Inverse design of heterodeformations for strain soliton networks in bilayer 2D materials

arXiv:2603.21446 · cond-mat.mtrl-sci · Submitted 2026-03-22 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.

Kai: I'm Kai.

Mira: And I'm Mira, here with Lev.

Kai: We're looking at the paper: "Inverse design of heterodeformations for strain soliton networks in bilayer 2D materials".

Mira: Here's the gist of the paper. Strain soliton networks strongly influence the structural and electronic properties of heterodeformed bilayer systems, yet their design remains challenging due to the high dimensionality of heterodeformation space and the absence of a direct map between deformation and network geometry In this work, we introduce a geometric framework that establishes a one-to-one mapping between heterodeformations and the geometry of the strain soliton network expressed as line vector–Burgers vector pairs The proposed framework enables the direct construction of heterodeformations from target networks, providing a systematic route for inverse design of moir´e interfaces beyond conventional twist-based approaches Geometric Framework and Mappings The core contribution of this research is the introduction of a geometric framework that establishes a one-to-one mapping between heterodeformations and the geometry of the strain soliton network expressed as line vector–Burgers vector pairs This framework constrains admissible networks by topology dictated by the generalized stacking fault energy landscape The authors demonstrate that while distinct heterodeformations can share identical moir´e Bravais lattices, they produce different soliton networks, reflecting an inherent many-to-one mapping when only translational symmetry is considered The soliton network is highlighted as encoding the full multilattice geometry of the interface, including topology and connectivity, which are not captured by the moir´e Bravais lattice alone Topological Constraints from GSFE The topology of a 2D soliton network is determined by the number of nearest-neighbor minima surrounding a local minimum in the generalized stacking fault energy (GSFE) landscape Specifically, for bilayer graphene, three nearest-neighbor minima surround a local minimum, suggesting a triangular network topology In contrast, for bilayer MoS2, six nearest-neighbor minima justify the hexagonal network topology The dimension of the soliton network is equal to the rank of F − I, which is determined by whether the transition matrix has full rank or not Inverse Design Procedure The inverse design problem seeks to identify the heterodeformation that produces a given strain soliton network The procedure involves starting with a desired strain soliton network described by line vector–Burgers vector pairs and seeking a unique heterodeformation This is achieved by relating the Nye tensor, which describes dislocation density, to the deformation gradient F through equation (19) The transition matrix Q can be obtained directly from the lattice coordinates of the vector pairs in equation (20) Application to Bilayer Systems The framework is applied to both bilayer graphene and bilayer MoS2, demonstrating its generality across different van der Waals homostructures For instance, the inverse design constructs a pure twist in bilayer graphene that results in a screw dislocation network Similarly, complex moir´e networks can be constructed by choosing non-integer rational values for line vectors to generate specific soliton topologies The algorithm provides the heterodeformation gradient F and the primitive simulation cell vectors l1 and l2, which are derived using Smith normal form bicrystallography The proposed framework establishes a geometric foundation for connecting GSFE landscapes, network topology, and heterodeformation It shifts the paradigm from forward construction to an inverse perspective where the target soliton network prescribes the heterodeformation This systematic approach offers a practical route for engineering interfacial structures with prescribed topology and symmetry The framework is general and applicable to arbitrary bilayer systems, enabling the rational design of moir´e systems with controlled structural and electronic characteristics Page 1 --- Strain soliton networks strongly influence the structural and electronic properties of heterodeformed bilayer systems, yet their design remains challenging due to the high dimensionality of heterodeformation space and the absence of a direct map between deformation and network geometry In this work, we introduce a geometric framework that establishes a one-to-one mapping between heterodeformations and the geometry of the strain soliton network expressed as line vector–Burgers vector pairs The admissible networks are constrained by topology dictated by the generalized stacking fault energy landscape We show that the moir´e Bravais lattice, corresponding to a uniform heterodeformation, alone is insufficient to characterize the interface distinct heterodeformations can share identical moir´e Bravais lattices while producing different soliton networks The soliton network encodes the full multilattice geometry of the interface, including topology and connectivity, which are not captured by the moir´e Bravais lattice alone The proposed framework enables the direct construction of heterodeformations from target networks, providing a systematic route for inverse design of moir´e interfaces beyond conventional twist-based approaches Page 2 --- In the era of micro- and nano-electromechanical devices, 2D materials have garnered significant attention from the scientific community due to their superior thermal, mechanical, and electronic properties

1–3: The observation of flat bands at magic angle twisted bilayer two-dimensional (2D) materials gives rise to interesting quantum properties such as quantum hall effect, Mott insulating effect, superconductivity, etc.

4, 5: which makes the use of heterodeformed1 bilayer 2D materials versatile for numerous applications

6, 7: A rigid twisted and/or heterostrained bilayer 2D material undergoes structural reconstruction where atoms are displaced from their positions in the rigidly heterodeformed configuration

8, 9: This reconstruction is mediated by the formation of an array of interface dislocations, also referred to as strain solitons

9–13: These solitons are characterized by their Burgers vector and line direction In twisted bilayer graphene, they typically organize into straight triangular networks

15: whereas heterostrained bilayers may exhibit distorted triangular

15, 16: or even spiral networks

17: The structure of each soliton depends on the alignment between its Burgers vector and line direction, imparting screw-like or edge-like character Page 3 --- The geometry of the soliton network is not merely a structural detail.

Mira: In continuum models of flat band physics

20–24: , the effective interlayer coupling is assumed to be periodic with respect to the 2D moir´e Bravais lattice generated by the imposed heterodeformation It does not specify the point group symmetry of the interlayer coupling Only when the moir´e is viewed in its entirety, i.e., the moir´e Bravais lattice along with the atoms of both layers interpreted as its basis atoms, do the internal shifts determine the point group symmetry of the interlayer coupling We will use the term moir´e lattice to refer to the entire multilattice As an example, Figure 1 shows three distinct heterodeformed configurations of bilayer graphene that share a common 2D moir´e Bravais lattice, whose primitive unit cell is shown by a white dashed parallelogram Although the three moir´e lattices have identical translational symmetries, they have distinct point group symmetries Consequently, two heterodeformations producing identical 2D moir´e Bravais lattices may nevertheless yield continuum Hamiltonians with distinct point group symmetries and band structures Moreover, anisotropic band structure observed in distorted, heterostrained bilayer configurations can be attributed to the strain-induced distortion of the moir´e lattice and the resulting modification of the moir´e Brillouin zone

25: Page 4 --- Heterodeformation is an umbrella term used to refer to relative twists and relative strains It is well established that heterodeformation controls the characteristic size and anisotropy of moir´e lattices in bilayer 2D materials Experiments and simulations show that increasing twist angle or heterostrain modifies domain-wall width and reduces the moir´e length scale

15, 17, 28: while numerous theoretical studies demonstrate that the moir´e lattice unit cell can be tuned continuously by adjusting the relative deformation However, these approaches constitute a forward-design rather than an inverse-design paradigm as their input is the imposed heterodeformation (e.g., heterostrain and/or twist), and the output that is usually emphasized is the 2D moir´e Bravais lattice Since the forward design is a many-to-one mapping—as demonstrated in Figure 1—it does not fully establish the relationship between the moir´e and heterodeformation, which is a key requirement for experiments

34, 2603.21446: Therefore, our goal is to develop an inverse design framework wherein the input is the target moir´e—fully described by not only the Bravais lattice vectors but also the soliton network—and the output is the heterodeformation We will show that the inverse design is a one-to-one mapping and, therefore, it is mathematically well-posed Page 5 --- Lowercase bold letters are used to denote vectors while uppercase bold letters are used to denote second-order tensors and matrices, unless stated otherwise The symbol · denotes the inner product of two vectors The Cartesian coordinates of a vector x are represented as a column matrix x = (x1,..., xn) T where T denotes transpose and the curl of a second-order tensor T is a tensor whose components are given by

curl(T): ij = ϵipqTjq,p The outer product of two vectors, denoted by a⊗b, is a tensor with components

a ⊗ b: ij = aibj An n-dimensional Bravais lattice is defined as A = A + x where x = Ax and x ∈ Z n (1) The volume of a primitive unit cell of A is equal to det A We use an uppercase calligraphic letter to denote a lattice; the same letter in bold denotes its structure matrix The lattice coordinates of a x ∈ R n are denoted by x = (x1,..., xn) T and given by x = A−1x If lattice A is subjected to a uniform deformation x′ = F x using a constant deformation gradient F, the resulting deformed lattice is denoted by F A Reciprocal lattice vectors of lattice A are the lattice vectors of its dual lattice given by A∗ = y where y = A−Ty and y ∈ Z n (2) whose structure matrix is A−T The dual lattice coordinates of y are denoted by y If x ∈ A and y ∈ A∗, then x · y is an integer We reserve the term lattice to represent a generic multilattice formed by a Bravais lattice A and a collection of basis atom positions (shift vectors) S = sα, i.e.

Kai: A + S:= A + sα where x ∈ R n, sα ∈ S (3) If S = 0, we recover the Bravais lattice Page 6 --- In this work, we focus on 2D homobilayers of graphene and MoS2, two prototypical systems that exhibit distinct soliton network topologies The structure matrices of the 2D triangular Bravais lattices of graphene and MoS2 are of the form A = a 2

0 −√3/2-1/sqrt(3): (4) Page 7 --- Strain Soliton AA Stacking AB Stacking BA Stacking Strain Soliton (a) bilayer graphene (b) bilayer MoS2 AA’ Stacking AB’ Stacking A’B Stacking Strain Soliton Page 8 --- (a) (b) Fig.

Title and authors: Mira: The paper "Inverse design of heterodeformations for strain soliton networks in bilayer 2D materials" is focused on building a systematic way to go from a desired network structure back to the physical deformation <ref:2603.21446#pg1,Inverse design of heterodeformations for strain soliton networks in bilayer 2D materials>.

Kai: And the authors, Tusher Ahmed and Nikhil Chandra Admal, they are tackling this by proposing a geometric framework that connects deformations to network geometry through line vector Burgers vector pairs.

Lev: I always find it interesting when someone tries to establish an inverse design problem like this instead of just a forward construction method.

Kai: They argue that the inverse design is actually a one-to-one mapping, which means it's mathematically well-posed, making the construction of moire interfaces with prescribed topology possible.

Mira: It’s shifting the paradigm away from just imposing a heterodeformation and seeing what network you get to one where you start with the target network.

Lev: That would be huge for experimental work, because right now, when we twist or strain things, we usually emphasize the resulting moire lattice more than the actual soliton configuration.

Kai: So, they're aiming to provide a systematic route for inverse design of moire interfaces beyond conventional twist-based approaches.

The paper's summary: Kai: To summarize, the paper introduces this geometric framework that allows you to construct heterodeformations directly from target networks.

Mira: They also show how the topology of a 2D soliton network is actually determined by counting how many nearest-neighbor minima surround a local minimum in the generalized stacking fault energy landscape <ref:2603.21446#pg1,the generalized stacking fault energy landscape>.

Lev: So, if you know the energy landscape features, you can predict whether you're looking at a triangular or hexagonal network before you even deform anything.

Kai: Specifically, they show that for bilayer graphene, three nearest-neighbor minima around a local minimum suggests a triangular network topology.

Mira: Whereas for bilayer MoS2, six nearest-neighbor minima justify the hexagonal network topology because of how the stacking fault energy landscape is structured there.

Lev: That links the fundamental energy landscape directly to the resulting structural arrangement of those interface dislocations.

The paper's improvements: Kai: One big improvement they suggest is using Smith normal form bicrystallography to compute a rational transition matrix Q directly from the lattice coordinates of those target soliton network vectors.

Mira: That bypasses what they see as error-prone numerical sequences by getting that transition matrix rational and mathematically well-posed for periodic boundary conditions.

Lev: For real hardware, I'd be worried about how much overhead that computation adds, but having a rational matrix is definitely a plus for simulation setup.

Kai: They also provide a tool within an AI framework that uses Smith normal form bicrystallography to determine the smallest simulation box vectors, l1 and l2, based on whether the input line vectors have integer lattice coordinates.

Mira: That ensures you correctly capture either a simple network or a complex network representation when setting up your simulation cell.

Lev: That's practical stuff; knowing the simulation cell size upfront saves a lot of time in setting up those large-scale simulations for error correction testing.

Conclusion: Kai: So, to wrap this up, the paper establishes a geometric foundation connecting the stacking fault energy landscapes, network topology, and heterodeformation.

Mira: They shift the approach from simply building forward to an inverse perspective where the target soliton network dictates the required heterodeformation.

Lev: It’s a systematic route for engineering interfacial structures with prescribed topology and symmetry, which is what we need when designing things for actual fabrication.

Kai: This framework is general and applicable to arbitrary bilayer systems, allowing for the rational design of moire systems with controlled structural and electronic characteristics.

Mira: The main implication is that we can now rationally design interfaces based on the network structure rather than just tuning a twist angle and hoping it works.

Lev: It’s a solid theoretical tool, but getting it into hardware runs is always the next hurdle we need to look at.

Kai: Well, that covers what they did here with "Inverse design of heterodeformations for strain soliton networks in bilayer 2D materials <ref:2603.21446#pg1,Inverse design of heterodeformations for strain soliton networks in bilayer 2D materials>." We'll be looking at how these network constraints affect the actual electronic properties in our next piece.

Md Tusher Ahmed, Nikhil Chandra Admal

Institute for Advanced Manufacturing, The University of Texas Rio Grande Valley · Department of Mechanical Science and Engineering, University of Illinois Urbana-Champaign

cond-mat.mtrl-sci

Submitted: 2026-03-22

Updated: 2026-03-22

DOI: 10.1038/s41699-026-00741-x

Code: https://github.com/admal-research-group/oILAB

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 99/100

The gist: Strain soliton networks strongly influence the structural and electronic properties of heterodeformed bilayer systems…

Key concepts

Strain Soliton Network
These are arrays of interface dislocations that form when bilayer 2D materials are heterostrained or twisted. They are characterized by their line direction and Burgers vector, which dictate the local structural distortion and influence properties like friction and ferroelectricity.
Moiré Bravais Lattice
This is the underlying periodic lattice structure formed by a uniform heterodeformation (like a twist). While it describes the overall periodicity, it does not capture all the topological details of how individual strain solitons are arranged.
Generalized Stacking Fault Energy (GSFE) Landscape
The GSFE landscape dictates the topology of the soliton network. The arrangement of nearest-neighbor minima in this energy map determines whether a system will naturally form a triangular or hexagonal network, linking microscopic energy states to macroscopic structural connectivity.

Terminology

Summary

Strain soliton networks strongly influence the structural and electronic properties of heterodeformed bilayer systems, yet their design remains challenging due to the high dimensionality of heterodeformation space and the absence of a direct map between deformation and network geometry In this work, we introduce a geometric framework that establishes a one-to-one mapping between heterodeformations and the geometry of the strain soliton network expressed as line vector–Burgers vector pairs The proposed framework enables the direct construction of heterodeformations from target networks, providing a systematic route for inverse design of moir´e interfaces beyond conventional twist-based approaches

Geometric Framework and Mappings

The core contribution of this research is the introduction of a geometric framework that establishes a one-to-one mapping between heterodeformations and the geometry of the strain soliton network expressed as line vector–Burgers vector pairs This framework constrains admissible networks by topology dictated by the generalized stacking fault energy landscape The authors demonstrate that while distinct heterodeformations can share identical moir´e Bravais lattices, they produce different soliton networks, reflecting an inherent many-to-one mapping when only translational symmetry is considered The soliton network is highlighted as encoding the full multilattice geometry of the interface, including topology and connectivity, which are not captured by the moir´e Bravais lattice alone

Topological Constraints from GSFE

The topology of a 2D soliton network is determined by the number of nearest-neighbor minima surrounding a local minimum in the generalized stacking fault energy (GSFE) landscape Specifically, for bilayer graphene, three nearest-neighbor minima surround a local minimum, suggesting a triangular network topology In contrast, for bilayer MoS2, six nearest-neighbor minima justify the hexagonal network topology The dimension of the soliton network is equal to the rank of F − I, which is determined by whether the transition matrix has full rank or not

Inverse Design Procedure

The inverse design problem seeks to identify the heterodeformation that produces a given strain soliton network The procedure involves starting with a desired strain soliton network described by line vector–Burgers vector pairs and seeking a unique heterodeformation This is achieved by relating the Nye tensor, which describes dislocation density, to the deformation gradient F through equation (19) The transition matrix Q can be obtained directly from the lattice coordinates of the vector pairs in equation (20)

Application to Bilayer Systems

The framework is applied to both bilayer graphene and bilayer MoS2, demonstrating its generality across different van der Waals homostructures For instance, the inverse design constructs a pure twist in bilayer graphene that results in a screw dislocation network Similarly, complex moir´e networks can be constructed by choosing non-integer rational values for line vectors to generate specific soliton topologies The algorithm provides the heterodeformation gradient F and the primitive simulation cell vectors l1 and l2, which are derived using Smith normal form bicrystallography

Conclusion

The proposed framework establishes a geometric foundation for connecting GSFE landscapes, network topology, and heterodeformation It shifts the paradigm from forward construction to an inverse perspective where the target soliton network prescribes the heterodeformation This systematic approach offers a practical route for engineering interfacial structures with prescribed topology and symmetry The framework is general and applicable to arbitrary bilayer systems, enabling the rational design of moir´e systems with controlled structural and electronic characteristics

--- Page 1 ---

Strain soliton networks strongly influence the structural and electronic properties of heterodeformed bilayer systems, yet their design remains challenging due to the high dimensionality of heterodeformation space and the absence of a direct map between deformation and network geometry In this work, we introduce a geometric framework that establishes a one-to-one mapping between heterodeformations and the geometry of the strain soliton network expressed as line vector–Burgers vector pairs The admissible networks are constrained by topology dictated by the generalized stacking fault energy landscape We show that the moir´e Bravais lattice, corresponding to a uniform heterodeformation, alone is insufficient to characterize the interface distinct heterodeformations can share identical moir´e Bravais lattices while producing different soliton networks The soliton network encodes the full multilattice geometry of the interface, including topology and connectivity, which are not captured by the moir´e Bravais lattice alone The proposed framework enables the direct construction of heterodeformations from target networks, providing a systematic route for inverse design of moir´e interfaces beyond conventional twist-based approaches

--- Page 2 ---

In the era of micro- and nano-electromechanical devices, 2D materials have garnered significant attention from the scientific community due to their superior thermal, mechanical, and electronic properties [1–3] The observation of flat bands at magic angle twisted bilayer two-dimensional (2D) materials gives rise to interesting quantum properties such as quantum hall effect, Mott insulating effect, superconductivity, etc. [4, 5] which makes the use of heterodeformed1 bilayer 2D materials versatile for numerous applications [6, 7] A rigid twisted and/or heterostrained bilayer 2D material undergoes structural reconstruction where atoms are displaced from their positions in the rigidly heterodeformed configuration [8, 9] This reconstruction is mediated by the formation of an array of interface dislocations, also referred to as strain solitons [9–13] These solitons are characterized by their Burgers vector and line direction In twisted bilayer graphene, they typically organize into straight triangular networks [15] whereas heterostrained bilayers may exhibit distorted triangular [15, 16] or even spiral networks [17] The structure of each soliton depends on the alignment between its Burgers vector and line direction, imparting screw-like or edge-like character

--- Page 3 ---

The geometry of the soliton network is not merely a structural detail. In continuum models of flat band physics [20–24], the effective interlayer coupling is assumed to be periodic with respect to the 2D moir´e Bravais lattice generated by the imposed heterodeformation It does not specify the point group symmetry of the interlayer coupling Only when the moir´e is viewed in its entirety, i. p. 1

Where the paper says it

In continuum models of flat band physics [20–24], the effective interlayer coupling is assumed to be periodic with respect to the 2D moir´e Bravais lattice generated by the imposed heterodeformation. In other words, the 2D moir´e Bravais lattice determines the translational symmetry of the interlayer coupling.

e., the moir´e Bravais lattice along with the atoms of both layers interpreted as its basis atoms, do the internal shifts determine the point group symmetry of the interlayer coupling We will use the term moir´e lattice to refer to the entire multilattice As an example, Figure 1 shows three distinct heterodeformed configurations of bilayer graphene that share a common 2D moir´e Bravais lattice, whose primitive unit cell is shown by a white dashed parallelogram Although the three moir´e lattices have identical translational symmetries, they have distinct point group symmetries Consequently, two heterodeformations producing identical 2D moir´e Bravais lattices may nevertheless yield continuum Hamiltonians with distinct point group symmetries and band structures Moreover, anisotropic band structure observed in distorted, heterostrained bilayer configurations can be attributed to the strain-induced distortion of the moir´e lattice and the resulting modification of the moir´e Brillouin zone [25]

--- Page 4 ---

Heterodeformation is an umbrella term used to refer to relative twists and relative strains It is well established that heterodeformation controls the characteristic size and anisotropy of moir´e lattices in bilayer 2D materials Experiments and simulations show that increasing twist angle or heterostrain modifies domain-wall width and reduces the moir´e length scale [15, 17, 28] while numerous theoretical studies demonstrate that the moir´e lattice unit cell can be tuned continuously by adjusting the relative deformation However, these approaches constitute a forward-design rather than an inverse-design paradigm as their input is the imposed heterodeformation (e.g., heterostrain and/or twist), and the output that is usually emphasized is the 2D moir´e Bravais lattice Since the forward design is a many-to-one mapping—as demonstrated in Figure 1—it does not fully establish the relationship between the moir´e and heterodeformation, which is a key requirement for experiments [34, 2603. p. 2

Where the paper says it

, heterostrain and/or twist), and the output that is usually emphasized is the 2D moir´e Bravais lattice. Since the forward design is a many-to-one mapping—as demonstrated in Figure 1—it does not fully establish the relationship between the moir´e and heterodeformation, which is a key requirement for experiments [34, 35].

p. 2
Where the paper says it

, heterostrain and/or twist), and the output that is usually emphasized is the 2D moir´e Bravais lattice. Since the forward design is a many-to-one mapping—as demonstrated in Figure 1—it does not fully establish the relationship between the moir´e and heterodeformation, which is a key requirement for experiments [34, 35].

21446] Therefore, our goal is to develop an inverse design framework wherein the input is the target moir´e—fully described by not only the Bravais lattice vectors but also the soliton network—and the output is the heterodeformation We will show that the inverse design is a one-to-one mapping and, therefore, it is mathematically well-posed

--- Page 5 ---

Lowercase bold letters are used to denote vectors while uppercase bold letters are used to denote second-order tensors and matrices, unless stated otherwise The symbol · denotes the inner product of two vectors The Cartesian coordinates of a vector x are represented as a column matrix x = (x1,..., xn) T where T denotes transpose and the curl of a second-order tensor T is a tensor whose components are given by [curl(T)]ij = ϵipqTjq,p The outer product of two vectors, denoted by a⊗b, is a tensor with components [a ⊗ b]ij = aibj An n-dimensional Bravais lattice is defined as A = A + x where x = Ax and x ∈ Z n (1) The volume of a primitive unit cell of A is equal to det A We use an uppercase calligraphic letter to denote a lattice; the same letter in bold denotes its structure matrix The lattice coordinates of a x ∈ R n are denoted by x = (x1,. p. 3

Where the paper says it

The curl of a second-order tensor T is a tensor whose components are given by [curl(T )]ij = ϵipqTjq,p. The outer product of two vectors, denoted by a⊗b, is a tensor with components [a ⊗ b]ij = aibj .

.., xn) T and given by x = A−1x If lattice A is subjected to a uniform deformation x′ = F x using a constant deformation gradient F, the resulting deformed lattice is denoted by F A Reciprocal lattice vectors of lattice A are the lattice vectors of its dual lattice given by A∗ = y where y = A−Ty and y ∈ Z n (2) whose structure matrix is A−T The dual lattice coordinates of y are denoted by y If x ∈ A and y ∈ A∗, then x · y is an integer We reserve the term lattice to represent a generic multilattice formed by a Bravais lattice A and a collection of basis atom positions (shift vectors) S = sα, i. p. 3
Where the paper says it

If lattice A is subjected to a uniform deformation x ′ = F x using a constant deformation gradient F, the resulting deformed lattice is denoted by F A. Reciprocal lattice vectors of lattice A are the lattice vectors of its dual lattice given by A∗ = y ∈ R n , such that y = A−Ty and y ∈ Z n , (2) whose structure matrix is A−T.

e. A + S:= A + sα where x ∈ R n, sα ∈ S (3) If S = 0, we recover the Bravais lattice

--- Page 6 ---

In this work, we focus on 2D homobilayers of graphene and MoS2, two prototypical systems that exhibit distinct soliton network topologies The structure matrices of the 2D triangular Bravais lattices of graphene and MoS2 are of the form A = a 2 [0 −√3/2-1/sqrt(3)] (4)

--- Page 7 ---

Strain Soliton

AA Stacking AB Stacking BA Stacking Strain Soliton (a) bilayer graphene (b) bilayer MoS2 AA’ Stacking AB’ Stacking A’B Stacking Strain Soliton

--- Page 8 ---

(a) (b) Fig. 1: Three heterodeformations of bilayer graphene, resulting in identical 2D moir´e Bravais lattices The three configurations, however, have distinct point group symmetries Structural relaxation of the three configurations in Figure 1 yields distinct soliton networks that inherit the translational and point group symmetries of the moir´e lattice In addition to the band structure, the soliton network also plays an important role in superlubricity and ferroelectricity The frictional response of a heterodeformed bilayer is determined by the mobility of interface dislocations [13] which depends on the Burgers vector, line direction, and network connectivity In ferroelectric systems like bilayer hBN, domain transitions under applied electric field are mediated by the deformation (bending/swirling) of interface dislocations [26, 27] which depends on the geometry of interface dislocations

--- Page 9 ---

It is well established that heterodeformation controls the characteristic size and anisotropy of moir´e lattices in bilayer 2D materials Experiments and simulations show that increasing twist angle or heterostrain modifies domain-wall width and reduces the moir´e length scale [15, 17, 28] while numerous theoretical studies demonstrate that the moir´e lattice unit cell can be tuned continuously by adjusting the relative deformation In particular, recent works [32, 33] treat heterodeformation as a control parameter for generating moir´e lattices However, these approaches constitute a forward-design rather than an inverse-design paradigm as their input is the imposed heterodeformation (e.g., heterostrain and/or twist), and the output that is usually emphasized is the 2D moir´e Bravais lattice Since the forward design is a many-to-one mapping—as demonstrated in Figure 1—it does not fully establish the relationship between the moir´e and heterodeformation, which is a key requirement for experiments [34, 2603.21446] Therefore, our goal is to develop an inverse design framework wherein the input is the target moir´e—fully described by not only the Bravais lattice vectors but also the soliton network—and the output is the heterodeformation We will show that the inverse design is a one-to-one mapping and, therefore, it is mathematically well-posed

--- Page 10 ---

Lowercase bold letters are used to denote vectors while uppercase bold letters are used to denote second-order tensors and matrices, unless stated otherwise The symbol · denotes the inner product of two vectors The Cartesian coordinates of a vector x are represented as a column matrix x = (x1,..., xn) T where T denotes transpose and the curl of a second-order tensor T is a tensor whose components are given by [curl(T)]ij = ϵipqTjq,p The outer product of two vectors, denoted by a⊗b is a tensor with components [a ⊗ b]ij = aibj An n-dimensional Bravais lattice is defined as A = A + x where x = Ax and x ∈ Z n (1) The volume of a primitive unit cell of A is equal to det A We use an uppercase calligraphic letter to denote a lattice; the same letter in bold denotes its structure matrix The lattice coordinates of a x ∈ R n are denoted by x = (x1,. p. 3

Where the paper says it

The curl of a second-order tensor T is a tensor whose components are given by [curl(T )]ij = ϵipqTjq,p. The outer product of two vectors, denoted by a⊗b, is a tensor with components [a ⊗ b]ij = aibj .

.., xn) T and given by x = A−1x If lattice A is subjected to a uniform deformation x′ = F x using a constant deformation gradient F, the resulting deformed lattice is denoted by F A Reciprocal lattice vectors of lattice A are the lattice vectors of its dual lattice given by A∗ = y where y = A−Ty and y ∈ Z n (2) whose structure matrix is A−T If x ∈ A and y ∈ A∗, then x · y is an integer We reserve the term lattice to represent a generic multilattice formed by a Bravais lattice A and a collection of basis atom positions (shift vectors) S = sα, i. p. 3
Where the paper says it

If lattice A is subjected to a uniform deformation x ′ = F x using a constant deformation gradient F, the resulting deformed lattice is denoted by F A. Reciprocal lattice vectors of lattice A are the lattice vectors of its dual lattice given by A∗ = y ∈ R n , such that y = A−Ty and y ∈ Z n , (2) whose structure matrix is A−T.

e. A + S:= A + sα where x ∈ R n, sα ∈ S (3) If S = 0, we recover the Bravais lattice

--- Page 11 ---

In this work, we focus on 2D homobilayers of graphene and MoS2, two prototypical systems that exhibit distinct soliton network topologies The structure matrices of the 2D triangular Bravais lattices of graphene and MoS2 are of the form A = a 2 [0 −√3/2-1/sqrt(3)] (4)

--- Page 15 ---

Strain Soliton

AA Stacking AB Stacking BA Stacking Strain Soliton

--- Page 16 ---

(a) bilayer graphene (b) bilayer MoS2 AA’ Stacking AB’ Stacking A’B Stacking Strain Soliton

--- Page 17 ---

Strain Soliton

--- Page 18 ---

(a) (b) Fig.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Develop an inverse design system that takes a target strain soliton network, described by line vector-Burgers vector pairs, as input and outputs a heterodeformation gradient F that produces this network upon structural relaxation. This allows for the direct construction of heterodeformations from target networks, providing a systematic route for inverse design of moire interfaces beyond conventional twist-based approaches.

  2. Implement an AI system capable of determining the periodicity and connectivity of strain soliton networks in bilayer 2D materials by analyzing the generalized stacking fault energy landscape (GSFE). The system can map GSFE extrema to network topology, allowing it to predict whether a network will be triangular or hexagonal based on the number of nearest-neighbor minima.

  3. Create an AI module that utilizes Smith normal form bicrystallography to compute rational transition matrices Q directly from the lattice coordinates of the target soliton network vectors. This bypasses error-prone numerical sequences, ensuring that the resulting transition matrix is rational and mathematically well-posed for constructing moire interfaces with periodic boundary conditions (PBCs).

  4. Design a simulation cell generation tool within an AI framework that uses SNF bicrystallography to determine the smallest simulation box vectors l1 and l2 based on whether the input line vectors have integer lattice coordinates, ensuring either a simple network or a complex network representation is correctly captured.

  5. Build an AI model that predicts the resulting continuum Hamiltonian point group symmetry from a given heterodeformation gradient F by analyzing how it relates to the moir´e Bravais lattice and its basis atoms, distinguishing between configurations that share identical moir´e lattices but have distinct point group symmetries.

Abstract

Strain soliton networks strongly influence the structural and electronic properties of heterodeformed bilayer systems, yet their design remains challenging due to the high dimensionality of heterodeformation space and the absence of a direct map between deformation and network geometry. In this work, we introduce a geometric framework that establishes a one-to-one mapping between heterodeformations and the geometry of the strain soliton network expressed as line vector-Burgers vector pairs. The admissible networks are constrained by topology dictated by the generalized stacking fault energy landscape. We show that the moiré Bravais lattice, corresponding to a uniform heterodeformation, alone is insufficient to characterize the interface: distinct heterodeformations can share identical moiré Bravais lattices while producing different soliton networks, reflecting an inherent many-to-one mapping when only translational symmetry is considered. In contrast, the soliton network encodes the full multilattice geometry of the interface, including topology and connectivity, which are not captured by the moiré Bravais lattice alone. The proposed framework enables the direct construction of heterodeformations from target networks, providing a systematic route for inverse design of moiré interfaces beyond conventional twist-based approaches.

Sources

Related papers