Square Cover: Concepts & Applications

• Square Cover is a framework defining diverse covering problems over square geometries, including random walk cover times, geometric set cover, and cyclic tilings.
• Researchers employ analytic, dynamic, and algorithmic methods to derive approximations, exact algorithms, and NP-completeness results within various square cover contexts.
• A unifying theme is the exploitation of square-based motifs to optimize coverage, whether in graph structures, polygon decompositions, or folded and tiled surfaces.

“Square Cover” is not a single technical notion. In the contemporary literature it denotes several distinct but structurally related objects: cover time on the regular square lattice and on Cartesian squares of graphs; geometric set cover and hitting set with axis-aligned squares; exact and approximate covering of segments, polygons, rectangles, or a square itself by squares; cover factors for folded squares; and “square-tiled cyclic covers” in Teichmüller dynamics. The common motif is that a square may serve as a graph geometry, a covering range, a target to be covered, or a tile arising from an algebraic covering construction.

1. Terminological scope

The main usages can be organized as follows.

Usage	Core object	Representative statement
Cover time on square lattice	Simple random walk on an $L\times L$ torus	$\tau(G_N)$ is compared to $(1/\pi)N(\ln N)^2$
Square of a graph	Cartesian product $G \square G$	$C(G \square G)$ is bounded in terms of $C(G)$
Geometric square set cover	Points and axis-aligned squares in $\mathbb R^2$	Dynamic $O(1)$-approximation with sublinear or subpolynomial updates
Covering by squares	Segments, polygons, rectangles, or a square covered by squares	Exact, approximation, and NP-completeness results
Square-tiled cyclic cover	Branched cyclic cover of $\mathbb P^1$ tiled by unit squares	The cover is tiled by $2N$ unit squares

In probabilistic graph theory, the square lattice appears as a regular degree-$\tau(G_N)$0 geometry and as the Cartesian square $\tau(G_N)$1. In computational geometry and approximation algorithms, “square cover” typically means selecting axis-aligned squares to cover points, segment endpoints, or polygonal regions. In flat-surface dynamics, the phrase may denote a square-tiled cyclic cover, namely a cyclic cover of the sphere endowed with a square tiling by pullback of a quadratic differential. These usages are all explicit in the literature, but they are not interchangeable (Mendonça, 2011, Abdullah, 2012, Agarwal et al., 2020, Forni et al., 2010, Dósa et al., 23 Jan 2026, Dhar et al., 2024).

2. Cover time on the regular square lattice and on Cartesian squares

For a finite, connected graph $\tau(G_N)$2 with $\tau(G_N)$3, the cover time $\tau(G_N)$4 is the maximum expected time, over starting vertices, for a simple random walk to visit every vertex at least once. On planar graphs, the lower-bound form investigated numerically is

$\tau(G_N)$5

and for the regular square lattice, where $\tau(G_N)$6,

$\tau(G_N)$7

The simulations were performed on regular square lattices with $\tau(G_N)$8 vertices under periodic boundary conditions, with $\tau(G_N)$9 ranging from $(1/\pi)N(\ln N)^2$0 to $(1/\pi)N(\ln N)^2$1, and each $(1/\pi)N(\ln N)^2$2 estimated as the sample mean over $(1/\pi)N(\ln N)^2$3 Monte Carlo runs. The square lattice is described as “borderline”: the data respect the lower bound in the simulated range, the ratio $(1/\pi)N(\ln N)^2$4 with $(1/\pi)N(\ln N)^2$5 is near $(1/\pi)N(\ln N)^2$6, but no direct asymptotic fit for a leading coefficient is reported, nor are explicit error bars or goodness-of-fit statistics given. At $(1/\pi)N(\ln N)^2$7, the empirical cover-time distribution has mean $(1/\pi)N(\ln N)^2$8, standard deviation $(1/\pi)N(\ln N)^2$9, skewness $G \square G$0, and excess kurtosis $G \square G$1; the histogram is fitted exploratorily by a rescaled Beta law with maximum-likelihood parameters $G \square G$2 and $G \square G$3 on the sampled support $G \square G$4 (Mendonça, 2011).

A related but distinct usage of “square” is the Cartesian square $G \square G$5. For graphs $G \square G$6 and $G \square G$7, the Cartesian product has vertex set $G \square G$8, and the “square of a graph” is $G \square G$9, not the distance-$C(G \square G)$0 graph. The product cover-time theorem gives

$C(G \square G)$1

and, if $C(G \square G)$2,

$C(G \square G)$3

where $C(G \square G)$4 and $C(G \square G)$5. For the two-dimensional torus $C(G \square G)$6, this yields

$C(G \square G)$7

which is within a factor of $C(G \square G)$8 of the sharp asymptotic

$C(G \square G)$9

More generally, prior work quoted in the thesis gives $C(G)$0 and $C(G)$1 for $C(G)$2, with $C(G)$3 and $C(G)$4 (Abdullah, 2012).

These two lines of work share the same degree-$C(G)$5 geometry but ask different questions. The Monte Carlo study concerns asymptotic sharpness of a conjectured planar lower bound on the square lattice, whereas the Cartesian-product theory studies how cover times of factors control the cover time of $C(G)$6. A plausible implication is that the square lattice is simultaneously a borderline case for asymptotic constants and the canonical balanced case in product-cover estimates.

3. Dynamic geometric set cover with squares

In computational geometry, square cover usually means geometric set cover or hitting set with axis-aligned squares. For unit-square set cover in $C(G)$7, the input is a point set $C(G)$8 and a family $C(G)$9 of axis-aligned unit squares, with objective

$\mathbb R^2$0

The dynamic model allows insertions and deletions of points and ranges. One framework gives an $\mathbb R^2$1-approximate fully dynamic data structure for unit-square set cover and hitting set with amortized update time $\mathbb R^2$2 for any fixed constant $\mathbb R^2$3, construction time $\mathbb R^2$4, $\mathbb R^2$5-time size queries, $\mathbb R^2$6-time membership queries, and $\mathbb R^2$7-time reporting. A second framework yields $\mathbb R^2$8-approximate partially dynamic structures with amortized update time $\mathbb R^2$9. The reduction from quadrants to unit squares incurs an approximation-factor blow-up of at most $O(1)$0, because every unit square intersects at most four unit grid cells (Agarwal et al., 2020).

A later account records sharper dynamic bounds. It states a data structure for $O(1)$1-approximate dynamic unit-square set cover with $O(1)$2 amortized update time, a data structure for $O(1)$3-approximate dynamic square set cover with $O(1)$4 randomized amortized update time for any fixed constant $O(1)$5, and a weighted unit-square result with $O(1)$6 amortized update time. The same source also emphasizes explicit cover maintenance rather than only reporting the approximate size (Chan et al., 2021).

The technical mechanisms are geometric decompositions rather than generic dynamic-set-cover machinery. The fully dynamic framework uses bootstrapping and an output-sensitive routine for quadrants; the partially dynamic one uses local modification on stable instances, where a single update changes the quasi-optimum by at most $O(1)$7. The later unit-square improvement to $O(1)$8 suggests that the unit-size restriction can be exploited much more aggressively than the arbitrary-size case. By contrast, weighted arbitrary-size square cover remains identified as a more difficult regime.

4. Squares as covering objects: segments, polygons, and rectangles

When squares are the covering shapes, the literature splits according to the target geometry. For line segments in the plane, the paper on “covering segments with unit squares” studies continuous and discrete variants in which an axis-parallel unit square covers a segment if it contains at least one endpoint. It defines continuous covering segments by unit squares (CCSUS), where the squares may be placed anywhere, and discrete covering segments by unit squares (DCSUS), where the candidate squares are part of the input. The results include an exact greedy $O(1)$9 algorithm for horizontal unit segments inside a unit-height strip, a $\mathbb P^1$0-approximation in $\mathbb P^1$1 for horizontal unit segments anywhere, a $\mathbb P^1$2-approximation in $\mathbb P^1$3 and a PTAS with ratio $\mathbb P^1$4 and time $\mathbb P^1$5 for unit horizontal and unit vertical segments, a $\mathbb P^1$6-approximation in $\mathbb P^1$7 for arbitrary segments, and a $\mathbb P^1$8-approximation for the discrete arbitrary case via multilevel LP relaxation. The same source states NP-completeness for CCSUS-H1 and for the arbitrary continuous and discrete variants even when the segments lie on a single line (Acharyya et al., 2016).

For orthogonal polygons without holes, the orthogonal polygon covering with squares (OPCS) problem asks for a minimum-cardinality set of axis-parallel squares whose union is exactly the polygon. The abstract of the 2024 paper reports a polynomial-time exact algorithm with worst-case running time $\mathbb P^1$9 and a knob-parameterized algorithm with runtime $2N$0, where a knob is a polygon edge whose both endpoints are convex. The detailed extraction, however, presents an $2N$1 exact algorithm and an $2N$2 parameterized bound, together with a decomposition formula

$2N$3

for separating maximal corner squares. Both accounts agree on the qualitative content: exact polynomial-time solvability for hole-free orthogonal polygons, improved performance when the number of knobs is small, and NP-completeness of the version with holes (OPCSH). The paper also states that OPCSH remains NP-complete even when square sizes are restricted to $2N$4 (Dhar et al., 2024).

A different exact classification appears for covering a $2N$5 rectangle by equal squares “on both sides in one layer.” The necessary and sufficient condition is

$2N$6

with $2N$7, $2N$8, and $2N$9. The proof uses the induced periodic square tiling, decomposition into strips, projection identities

$\tau(G_N)$00

and the resulting quadratic constraint

$\tau(G_N)$01

This yields a rational-parameter classification of achievable aspect ratios (Ozhegov, 2020).

Across these problems, the square acts as the movable primitive rather than the substrate. Exact solvability is rare and tends to depend on strong geometric structure: strip decompositions, maximal-square recursions, or explicit rational parametrizations. NP-completeness emerges rapidly once the target gains holes, arbitrary orientations, or discrete candidate sets.

5. Squares as targets: covering a square and covering folded squares

Another major use of “square cover” reverses the geometry: the square is the object to be covered. One recent formulation defines

$\tau(G_N)$02

and, for boundary cover,

$\tau(G_N)$03

The exact small-$\tau(G_N)$04 values reported are

$\tau(G_N)$05

$\tau(G_N)$06

where

$\tau(G_N)$07

Hence $\tau(G_N)$08: boundary and interior covering coincide for $\tau(G_N)$09 but diverge at $\tau(G_N)$10. A key lemma analyzes the intersection of a unit square with an $\tau(G_N)$11-shape on the target boundary; if the longer leg has length $\tau(G_N)$12, the maximal other leg is

$\tau(G_N)$13

with $\tau(G_N)$14. The boundary problem also satisfies the recurrence

$\tau(G_N)$15

and therefore

$\tau(G_N)$16

for all $\tau(G_N)$17 (Dósa et al., 23 Jan 2026).

A different target-side notion is the cover factor of a folded square. For a shape $\tau(G_N)$18, the one-fold cover factor $\tau(G_N)$19 is the least scale $\tau(G_N)$20 such that every single fold of $\tau(G_N)$21 is covered by some rigid motion of $\tau(G_N)$22; the origami cover factor $\tau(G_N)$23 allows arbitrary folds. For the unit square,

$\tau(G_N)$24

where $\tau(G_N)$25 is the largest positive real root of a degree-$\tau(G_N)$26 polynomial explicitly given in the source. The hardest single folds are exactly those for which the fold line intersects the relative interiors of two opposite sides. For arbitrary folds the best bounds stated are

$\tau(G_N)$27

The exact equality $\tau(G_N)$28 is left open (Aichholzer et al., 2014).

These two problems are conceptually inverse to set cover by squares. In the congruent-cover problem, unit squares are the coverers and a larger square is the target. In the origami problem, a square is compared against its own folded images via scaling. Both, however, reduce optimality to a small set of extremal geometric configurations: an $\tau(G_N)$29-shape balance in one case, and a finite family of minimum-enclosing-square contact patterns in the other.

6. Square-tiled cyclic covers and unifying themes

In Teichmüller dynamics, a “square cover” may mean a square-tiled cyclic cover. Fix $\tau(G_N)$30 and integers $\tau(G_N)$31 with $\tau(G_N)$32, $\tau(G_N)$33, and $\tau(G_N)$34. The cyclic cover

$\tau(G_N)$35

is a compact Riemann surface $\tau(G_N)$36. Pulling back the quadratic differential

$\tau(G_N)$37

from the “unit square pillow” yields a flat surface tiled by $\tau(G_N)$38 unit squares. The genus is

$\tau(G_N)$39

The pullback differential is a global square of a holomorphic $\tau(G_N)$40-form if and only if $\tau(G_N)$41 is even and all $\tau(G_N)$42 are odd; otherwise it is a genuine quadratic differential. The Veech group contains $\tau(G_N)$43 and has index in $\tau(G_N)$44. Explicit Eskin–Kontsevich–Zorich-type formulas are given for Lyapunov sums, and within the Abelian case the only square-tiled cyclic covers with maximally degenerate Kontsevich–Zorich spectrum are $\tau(G_N)$45 and $\tau(G_N)$46 (Forni et al., 2010).

The breadth of these usages shows that “Square Cover” is best treated as a family of domain-specific constructions rather than a unitary theory. Still, several recurrent patterns are visible. First, extremality is often encoded by a single explicit constant: $\tau(G_N)$47 for square-lattice cover time, $\tau(G_N)$48 for single-fold square cover, $\tau(G_N)$49 and $\tau(G_N)$50 for small-$\tau(G_N)$51 square covering. Second, strip or block decompositions recur across unrelated areas: random-walk cover-time proofs on products use block arguments, polygon-cover algorithms use strips and maximal corner squares, and the rectangle classification is formulated through strip tilings. Third, many open problems remain explicitly stated: limit laws for square-lattice cover times are largely unknown; the exact arbitrary-fold square cover factor $\tau(G_N)$52 is open; the conjecture $\tau(G_N)$53 is posed for covering a square by unit squares; polylogarithmic fully dynamic update time for arbitrary-size square set cover is still open; and lowering the high polynomial exponent in exact orthogonal-polygon covering is identified as a natural direction (Mendonça, 2011, Dósa et al., 23 Jan 2026, Aichholzer et al., 2014, Dhar et al., 2024).