## research papers

## Crystal

of bicontinuous cubic structures**Lu Han,**

^{a}Nobuhisa Fujita,^{b,}^{c}^{*}Hao Chen,^{d}Chenyu Jin,^{e}Osamu Terasaki^{f,}^{g}^{*}and Shunai Che^{a,}^{h}^{*}^{a}School of Chemical Science and Engineering, Tongji University, Shanghai 200092, People's Republic of China, ^{b}Institute of Multidisciplinary Research for Advanced Materials, Tohoku University, Sendai 980-8577, Japan, ^{c}JST, PRESTO, Saitama 332-0012, Japan, ^{d}Institut für Numerische und Angewandte Mathematik, Georg-August-Universität Göttingen, Lotzestr. 16-18, Göttingen 37083, Germany, ^{e}Max Planck Institute for Dynamics and Self-Organisation, Am Faßberg 17, Göttingen 37077, Germany, ^{f}Centre for High-resolution Electron Microscopy, School of Physical Science and Technology, ShanghaiTech University, Shanghai 201210, People's Republic of China, ^{g}Department of Materials and Environmental Chemistry, Stockholm University, Stockholm S-10691, Sweden, and ^{h}School of Chemistry and Chemical Engineering, Shanghai Jiao Tong University, Shanghai 200240, People's Republic of China^{*}Correspondence e-mail: nobuhisa.fujita.a4@tohoku.ac.jp, oterasaki@shanghaitech.edu.cn, chesa@tongji.edu.cn

Bicontinuous cubic structures in soft matter consist of two intertwining labyrinths separated by a partitioning layer. Combining experiments, numerical modelling and techniques in differential geometry, we investigate

defects in bicontinuous cubic structures. We first demonstrate that a twin boundary is most likely to occur at a plane that cuts the partitioning layer almost perpendicularly, so that the perturbation caused by remains minimal. This principle can be used as a criterion to identify potential twin boundaries, as demonstrated through detailed investigations of mesoporous silica crystals characterized by diamond and gyroid surfaces. We then discuss that a twin boundary can result from a stacking fault in the arrangement of inter-lamellar attachments at an early stage of structure formation. It is further shown that enhanced curvature fluctuations near the twin boundary would cost energy because of geometrical frustration, which would be eased by a crystal distortion that is experimentally observed.Keywords: twinning defects; electron crystallography; bicontinuous cubic structures; constant mean curvature surfaces; inorganic porous solids; crystal distortions.

### 1. Introduction

Bicontinuous cubic structures (BCSs) are crystal structures with cubic symmetry consisting of two continuous intertwining subvolumes, or labyrinths, separated by a non-self-intersecting partitioning layer. They are found in cell endomembrane systems (Landh, 1995), butterfly wing scales and beetle exoskeletons (Galusha *et al.*, 2008; Michielsen & Stavenga, 2008), lyotropic liquid crystals (LLCs) and related systems [mesoporous silica crystals (MSCs), self-assemblies, *etc*.] (Luzzati & Spegt, 1967; Hyde *et al.*, 1984; Alward *et al.*, 1986; Kresge *et al.*, 1992; Bates & Fredrickson, 1999; Wan & Zhao, 2007). Such structures are known to possess unique properties, such as structural colours (Galusha *et al.*, 2008; Michielsen & Stavenga, 2008), unusual mechanical and electronic features (Bruinsma, 1992; Fujita & Terasaki, 2005), and biological functions (Ellens *et al.*, 1989; Caffrey, 2000; Shah *et al.*, 2001).

Different sides of the partitioning layer can be filled with mutually immiscible sub-chains in *e.g.* atoms). The formation of BCSs in LLCs (Schwarz & Gompper, 2000*a*,*b*) is currently explained in terms of bending and stretching energies associated with the hydrophilic/hydrophobic interfaces, following the work by Helfrich (Helfrich, 1973), leading to two interpretations of the partitioning layer: either bounded by a pair of parallel surfaces with fixed distance off the TPMS, or by a pair of constant mean curvature (CMC) companions of the TPMS. The mean curvatures for the two CMC companions are equal in magnitude but opposite in sign. While the parallel surface model is more straightforward (Shearman *et al.*, 2007), competition and compromise between the two models have only recently been scrutinized (Chen & Jin, 2017).

Crystal defects, *e.g.* (Cahn, 1954), often carry a wealth of information on crystal growth and structural transformation. Contact reflection twins, a common form of involve two individual crystalline domains with an identical structure related by a reflection in their common boundary plane (termed twin boundary). Hereafter, the term `twinning' will only refer to contact reflection twins unless otherwise stated. Despite its importance, in BCSs has rarely been considered until recently.

MSCs are fabricated by the cooperative self-assembly of an LLC system with silicates as inorganic precursors (Kresge *et al.*, 1992; Wan & Zhao, 2007). In MSCs, BCSs are often formed as an intermediate structure between cylindrical and lamellar structures (Kaneda *et al.*, 2002; Gao *et al.*, 2006; Han *et al.*, 2009), where the silica wall traces the partitioning layer in the LLC phase before Recently, Han *et al.* (2011*b*) reported a BCS in spherical MSC particles with polyhedral cavities. More specifically, each facet of the cavity is a {111} surface of a single domain of a BCS with the geometry of a diamond (D) surface (termed D-BCS), while a twin boundary is observed at each polyhedral edge that bounds adjacent facets. More recently, a similar has also been observed in systems (Lin *et al.*, 2017). So far, however, no common law is known for describing the structure, formation and stability of twin boundaries in BCSs, as the conventional description of twins for discrete atomic crystals does not apply. The main difference here lies in the fact that BCSs are continuous structures that can be cut through by an arbitrary plane, in sharp contrast to conventional crystals.

Herein, a principle of BCS *et al.*, 2011*b*) and may serve as a practical criterion for preliminarily locating potential twin boundaries in general. It has been successfully applied to single out an observed twin boundary in another kind of BCS with the geometry of a gyroid (G) surface, termed G-BCS. Moreover, we point out that a D-twin boundary can be thought of as a stacking fault in a layered stacking of catenoidal channels, providing further insights into the formation of twin boundaries. Finally, analysis of Gaussian curvatures reveals how twin boundaries, as microscopic topological defects, may be the cause of crystal distortions manifested in the macroscopic morphology of twinned BCSs.

### 2. Crystallographic reconstruction of twinned MSCs

A twin boundary in conventional crystals can be described as the corresponding atomic plane specified by *x* as a fraction of *d _{hkl}*, the

*d*-spacing of the {

*hkl*} planes, to indicate the relative position of a twin boundary with respect to the reference planes. Hence, the twin boundary is specified as {

*hkl*} +

*x*. The reference plane with

*x*= 0 is taken to include flat points on the associated TPMS. For the D surface, for which the of the non-oriented [oriented] TPMS is [], the flat points correspond to the Wyckoff positions 4

*b*[32

*e*] with the [3

*m*]. On the other hand, the flat points for the G surface, for which the is [

*I*4

_{1}32], take the Wyckoff positions 16

*a*[16

*e*] with the [3]. Note that the surface normal vectors at the flat points are parallel to 〈111〉 in either case. Generally speaking, the reference {

*hkl*} plane may not be determined uniquely by the flat points. In the case of the G surface, in particular, the flat points can be classified according to whether their normal vectors are parallel to the {211} twin boundary or not. This in fact gives us two distinct choices of the reference {211} plane to describe a twin boundary of G-BCS. In this work, the reference plane is chosen such that it includes only those flat points whose surface normal vectors are not parallel to the plane and hence not normal to the plane normal 〈211〉.

Thanks to the stability of the inorganic skeletons under electron beam, MSCs are amenable to detailed structural analyses. A *et al.*, 2011*b*) is shown in Fig. 1(*a*). Fine details of the twin boundaries are revealed through high-resolution TEM (HRTEM) observations on sliced samples embedded in epoxy resin. Fig. 1(*b*) shows two structural sub-domains, each of which show the typical contrast from 〈110〉 directions of D-BCS. Here, alternating bright and dark contrast signifies the projection of electron-scattering intensities in the silica wall. The sub-domains are the twin individuals contacting each other at a {111} plane (running vertically) whose contact angle with one of the {100} planes is 54.7°. Accordingly, the Fourier transform [Fig. 1(*c*)] of the HRTEM image shows two sets of diffraction spots, in which the 111 reflections overlap with each other. To decide the offset of the twin boundary, an HRTEM image and an electron diffraction (ED) pattern of a twin individual were simulated using the software *MesoPoreImage* (Ohsuna *et al.*, 2011) using a 3-term nodal approximation (Gandy *et al.*, 2001). The simulated HRTEM image was then combined with its mirror image at different offsets (see Fig. S2 in the Supporting information); the experimental contrast is best reproduced with a twin boundary at {111} + 0.5 [inset of Fig. 1(*b*)]. The geometrical relationship in Fig. 1(*c*) is clarified in Fig. 1(*d*) where simulated ED patterns for the twin individuals are superposed. (Note, the Fourier transform of a TEM image is generally affected by the contrast transfer function while an ED pattern is not, so their peak intensities are not directly related.) The two labyrinths can be sketched using skeletal graphs with 4-connected nodes at the 2*a* positions for the with the . Around the twin boundary, the red skeleton maintains the 4-connectivity, while the blue skeleton involves 5-connected nodes at the boundary because of a modified topology [Fig. 1(*e*)].

Han *et al.* (2011*a*,*b*) reconstructed an average silica wall structure within an individual domain of D-BCS using an electron crystallographic technique (Sakamoto *et al.*, 2000; Miyasaka & Terasaki, 2010; Willhammar *et al.*, 2012). Here, the structure factors extracted from a series of HRTEM images taken along high-symmetric zone axes were used to obtain the electrostatic potential map. Then, the side surfaces of the silica wall were determined as the equipotential surfaces that minimize an energy density associated with the curvature. To be specific, the Helfrich free-energy density, *f* = κ_{1}(〈*H*^{2}〉 − 〈*H*〉^{2}) + κ_{2}〈*K*〉, was employed, where *H* and *K* are the mean and Gaussian curvatures, respectively, and is the area-weighted average of *x* on the surface *S*. Although it is difficult to determine the two coefficients, κ_{1} and κ_{2}, *a priori*, the authors heuristically assumed κ_{2} = 0, thereby only the mean curvature fluctuations contribute to the energy, as if the side surfaces favoured CMC surfaces. (If 〈*H*〉 is identified with a predefined constant *H*_{0}, reduces to the modified Willmore functional to be discussed later.) The threshold of the equipotential was thus determined at 75% within the min–max range of the electrostatic potential map (see Table S1 and Fig. S3 in the Supporting information). Remarkably, this crude approximation could fairly reproduce the high-resolution electron micrographs [Fig. 1(*b*)]. The reconstructed silica wall domain [Fig. 1(*g*)] is now mirrored at {111} + 0.5 to mimic the D-twin boundary [Fig. 1(*f*)].

### 3. In favour of smoothness

The D-twin model above, obtained by simply putting two reconstructed D-BCS together [Fig. 1(*f*)], seems fairly smooth although not quite, suggesting that the silica wall in the un-twinned BCS intersects with the twin boundary almost perpendicularly. Only a small perturbation seems necessary to restore smoothness. In other words, for the preference of minimal perturbation, a potential twin boundary should, in general, intersect the BCS partitioning layer almost perpendicularly.

The perpendicularity of a surface with a plane can be quantified by the dot products |**n**_{s}·**n**_{b}| over the intersection, where **n**_{s} denotes the unit normal vectors of the surface and **n**_{b} denotes the unit normal vector of the plane. By checking the distribution of |**n**_{s}·**n**_{b}| slice by slice throughout the reconstructed D-BCS along different crystal axes, we find that the {110} + 0.5 plane is perfectly perpendicular to the equipotential surfaces [Figs. 2(*a*) and S4]. This is no surprise since it is actually a mirror plane in D-BCS. The second-best candidate is the {111} + 0.5 plane [Figs. 2(*b*) and S5], which corresponds to the observed twin boundaries.

The D surface and its CMC companions may also be used to approximate the median and side surfaces of the partitioning layer of D-BCS. These surfaces can be numerically constructed using Brakke's *Surface Evolver* (Brakke, 1992), an efficient gradient descender (see Appendix *A*), and lend themselves to geometrical analyses. The elevation angles of surface normal vectors with respect to {111} at uniformly sampled points on the surfaces are plotted in Fig. 2(*d*), which again confirms that {111} + 0.5 is the most preferable in terms of perpendicularity. It is instructive to note that the plots in Fig. 2(*d*) reflect symmetry elements of the surfaces. Because of the inversion symmetry, the black dots are distributed symmetrically with respect to the vertical lines at integer offsets (*e.g.* *x* = 0), which correspond to the planes containing flat points (*i.e.* inversion centres), whereas the same vertical mirrors exchange the blue and red dots corresponding to the two CMC companions. Also, the distribution of dots of each colour is symmetric with respect to any half-integer offset point (*e.g.* *x* = 0.5) on the horizontal axis (*i.e.* the zero angle line) since twofold rotational axes lie in the relevant {111} plane.

Generally speaking, twin boundaries tend to favour low indices. In the case of conventional crystals, twin boundaries usually correspond to dense atomic planes costing low boundary energies. This naturally makes the density variation high along the normal axis to the twin boundary. In the case of BCS (*c*) and S6]. This has two implications. On the one hand, a high pore fraction indicates flexibility of the structure subject to perturbation; on the other hand, it also results in a high density variation along the boundary normal. (The projected density onto an axis through the origin is directly connected with X-ray structure factors lying on the same axis, whereas corresponding electron-structure factors are somewhat more loosely correlated.) For our D-BCS, the strongest reflection is {110}, associated with the mirror plane; the second strongest is {111}, associated with the observed D-twin boundary (Fig. S1 and Table S1). In order to identify potential twin boundaries, it is therefore important to compare the perpendicularities of different planes that show large density variations along their normal axes.

### 4. Numerical modelling of minimal and CMC twins

We now investigate structural perturbations caused by the *Surface Evolver* (Brakke, 1992). A single twin boundary is approximated by parallel twin boundaries at a distance comparable to the experiment. Free boundary conditions are imposed on the boundary planes so that the surface can be extended through reflection. The surface integral of (*H*−*H*_{0})^{2}, where *H* is the mean curvature and *H*_{0} is a prescribed mean curvature, can be used as the cost function such that a surface with a mean curvature *H*_{0} is obtained if it is numerically reduced down to 0 (Große-Brauckmann, 1997). The latter integral is a modification of the so-called Willmore functional (Hsu *et al.*, 1992), for which the integrand is *H*^{2}. Figs. 2(*e*) and 2(*f*) show constructed minimal (*H*_{0} = 0) and CMC (*H*_{0} = 1.1/*a*_{D}) surfaces, respectively, for modelling a D-twin boundary, where *a*_{D} is the cubic constant of the D surface under the .

The perturbation caused by (*e*) and 2(*f*), where blue and red indicate the normal deviations of the twinned surface from its un-twinned original in opposite senses as drawn using the *CloudCompare* package (Girardeau-Montaut, 2015). Note that the deviations are only prominent in the vicinity of the twin boundary and decay fast within a short distance. In fact, a similar decay has also been reported for a simulated twist grain boundary in LLC within the Ginzburg–Landau scheme (Belushkin & Gompper, 2009).

It is interesting to note that the opposite sides of the partitioning layer can be asymmetric, hence one side can be more strongly perturbed by (*b*) that the intersection of a {111} + 0.5 plane and the silica wall (tracing the partitioning layer) consists of rings arranged on a triangular The inner and outer circles of a ring correspond to larger and smaller necks in the CMC model, respectively. Clearly, with larger neck radius, the surface is less perpendicular to the twin boundary, hence larger perturbation is entailed. An obvious disparity between the CMC pair with regard to the intersection angles is also shown in Fig. 2(*d*) at *x* = 0.5.

### 5. Extrapolation of the theory to G-twins

We now extend our discussion to G-BCS. By carefully inspecting HRTEM images of MSC particles identified as G-BCS, a twin boundary, termed G twin, is revealed. HRTEM images [Figs. 3(*a*) and 3(*b*)] taken from a common 〈311〉 axis show that the twin boundary can be indexed as {211} and that its contact angle with a {110} plane on each side is around 73.2°. In the inset of Fig. 3(*b*) (and Fig. S7), a simulated HRTEM image of G-BCS, modelled using a 4-term nodal approximation (Ohsuna *et al.*, 2011), is merged with its mirror image at {211} + 0.5, fairly reproducing the observation. A superposition of simulated ED patterns [Fig. 3(*d*)] also agrees with the Fourier transform shown in Fig. 3(*c*). Remarkably, we find that a twin boundary that was reported in a copolymeric G-BCS (Vignolini *et al.*, 2012) has the same twin orientational relationship as the present one.

The two disjoint labyrinths of G-BCS can be described using 3-connected nets whose nodes correspond to the 16*b* positions [for ] with the 32. Both the red and blue skeletons are reflected with respect to the G-twin boundary, yielding 3- and 5-connected junctions on the boundary after fusing every close pair of nodes [Fig. 3(*e*)]. The modified topology also bears a fault at which the two networks swap their handedness. As in the case of the D-BCS, we adopt the G-BCS that was reconstructed by Han *et al.* (2011*a*) [Fig. 3(*g*), Table S2 and Fig. S8]. Recall that the side surfaces of the silica wall were approximated as the equipotential surfaces that minimize the Helfrich energy density with κ_{2} = 0, wherein the threshold value was determined to be 53% within the min–max range of the equipotential. We find that two mirrored domains at {211} + 0.5 meet almost smoothly [Fig. 3(*f*)]. Indeed, the {211} + 0.5 planes of G-BCS are the lowest in terms of the dot product |**n**_{s}·**n**_{b}| for the reconstructed side surfaces [Fig. 4(*a*)], accommodating the highest sectional pore fraction [Fig. 4(*c*)].

If the G surface and its CMC companions are employed as geometrical approximations to the median and side surfaces of G-BCS, respectively, it is again found that elevation angles of the surface normal vectors with respect to {211} are the lowest at half-integer offsets [Fig. 4(*d*)]. These altogether suggest that {211} + 0.5 is the best candidate of the twin boundary in terms of perpendicularity. Note that the plots in Fig. 4(*d*) are symmetrical with respect to vertical lines at both integer and half-integer offsets, provided that the blue and red colours are exchanged. This, as in Fig. 2(*d*), is attributed to the inversion centres (*i.e.* the flat points) lying in the {211} planes at these offsets. In the meantime, the symmetry of each coloured dot distribution at any integer [half-integer] offset position on the horizontal axis is attributed to twofold screw [rotational] axes along 〈110〉 lying in the {211} planes at the relevant offset. Furthermore, an additional mirror symmetry of the plots with respect to the horizontal axis, provided that the red and blue colours are exchanged, is attributed to the *d*-glide planes that orthogonally intersect {211}.

For modelling the G-twin boundary, the modified Willmore functional is reduced to zero using *Surface Evolver* for the minimal (*H*_{0} = 0) and CMC [] surfaces where *a*_{G} is the cubic constant of the G surface under both the space groups and *I*4_{1}32. The model surfaces are bounded by two parallel twin boundaries at a distance comparable to the size scale of the experiment [Figs. 3(*a*) and 3(*b*), and Figs. S9 and S10]. Again, the deviations from their un-twinned counterparts are concentrated at the twin boundaries [Figs. 4(*e*) and 4(*f*)]. The two CMC surfaces can be swapped by a glide operation that does not change the twin boundaries [as reflected in Fig. 4(*d*)], hence the two side surfaces of the partitioning layer are perturbed equally.

### 6. BCS as stacking fault

Defects are no stranger to MSCs. In particular, cage-type MSCs are built of spherical micelle/silica composites in analogy with atomic crystals of atoms. It is then no surprise that *et al.*, 2006; Sakamoto *et al.*, 2009). In contrast, a collective self-organization of a molecular assembly is essential in the formation of a bicontinuous MSC, where a prior generation of separate building blocks, *i.e.* micelles/silica composites, may not be necessary. In the present synthesis system, the formation of a BCS takes place through a structural relaxation of an inverse bilayer consisting of molecules during the silica condensation. The head groups together with additional components, including water and silicate oligomers, form the polar region, which settles down along a TPMS as its median surface. This process should also take place while D and G twins are formed. It has been suggested that bicontinuous phases can be transformed from lamellar phases by forming so-called inter-lamellar attachments (ILAs) between adjacent bilayers (Squires *et al.*, 2002; Conn *et al.*, 2006; Tang *et al.*, 2015). More recently, direct imaging has also shown that ILAs emerge as open necks between parallel planes (Demurtas *et al.*, 2015).

The central cavity of our spherical D-twin particles is most likely to be a reminiscence of a sparse centre in a concentric multilayer vesicle in an initial lamellar stage [*cf*. Conn *et al.* (2006) and Fig. S12]. Then the creation of ILAs initiates the inverse bilayers to rupture so as to diminish the local curvatures, whereby a {111} stacking of necks within each individual D domain is formed. The in-plane positions of necks in each {111} layer of D-BCS define a triangular while the layer stacking exhibits a cyclic pattern …*ABCABC*… in terms of the neck positions [Fig. 5(*a*)]. The 3D arrangement of the parallel necks gives a simple cubic (corresponding to the ). Then, like a in a conventional crystal, a D-twin boundary may arise as a stacking fault, with the ABC pattern reverted at some layer *C* to form a palindromic sequence …*ABCABCBACBA*…. It is also instructive to consider an alternative stacking pattern, …*BCBCBC*…, where the 3D arrangement of parallel necks defines a composite of two simple hexagonal lattices similar to a hexagonal close-packed but with the constant along the threefold axis being halved. The same arrangement is found in Schwarz's H surfaces (space group, *P*6_{3}*mmc* []), for which the labyrinths can be represented as interwoven 5-connected nets [Fig. 5(*b*)]. This in fact allows us to think of an H surface as the simplest polysynthetic D twins, with the shortest distance between adjacent twin boundaries. The neighbourhood of a D-twin boundary hence resembles a slab of an H surface.

Surprisingly, the ILA formation finds correspondence in differential geometry. Traizet constructed virtually all TPMSs near the so-called `catenoid limit' by periodically opening nodes among horizontal planes (Traizet, 2008). Here, `opening nodes' means a desingularization of singular points (*i.e.* nodes) into small catenoids, in analogy with the opening of ILAs into necks. In particular, Traizet shows the necessity of a balanced condition, namely that all necks are in equilibrium under effective interactions, which are attractive between necks within the same level while repulsive across adjacent levels, with no interaction across levels farther apart. Their strengths are inversely proportional to the distance between the necks while proportional to their sizes, similar to electrostatic forces in the 2D universe. The balanced condition is satisfied for the D twin at the catenoid limit (Chen, 2019), ensuring the stability of twinned minimal surfaces, though the proof does not work away from that limit.

The formation of a G-twin boundary is also likely to be related to a structural transformation process. In copolymeric systems, it is reported that the transformation from lamellar to G-BCS passes through an intermediate state known as hexagonally perforated layers (HPL) (Foerster *et al.*, 1994; Khandpur *et al.*, 1995). The HPL phase can be considered as a hexagonally modulated lamellar phase maintaining an epitaxial relationship with {211} planes of G-BCS. A transient state like HPL is also likely in the pathway from lamellar to G-BCS in LLC systems, considering that a stacking fault in HPL could lead to a fault that is an essential ingredient of the {211} G-twin boundary.

In fact, G-BCS also allows an ILA argument as in the case of D-BCS (Chen, 2019). More specifically, the G surface is derived if catenoidal necks are introduced between an infinite array of parallel planes, such that necks on the same layer are arranged on a rhombic along a {110} plane of G. One would then consider a transformation from lamellar to G-BCS, during which a stacking fault might arise to generate another type of twin boundary at a {220} + 0.5 plane. Indeed, the {220} + 0.5 planes are the secondary choice in terms of the perpendicularity principle [*cf*. Figs. 4(*b*) and S10]. So far, however, no {220} epitaxial relation has been reported in lamellar/G phase transformations but in cylindrical/G phase transformations (Honda & Kawakatsu, 2006; Schulz *et al.*, 1994).

Again, it is worth mentioning that the simplest polysynthetic G twins with the shortest distance between adjacent {211} twin boundaries correspond to a TPMS represented in Fig. 5(*c*) (centre). The latter surface is an orthorhombically deformed D surface (oDb surface), with the *Cmme* [*Imma*], for which the Weierstrass parameterization is available (Fogden & Hyde, 1992). The neighbourhood of the G-twin boundary hence resembles a slab of an oDb surface.

### 7. Crystal distortions associated with twinning

Our MSC samples often exhibit, albeit with varying amounts, crystal distortions that may be associated with (*a*), the {111} facets surrounding the central cavity are slightly bent inward (see also Fig. S13). If the central cavity were a regular icosahedron, two adjacent triangular facets would subtend a dihedral angle of 138.19°, whereas the ideal dihedral angle that originates from {111} facets of the cubic D-BCS would be 2 × 70.53° = 141.06° with 70.53° being the dihedral angle of a regular tetrahedron. The angle mismatch could be absorbed if the inner facets are bent outward, which opposes our observation.

HRTEM images [Figs. 1(*b*) and S13] around a D-twin boundary are closely inspected to reveal some crystal distortions that might be responsible for the unusual bending. It is observed near the twin boundary that (i) the period normal to the twin boundary is enlarged by ∼4.5% and (ii) that the periods along the twin boundary shrink by ∼1%. In a sample with thicker shell, the shrinkage along the twin boundary is even larger, leading to more pronounced bending (Fig. S14). On the other hand, the HRTEM image of the G-twin boundary hardly shows bending of the crystal plane, probably because of the small particle size (Fig. S15). Still, the periodicity normal to the twin boundary is enlarged by ∼5.5%.

Generally speaking, the geometry of a BCS is dominated by the geometry of the underlying molecules, which in our case are surfactants. The *v* is the volume occupied by the hydrophobic chain, α is the interfacial area per head group and *l* is the effective length of the hydrophobic chain. Following the work by Hyde (1989), it can be argued that the median surface of an inverse bilayer is modelled as a minimal surface for which the area-weighted average of Gaussian curvature is written as , where *t* is the of the polar region (including ionic head groups, water and silicate oligomers) (Hyde, 1989). Note that this theory assumes that the curvature fluctuation along the surface is small enough.

Recall that the median surface of twinned D or G surfaces locally resembles an H or oDb surface, respectively, at the twin boundary. All these four minimal surfaces are surfaces of genus three (*i.e.* *g* = 3). It then follows from the Gauss–Bonnet theorem that 〈*K*〉 = 4π(1 − *g*)/*A* = −8π/*A*, where *A* is the surface area per of the oriented surface. Figs. 6(*a*) or 6(*c*) show that 〈*K*〉 is enhanced by ∼0.2% in H compared with D or by ∼1.2% in oDb compared with G, respectively, provided that the reference parameters [*i.e.* *a, b* and *c* in Figs. 5(*b*) and 5(*c*)] are fixed. The surface area per is reduced accordingly. The formal reinforcement of curvature at the twin boundary is, however, so small that it can be readily amended by swelling the only slightly. Moreover, Figs. 6(*a*) or 6(*c*) show that the dependence of 〈*K*〉 on the aspect ratios of the is extremely weak.

In the meantime, the relative variance of the curvature, Δ*K*^{2}/〈*K*〉^{2} (= 〈*K*^{2}〉/〈*K*〉^{2} − 1), is markedly enhanced in the non-cubic variants as compared with their cubic counterparts, provided that the original aspect ratios of the (as determined from the reference constants *a*, *b* and *c*) are maintained. Fig. 6(*b*) shows that Δ*K*^{2}/〈*K*〉^{2} for the H surface doubles that for the D surface. This buildup of curvature fluctuations around the twin boundary, also visualized in Fig. 6(*e*), would give rise to a strong frustration between the surface and molecular geometries, and is most likely to cost curvature energy. Still, the frustration could be eased by modifying the unit-cell geometry, *i.e.* by increasing *c*/*a* in the direction from H to H_{x}. Indeed, this is in accordance with the crystal distortion observed experimentally at the D-twin boundary. Similarly, the curvature fluctuations at the G-twin boundary are enhanced by about 20% [Figs. 6(*d*) and 6(*e*)], which could be eased if the unit-cell geometry is changed along the green arrow in Fig. 6(*d*). This is again consistent with the observed small increase of *a* relative to *b* and *c* at the twin boundary. For drawing Figs. 6(*a*)–6(*d*), all the three quantities, *A*, 〈*K*〉 and Δ*K*^{2}, have been accurately computed for the TPMSs using the Weierstrass integration formulae (Fogden & Hyde, 1992).

Apart from the local curvature fluctuations, the fluctuations of labyrinthine diameter would also be enhanced strongly by the non-cubic geometries at the twin boundaries (Schröder-Turk *et al.*, 2006). The latter effect is considered to be the second source of geometrical frustrations associated with the packing inhomogeneity of the molecules, which need to be stretched or compressed to fill in the given space. Note that at any point *p* on the surface the distance function *d*(*p*) is defined as the radius of the largest sphere that fits within one labyrinth and touches the surface tangentially at *p*. The labyrinthine diameter at *p* is then measured by doubling *d*(*p*). An intriguing fact here is that the increase of Δ*d*^{2}/〈*d*〉^{2} (the relative fluctuation of the labyrinthine diameter) around the D-twin boundary, which resembles an H surface, can be eased not by increasing but by decreasing *c*/*a* [see Fig. 6(*b*)]. This suggests that the packing inhomogeneity is less dominant in determining the surface geometry of the BCS in the present system.

To conclude, we have presented two kinds of twin boundaries in bicontinuous MSCs; namely, one in D-BCS at {111} + 0.5 and the other in G-BCS at {211} + 0.5 planes. The 3D reconstruction of the silica wall has allowed us to elucidate detailed structural characteristics of the twinned BCSs with the aid of geometrical methods to construct and analyze the characteristic surfaces. The finding of primary importance is that the twin boundaries occur at crystallographic planes that intersect the relevant BCS most perpendicularly. Moreover, we presented the novel view that a D-twin boundary is a stacking fault in the arrangement of catenoidal necks. The local geometries of the twinned surfaces have been modelled using non-cubic variants of the TPMSs, whereby the dominant role of the local curvature frustrations in determining the surface geometry has been revealed. The present results may provide further insights into the formation processes of BCSs in natural and synthetic self-assembly systems.

### APPENDIX A

### Methods

The MSCs were synthesized in the work by Han *et al.* (2011*a*). The typical synthesis composition is *N*-stearoyl-L-glutamic acid : Brij-56 [C_{16}H_{33}(OCH_{2}CH_{2})_{10}OH] : 3-aminopropyl trimethoxysilane : tetraethyl orthosilicate : H_{2}O = 1 : *x* : 2 : 15 : 2335, where *x* = 1.45 for D and 1.75 for G. The surfactant-free materials for HRTEM analyses were obtained by at 550°C in air for 6 h.

HRTEM was performed using a JEOL JEM-2100 microscope that was equipped with a LaB_{6} gun operating at 200 kV [spherical aberration coefficient (*C*_{s}) of the objective lens = 1.0 mm, point resolution = 2.3 Å]. Images were recorded using a TENGRA CCD camera (resolution of 2304 × 2304 pixels with a 2:1 fibre-optical taper and an effective pixel size of 8 µm^{2}) at 50 000–120 000 times magnification under low-dose conditions. HRTEM is the ultimate tool for analysing defects as the structure can be directly observed. Since both phases and amplitudes of factors can be obtained by electron crystallography from the Fourier transform of the HRTEM images, the 3D electrostatic potential distribution map can be directly reconstructed to elucidate the characteristic structural features.

To obtain a minimal or CMC surface with *Surface Evolver*, we first prepared an initial surface that is homeomorphic to the target surface by appropriately opening channels between equally spaced parallel planes [following Traizet's idea; Traizet (2008)] under prescribed periodic boundary conditions. Then, the surfaces used for the scatter plots of Figs. 2(*d*) and 4(*d*) were obtained by minimizing the area under fixed volume ratios between the two labyrinths. For the minimal surface the volume ratio was constrained to be 1, while for each of the two CMC companions it was fixed to be either 3 or 1/3.

We then used translated copies of the (*e*), 2(*f*), 4(*e*) and 4(*f*) were obtained by minimizing the modified Willmore functionals. The prescribed mean curvature *H*_{0} was set to 0 for Figs. 2(*e*) or 4(*e*), 1.1/*a*_{D} for Fig. 2(*f*) or 0.6/*a*_{G} for Fig. 4(*f*), where *a*_{D} or *a*_{G} stands for the cubic constants of the D or G surface, respectively. *Surface Evolver* managed to reduce the modified Willmore functionals down to 10^{−25} even when the twin boundaries were at a distance 15 × *d*_{111} for D twin or 15 × *d*_{211} for G twin, which was comparable to the scale of the experiment.

### Supporting information

Supporting information. DOI: https://doi.org//10.1107/S2052252519017287/zx5019sup1.pdf

### Acknowledgements

The authors are indebted to Keiichi Miyasaka for his assistance in the 3D reconstruction and visualization of the MSC structures, and to Stephen Hyde and Sven Lidin for their helpful comments on the manuscript. They also appreciate the discussions with Karsten Große-Brauckmann and Ken Brakke. CJ acknowledges the support of Corinna C. Maaß and Stephan Herminghaus. LH synthesized the materials, carried out electron-microscopy observations and analysed the 3D reconstructed MSC structures, under the supervision and support of OT and SC. Numerical modelling was mostly conducted by CJ and HC using *Surface Evolver*. HC introduced the stacking fault argument along with a new perspective of differential geometry. The analysis of Gaussian curvatures was carried out by NF to explain the observed crystal distortions. The manuscript was originally drafted by LH, and further enriched by LH, HC, CJ and NF. NF led the preparation of the final manuscript. All the authors discussed the results and commented on the manuscript.

### Funding information

This work has been financially supported by the National Natural Science Foundation of China (Grant Nos. 21922304 and 21873072; LH), the National Excellent Doctoral Dissertation of the People's Republic of China (Grant No. 201454; LH), the Shanghai Rising-Star Program (Grant No. 17QA1401700; LH), Japan Science and Technology Agency (JST) through PRESTO (Grant No. JPMJPR1519; NF), Deutsche Forschungsgemeinschaft (Grant No. 398759432; HC), Center for High-resolution

(CEM) in SPST, ShanghaiTech University (Grant No. EM02161943; OT) and The Foreign 1000 Talents Plan of China (OT).### References

Alward, D. B., Kinning, D. J., Thomas, E. L. & Fetters, L. J. (1986). *Macromolecules*, **19**, 215–224. CrossRef CAS Web of Science Google Scholar

Bates, F. S. & Fredrickson, G. H. (1999). *Phys. Today*, **52**, 32–38. Web of Science CrossRef CAS Google Scholar

Belushkin, M. & Gompper, G. (2009). *J. Chem. Phys.* **130**, 134712. Web of Science CrossRef PubMed Google Scholar

Brakke, K. A. (1992). *Exp. Math.* **1**, 141–165. CrossRef Google Scholar

Bruinsma, R. (1992). *J. Phys. II France*, **2**, 425–451. CrossRef CAS Google Scholar

Caffrey, M. (2000). *Curr. Opin. Struct. Biol.* **10**, 486–497. Web of Science CrossRef PubMed CAS Google Scholar

Cahn, R. (1954). *Adv. Phys.* **3**, 363–445. CrossRef Web of Science Google Scholar

Chen, H. (2019). *Exp. Math.* **28**, 404–419. Web of Science CrossRef Google Scholar

Chen, H. & Jin, C. (2017). *Interface Focus*, **7**, 20160114. Web of Science CrossRef PubMed Google Scholar

Conn, C. E., Ces, O., Mulet, X., Finet, S., Winter, R., Seddon, J. M. & Templer, R. H. (2006). *Phys. Rev. Lett.* **96**, 108102. Web of Science CrossRef PubMed Google Scholar

Demurtas, D., Guichard, P., Martiel, I., Mezzenga, R., Hébert, C. & Sagalowicz, L. (2015). *Nat. Commun.* **6**, 8915. Web of Science CrossRef PubMed Google Scholar

Ellens, H., Siegel, D. P., Alford, D., Yeagle, P. L., Boni, L., Lis, L. J., Quinn, P. J. & Bentz, J. (1989). *Biochemistry*, **28**, 3692–3703. CrossRef CAS PubMed Web of Science Google Scholar

Foerster, S., Khandpur, A. K., Zhao, J., Bates, F. S., Hamley, I. W., Ryan, A. J. & Bras, W. (1994). *Macromolecules*, **27**, 6922–6935. CrossRef CAS Google Scholar

Fogden, A. & Hyde, S. T. (1992). *Acta Cryst.* A**48**, 575–591. CrossRef Web of Science IUCr Journals Google Scholar

Fujita, N. & Terasaki, O. (2005). *Phys. Rev. B*, **72**, 085459. Web of Science CrossRef Google Scholar

Galusha, J. W., Richey, L. R., Gardner, J. S., Cha, J. N. & Bartl, M. H. (2008). *Phys. Rev. E*, **77**, 050904. Web of Science CrossRef Google Scholar

Gandy, P. J., Bardhan, S., Mackay, A. L. & Klinowski, J. (2001). *Chem. Phys. Lett.* **336**, 187–195. Web of Science CrossRef CAS Google Scholar

Gao, C., Sakamoto, Y., Sakamoto, K., Terasaki, O. & Che, S. (2006). *Angew. Chem. Int. Ed.* **45**, 4295–4298. Web of Science CrossRef CAS Google Scholar

Girardeau-Montaut, D. (2015). *Cloud Compare* (http://www.cloudcompare.org). Google Scholar

Große-Brauckmann, K. (1997). *Exp. Math.* **6**, 33–50. Google Scholar

Han, L., Miyasaka, K., Terasaki, O. & Che, S. (2011*a*). *J. Am. Chem. Soc.* **133**, 11524–11533. Web of Science CrossRef CAS PubMed Google Scholar

Han, L., Xiong, P., Bai, J. & Che, S. (2011*b*). *J. Am. Chem. Soc.* **133**, 6106–6109. Web of Science CrossRef CAS PubMed Google Scholar

Han, Y., Zhang, D., Chng, L. L., Sun, J., Zhao, L., Zou, X. & Ying, J. Y. (2009). *Nat. Chem.* **1**, 123–127. Web of Science CrossRef CAS PubMed Google Scholar

Helfrich, W. (1973). *Z. Naturforsch. C*, **28**, 693–703. CrossRef CAS PubMed Google Scholar

Honda, T. & Kawakatsu, T. (2006). *Macromolecules*, **39**, 2340–2349. Web of Science CrossRef CAS Google Scholar

Hsu, L., Kusner, R. & Sullivan, J. (1992). *Exp. Math.* **1**, 191–207. CrossRef Google Scholar

Hyde, S. T. (1989). *J. Phys. Chem.* **93**, 1458–1464. CrossRef CAS Web of Science Google Scholar

Hyde, S. T., Andersson, S., Ericsson, B. & Larsson, K. (1984). *Z. Kristallogr.* **168**, 213–219. CrossRef CAS Web of Science Google Scholar

Kaneda, M., Tsubakiyama, T., Carlsson, A., Sakamoto, Y., Ohsuna, T., Terasaki, O., Joo, S. & Ryoo, R. (2002). *J. Phys. Chem. B*, **106**, 1256–1266. Web of Science CrossRef CAS Google Scholar

Khandpur, A. K., Foerster, S., Bates, F. S., Hamley, I. W., Ryan, A. J., Bras, W., Almdal, K. & Mortensen, K. (1995). *Macromolecules*, **28**, 8796–8806. CrossRef CAS Web of Science Google Scholar

Kresge, C. T., Leonowicz, M., Roth, W., Vartuli, J. C. & Beck, J. (1992). *Nature*, **359**, 710–712. CrossRef CAS Web of Science Google Scholar

Landh, T. (1995). *FEBS Lett.* **369**, 13–17. CrossRef CAS PubMed Web of Science Google Scholar

Lin, Z., Liu, S., Mao, W., Tian, H., Wang, N., Zhang, N., Tian, F., Han, L., Feng, X. & Mai, Y. (2017). *Angew. Chem. Int. Ed.* **56**, 7135–7140. Web of Science CrossRef CAS Google Scholar

Luzzati, V. & Spegt, P. A. (1967). *Nature*, **215**, 701–704. CrossRef CAS Web of Science Google Scholar

Michielsen, K. & Stavenga, D. G. (2008). *J. R. Soc. Interface*, **5**, 85–94. Web of Science CrossRef PubMed CAS Google Scholar

Miyasaka, K., Han, L., Che, S. & Terasaki, O. (2006). *Angew. Chem. Int. Ed.* **45**, 6516–6519. Web of Science CrossRef CAS Google Scholar

Miyasaka, K. & Terasaki, O. (2010). *Angew. Chem. Int. Ed.* **49**, 8867–8871. Web of Science CrossRef CAS Google Scholar

Ohsuna, T., Sakamoto, Y., Terasaki, O. & Kuroda, K. (2011). *Solid State Sci.* **13**, 736–744. Web of Science CrossRef CAS Google Scholar

Sakamoto, Y., Han, L., Che, S. & Terasaki, O. (2009). *Chem. Mater.* **21**, 223–229. Web of Science CrossRef CAS Google Scholar

Sakamoto, Y., Kaneda, M., Terasaki, O., Zhao, D., Kim, J., Stucky, G. D., Shin, H. J. & Ryoo, R. (2000). *Nature*, **408**, 449–453. Web of Science CrossRef PubMed CAS Google Scholar

Schröder-Turk, G. E., Fogden, A. & Hyde, S. T. (2006). *Eur. Phys. J. B*, **54**, 509–524. Google Scholar

Schulz, M. F., Bates, F. S., Almdal, K. & Mortensen, K. (1994). *Phys. Rev. Lett.* **73**, 86–89. CrossRef PubMed CAS Web of Science Google Scholar

Schwarz, U. S. & Gompper, G. (2000*a*). *J. Chem. Phys.* **112**, 3792–3802. Web of Science CrossRef CAS Google Scholar

Schwarz, U. S. & Gompper, G. (2000*b*). *Phys. Rev. Lett.* **85**, 1472–1475. Web of Science CrossRef PubMed CAS Google Scholar

Shah, J. C., Sadhale, Y. & Chilukuri, D. M. (2001). *Adv. Drug Deliv. Rev.* **47**, 229–250. Web of Science CrossRef PubMed CAS Google Scholar

Shearman, G. C., Khoo, B. J., Motherwell, M.-L., Brakke, K. A., Ces, O., Conn, C. E., Seddon, J. M. & Templer, R. H. (2007). *Langmuir*, **23**, 7276–7285. Web of Science CrossRef PubMed CAS Google Scholar

Squires, A. M., Templer, R. H., Seddon, J. M., Woenckhaus, J., Winter, R., Finet, S. & Theyencheri, N. (2002). *Langmuir*, **18**, 7384–7392. Web of Science CrossRef CAS Google Scholar

Tang, T. D., Brooks, N. J., Ces, O., Seddon, J. M. & Templer, R. H. (2015). *Soft Matter*, **11**, 1991–1997. Web of Science CrossRef CAS PubMed Google Scholar

Traizet, M. (2008). *J. Differ. Geom.* **79**, 243–275. CrossRef Google Scholar

Vignolini, S., Yufa, N. A., Cunha, P. S., Guldin, S., Rushkin, I., Stefik, M., Hur, K., Wiesner, U., Baumberg, J. J. & Steiner, U. (2012). *Adv. Mater.* **24** OP23–OP27. Google Scholar

Wan, Y. & Zhao, D. (2007). *Chem. Rev.* **107**, 2821–2860. Web of Science CrossRef PubMed CAS Google Scholar

Willhammar, T., Sun, J., Wan, W., Oleynikov, P., Zhang, D., Zou, X., Moliner, M., Gonzalez, J., Martínez, C., Rey, F. & Corma, A. (2012). *Nat. Chem.* **4**, 188–194. Web of Science CrossRef CAS PubMed Google Scholar

This is an open-access article distributed under the terms of the Creative Commons Attribution (CC-BY) Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original authors and source are cited.