb Z$07; moreover the total number of loops is congruent to $\mathbb Z\times \mathbb Z$08, and its minimal possible value is $\mathbb Z\times \mathbb Z$09. In that symmetric annular reformulation, hitomezashi loops become Seifert circles of a diagram of the torus link $\mathbb Z\times \mathbb Z$10 (Ren et al., 2023).

The cylindrical theory sharpens the zero-drift case. For $\mathbb Z\times \mathbb Z$11 with $\mathbb Z\times \mathbb Z$12, let $\mathbb Z\times \mathbb Z$13 be the range of cyclic partial sums of $\mathbb Z\times \mathbb Z$14. A contiguous block $\mathbb Z\times \mathbb Z$15 is minimally positively overflowing if $\mathbb Z\times \mathbb Z$16 and no proper interior subblock has that property. The main theorem states that loops of homology $\mathbb Z\times \mathbb Z$17 are in bijection with minimally positively overflowing subsequences, and loops of homology $\mathbb Z\times \mathbb Z$18 are in bijection with minimally negatively overflowing subsequences. If $\mathbb Z\times \mathbb Z$19, no nontrivial cylindrical loops exist. Lifting toroidal patterns to the cylinder yields the final criterion for the previously unresolved case $\mathbb Z\times \mathbb Z$20: if $\mathbb Z\times \mathbb Z$21, loops with homology $\mathbb Z\times \mathbb Z$22 exist; if $\mathbb Z\times \mathbb Z$23, no nontrivial loops exist; if $\mathbb Z\times \mathbb Z$24, loops with homology $\mathbb Z\times \mathbb Z$25 exist. This completes the classification and counting problem left open in earlier toroidal work (Xie, 5 Sep 2025).

These cylindrical and toroidal results change the role of loop data. In the plane, the key bounded object is an enclosing rectangle and the relevant combinatorics is encoded by equal-height excursions or Dyck paths. On the cylinder, the decisive datum is a minimally overflowing subsequence relative to the range $\mathbb Z\times \mathbb Z$26. On the torus, global imbalance and homology replace enclosing width and height as the primary invariants.

6. Encodings, duality, friezes, and non-square grids

Several papers treat loops through symbolic encodings rather than direct structural classification. One square-grid specification uses two binary words $\mathbb Z\times \mathbb Z$27 and $\mathbb Z\times \mathbb Z$28, repeated periodically to encode vertical and horizontal lines of stitching. Duality is defined by bit-complement,

$\mathbb Z\times \mathbb Z$29

reflecting the fact that the reverse side of running stitch is the complementary pattern. In this framework, traditional motifs such as kuchizashi, jūjizashi, kakinohanazashi, sanjū kakinohanazashi, and igetazashi are analyzed through their largest loops, and the universal loop constraints

$\mathbb Z\times \mathbb Z$30

are imported explicitly. The same paper introduces the Pell persimmon family

$\mathbb Z\times \mathbb Z$31

where the Pell words $\mathbb Z\times \mathbb Z$32 satisfy

$\mathbb Z\times \mathbb Z$33

and states the Persimmon-Snowflake Conjecture: the largest polyomino in the order-$\mathbb Z\times \mathbb Z$34 Pell persimmon pattern is the Fibonacci snowflake of order $\mathbb Z\times \mathbb Z$35. The paper is explicit that it does not give a necessary-and-sufficient condition on $\mathbb Z\times \mathbb Z$36 for loop existence or enumeration (Seaton et al., 2022).

A different symbolic approach models a hitomezashi strip as a two-sided frieze encoded by a periodic binary word $\mathbb Z\times \mathbb Z$37 for the vertical lines and a finite binary word $\mathbb Z\times \mathbb Z$38 for the horizontal lines, with the reverse side forced to be the complementary pattern $\mathbb Z\times \mathbb Z$39. The paper does not use the exact phrase “Hitomezashi loops,” but it observes that at each interior grid vertex one horizontal stitch and one vertical stitch form a right-angle corner, so every interior stitched vertex has degree $\mathbb Z\times \mathbb Z$40. It follows that each connected component is forced to be either a closed cycle or an open curve meeting the top and/or bottom boundary of the strip. The paper does not formulate a theorem in these topological terms; its main theorem is instead a symmetry classification of two-sided frieze groups, proving that 13 of the 31 groups are realizable in hitomezashi and 18 are impossible (Seaton, 30 Jan 2026).

The isometric-grid variant introduces three stitch directions rather than two. In the dilute version, only alternate lines of stitching in each direction are present, each vertex visited by the stitching has degree two, and one quarter of the vertices are empty. The paper presents periodic examples with wallpaper symmetries p6mm and p3m1 and a recursive word construction

$\mathbb Z\times \mathbb Z$41

used in all three directions as $\mathbb Z\times \mathbb Z$42. The resulting patterns appear empirically to contain the outlines of the von Koch snowflake iterates, together with lower-order iterates and flower-like motifs. The paper is explicit that this is a preliminary investigation: it gives the degree-two local mechanism that organizes the stitched graph into loop components, but it does not prove a general loop-classification theorem or an all-order proof of the snowflake construction (Seaton, 28 Feb 2025).

Across these variants, the notion of a hitomezashi loop is stable at the local level—alternating degree-two stitched geometry—but not at the global one. In the classical planar square-grid literature it means a cycle or bounded connected component; in cylindrical and toroidal work it becomes a directed circuit with a homology class; in frieze and isometric-grid settings the loop interpretation is sometimes explicit and sometimes inferred from degree-two local structure. That distinction is not merely terminological: it tracks the shift from planar enclosure to periodic winding, from parity-based coordinates to symbolic encodings, and from exact congruence theorems to exploratory symmetry and construction problems.