Defect Organization in Coexisting Hexagonal and Square Lattices on Ellipsoids

We used simulated-annealing Langevin dynamics [27], implemented in LAMMPS [28], to search for low-energy configurations of $N=1000$ particles constrained to an ellipsoidal surface. The particles interacted through the Hertzian potential

where $r$ is the center-to-center distance, $\sigma$ is the particle diameter, and $\epsilon$ sets the energy scale [17, 18]. Reduced units were defined by $\sigma$, $\epsilon$, and the particle mass $m$, giving the time unit $\tau=\sqrt{m\sigma^{2}/\epsilon}$. The integration time step was $\Delta t=0.5\tau$.

Independent initial configurations were generated on the prescribed ellipsoidal surface at $k_{\rm B}T_{0}/\epsilon=10^{-2}$. The surface constraint was enforced using the RATTLE algorithm [29, 30], with a convergence tolerance of $10^{-4}$ and a maximum of 20 iterations per time step. Each initial configuration was equilibrated for $10^{6}$ steps at $T_{0}$. The temperature was subsequently reduced according to

with $10^{4}$ integration steps performed at each temperature. The Langevin damping time was $10\Delta t$. Each annealing trajectory was followed by a zero-temperature energy minimization.

For each state point, we performed 20 independent annealing trajectories and ranked the resulting configurations by their final potential energies. The lowest-energy configuration was used in the morphology diagram, whereas the ten lowest-energy configurations from the same set of 20 trajectories were retained for the latitude-resolved statistical analysis.

These selected configurations arise from independent annealing trajectories rather than from consecutive frames of a single trajectory.

The ellipsoidal surface was parameterized as

where $A$ and $B$ are the polar and equatorial semiaxes, respectively, $0\leq\theta\leq\pi$ is the parametric colatitude, and $0\leq\varphi<2\pi$. Thus,

The cases $A/B>1$, $A/B=1$, and $A/B<1$ correspond to prolate ellipsoids, a sphere, and oblate ellipsoids, respectively.

The dimensionless surface density was

where $S$ is the total ellipsoidal area. For each prescribed pair of $\rho^{*}$ and $A/B$, the semiaxes were rescaled such that $S=N\sigma^{2}/\rho^{*}$. The spheroidal area was evaluated analytically as

For the above parameterization, the Gaussian curvature is

and the surface-area element integrated over the azimuthal angle is

These expressions were used both to calculate the curvature profile and to determine the exact area of each latitude bin [31].

Local structures were identified from the bond network between neighboring particles. Two particles were defined as bonded when their center-to-center separation $r_{ij}$ satisfied $r_{ij}<\sigma$. Closed loops in the resulting bond network were classified by their number of boundary bonds: a loop bounded by three bonds was identified as a triangular cell, whereas a loop bounded by four bonds was identified as a quadrilateral cell. These cells were then used to distinguish the local Hex and Sq environments.

Particles surrounded by both types of cells were assigned to the Hex-Sq interface.

For particle $j$, let $n_{j}^{(3)}$ and $n_{j}^{(4)}$ denote the numbers of triangular and quadrilateral cells incident on it, respectively. Its winding number was calculated as

In a purely Hex environment, this expression reduces to $w_{j}=1-Z_{j}/6$, where $Z_{j}$ is the number of bonded neighbors. In a purely Sq environment, it reduces to $w_{j}=1-Z_{j}/4$. As a consistency check, the identified defects satisfy the Euler constraint

on the closed ellipsoidal surface.

Here the winding number is the dimensionless local topological charge. The corresponding angular deficit is

For example, fivefold and sevenfold sites in a purely Hex environment carry winding numbers $+1/6$ and $-1/6$, respectively.

The surface was divided into nonoverlapping latitude bins of width $\Delta\theta=5^{\circ}$, spanning $\theta=0$ to $\theta=\pi$, with each bin labeled by its central latitude $\theta$. The area associated with each latitude bin was

and the corresponding area-averaged Gaussian curvature was

The signed topological-charge surface density was defined as

where the sum runs over particles within the corresponding latitude bin. For each latitude bin, positive and negative winding numbers were summed algebraically, so negative winding numbers reduce $\rho(\theta)$. Consequently, $\rho(\theta)$ is a signed topological-charge density rather than a positive-defect number density. The normalized signed topological-charge density and curvature profiles were then defined as

where

and positive and negative winding numbers enter the normalization algebraically. Because the latitude bins cover the complete closed surface, $W=\sum_{j}w_{j}=2$, consistent with the Euler constraint in Eq. 10.

The mismatch between the two profiles was measured using

where $r_{\rho K}$ is the Pearson correlation coefficient between the 36 latitude-bin values of $\rho(\theta)$ and $K(\theta)$, and $\sigma_{K}$ is the unweighted standard deviation of the corresponding 36 values of $K(\theta)$. The correlation coefficient and $D_{\rho K}$ were calculated separately for each realization and then averaged over the ten lowest-energy configurations selected from the same 20 independent annealing trajectories. Error bars and shaded regions represent the standard error of the mean.

Particle configurations were visualized using OVITO [32].

Figure 1 summarizes the lowest-energy configurations found over the $(\rho^{*},A/B)$ parameter space. At each state point, we performed 20 independent annealing runs and selected the configuration with the lowest final energy. We used $\Delta\rho^{*}=0.01$, aspect ratios $A/B=1.1,\,1.2,\ldots,2.0$ for prolate ellipsoids, and the reciprocal series $A/B=1/2.0,\,1/1.9,\ldots,1/1.1$ for oblate ellipsoids.

The diagram contains three broad regimes. Lower densities generally favor Hex-dominant configurations, higher densities favor Sq-dominant configurations, and intermediate morphologies occur near the crossover between them. The precise boundaries depend on the aspect ratio $A/B$. The mixed states have lower global symmetry and stronger phase separation than the states on either side.

Ellipsoidal confinement preserves the main spherical motifs but changes their placement and connectivity. The Hex-dominant states include (i) Hex scars [Fig. 2], (ii) scar-domain coexistence [Fig. 7], (iii) Sq-rich rectangular domains [Fig. 4], (iv) domain–counter-domain coexistence [Fig. 9], (v) counter domains [Fig. 8], (vi) vertex-contacted domains [Fig. 6], and (vii) biangular domains [Fig. 10]. The Sq-dominant states include (ix) Hex-rich triangular domains [Fig. 11], (x) bridged states [Fig. 13], and (xi) linear and open scars [Fig. 14]. The following sections focus on the geometric rules that organize these motifs. We omit the disordered (viii) asymmetric mixed state, characterized by maze-like intertwining without global symmetry[16], because it lacks a well-defined topological motif for systematic classification.

Hex scars.—Figure 2 shows the Hex-scar states at $\rho^{*}=2.10$. In a Hex-dominant background, we use “Hex scar” to denote a grain-boundary chain composed mainly of alternating five- and sevenfold coordinated sites and carrying a net positive winding number. On a sphere, the twelve net-positive scar complexes adopt an approximately icosahedral arrangement [Fig. 2(a)] [10].

Deforming the sphere breaks this global symmetry. On prolate ellipsoids, the scars form a helical sequence around the long axis. Increasing $A/B$ stretches this sequence but does not change its basic organization. On weakly oblate ellipsoids, the helical pattern remains. For stronger oblateness, $A/B\leq 1/1.4$, the scars form two nearly symmetric rings on opposite sides of the equator. Their projection along the polar axis resembles the twelve marks of a clock [upper right inset of Fig. 2(g)].

The helical and ring-like patterns have a common origin. Rotational symmetry makes all longitudes equivalent, so defect repulsion favors azimuthal separation. Gaussian curvature varies with latitude and biases the positive charge toward the curvature maxima. Combining azimuthal separation with curvature-selected latitudes produces a helix on a prolate ellipsoid and rings near the equator of an oblate ellipsoid.

Figure 3 quantifies this relation using the latitude-resolved profiles defined in Eq. 14. We compare the normalized signed topological-charge density, $\widetilde{\rho}(\theta)=\rho(\theta)/\langle\rho\rangle$, with the normalized Gaussian curvature, $\widetilde{K}(\theta)=K(\theta)/\langle K\rangle$, in latitude-resolved surface bins. Normalization by the respective surface averages allows the comparison to probe spatial co-localization rather than absolute magnitude.

For weakly deformed prolate ellipsoids, the high-curvature polar regions remain sufficiently broad to accommodate both curvature-driven scar localization and scar-scar separation. The two profiles are therefore similar. As $A/B$ increases, the Gaussian curvature becomes concentrated within progressively smaller polar caps. Although these caps attract the net-positive scar complexes, confining all scars within them would increase crowding and mutual elastic repulsion. Therefore, some scars are found near the relatively low-curvature equator, causing the two profiles to separate rapidly for $A/B\gtrsim 1.6$. We quantify this mismatch using $D_{\rho K}=(1-r_{\rho K})\sigma_{K}/[2\langle K\rangle]$, where $r_{\rho K}$ is the Pearson correlation coefficient between the two profiles and $\sigma_{K}$ is the standard deviation of $K(\theta)$.

On oblate ellipsoids, the high-curvature region forms an extended equatorial belt rather than two compact caps. This belt provides sufficient azimuthal space for scars to separate while remaining close to the curvature maximum. Consequently, $D_{\rho K}$ remains small throughout the oblate series. Thus, topology fixes the net charge but not its local distribution; the observed scar profile reflects competition among curvature-driven localization, defect repulsion, and the available area of the high-curvature region.

Sq-rich rectangular domains.—Near the Hex-Sq transition, Sq-rich rectangular domains form within the Hex-dominant background [16]. Each of the four domain vertices carries charge $+1/12$, giving a total charge of $+1/3$ per domain. On the sphere, six such domains form an approximately octahedral arrangement [Fig. 4(a)].

On prolate ellipsoids, the six domains form a helix around the long axis [Figs. 4(b)–(d)]. Curvature draws the domains toward the poles, while their finite size and mutual repulsion maintain azimuthal separation. Increasing $A/B$ stretches the helix and intensifies crowding within the polar caps.

Oblate ellipsoids instead support a single domain ring [Figs. 4(e)–(g)]. As the ellipsoid flattens, the domain ring progressively approaches the equator and, for $A/B\lesssim 1/1.9$, becomes fully confined to the equatorial region [Fig. 4(g)]. The extended high-curvature belt provides enough azimuthal space for all six domains to remain separate.

The latitude-resolved profiles in Fig. 5 show that $\widetilde{\rho}(\theta)$ increasingly departs from $\widetilde{K}(\theta)$ once $A/B$ exceeds approximately 1.3. As the prolate ellipsoid becomes more elongated, Gaussian curvature becomes increasingly localized within the polar caps, while the finite-sized domains cannot remain confined to these narrow high-curvature regions without substantial crowding. The positive defects therefore spread progressively toward lower-curvature latitudes. With further increasing $A/B$, the polar caps eventually become too confined to accommodate the domains while maintaining their separation. At $A/B\geq 1.6$, neighboring domains near the poles begin to contact at their vertices, marking a change in domain connectivity. We refer to these structures as vertex-contacted-domain (VCD) states.

Vertex-contacted domains.—Figure 6 shows representative VCD configurations at $\rho^{*}=2.24$, including the local structures formed at the contacted vertices.

An isolated domain vertex is surrounded by one quadrilateral cell and four triangular cells and carries $w=+1/12$. As highlighted by the red dots in Fig. 6, contact reconstructs the local environment into two quadrilaterals and three triangles, for which $w=0$. Each contact therefore locally neutralizes two $+1/12$ corner charges from the polar region. Because topology fixes the net charge on the closed surface, additional positive defects appear at lower-curvature latitudes, as marked by the black dashed lines in Fig. 6.

Consistent with the VCD reconstruction, the deviation between the signed defect-charge and curvature profiles evolves nonmonotonically with increasing prolateness. For $1.6\leq A/B\leq 1.8$, the occurrence of vertex contacts and compensating defects coincides with a short plateau in the deviation index. This behavior indicates that vertex sharing partially relaxes the curvature-driven localization of the domains, allowing the defect structure to reorganize without placing all positive vertices independently at the polar curvature maxima. When $A/B$ exceeds approximately $1.9$, however, the deviation increases again. Thus, vertex contact delays the separation between the defect and curvature distributions but cannot prevent it once the curvature inhomogeneity becomes sufficiently strong. By contrast, on oblate ellipsoids, the domains remain separated along the high-curvature belt, and the signed defect-charge profile continues to follow the curvature profile more closely.

Coexistence of Hex scars and Sq-rich rectangular domains.—Between the scar-only and domain-only limits, Fig. 7 reveals a coexistence regime in which scars and domains share the required topological charge, with Fig. 7(a) providing a detailed view of the corresponding state-(ii) region in Fig. 1. With increasing $\rho^{*}$, the number of scars decreases while the number of domains increases. Density therefore controls the scar-to-domain conversion, whereas ellipsoidal anisotropy determines the observed placement of the two defect motifs.

This spatial selection is most apparent on prolate ellipsoids. Domains are preferentially trapped by the high-curvature polar regions, while scars are pushed toward lower-curvature latitudes. In the “10 scars + 1 domain” state, which occurs mainly at low $\rho^{*}$ and large $A/B$, one domain occupies one polar region and one scar appears near the opposite pole [Fig. 7(b)]. The remaining nine scars form three latitude bands: one around the equator and two approximately symmetric bands on either side of it. As $\rho^{*}$ increases, the system reaches the most extended coexistence state, “8 scars + 2 domains”. This pattern matches the prolate geometry particularly well: two Sq-rich rectangular domains sit at the two poles, while the eight scars are arranged into two nearly symmetric off-equatorial bands [Fig. 7(c)]. Its large area in the phase diagram indicates that this two-pole domain arrangement provides an especially efficient compromise between curvature localization and finite-size packing.

The intermediate “6 scars + 3 domains” state occupies the smallest region of the phase diagram. It is found mainly under oblate conditions, where three domains are distributed near the equatorial region and the six scars remain more evenly dispersed outside the domains [Fig. 7(d)]. This configuration is less compatible with the dominant ellipsoidal symmetry and therefore appears only in a narrow parameter window. By contrast, the “4 scars + 4 domains” state again occupies a substantial region, mainly at higher $\rho^{*}$ on prolate ellipsoids. In this case, two domains are placed near each polar region, while the four scars form a spiral-like arrangement around the equatorial zone [Fig. 7(e)]. Finally, the “2 scars + 5 domains” state is favored on oblate ellipsoids. The five domains form an approximately pentagonal arrangement around the equatorial belt, highlighted by the dashed guide in Fig. 7(f), while the two residual scars are located in opposite hemispheres. These representative structures show that the coexistence states are not arbitrary mixtures of scars and domains; rather, each pattern reflects a specific compromise among defect charge, curvature localization, and packing constraints.

Counter domains.—We use “counter domain” for a composite motif in which the local lattice symmetry switches from the surrounding background to that of a host domain and then switches back within the host; the complete motif carries total winding number $w=+1/2$ [16]. Four counter domains form an approximately tetrahedral arrangement on the sphere [Fig. 8(a)].

Under ellipsoidal confinement, this tetrahedral organization is deformed by the anisotropic curvature field. In prolate ellipsoids [Fig. 8(b)], the counter-domain pattern is elongated along the major axis, giving rise to a stretched tetrahedral configuration. In oblate ellipsoids [Fig. 8(c)], the same structure is compressed along the symmetry axis and expands preferentially within the equatorial plane. At sufficiently large eccentricity, this confinement becomes strong enough to drive all counter domains onto the equatorial belt, as shown in Fig. 8(d). This behavior reflects the coupling between defect charge and curvature: charged vertices are attracted toward curvature extrema, whereas topological conservation and defect-defect repulsion drive a collective reorganization of the entire structure.

Coexistence of Sq-rich rectangular domains and counter domains.—Curvature inhomogeneity also permits the two domain types to coexist [Fig. 9]. Counter domains preferentially occupy the higher-curvature regions, whereas Sq-rich rectangular domains remain at lower-curvature latitudes. Although a similar spatial arrangement is geometrically possible on a uniformly curved sphere, it is not energetically favored. On the prolate ellipsoid, the curvature gradient stabilizes this spatial separation, allowing the coexistence pattern to emerge as the ground-state structure.

Biangular domains.—A biangular domain extends this nested construction by combining a counter domain with a counter–counter domain, corresponding to a further reversal of local lattice symmetry, giving net winding number $w=+2/3$ [16]. These higher-order structures are also not lowest-energy states on the sphere but appear among the lowest-energy configurations found on strongly oblate ellipsoids. Three biangular domains occupy the high-curvature equatorial belt, while the lower-curvature poles remain largely Hex ordered (Fig. 10).

Hex-rich triangular domains.—The Sq-dominant regime contains eight Hex-rich triangular domains, each carrying a net charge of $+1/4$. On the sphere, their centers occupy the vertices of an inscribed cube [Fig. 11(a)] [20].

Weakly prolate ellipsoids split the domains into two four-domain rings near the poles [Fig. 11(b)]. With increasing $A/B$, these rings merge into a pole-to-pole helical arrangement [Fig. 11(c)]. Weakly oblate ellipsoids likewise exhibit a distorted helical distribution [Fig. 11(d)]. Under stronger flattening, all eight triangular domains migrate toward the equator and form a single ring [Fig. 11(e)]. The top-view inset reveals that these equatorial domains assemble into a compass-rose-like pattern.

Figure 12 provides a quantitative test of curvature selection for Hex-rich triangular domains in the Sq-dominant regime at $\rho^{*}=2.35$. Only the oblate $A/B=1/1.1$ and prolate $A/B=1.1$ and $1.2$ conditions are included because clean configurations of Hex-rich triangular domains occur over a relatively narrow part of the state diagram. The normalized signed topological-charge density is enhanced toward the equator on the oblate ellipsoid and toward the poles on the prolate ellipsoids, consistent with the locations of the corresponding curvature maxima. These signed topological-charge profiles are nevertheless broader and more strongly fluctuating than the curvature profiles because the charge is carried by a finite number of domain-associated point defects. We therefore use these profiles as a consistency check rather than inferring a systematic deformation trend from the limited data for Hex-rich triangular domains.

Bridged states.—A bridged state consists of neighboring Hex-rich triangular domains connected by a parallelogram-shaped Hex bridge. The bridge shares vertices with both domains, changes their relative orientation, and relocates the $+1/12$ point defects from adjacent domain corners to the exposed bridge vertices [20]. The increased separation between these defects reduces their elastic interaction energy. In this limited sense, the bridged state resembles the VCD state, since both use shared vertices to rearrange point defects and lower the elastic energy, although their local topologies remain distinct. In a VCD contact, two $+1/12$ domain-corner charges are converted into locally neutral vertices with $w=0$, with compensating positive charge appearing elsewhere on the surface. In a bridged motif, by contrast, the $+1/12$ defects are not locally neutralized but are relocated from adjacent triangular-domain corners to the exposed bridge vertices. The fully connected spherical network is not retained on an ellipsoid; only partially bridged states occur within a narrow region of the state diagram.

At $A/B\lesssim 1.4$, four paired-domain complexes form an approximately tetrahedral arrangement [Fig. 13(a)]. For $1.5\lesssim A/B\lesssim 1.8$, two triangular domains occupy the poles and three paired complexes are separated azimuthally by about $2\pi/3$ [Fig. 13(b)]. At $A/B>1.9$, each pole is occupied by a triangular domain and surrounded by a ring of three domains and three bridges [Fig. 13(c)]. Oblate confinement instead produces two latitudinal bridging rings rotated by approximately $\pi/4$ [Fig. 13(d)].

Linear and open scars.—At large Sq fraction, a linear scar consists of a sequence of positive and negative winding-number defects with a net charge of $+1/4$ [20]. Despite the shared term, this Sq-background scar is topologically distinct from the five–seven-coordination scar in a Hex-dominant background. As the Hex fraction increases, the chain opens into a Hex-rich corridor bounded by a neutral interface. The morphology changes, but the net charge remains $+1/4$; eight scars therefore supply the required total charge of $+2$ [Fig. 14(a)].

On weakly prolate ellipsoids, the scars trace a pole-to-pole path [Fig. 14(b)]. Stronger prolateness separates the polar open scars and organizes the remaining defects into two hemispherical rings [Fig. 14(c)]. Moderate oblateness retains a distorted pole-to-pole path [Fig. 14(d)], whereas strong flattening produces an equatorial ring [Fig. 14(e)].

Figure 15 compares the same normalized defect-charge and curvature profiles, $\widetilde{\rho}(\theta)$ and $\widetilde{K}(\theta)$, for linear and open scars in the Sq-dominant background. Along the prolate series, the curvature maxima progressively contract into narrow polar caps, whereas the defect charge associated with linear and open scars remains distributed over broader latitude ranges. The resulting defect-curvature mismatch therefore increases overall with prolateness. Along the oblate series, both the curvature and the signed topological-charge density remain concentrated around the extended equatorial belt, and the deviation index remains small. The prolate-oblate contrast is thus common to both lattice backgrounds: compact polar curvature maxima promote defect spreading, whereas an extended equatorial maximum allows defects to separate while remaining curvature-localized.