Statements and Declarations The authors do not declare financial or non-financial interests that are directly or indirectly related to the work submitted for publication.
Data availability The manuscript has no associated data.
Acknowledgments MH’s and ML’s research was partially supported by Simons Targeted Grant Award No. 896630. AW’s, and ML’s research was supported in part by NSF DMREF Award No. 1922165. ML’s research was also supported in part by grant NSF PHY-2309135 to the Kavli Institute for Theoretical Physics (KITP). AW’s research was also supported in part by grant NSF DMS-2406981. The authors would like to thank Ziyan (Zoe) Zhu for helpful comments and stimulating discussions.
Introduction
Generally, phonons are understood as generalized eigenvectors of a stability operator acting on ℓ 2 \ell^2 ℓ 2 , associated with a mechanical energy. More precisely, expanding a mechanical energy functional w.r.t. equilibrium positions { R e q j } j ∈ J \{R_{\mathrm{eq}}^j\}_{j\in\mathcal{J}} { R eq j } j ∈ J , perturbed by lattice deformations δ u : { R e q j } j ∈ J → R d \delta u:\{R_{\mathrm{eq}}^j\}_{j\in\mathcal{J}}\to\mathbb{R}^d δ u : { R eq j } j ∈ J → R d ,
E ( { R e q j + δ u ( R e q j ) } j ∈ J ) = E ( { R e q j } j ∈ J ) + δ 2 E ( { R e q j } j ∈ J ) [ δ u , δ u ] + ⋯ , E\left(\{R_{\mathrm{eq}}^j+\delta u(R_{\mathrm{eq}}^j)\}_{j\in\mathcal{J}}\right)=E\left(\{R_{\mathrm{eq}}^j\}_{j\in\mathcal{J}}\right)+\delta^2E\left(\{R_{\mathrm{eq}}^j\}_{j\in\mathcal{J}}\right)[\delta u,\delta u]+\cdots, E ( { R eq j + δ u ( R eq j ) } j ∈ J ) = E ( { R eq j } j ∈ J ) + δ 2 E ( { R eq j } j ∈ J ) [ δ u , δ u ] + ⋯ ,
we define phonons as generalized eigenvectors of the stability operator δ 2 E ( { R e q j } j ∈ J ) \delta^2E\left(\{R_{\mathrm{eq}}^j\}_{j\in\mathcal{J}}\right) δ 2 E ( { R eq j } j ∈ J ) acting on ℓ 2 \ell^2 ℓ 2 . Crucially, this requires to define an energy functional yielding a well-defined stability operator acting on ℓ 2 \ell^2 ℓ 2 . Notice that δ 2 E ( { R e q j } j ∈ J ) \delta^2E\left(\{R_{\mathrm{eq}}^j\}_{j\in\mathcal{J}}\right) δ 2 E ( { R eq j } j ∈ J ) is necessarily positive (semi-)definite.
In the case of periodic media described by a (multi)lattice, the unperturbed lattice positions can be shown to be stable under global ℓ 2 \ell^2 ℓ 2 or local perturbations. An energy functional, in the simplest case, is given by the sum of harmonic springs. Then the associated phonon Hamiltonian can be computed as the root of the second variation of the mechanical energy. This is due to the fact that, in this case, we model lattice vibrations via harmonic oscillators at every lattice position. Due to the lattice periodicity, Bloch-Floquet theory enables us to analyze the spectral properties of the associated phonons. More precisely, in this case, we define the phonon Hamiltonian as the (square root of) the stability operator associated with the energy functional.
In general aperiodic media, however, the story is not as straight-forward. First, we need to study a mechanical energy to obtain a relaxed lattice configuration. This relaxed lattice configuration will generally not be periodic, so Bloch-Floquet theory is not applicable. Moreover, while it is possible to again model the mechanical energy via harmonic springs, analytically or computationally determining the associated energy minimum is generally challenging. Our goal is thus twofold:
Determine an analytically/computationally accessible mechanical energy to compute an energy minimizer.
Find an appropriate phonon model that allows us to treat phonons as ℓ 2 \ell^2 ℓ 2 -waves.
In order to analyze this question, we shall turn to incommensurate heterostructures. These arise when two or more two-dimensional crystal layers are stacked on top of each other at a relative twist, forming a quasiperiodic medium. Due to the periodicity of the individual layers, quasiperiodic macroscopic patterns, referred to as moiré pattern , emerge. Such systems have gained a lot of attention, when it turned out that, at certain small twist angles, known as magic angles , special electronic properties are unlocked. More precisely, Bistritzer and MacDonald [5 ] derived a reduced model for twisted bilayer graphene that predicted the occurrence of nearly flat electronic bands at a twist angle of ∼ 1.1 ∘ \sim1.1^\circ ∼ 1. 1 ∘ , which in turn yields a highly localized density of states. As a consequence of the flat dispersion, the interaction between electrons become dominant and various behavior typical for strongly correlated systems has been observed to emerge, see [6 ] . One of the mysterious phases is the superconducting phase, for it occurs at a very low electron density compared to the transition temperature, which is rather typical for unconventional superconductors. However, the mechanism underlying the superconducting phase remains a puzzle.
Spectroscopic measurements [31 ] seem to hint at unconventional superconductivity, while electronic transport [24 ] shows independence of the critical temperature w.r.t. Coulomb screening. This, in turn, suggests that phonons may play a critical role in understanding superconductivity.
One of the main obstacles in understanding the role of phonons, is the large number of – even intersecting – phonon bands. Physicists have thus come up with certain selection rules to pick only a few bands and projected them onto a low-energy regime, similar as previously done for electrons by Bistritzer and MacDonald, see [22 , 25 ] . In addition, these computations cutoff the phonon energy at zero and ignore possible phonon bands with negative dispersion, see, e.g., [26 , 39 ] . Our goal is to provide a mathematical foundation for phonons in incommensurate multilayer structures, and to study whether or not this negative dispersion is a model relic or if it could be avoided by adjusting the model.
In order to find an energy functional, we can generally study (at least) two models
Minimize a global energy w.r.t. ℓ 2 \ell^2 ℓ 2 -perturbations.
Minimize a local energy in any finite subset of the medium.
The first approach has the advantage that we would like to model phonons as ℓ 2 \ell^2 ℓ 2 -normalized waves propagating through an aperiodic medium. However, the main disadvantage here is that, in order to construct a relaxed configuration, one typically constrains the problem to a supercell, corresponding, e.g., to a moiré cell, and chooses, e.g., periodic boundary conditions, see [13 , 14 , 38 ] . This is due to the fact that displacement functions are parametrized by their discrete lattice positions and continuum approximations that are commonly used for periodic systems, such as the Cauchy-Born approximation, are not accessible to the generic aperiodicity. Computations over large supercells are challenging, as they scale with O θ → 0 ( 1 θ 2 ) O_{\theta\to0}\left(\frac{1}{\theta^2}\right) O θ → 0 ( θ 2 1 ) for twist angles θ \theta θ .
In order to minimize local energies in a quasiperiodic medium, Aubry and Le Daeron [1 ] studied the Frenkel-Kontorova model. In this model, lattice displacements on Z \mathbb{Z} Z are modeled via harmonic springs, and an external incommensurate potential is added. Then, under the assumption of an additional recurrence condition, it was shown that a minimizer can be parametrized by a periodic function with the same periodicity as the external potential. This hull function, see also [2 , 3 , 4 ] , is either continuous, when the external potential is very weak, or discontinuous w.r.t. a Cantorus when the external potential is very strong. Analogous results were obtained for 1D rippling chains [10 ] by applying Aubry-Mather theory.
Motivated by these 1D results, Cazeaux, Massatt and one of the authors [9 ] , see also [7 ] , developed an energy density for continuous functions that are periodic w.r.t. to the local stacking configuration . We shall note that establishing a two- and higher-dimensional Aubry-Mather theory remains an open problem. In particular, this approach introduces a third model for an energy functional:
Minimize an energy density w.r.t. stacking-periodic displacements.
From the previous discussion, we have that such a model replaces the second model above in the case of the 1D Frenkel-Kontorova model, while, in 2D, it is a distinct model.
The main advantage of this approach is that, instead of sampling over supercells, it suffices to sample over stacking configurations, making the minimization problem analytically and computationally more accessible; the predicted results are in agreement with the experimental observations [8 ] . However, since this energy functional is derived for energy minimizers only, and since these are assumed to be periodic, and thus non-decaying in ℓ 2 \ell^2 ℓ 2 , we cannot treat phonons in the same framework.
In order to pass from a model based on stacking-periodic perturbations to ℓ 2 \ell^2 ℓ 2 -perturbations, we employ an idea pursued in [17 ] . There, a theory is developed to study energy minimizers for interacting Dirac particles. They start by constructing a translation-invariant minimizer in the absence of an electric background field for which the energy density is stable w.r.t. translation-invariant perturbations . Using this minimizer as a reference state, they then consider the offset energy and minimize it w.r.t. Hilbert-Schmidt perturbations. These two minimizers are shown to coincide.
In this work, we adopt these ideas in the following way: We start by recalling results established [18 ] , where we generalized the ideas in [9 ] to obtain an atomistic energy based on displacements that are periodic w.r.t. the local stacking configuration. Crucially, this energy functional is obtained via a thermodynamic limit, analogously to [17 ]
e ( t o t ) ( u ) : = lim r → ∞ E ∣ B r ( u ) ∣ B r ∣ . e^{(\mathrm{tot})}(\mathbf{u}) := \lim_{r \to\infty} \frac{E\big|_{B_r}(\mathbf{u})}{|B_r|}. e ( tot ) ( u ) := r → ∞ lim ∣ B r ∣ E B r ( u ) .
We also establish conditions on the many-body potentials that allow us to employ the results in [18 ] . For more details, we refer to § 2.6 for the assumptions on the potentials, and to § 3 to show well-definedness of the energy functional.
In § 4 , we continue to prove the existence a locally unique minimizer u ( e q ) \mathbf{u}^{(\mathrm{eq})} u ( eq ) of e ( t o t ) e^{(\mathrm{tot})} e ( tot ) which is stable w.r.t. perturbations that are periodic w.r.t. the local environment. This minimizer lets us compute the offset energy w.r.t. the total energy of u ( e q ) \mathbf{u}^{(\mathrm{eq})} u ( eq )
Δ E u ( e q ) ( t o t ) ( δ u ) : = lim r → ∞ ( E ∣ B r ( u ( e q ) + δ u ) − E ∣ B r ( u ( e q ) ) ) . \Delta E_{\mathbf{u}^{(\mathrm{eq})}}^{(\mathrm{tot})}(\delta\mathbf{u}) := \lim_{r\to\infty}\left(E\big|_{B_r}\left(\mathbf{u}^{(\mathrm{eq})}+\delta\mathbf{u}\right)-E\big|_{B_r}\left(\mathbf{u}^{(\mathrm{eq})}\right)\right). Δ E u ( eq ) ( tot ) ( δ u ) := r → ∞ lim ( E B r ( u ( eq ) + δ u ) − E B r ( u ( eq ) ) ) .
Definition of phonon Hamiltonian A natural question to ask is whether u ( e q ) \mathbf{u}^{(\mathrm{eq})} u ( eq ) is still a minimizer for the offset energy, i.e., whether 0 0 0 is a minimizer of Δ E u ( e q ) ( t o t ) \Delta E_{\mathbf{u}^{(\mathrm{eq})}}^{(\mathrm{tot})} Δ E u ( eq ) ( tot ) . Assuming sufficient regularity of the involved potentials, expanding in orders of δ u \delta\mathbf{u} δ u yields that an appropriate space to study Δ E ( t o t ) \Delta E^{(\mathrm{tot})} Δ E ( tot ) is given by ℓ 1 \ell^1 ℓ 1 , see 2.5 below.
The goal of this work is thus to study when the minimization problem associated with e ( t o t ) e^{(\mathrm{tot})} e ( tot ) is consistent with that associated with Δ E u ( e q ) ( t o t ) \Delta E_{\mathbf{u}^{(\mathrm{eq})}}^{(\mathrm{tot})} Δ E u ( eq ) ( tot ) . When it is, we can rigorously pass from the configuration space framework to the ℓ 2 \ell^2 ℓ 2 -phonon framework, and the phonon Hamiltonian is defined as
H p h : = δ 2 Δ E u ( e q ) ( t o t ) ( 0 ) : ℓ 2 → ℓ 2 . (1) H_{\mathrm{ph}} := \sqrt{\delta^2\Delta E_{\mathbf{u}^{(\mathrm{eq})}}^{(\mathrm{tot})}(0)}:\ell^2\to\ell^2. \tag*{(1)} H ph := δ 2 Δ E u ( eq ) ( tot ) ( 0 ) : ℓ 2 → ℓ 2 . (1)
As we will see, we cannot employ standard perturbative methods and, instead, will provide sufficient criteria to ensure or exclude phonon stability. These criteria need to be numerically verified.
We prove in Section 2.5 that
δ W ~ p e r 1 , 2 e ( t o t ) ( u ( e q ) ) = 0 ⇒ δ ℓ 1 Δ E u ( e q ) ( t o t ) ( 0 ) = 0 , \delta\widetilde{W}_{\mathrm{per}}^{1,2}e^{(\mathrm{tot})}\left(\mathbf{u}^{(\mathrm{eq})}\right)=0
\quad\Rightarrow\quad
\delta^{\ell^1}\Delta E_{\mathbf{u}^{(\mathrm{eq})}}^{(\mathrm{tot})}(0)=0, δ W per 1 , 2 e ( tot ) ( u ( eq ) ) = 0 ⇒ δ ℓ 1 Δ E u ( eq ) ( tot ) ( 0 ) = 0 ,
i.e., that a critical point of e u ( e q ) ( t o t ) e_{\mathbf{u}^{(\mathrm{eq})}}^{(\mathrm{tot})} e u ( eq ) ( tot ) under moiré-periodic perturbations remains a critical point of Δ E ( t o t ) \Delta E^{(\mathrm{tot})} Δ E ( tot ) under ℓ 1 \ell^1 ℓ 1 -perturbations. In particular, this means that δ 2 Δ E u ( e q ) ( t o t ) ( 0 ) : ( ℓ 2 ) 2 → R \delta^2\Delta E_{\mathbf{u}^{(\mathrm{eq})}}^{(\mathrm{tot})}(0):(\ell^2)^2\to\mathbb{R} δ 2 Δ E u ( eq ) ( tot ) ( 0 ) : ( ℓ 2 ) 2 → R is well-defined. Equivalently, by duality, δ 2 Δ E u ( e q ) ( t o t ) ( 0 ) ℓ 2 → ℓ 2 \delta^2\Delta E_{\mathbf{u}^{(\mathrm{eq})}}^{(\mathrm{tot})}(0)\ell^2\to\ell^2 δ 2 Δ E u ( eq ) ( tot ) ( 0 ) ℓ 2 → ℓ 2 is well-defined. We refer to the Section 2 for the chosen notation, and to the respective sections for the precise statements.
The next question thus is whether stability of e ( t o t ) e^{(\mathrm{tot})} e ( tot ) is inherited by Δ E u ( e q ) ( t o t ) \Delta E_{\mathbf{u}^{(\mathrm{eq})}}^{(\mathrm{tot})} Δ E u ( eq ) ( tot ) . To answer this question, we start in Section 5 by proving that, for certain choices of interlayer pair potentials
δ 2 e ( t o t ) ( 0 ) ≥ κ > 0 ∧ δ 2 ∣ δ u = 0 Δ E 0 ( t o t ) ( δ u ) < 0. \delta^2e^{(\mathrm{tot})}(0)\ge\kappa>0
\quad\wedge\quad
\left.\delta^2\right|_{\delta\mathbf{u}=0}\Delta E_0^{(\mathrm{tot})}(\delta\mathbf{u})<0. δ 2 e ( tot ) ( 0 ) ≥ κ > 0 ∧ δ 2 δ u = 0 Δ E 0 ( tot ) ( δ u ) < 0.
This means that the undeformed union of lattices is stable under periodic perturbations, while it is not under ℓ 2 \ell^2 ℓ 2 -perturbations. It matches the intuition that we expect the undeformed lattice not to be stable. The condition δ 2 e ( t o t ) ( 0 ) ≥ κ \delta^2e^{(\mathrm{tot})}(0)\ge\kappa δ 2 e ( tot ) ( 0 ) ≥ κ is typically referred to as stability of the lattice . In the case of a single layer, this has been shown to be a necessary condition for stability of minimizers, see [33 Appendix C] . Including the relaxed lattice positions, we prove that, in the case of pair interlayer potentials v v v ,
inf ∥ δ v ∥ ℓ 2 = 1 δ 2 ∣ δ u = 0 Δ E u ( e q ) ( t o t ) ( δ u ) [ δ v , δ v ] ≤ inf a ∈ S 2 ∫ R 2 d x ∂ a 2 v ( x + u 2 ( e q ) ( A M A 1 − 1 x ) − u 1 ( e q ) ( A M A 2 − 1 x ) ) . \inf_{\|\delta v\|_{\ell^2}=1}\left.\delta^2\right|_{\delta\mathbf{u}=0}\Delta E_{\mathbf{u}^{(\mathrm{eq})}}^{(\mathrm{tot})}(\delta\mathbf{u})[\delta v,\delta v]
\le
\inf_{a\in S^2}\int_{\mathbb{R}^2} \mathrm{d}x\,\partial_a^2v\left(x+u_2^{(\mathrm{eq})}\left(A_{\mathcal{M}}A_1^{-1}x\right)-u_1^{(\mathrm{eq})}\left(A_{\mathcal{M}}A_2^{-1}x\right)\right). ∥ δ v ∥ ℓ 2 = 1 inf δ 2 δ u = 0 Δ E u ( eq ) ( tot ) ( δ u ) [ δ v , δ v ] ≤ a ∈ S 2 inf ∫ R 2 d x ∂ a 2 v ( x + u 2 ( eq ) ( A M A 1 − 1 x ) − u 1 ( eq ) ( A M A 2 − 1 x ) ) .
Thus, a sufficient condition for instability is a negative r.h.s. This instability criterion stems from constructing an approximate minimizer for which the monolayer phonon energy vanishes while the interlayer phonon energy is minimized.
Conversely, a sufficient condition for phonon stability is given by
∑ R 1 ∈ R 1 D 2 v ( x − R 1 + u 2 ( e q ) ( A M A 1 − 1 x ) − u 1 ( e q ) ( R 1 + A M A 2 − 1 x ) ) ≥ 0 (2) \sum_{R_1\in\mathcal{R}_1} D^2v\left(x-R_1+u_2^{(\mathrm{eq})}\left(A_MA_1^{-1}x\right)-u_1^{(\mathrm{eq})}\left(R_1+A_MA_2^{-1}x\right)\right)\ge0 \tag*{(2)} R 1 ∈ R 1 ∑ D 2 v ( x − R 1 + u 2 ( eq ) ( A M A 1 − 1 x ) − u 1 ( eq ) ( R 1 + A M A 2 − 1 x ) ) ≥ 0 (2)
for all x ∈ Γ 1 x\in\Gamma_1 x ∈ Γ 1 , in the sense of quadratic forms. Here, R 1 \mathcal{R}_1 R 1 refers to the lattice of layer 1, and Γ 1 \Gamma_1 Γ 1 is the corresponding unit cell. More precisely, we show in § 5 that this condition implies positive (semi-)definiteness of δ 2 Δ E u ( e q ) ( t o t ) \delta^2\Delta E_{u^{(\mathrm{eq})}}^{(\mathrm{tot})} δ 2 Δ E u ( eq ) ( tot ) . While we explain the ideas only for interlayer pair potentials, we believe that one can generalize them to, more general, empirical many-body potentials which have become computationally more accessible and which promise to yield yield highly accurate calculations, thanks to the advances made in machine-learned empirical many-body, see, e.g., [11 , 28 , 36 ] .
Relating the present energy functional to an appropriate Allen-Cahn functional, we argue in § 6 that, for a purely twisted bilayer system for which the layer coupling is mediated through a pair potential v v v , at (twist angles) → 0 \to0 → 0 , the Allen-Cahn minimizers converges to the potential wells of the misfit energy Φ 0 ( x ) : = ∑ n ∈ Z 2 v ( A ( x − n ) ) det A \Phi_0(x):=\sum_{n\in\mathbb{Z}^2}\frac{v(A(x-n))}{\det A} Φ 0 ( x ) := ∑ n ∈ Z 2 d e t A v ( A ( x − n )) . For the Allen-Cahn theory to be applicable, we need to require that at the potential wells x A B , x B A x_{\mathrm{AB}},x_{\mathrm{BA}} x AB , x BA , we have that
D 2 Φ 0 ( x B A / A B ) > 0. (3) D^2\Phi_0(x_{\mathrm{BA/AB}})>0. \tag*{(3)} D 2 Φ 0 ( x BA/AB ) > 0. (3)
This, in turn, means that we need to establish strict inequality in (2) . We establish (1.3) for 2D pair potentials for graphene, at the equilibrium distance $d_{\mathrm{BG}}=3.35$ Å between the layers. Our choices of pair potentials are Lennard-Jones, Morse and Kolmogorov-Crespi potentials [21 ] , see {xref}Section 6.2. Consequently, in the limit of zero twist angle and for sufficiently small lattice constants, in the sense of a continuum limit, the phonon Hamiltonian $H_{\mathrm{ph}}\ge 0$, see {xref}(1.1)`, is well-defined.
Preliminaries and notation
Lattices
For j = 1 , 2 j=1,2 j = 1 , 2 , let A j ∈ G L 2 ( R ) A_j\in\mathrm{GL}_2(\mathbb{R}) A j ∈ GL 2 ( R ) be the matrix with columns consisting of the primitive translation vectors associated with layer j j j , with layer unit cells Γ j : = A j [ 0 , 1 ) 2 \Gamma_j:=A_j[0,1)^2 Γ j := A j [ 0 , 1 ) 2 . We define the direct lattices and their truncations
R j : = A j Z 2 , R j ( N ) : = A j ( Z 2 ∩ [ − N , N ] 2 ) . (4) \mathcal{R}_j:=A_j\mathbb{Z}^2,\qquad\mathcal{R}_j(N):=A_j\left(\mathbb{Z}^2\cap[-N,N]^2\right). \tag*{(4)} R j := A j Z 2 , R j ( N ) := A j ( Z 2 ∩ [ − N , N ] 2 ) . (4)
Moreover, we define the reciprocal primitive translation vectors and the associated reciprocal lattice
B j : = 2 π A j − T , R j ∗ : = B j Z 2 . B_j := 2\pi A_j^{-T}, \qquad\mathcal{R}_j^* := B_j\mathbb{Z}^2. B j := 2 π A j − T , R j ∗ := B j Z 2 .
Next, we define the primitive moiré translation vectors and the corresponding reciprocal vectors
A M : = ( A 1 − 1 − A 2 − 1 ) − 1 , B M : = 2 π A M − T = B 1 − B 2 . A_M := \left(A_1^{-1}-A_2^{-1}\right)^{-1}, \qquad B_M := 2\pi A_M^{-T}=B_1-B_2. A M := ( A 1 − 1 − A 2 − 1 ) − 1 , B M := 2 π A M − T = B 1 − B 2 .
Then the corresponding direct and reciprocal lattices are given by
R M : = A M Z 2 , R M ∗ : = B M Z 2 , \mathcal{R}_M := A_M\mathbb{Z}^2, \qquad\mathcal{R}_M^* := B_M\mathbb{Z}^2, R M := A M Z 2 , R M ∗ := B M Z 2 ,
and they have the unit cells
Γ M : = A M [ 0 , 1 ) 2 , Γ M ∗ : = B M [ − 1 / 2 , 1 / 2 ) 2 . \Gamma_M := A_M[0,1)^2, \qquad\Gamma_M^* := B_M[-1/2,1/2)^2. Γ M := A M [ 0 , 1 ) 2 , Γ M ∗ := B M [ − 1/2 , 1/2 ) 2 .
Assumption 1. ( R 1 , R 2 ) (\mathcal{R}_1,\mathcal{R}_2) ( R 1 , R 2 ) is incommensurate, i.e. ,
G 1 + G 2 = 0 for G j ∈ R j ∗ iff G 1 = G 2 = 0. G_1+G_2=0 \text{ for } G_j\in\mathcal{R}_j^* \text{ iff } G_1=G_2=0. G 1 + G 2 = 0 for G j ∈ R j ∗ iff G 1 = G 2 = 0.
Local disregistry An important concept that allows us to analyze the quasiperiodicity of twisted bi- and multilayer structures is the following notion of a local environment, which we will refer to as (local) disregistry . For x ∈ R 2 x\in\mathbb{R}^2 x ∈ R 2 , define
{ x } M : = ∑ R M ∈ R M ( x − R M ) 1 Γ M ( x − R M ) \{x\}_M := \sum_{R_M\in\mathcal{R}_M}(x-R_M)\mathbf{1}_{\Gamma_M}(x-R_M) { x } M := R M ∈ R M ∑ ( x − R M ) 1 Γ M ( x − R M )
and
⌊ x ⌋ M : = x − { x } M . \lfloor x\rfloor_M := x-\{x\}_M. ⌊ x ⌋ M := x − { x } M .
Observe that
1 Γ M ( x − R M ) = 0 ⟺ x ∈ Γ M + R M . \mathbf{1}_{\Gamma_M}(x-R_M)=0 \quad\Longleftrightarrow\quad x\in\Gamma_M+\mathcal{R}_M. 1 Γ M ( x − R M ) = 0 ⟺ x ∈ Γ M + R M .
Analogously, we define the decomposition w.r.t. the lattices R j \mathcal{R}_j R j
x = ⌊ x ⌋ j + { x } j x=\lfloor x\rfloor_j+\{x\}_j x = ⌊ x ⌋ j + { x } j
with ⌊ x ⌋ j ∈ R j \lfloor x\rfloor_j\in\mathcal{R}_j ⌊ x ⌋ j ∈ R j , { x } j ∈ Γ j \{x\}_j\in\Gamma_j { x } j ∈ Γ j . We refer to { x } k \{x\}_k { x } k , k ∈ { 1 , 2 , M } k\in\{1,2,M\} k ∈ { 1 , 2 , M } as the (local disregistry) w.r.t. layer k k k , if k = 1 , 2 k=1,2 k = 1 , 2 , and w.r.t. the moiré lattice if k = M k=M k = M .
In order to connect these different notions of disregistry, we also introduce the disregistry matrices
D j → 3 − j : = I − A 3 − j A j − 1 . D_{j\to3-j}:=I-A_{3-j}A_j^{-1}. D j → 3 − j := I − A 3 − j A j − 1 .
We adopt [18 Lemma 3.1] .
Lemma 2.1. The disregistry matrices have the following properties.
(1) D j → 3 − j = ( − 1 ) j A 3 − j A M − 1 D_{j\to3-j}=(-1)^jA_{3-j}A_M^{-1} D j → 3 − j = ( − 1 ) j A 3 − j A M − 1 ,
(2) D j → 3 − j : Γ M → ( − 1 ) j Γ 3 − j D_{j\to3-j}:\Gamma_M\to(-1)^j\Gamma_{3-j} D j → 3 − j : Γ M → ( − 1 ) j Γ 3 − j is a bijection,
(3) D j → 3 − j : R M → R 3 − j D_{j\to3-j}:\mathcal{R}_M\to\mathcal{R}_{3-j} D j → 3 − j : R M → R 3 − j is an isomorphism,
(4) D j → 3 − j T : R 3 − j ∗ → R M ∗ D_{j\to3-j}^{T}:\mathcal{R}_{3-j}^{*}\to\mathcal{R}_M^{*} D j → 3 − j T : R 3 − j ∗ → R M ∗ is an isomorphism,
(5) { R j } 3 − j = D j → 3 − j { R j } M + A 2 ( 1 1 ) δ j , 1 \{R_j\}_{3-j}=D_{j\to3-j}\{R_j\}_M+A_2\begin{pmatrix}1\\1\end{pmatrix}\delta_{j,1} { R j } 3 − j = D j → 3 − j { R j } M + A 2 ( 1 1 ) δ j , 1 for all R j ∈ R j R_j\in\mathcal{R}_j R j ∈ R j .
Notice that (5) provides a linear transformation between the different notions of disregistry. However, it cannot be extended to general x ∈ R 2 x\in\mathbb{R}^2 x ∈ R 2 , as we have previously discussed in [18 ] . Instead, the proper generalization of (5) is given by an appropriate ergodic theorem which we will state below.
Sublattice degrees of freedom
In addition to the primitive lattice vectors introduced above, we also now introduce basis vectors . These allows us to include various degrees of freedom in order to describe general multilattices, e.g., the honeycomb structure of graphene, or to include higher orbitals.
In particular, for j = 1 , 2 j=1,2 j = 1 , 2 , let A j \mathcal{A}_j A j be finite sets, and to each α j ∈ A j \alpha_j\in\mathcal{A}_j α j ∈ A j assign a shift vector τ j ( α j ) ∈ R 2 × { 0 } \tau_j^{(\alpha_j)}\in\mathbb{R}^2\times\{0\} τ j ( α j ) ∈ R 2 × { 0 } . In addition, let γ j ∈ Γ j \gamma_j\in\Gamma_j γ j ∈ Γ j be the position of the origin of layer j j j . We identify γ j \gamma_j γ j with its embedding ( γ j , 0 ) (\gamma_j,0) ( γ j , 0 ) into R 3 \mathbb{R}^3 R 3 .
Analogously to (4) , we define
Ω j : = R j × A j , Ω j ( N ) : = R j ( ⌊ ∣ Γ 3 − j ∣ 1 / 2 N ⌋ ) × A j . \Omega_j:=\mathcal{R}_j\times\mathcal{A}_j,\qquad\Omega_j(N):=\mathcal{R}_j\left(\left\lfloor|\Gamma_{3-j}|^{1/2}N\right\rfloor\right)\times\mathcal{A}_j. Ω j := R j × A j , Ω j ( N ) := R j ( ⌊ ∣ Γ 3 − j ∣ 1/2 N ⌋ ) × A j .
In addition, denoting by ⊔ \sqcup ⊔ the disjoint union, we also abbreviate
Ω : = Ω 1 ⊔ Ω 2 , Ω ( N ) : = Ω 1 ( N ) ⊔ Ω 2 ( N ) . \Omega:=\Omega_1\sqcup\Omega_2,\qquad\Omega(N):=\Omega_1(N)\sqcup\Omega_2(N). Ω := Ω 1 ⊔ Ω 2 , Ω ( N ) := Ω 1 ( N ) ⊔ Ω 2 ( N ) .
For convenience, we also define
ω j + x : = ( R j + x , α j ) ∀ ω j = ( R j , α j ) ∈ Ω j , ∀ x ∈ R 2 . \omega_j+x:=(R_j+x,\alpha_j)\quad\forall\omega_j=(R_j,\alpha_j)\in\Omega_j,\ \forall x\in\mathbb{R}^2. ω j + x := ( R j + x , α j ) ∀ ω j = ( R j , α j ) ∈ Ω j , ∀ x ∈ R 2 .
To define the relaxed lattice positions, let C p e r ( Γ M ; R 3 ) \mathcal{C}_{\mathrm{per}}(\Gamma_M;\mathbb{R}^3) C per ( Γ M ; R 3 ) denote the space of continuous periodic R 3 \mathbb{R}^3 R 3 -valued functions defined on Γ M \Gamma_M Γ M , and let u j ∈ C p e r ( Γ M ; R 3 ) u_j\in\mathcal{C}_{\mathrm{per}}(\Gamma_M;\mathbb{R}^3) u j ∈ C per ( Γ M ; R 3 ) . Then we assume that the total lattice positions (in Lagrangian coordinates) are given by
Y j ( ω j + γ j ) = γ j + R j + τ j ( α j ) + u j ( ω j + γ j ) . Y_j(\omega_j+\gamma_j)=\gamma_j+R_j+\tau_j^{(\alpha_j)}+u_j(\omega_j+\gamma_j). Y j ( ω j + γ j ) = γ j + R j + τ j ( α j ) + u j ( ω j + γ j ) .
We extend these to all x ∈ R 2 x\in\mathbb{R}^2 x ∈ R 2 by
Y j ( x , α j ) = x + τ j ( α j ) + u j ( x , α j ) . Y_j(x,\alpha_j)=x+\tau_j^{(\alpha_j)}+u_j(x,\alpha_j). Y j ( x , α j ) = x + τ j ( α j ) + u j ( x , α j ) .
Multivector notation We abbreviate
ω j ℓ : = ( ω j ( 1 ) , … , ω j ( ℓ ) ) , R j ℓ : = ( R j ( 1 ) , … , R j ( ℓ ) ) \boldsymbol{\omega}_j^\ell:= \left(\omega_j^{(1)},\ldots,\omega_j^{(\ell)}\right), \quad\mathbf{R}_j^\ell:= \left(R_j^{(1)},\ldots,R_j^{(\ell)}\right) ω j ℓ := ( ω j ( 1 ) , … , ω j ( ℓ ) ) , R j ℓ := ( R j ( 1 ) , … , R j ( ℓ ) )
where ω j ( n ) = ( R j ( n ) , α j ( n ) ) ∈ Ω j \omega_j^{(n)} = \left(R_j^{(n)},\alpha_j^{(n)}\right) \in\Omega_j ω j ( n ) = ( R j ( n ) , α j ( n ) ) ∈ Ω j . Throughout this work, we also abbreviate
u = ( u j ) j = 1 2 , δ u = ( δ u j ) j = 1 2 , γ = ( γ j ) j = 1 2 , α = ( α j ) j = 1 2 . \mathbf{u} = \left(u_j\right)_{j=1}^2,\quad\delta\mathbf{u} = \left(\delta u_j\right)_{j=1}^2,\quad\gamma= \left(\gamma_j\right)_{j=1}^2,\quad\alpha= \left(\alpha_j\right)_{j=1}^2. u = ( u j ) j = 1 2 , δ u = ( δ u j ) j = 1 2 , γ = ( γ j ) j = 1 2 , α = ( α j ) j = 1 2 .
Mechanical energy
We now introduce the formal expressions for the energy density, as well as for the offset energy. In the next section, we will state sufficient conditions to ensure the existence of the respective functionals. We will also prove the convergence to the respective limit functionals.
In the following and throughout this work, let j ∈ { 1 , 2 } j \in\{1,2\} j ∈ { 1 , 2 } if not specified otherwise.
Monolayer potential
Let
V j , α j ( m o n o ) : ( R 3 ) Ω j → R V_{j,\alpha_j}^{(\mathrm{mono})} : \left(\mathbb{R}^3\right)^{\Omega_j} \to\mathbb{R} V j , α j ( mono ) : ( R 3 ) Ω j → R
with α j ∈ A j \alpha_j \in\mathcal{A}_j α j ∈ A j be given. In addition, we abbreviate the (relative) finite difference stencils
Δ j [ u j ] ( x , α j ) : = ( u j ( ω j + x ) − u j ( x , α j ) ) ω j ∈ Ω j , \Delta_j[u_j](x,\alpha_j) := \left(u_j(\omega_j+x)-u_j(x,\alpha_j)\right)_{\omega_j\in\Omega_j}, Δ j [ u j ] ( x , α j ) := ( u j ( ω j + x ) − u j ( x , α j ) ) ω j ∈ Ω j ,
For γ j ∈ Γ j \gamma_j \in\Gamma_j γ j ∈ Γ j and N ∈ N N \in\mathbb{N} N ∈ N , we then introduce the monolayer energy density
e j , N , γ j ( m o n o ) ( u j ) : = ∣ A j ∣ ∣ Γ j ∣ ∣ Ω j ( N ) ∣ ∑ ω j ∈ Ω j ( N ) V j , α j ( m o n o ) ( Δ j [ u j ] ( ω j + γ j ) ) , e_{j,N,\gamma_j}^{(\mathrm{mono})}(u_j) := \frac{|\mathcal{A}_j|}{|\Gamma_j||\Omega_j(N)|}\sum_{\omega_j\in\Omega_j(N)}V_{j,\alpha_j}^{(\mathrm{mono})}\left(\Delta_j[u_j](\omega_j+\gamma_j)\right), e j , N , γ j ( mono ) ( u j ) := ∣ Γ j ∣∣ Ω j ( N ) ∣ ∣ A j ∣ ω j ∈ Ω j ( N ) ∑ V j , α j ( mono ) ( Δ j [ u j ] ( ω j + γ j ) ) ,
where
∣ ⋃ ω j ∈ Ω j ( N ) ( Γ j + R j + τ j ( α j ) ) ∣ = ∣ Γ 1 ∣ ∣ Γ 2 ∣ ( 2 N + 1 ) 2 + O N → ∞ ( N ) = ∣ Γ j ∣ ∣ Ω j ( N ) ∣ ∣ A j ∣ + O N → ∞ ( N ) \left|\bigcup_{\omega_j\in\Omega_j(N)}\left(\Gamma_j+R_j+\tau_j^{(\alpha_j)}\right)\right|=|\Gamma_1||\Gamma_2|(2N+1)^2+O_{N\to\infty}(N)=\frac{|\Gamma_j||\Omega_j(N)|}{|\mathcal{A}_j|}+O_{N\to\infty}(N) ω j ∈ Ω j ( N ) ⋃ ( Γ j + R j + τ j ( α j ) ) = ∣ Γ 1 ∣∣ Γ 2 ∣ ( 2 N + 1 ) 2 + O N → ∞ ( N ) = ∣ A j ∣ ∣ Γ j ∣∣ Ω j ( N ) ∣ + O N → ∞ ( N )
is the area of the corresponding truncated region covered by the nuclei in layer j j j .
We formally also define the limit energy density
e j ( m o n o ) ( u j ) : = 1 ∣ Γ j ∣ ∑ α j ∈ A j ∫ Γ M d x V j , α j ( m o n o ) ( Δ j [ u j ] ( x , α j ) ) . e_j^{(\mathrm{mono})}(u_j) := \frac{1}{|\Gamma_j|}\sum_{\alpha_j\in\mathcal{A}_j}\int_{\Gamma_M} dx\,V_{j,\alpha_j}^{(\mathrm{mono})}\left(\Delta_j[u_j](x,\alpha_j)\right). e j ( mono ) ( u j ) := ∣ Γ j ∣ 1 α j ∈ A j ∑ ∫ Γ M d x V j , α j ( mono ) ( Δ j [ u j ] ( x , α j ) ) .
which we will prove to coincide with lim N → ∞ e j , N , γ j ( m o n o ) ( u j ) \lim_{N \to\infty} e_{j,N,\gamma_j}^{(\mathrm{mono})}(u_j) lim N → ∞ e j , N , γ j ( mono ) ( u j ) . In this context, it is useful to also abbreviate the site potential density
ϕ j ( m o n o ) [ u j ] : R 2 → R , x ↦ 1 ∣ Γ j ∣ ∑ α j ∈ A j V j , α j ( m o n o ) ( Δ j [ u j ] ( x , α j ) ) . \phi_j^{(\mathrm{mono})}[u_j] : \mathbb{R}^2 \to\mathbb{R},\quad x \mapsto\frac{1}{|\Gamma_j|}\sum_{\alpha_j \in A_j} V_{j,\alpha_j}^{(\mathrm{mono})}\left(\Delta_j[u_j](x,\alpha_j)\right). ϕ j ( mono ) [ u j ] : R 2 → R , x ↦ ∣ Γ j ∣ 1 α j ∈ A j ∑ V j , α j ( mono ) ( Δ j [ u j ] ( x , α j ) ) .
which, in turn, allows us to write
e j ( m o n o ) ( u j ) = ∫ Γ M d x ϕ j ( m o n o ) [ u j ] ( x ) . e_j^{(\mathrm{mono})}(u_j) = \int_{\Gamma_M} \mathrm{d}x\,\phi_j^{(\mathrm{mono})}[u_j](x). e j ( mono ) ( u j ) = ∫ Γ M d x ϕ j ( mono ) [ u j ] ( x ) .
Similarly, we define the offset site potential
Δ Φ j ( m o n o ) [ δ u j ; u j ] : = ∣ Γ j ∣ ( ϕ j ( m o n o ) [ u j + δ u j ] − ϕ j ( m o n o ) [ u j ] ) = ∑ α j ∈ A j ( V j , α j ( m o n o ) ( Δ j [ u j + δ u j ] ( ⋅ , α j ) ) − V j , α j ( m o n o ) ( Δ j [ u j ] ( ⋅ , α j ) ) ) . \begin{aligned}
\Delta\Phi_j^{(\mathrm{mono})}[\delta u_j;u_j] := |\Gamma_j|\left(\phi_j^{(\mathrm{mono})}[u_j+\delta u_j]-\phi_j^{(\mathrm{mono})}[u_j]\right) \\
&= \sum_{\alpha_j \in A_j}\left(V_{j,\alpha_j}^{(\mathrm{mono})}\left(\Delta_j[u_j+\delta u_j](\cdot,\alpha_j)\right)-V_{j,\alpha_j}^{(\mathrm{mono})}\left(\Delta_j[u_j](\cdot,\alpha_j)\right)\right).
\end{aligned} Δ Φ j ( mono ) [ δ u j ; u j ] := ∣ Γ j ∣ ( ϕ j ( mono ) [ u j + δ u j ] − ϕ j ( mono ) [ u j ] ) = α j ∈ A j ∑ ( V j , α j ( mono ) ( Δ j [ u j + δ u j ] ( ⋅ , α j ) ) − V j , α j ( mono ) ( Δ j [ u j ] ( ⋅ , α j ) ) ) .
and the offset energy
Δ E j ; u j ( m o n o ) ( δ u j ) : = ∑ R j ∈ R j Δ Φ j ( m o n o ) [ δ u j ; u j ] ( R j ) . \Delta E_{j;u_j}^{(\mathrm{mono})}(\delta u_j) := \sum_{R_j \in\mathcal{R}_j}\Delta\Phi_j^{(\mathrm{mono})}[\delta u_j;u_j](R_j). Δ E j ; u j ( mono ) ( δ u j ) := R j ∈ R j ∑ Δ Φ j ( mono ) [ δ u j ; u j ] ( R j ) .
Interlayer coupling
Definition 2.2. A function
V : ( R 3 ) Ω j → R V : (\mathbb{R}^3)^{\Omega_j} \to\mathbb{R} V : ( R 3 ) Ω j → R is translation-invariant iff for any ( a ω j ) ω j ∈ Ω j ∈ ( R 3 ) Ω j (a_{\omega_j})_{\omega_j\in\Omega_j} \in(\mathbb{R}^3)^{\Omega_j} ( a ω j ) ω j ∈ Ω j ∈ ( R 3 ) Ω j and any R j ∈ R j R_j \in\mathcal{R}_j R j ∈ R j , we have that
V ( ( a ω j + R j ) ω j ∈ Ω j ) = V ( ( a ω j ) ω j ∈ Ω j ) . V\left((a_{\omega_j+R_j})_{\omega_j\in\Omega_j}\right)=V\left((a_{\omega_j})_{\omega_j\in\Omega_j}\right). V ( ( a ω j + R j ) ω j ∈ Ω j ) = V ( ( a ω j ) ω j ∈ Ω j ) . In addition, V V V is even iff V ( x ) = V ( − x ) V(x)=V(-x) V ( x ) = V ( − x ) .
Let
V j , α j ( i n t e r ) : ( R 3 ) Ω 3 − j → R , V_{j,\alpha_j}^{(\mathrm{inter})} : (\mathbb{R}^3)^{\Omega_{3-j}} \to\mathbb{R}, V j , α j ( inter ) : ( R 3 ) Ω 3 − j → R ,
with α j ∈ A j \alpha_j \in A_j α j ∈ A j be even and translation-invariant. As a special instance, let v α : R 3 → R v_\alpha: \mathbb{R}^3 \to\mathbb{R} v α : R 3 → R , with α k ∈ A k \alpha_k \in A_k α k ∈ A k , k = 1 , 2 k=1,2 k = 1 , 2 , α : = ( α 1 , α 2 ) \alpha:= (\alpha_1,\alpha_2) α := ( α 1 , α 2 ) , denote an even pair potential, and consider
V j , α j ( i n t e r ) ( ( a ω 3 − j ) ω 3 − j ∈ Ω 3 − j ) = 1 2 ∑ ω 3 − j ∈ Ω 3 − j v α ( a ω 3 − j ) . V_{j,\alpha_j}^{(\mathrm{inter})}\left((a_{\omega_{3-j}})_{\omega_{3-j}\in\Omega_{3-j}}\right)=\frac{1}{2}\sum_{\omega_{3-j}\in\Omega_{3-j}}v_\alpha(a_{\omega_{3-j}).} V j , α j ( inter ) ( ( a ω 3 − j ) ω 3 − j ∈ Ω 3 − j ) = 2 1 ω 3 − j ∈ Ω 3 − j ∑ v α ( a ω 3 − j ) .
Such interatomic pair potentials v α v_\alpha v α have been constructed using DFT, e.g., in [23 , 27 , 35 ] .
The fact that we are using the ordered index α = ( α 1 , α 2 ) \alpha=(\alpha_1,\alpha_2) α = ( α 1 , α 2 ) to label pair potentials, as opposed to, e.g., ( α j , α 3 − j ) (\alpha_j,\alpha_{3-j}) ( α j , α 3 − j ) , is to indicate that pair interactions only depend on the pair of involved atoms, as opposed to the order of the pair. Whenever there is a α j \alpha_j α j or α 3 − j \alpha_{3-j} α 3 − j on the LHS of an equation, and α 1 \alpha_1 α 1 or α 2 \alpha_2 α 2 on the RHS, the corresponding equation is to be read individually for j = 1 j=1 j = 1 or j = 2 j=2 j = 2 , to avoid confusion.
We abbreviate the (total) finite difference stencils
Δ j , 3 − j [ u ] ( x , α j ) : = ( Y 3 − j ( ω 3 − j + A 3 − j A j − 1 x + γ 3 − j ) − Y j ( x + γ j , α j ) ) ω 3 − j ∈ Ω 3 − j . \Delta_{j,3-j}[\mathbf{u}](x,\alpha_j) := \left(Y_{3-j}(\omega_{3-j}+A_{3-j}A_j^{-1}x+\gamma_{3-j})-Y_j(x+\gamma_j,\alpha_j)\right)_{\omega_{3-j}\in\Omega_{3-j}}. Δ j , 3 − j [ u ] ( x , α j ) := ( Y 3 − j ( ω 3 − j + A 3 − j A j − 1 x + γ 3 − j ) − Y j ( x + γ j , α j ) ) ω 3 − j ∈ Ω 3 − j .
With that, we define the truncated interlayer coupling energy density of the j j j -th layer
e j , N , γ ( i n t e r ) ( u ) : = ∣ A j ∣ ∣ Γ j ∣ ∣ Ω j ( N ) ∣ ∑ ω j ∈ Ω j ( N ) V j , α j ( i n t e r ) ( Δ j , 3 − j [ u ] ( ω j ) ) . e_{j,N,\gamma}^{(\mathrm{inter})}(\mathbf{u}) := \frac{|\mathcal{A}_j|}{|\Gamma_j||\Omega_j(N)|}\sum_{\omega_j\in\Omega_j(N)}V_{j,\alpha_j}^{(\mathrm{inter})}\left(\Delta_{j,3-j}[\mathbf{u}](\omega_j)\right). e j , N , γ ( inter ) ( u ) := ∣ Γ j ∣∣ Ω j ( N ) ∣ ∣ A j ∣ ω j ∈ Ω j ( N ) ∑ V j , α j ( inter ) ( Δ j , 3 − j [ u ] ( ω j ) ) .
and its formal limit
e j , γ ( i n t e r ) ( u ) : = 1 ∣ Γ j ∣ ∑ α j ∈ A j ∫ Γ j d x V j , α j ( i n t e r ) ( Δ j , 3 − j [ u ] ( x , α j ) ) . e_{j,\gamma}^{(\mathrm{inter})}(\mathbf{u}) := \frac{1}{|\Gamma_j|}\sum_{\alpha_j\in\mathcal{A}_j}\int_{\Gamma_j} \mathrm{d}x\,V_{j,\alpha_j}^{(\mathrm{inter})}\left(\Delta_{j,3-j}[\mathbf{u}](x,\alpha_j)\right). e j , γ ( inter ) ( u ) := ∣ Γ j ∣ 1 α j ∈ A j ∑ ∫ Γ j d x V j , α j ( inter ) ( Δ j , 3 − j [ u ] ( x , α j ) ) .
Again, we define the site potential density
ϕ j , γ ( i n t e r ) [ u ] : R 2 → R , x ↦ 1 ∣ Γ j ∣ ∑ α j ∈ A j V j , α j ( i n t e r ) ( Δ j , 3 − j [ u ] ( x , α j ) ) , \phi_{j,\gamma}^{(\mathrm{inter})}[\mathbf{u}]:\mathbb{R}^{2}\to\mathbb{R},\ x\mapsto\frac{1}{|\Gamma_j|}\sum_{\alpha_j\in\mathcal{A}_j}V_{j,\alpha_j}^{(\mathrm{inter})}\left(\Delta_{j,3-j}[\mathbf{u}](x,\alpha_j)\right), ϕ j , γ ( inter ) [ u ] : R 2 → R , x ↦ ∣ Γ j ∣ 1 α j ∈ A j ∑ V j , α j ( inter ) ( Δ j , 3 − j [ u ] ( x , α j ) ) ,
and the offset site potential
Δ Φ j , γ ( i n t e r ) [ δ u ; u ] : = ∣ Γ j ∣ ( ϕ j , γ ( i n t e r ) [ u + δ u ] − ϕ j , γ ( i n t e r ) [ u ] ) = ∑ α j ∈ A j ( V j , α j ( i n t e r ) ( Δ j , 3 − j [ u + δ u ] ( ⋅ , α j ) ) − V j , α j ( i n t e r ) ( Δ j , 3 − j [ u ] ( ⋅ , α j ) ) ) . \begin{aligned}
\Delta\Phi_{j,\gamma}^{(\mathrm{inter})}[\delta\mathbf{u};\mathbf{u}]:=|\Gamma_j|\left(\phi_{j,\gamma}^{(\mathrm{inter})}[\mathbf{u}+\delta\mathbf{u}]-\phi_{j,\gamma}^{(\mathrm{inter})}[\mathbf{u}]\right) \\
&=\sum_{\alpha_j\in\mathcal{A}_j}\left(V_{j,\alpha_j}^{(\mathrm{inter})}\left(\Delta_{j,3-j}[\mathbf{u}+\delta\mathbf{u}](\cdot,\alpha_j)\right)-V_{j,\alpha_j}^{(\mathrm{inter})}\left(\Delta_{j,3-j}[\mathbf{u}](\cdot,\alpha_j)\right)\right).
\end{aligned} Δ Φ j , γ ( inter ) [ δ u ; u ] := ∣ Γ j ∣ ( ϕ j , γ ( inter ) [ u + δ u ] − ϕ j , γ ( inter ) [ u ] ) = α j ∈ A j ∑ ( V j , α j ( inter ) ( Δ j , 3 − j [ u + δ u ] ( ⋅ , α j ) ) − V j , α j ( inter ) ( Δ j , 3 − j [ u ] ( ⋅ , α j ) ) ) .
This allows us to define the interlayer offset energy
Δ E j , γ ; u ( i n t e r ) ( δ u ) : = ∑ R j ∈ R j Δ Φ j , γ ( i n t e r ) [ δ u ; u ] ( R j ) . \Delta E_{j,\gamma;\mathbf{u}}^{(\mathrm{inter})}(\delta\mathbf{u}):=\sum_{R_j\in\mathcal{R}_j}\Delta\Phi_{j,\gamma}^{(\mathrm{inter})}[\delta\mathbf{u};\mathbf{u}](R_j). Δ E j , γ ; u ( inter ) ( δ u ) := R j ∈ R j ∑ Δ Φ j , γ ( inter ) [ δ u ; u ] ( R j ) .
Total energy
We abbreviate the truncated total energy density
e N , γ ( t o t ) ( u ) : = ∑ j = 1 2 ( e j , N , γ j ( m o n o ) ( u j ) + e j , N , γ ( i n t e r ) ( u ) ) . e_{N,\gamma}^{(\mathrm{tot})}(\mathbf{u}):=\sum_{j=1}^{2}\left(e_{j,N,\gamma_j}^{(\mathrm{mono})}(u_j)+e_{j,N,\gamma}^{(\mathrm{inter})}(\mathbf{u})\right). e N , γ ( tot ) ( u ) := j = 1 ∑ 2 ( e j , N , γ j ( mono ) ( u j ) + e j , N , γ ( inter ) ( u ) ) .
the limit energy
e γ ( t o t ) ( u ) : = ∑ j = 1 2 ( e j ( m o n o ) ( u j ) + e j , γ ( i n t e r ) ( u ) ) . e_{\gamma}^{(\mathrm{tot})}(\mathbf{u}):=\sum_{j=1}^{2}\left(e_j^{(\mathrm{mono})}(u_j)+e_{j,\gamma}^{(\mathrm{inter})}(\mathbf{u})\right). e γ ( tot ) ( u ) := j = 1 ∑ 2 ( e j ( mono ) ( u j ) + e j , γ ( inter ) ( u ) ) .
as well as the total offset energy
Δ E γ ; u ( t o t ) ( δ u ) : = ∑ j = 1 2 ( Δ E j ; u j ( m o n o ) ( δ u j ) + Δ E j , γ ; u ( i n t e r ) ( δ u ) ) . \Delta E_{\gamma;\mathbf{u}}^{(\mathrm{tot})}(\delta\mathbf{u}) := \sum_{j=1}^{2} \left(\Delta E_{j;u_j}^{(\mathrm{mono})}(\delta u_j) + \Delta E_{j,\gamma;\mathbf{u}}^{(\mathrm{inter})}(\delta\mathbf{u})\right). Δ E γ ; u ( tot ) ( δ u ) := j = 1 ∑ 2 ( Δ E j ; u j ( mono ) ( δ u j ) + Δ E j , γ ; u ( inter ) ( δ u ) ) .
Below, we will show the convergence of Δ E u ( t o t ) ( δ u ) \Delta E_{\mathbf{u}}^{(\mathrm{tot})}(\delta\mathbf{u}) Δ E u ( tot ) ( δ u ) and of
lim N → ∞ e N , γ ( t o t ) ( u ) = e γ ( t o t ) ( u ) \lim_{N\to\infty} e_{N,\gamma}^{(\mathrm{tot})}(\mathbf{u}) = e_{\gamma}^{(\mathrm{tot})}(\mathbf{u}) N → ∞ lim e N , γ ( tot ) ( u ) = e γ ( tot ) ( u )
for displacements u \mathbf{u} u and δ u \delta\mathbf{u} δ u in appropriate function spaces, and sufficiently regular potentials V ( m o n o ) V^{(\mathrm{mono})} V ( mono ) , V ( i n t e r ) V^{(\mathrm{inter})} V ( inter ) .
Below, if the explicit γ \gamma γ -dependence is not important, we will also omit it in the notation of e ( t o t ) e^{(\mathrm{tot})} e ( tot ) , Δ E ( t o t ) \Delta E^{(\mathrm{tot})} Δ E ( tot ) , etc.
Assumptions and function spaces
We largely follow [33 ] and adapt the assumptions to the present context.
Displacements
We start by defining the appropriate function spaces for the energy minimizers. Observe that e j ( m o n o ) ( u j ) e_j^{(\mathrm{mono})}(u_j) e j ( mono ) ( u j ) is invariant under the shift u j u_j u j , and that shifting u → u + ( t 1 , t 2 ) \mathbf{u} \to\mathbf{u} + (t_1,t_2) u → u + ( t 1 , t 2 ) with t j ∈ R 2 t_j \in\mathbb{R}^2 t j ∈ R 2 in e j ( i n t e r ) ( u ) e_j^{(\mathrm{inter})}(\mathbf{u}) e j ( inter ) ( u ) amounts to redefining the origins γ j → { γ j + t j } 3 − j \gamma_j \to\{\gamma_j+t_j\}_{3-j} γ j → { γ j + t j } 3 − j . Consequently, we assume that
\fint_{\Gamma_M} \, \mathrm{d}x \, u_j(x) = 0.
Let
\overline{W}_{\mathrm{per}}^{1,2} := \overline{W}_{\mathrm{per}}^{1,2}(\Gamma_M;(\mathbb{R}^2)^{\mathcal{A}_1}\times(\mathbb{R}^2)^{\mathcal{A}_2}) := \{ \mathbf{u}|_{\Gamma_M} \,\middle|\, \mathbf{u}\in L_{\mathrm{loc}}^2(\mathbb{R}^2;(\mathbb{R}^2)^{\mathcal{A}_1}\times(\mathbb{R}^2)^{\mathcal{A}_2}),
u ( x ) = u ( x + R M ) ∀ R M ∈ R M , a.e. x ∈ R 2 , \mathbf{u}(x)=\mathbf{u}(x+R_M)\quad\forall R_M\in\mathcal{R}_M,\ \text{a.e. }x\in\mathbb{R}^2, u ( x ) = u ( x + R M ) ∀ R M ∈ R M , a.e. x ∈ R 2 ,
\mathbf{u}|_{\Gamma_M}\in W^{1,2}(\Gamma_M;(\mathbb{R}^2)^{\mathcal{A}_1}\times(\mathbb{R}^2)^{\mathcal{A}_2}),\ \fint_{\Gamma_M}\,\mathrm{d}x\,\mathbf{u}(x)=0\}.
We endow W ‾ p e r k , p \overline{W}_{\mathrm{per}}^{k,p} W per k , p with the W ˙ k , p \dot{W}^{k,p} W ˙ k , p -topology, i.e.,
∥ u ∥ W ‾ p e r k , p : = ∥ ∇ k u ∥ p . \|\mathbf{u}\|_{\overline{W}_{\mathrm{per}}^{k,p}} := \|\nabla^k\mathbf{u}\|_p. ∥ u ∥ W per k , p := ∥ ∇ k u ∥ p .
Analogously, we define W ‾ p e r 1 , 2 ( Γ M ; ( R 2 ) A j ) \overline{W}_{\mathrm{per}}^{1,2}(\Gamma_M;(\mathbb{R}^2)^{\mathcal{A}_j}) W per 1 , 2 ( Γ M ; ( R 2 ) A j ) , j = 1 , 2 j=1,2 j = 1 , 2 , as well as W ‾ p e r n , p \overline{W}_{\mathrm{per}}^{n,p} W per n , p , etc. Next, let
\mathcal{D}_\kappa:= \{ \mathbf{u}=(u_1,u_2)\in\overline{W}_{\mathrm{per}}^{1,2}(\Gamma_M;(\mathbb{R}^2)^{\mathcal{A}_1}\times(\mathbb{R}^2)^{\mathcal{A}_2})\ \middle|\ \forall j=1,2:\ \|\nabla u_j\|_\infty\leq\kappa,
sup R j ∈ R j ∥ u j ( ⋅ + R j ) − u j ∥ ∞ , sup R j ∈ R j ∥ u j ( A j A 3 − j − 1 ⋅ + R j ) − u 3 − j ∥ ∞ ≤ κ } . \sup_{R_j\in\mathcal{R}_j}\|u_j(\cdot+R_j)-u_j\|_\infty,\ \sup_{R_j\in\mathcal{R}_j}\|u_j(A_jA_{3-j}^{-1}\cdot+R_j)-u_{3-j}\|_\infty\leq\kappa\}. R j ∈ R j sup ∥ u j ( ⋅ + R j ) − u j ∥ ∞ , R j ∈ R j sup ∥ u j ( A j A 3 − j − 1 ⋅ + R j ) − u 3 − j ∥ ∞ ≤ κ } .
Interpolants
When discussing the Cauchy-Born approximation, we will restrict ourselves to a single sublattice degree of freedom per layer to simplify the analysis. This will allow us to immediately adapt ideas from [33 ] . To extent the ideas to multilattices, we refer to [32 ] .
Let ζ ∈ C c ∞ ( R 2 ) \zeta\in C^\infty_c(\mathbb{R}^2) ζ ∈ C c ∞ ( R 2 ) be such that supp ζ ∩ Z 2 = { 0 } \operatorname{supp}\zeta\cap\mathbb{Z}^2 = \{0\} supp ζ ∩ Z 2 = { 0 } , ζ ( 0 ) = 1 \zeta(0)=1 ζ ( 0 ) = 1 , ζ ≥ 0 \zeta\ge0 ζ ≥ 0 , ζ ( x ) = ζ ( − x ) \zeta(x)=\zeta(-x) ζ ( x ) = ζ ( − x ) , and satisfying
∑ n ∈ Z 2 ζ ( x − n ) ( a + b ⋅ n ) = a + b ⋅ x ∀ a ∈ R , b , x ∈ R 2 . \sum_{n\in\mathbb{Z}^2} \zeta(x-n)(a+b\cdot n)=a+b\cdot x \quad\forall a\in\mathbb{R},\ b,x\in\mathbb{R}^2. n ∈ Z 2 ∑ ζ ( x − n ) ( a + b ⋅ n ) = a + b ⋅ x ∀ a ∈ R , b , x ∈ R 2 .
In particular, we have that ∫ R 2 d x ζ ( x ) = 1 \int_{\mathbb{R}^2} \mathrm{d}x\,\zeta(x)=1 ∫ R 2 d x ζ ( x ) = 1 . We also abbreviate ζ j : = ζ ( A j − 1 ⋅ ) \zeta_j:=\zeta(A_j^{-1}\cdot) ζ j := ζ ( A j − 1 ⋅ ) .
Given a function v j : R j → R v_j:\mathcal{R}_j\to\mathbb{R} v j : R j → R , we construct the first order interpolant
v j ( x ) : = ∑ R j ∈ R j ζ j ( x − R j ) v j ( R j ) , (5) v_j(x):=\sum_{R_j\in\mathcal{R}_j}\zeta_j(x-R_j)v_j(R_j), \tag*{(5)} v j ( x ) := R j ∈ R j ∑ ζ j ( x − R j ) v j ( R j ) , (5)
as well as the quasi-interpolant
v ~ j : = 1 ∣ Γ j ∣ ζ j ∗ v j . (6) \widetilde{v}_j:=\frac{1}{|\Gamma_j|}\zeta_j*v_j. \tag*{(6)} v j := ∣ Γ j ∣ 1 ζ j ∗ v j . (6)
A useful property of convolving with ζ j \zeta_j ζ j is that finite differences can be rewritten as convolution with a directional gradient
v ~ j ( x + R j ) − v ~ j ( x ) = ∫ 0 1 d t ∇ R j v ~ j ( x + t R j ) = 1 ∣ Γ j ∣ ∫ R 2 d y ∫ 0 1 d t ζ j ( x + t R j − y ) ∇ R j v ( y ) . (7) \begin{aligned}
\widetilde{v}_j(x+R_j)-\widetilde{v}_j(x)=\int_0^1\mathrm{d}t\,\nabla_{R_j}\widetilde{v}_j(x+tR_j) \\
&=\frac{1}{|\Gamma_j|}\int_{\mathbb{R}^2}\mathrm{d}y\int_0^1\mathrm{d}t\,\zeta_j(x+tR_j-y)\nabla_{R_j}v(y). \tag*{(7)}
\end{aligned} v j ( x + R j ) − v j ( x ) = ∫ 0 1 d t ∇ R j v j ( x + t R j ) = ∣ Γ j ∣ 1 ∫ R 2 d y ∫ 0 1 d t ζ j ( x + t R j − y ) ∇ R j v ( y ) . (7)
Potentials
Smoothness and decay We define the spaces for finite difference stencils
D j κ : = { a = ( a ω j ) ω j ∈ Ω j ∣ a ω j ∈ R 2 , ∥ a ∥ D j : = sup ω j ∈ Ω j , ω j ≠ ( 0 , α j ) ∣ a ω j ∣ ∣ R j ∣ ≤ κ } . \mathcal{D}_j^\kappa:=\left\{a=(a_{\omega_j})_{\omega_j\in\Omega_j}\mid a_{\omega_j}\in\mathbb{R}^2,\ \lVert a\rVert_{\mathcal{D}_j}:=\sup_{\substack{\omega_j\in\Omega_j,\\ \omega_j\ne(0,\alpha_j)}}\frac{|a_{\omega_j}|}{|R_j|}\le\kappa\right\}. D j κ := ⎩ ⎨ ⎧ a = ( a ω j ) ω j ∈ Ω j ∣ a ω j ∈ R 2 , ∥ a ∥ D j := ω j ∈ Ω j , ω j = ( 0 , α j ) sup ∣ R j ∣ ∣ a ω j ∣ ≤ κ ⎭ ⎬ ⎫ .
Let V m = ( v ( 1 ) , … , v ( m ) ) \mathbf{V}^m=(\mathbf{v}^{(1)},\ldots,\mathbf{v}^{(m)}) V m = ( v ( 1 ) , … , v ( m ) ) and v ( ℓ ) = ( v 1 ( ℓ ) , v 2 ( ℓ ) ) \mathbf{v}^{(\ell)}=(v_1^{(\ell)},v_2^{(\ell)}) v ( ℓ ) = ( v 1 ( ℓ ) , v 2 ( ℓ ) ) , where v j ( ℓ ) ∈ W ‾ p e r ( Γ M ; ( R 2 ) A j ) v_j^{(\ell)}\in\overline{W}_{\mathrm{per}}(\Gamma_M;(\mathbb{R}^2)^{\mathcal{A}_j}) v j ( ℓ ) ∈ W per ( Γ M ; ( R 2 ) A j ) . We abbreviate the interlayer difference stencils
( Δ ω 3 − j ( j , 3 − j ) v ( ℓ ) ) ( x , α j ) : = v 3 − j ( ℓ ) ( ω 3 − j + γ 3 − j + A 3 − j A j − 1 x ) − v j ( ℓ ) ( x + γ j , α j ) , \left(\Delta_{\omega_{3-j}}^{(j,3-j)}\mathbf{v}^{(\ell)}\right)(x,\alpha_j):=v_{3-j}^{(\ell)}(\omega_{3-j}+\gamma_{3-j}+A_{3-j}A_j^{-1}x)-v_j^{(\ell)}(x+\gamma_j,\alpha_j), ( Δ ω 3 − j ( j , 3 − j ) v ( ℓ ) ) ( x , α j ) := v 3 − j ( ℓ ) ( ω 3 − j + γ 3 − j + A 3 − j A j − 1 x ) − v j ( ℓ ) ( x + γ j , α j ) ,
( Δ ω 3 − j ( j , 3 − j ) V m ) ( x , α j ) : = ( ( Δ ω 3 − j ( j , 3 − j ) v ( ℓ ) ) ( x , α j ) ) ℓ = 1 m . \left(\Delta_{\omega_{3-j}}^{(j,3-j)}\mathbf{V}^m\right)(x,\alpha_j):=\left(\left(\Delta_{\omega_{3-j}}^{(j,3-j)}\mathbf{v}^{(\ell)}\right)(x,\alpha_j)\right)_{\ell=1}^m. ( Δ ω 3 − j ( j , 3 − j ) V m ) ( x , α j ) := ( ( Δ ω 3 − j ( j , 3 − j ) v ( ℓ ) ) ( x , α j ) ) ℓ = 1 m .
Similarly, denoting V j m : = ( v j ( 1 ) , … , v j ( m ) ) \mathbf{V}_j^m := (v_j^{(1)},\ldots,v_j^{(m)}) V j m := ( v j ( 1 ) , … , v j ( m ) ) , we introduce the monolayer difference stencils
( Δ ω j ( j ) v j ( ℓ ) ) ( x , α j ) : = v j ( ℓ ) ( ω j ( ℓ ) + x ) − v j ( ℓ ) ( x , α j ) , (\Delta_{\omega_j}^{(j)}v_j^{(\ell)})(x,\alpha_j) := v_j^{(\ell)}(\omega_j^{(\ell)}+x)-v_j^{(\ell)}(x,\alpha_j), ( Δ ω j ( j ) v j ( ℓ ) ) ( x , α j ) := v j ( ℓ ) ( ω j ( ℓ ) + x ) − v j ( ℓ ) ( x , α j ) ,
( Δ ω j ( j , m ) V j m ) ( x , α j ) : = ( ( Δ ω j ( ℓ ) ( j ) v j ( ℓ ) ) ( x , α j ) ) ℓ = 1 m . (\Delta_{\omega_j}^{(j,m)}\mathbf{V}_j^m)(x,\alpha_j) := \left((\Delta_{\omega_j^{(\ell)}}^{(j)}v_j^{(\ell)})(x,\alpha_j)\right)_{\ell=1}^{m}. ( Δ ω j ( j , m ) V j m ) ( x , α j ) := ( ( Δ ω j ( ℓ ) ( j ) v j ( ℓ ) ) ( x , α j ) ) ℓ = 1 m .
In addition, we abbreviate the partial derivatives
V j , α j ; ω j m ( m o n o ) ( a ) : = ( D ω j m V j , α j ( m o n o ) ) ( a ) : = ∂ m V j , α j ( m o n o ) ( a ) ∂ a ω j ( 1 ) ⋯ ∂ a ω j ( m ) , V_{j,\alpha_j;\omega_j^m}^{(\mathrm{mono})}(\mathbf{a}) := (D_{\omega_j^m}V_{j,\alpha_j}^{(\mathrm{mono})})(\mathbf{a}) := \frac{\partial^m V_{j,\alpha_j}^{(\mathrm{mono})}(\mathbf{a})}{\partial a_{\omega_j^{(1)}}\cdots\partial a_{\omega_j^{(m)}}}, V j , α j ; ω j m ( mono ) ( a ) := ( D ω j m V j , α j ( mono ) ) ( a ) := ∂ a ω j ( 1 ) ⋯ ∂ a ω j ( m ) ∂ m V j , α j ( mono ) ( a ) ,
V j , α j ; ω j m ( i n t e r ) ( a ) : = ( D ω j m V j , α j ( i n t e r ) ) ( a ) : = ∂ m V j , α j ( i n t e r ) ( a ) ∂ a ω j ( 1 ) ⋯ ∂ a ω j ( m ) . V_{j,\alpha_j;\omega_j^m}^{(\mathrm{inter})}(\mathbf{a}) := (D_{\omega_j^m}V_{j,\alpha_j}^{(\mathrm{inter})})(\mathbf{a}) := \frac{\partial^m V_{j,\alpha_j}^{(\mathrm{inter})}(\mathbf{a})}{\partial a_{\omega_j^{(1)}}\cdots\partial a_{\omega_j^{(m)}}}. V j , α j ; ω j m ( inter ) ( a ) := ( D ω j m V j , α j ( inter ) ) ( a ) := ∂ a ω j ( 1 ) ⋯ ∂ a ω j ( m ) ∂ m V j , α j ( inter ) ( a ) .
With that, we introduce the intralayer site potential norm
m ℓ , j ( m o n o ) ( ω j ℓ ) : = ( ∏ n = 1 ℓ ∣ R j ( n ) ∣ ) sup h ( ℓ ) ∈ ( R 2 ) ℓ , ∣ h ( ℓ ) ∣ = 1 , a ∈ D j κ ∑ α j ∈ A j ∣ V j , α j ; ω j ℓ ( m o n o ) ( a ) [ h ( ℓ ) ] ∣ , m_{\ell,j}^{(\mathrm{mono})}(\boldsymbol{\omega}_j^\ell) := \left(\prod_{n=1}^{\ell}|R_j^{(n)}|\right)\sup_{\substack{\mathbf{h}^{(\ell)}\in(\mathbb{R}^2)^\ell,\ |\mathbf{h}^{(\ell)}|=1,\\ \mathbf{a}\in\mathcal{D}_j^\kappa}}\sum_{\alpha_j\in\mathcal{A}_j}|V_{j,\alpha_j;\omega_j^\ell}^{(\mathrm{mono})}(\mathbf{a})[\mathbf{h}^{(\ell)}]|, m ℓ , j ( mono ) ( ω j ℓ ) := ( n = 1 ∏ ℓ ∣ R j ( n ) ∣ ) h ( ℓ ) ∈ ( R 2 ) ℓ , ∣ h ( ℓ ) ∣ = 1 , a ∈ D j κ sup α j ∈ A j ∑ ∣ V j , α j ; ω j ℓ ( mono ) ( a ) [ h ( ℓ ) ] ∣ ,
m ℓ , 3 − j ( i n t e r ) ( ω j ℓ ) : = sup h ( ℓ ) ∈ ( R 2 ) ℓ , ∣ h ( ℓ ) ∣ = 1 , a ∈ D j κ , x ∈ Γ M ∑ α 3 − j ∈ A 3 − j ∣ V 3 − j , α 3 − j ; ω j ℓ ( i n t e r ) ( Δ 3 − j , j [ 0 ] ( x , α 3 − j ) + a ) [ h ( ℓ ) ] ∣ . m_{\ell,3-j}^{(\mathrm{inter})}(\boldsymbol{\omega}_j^\ell) := \sup_{\substack{\mathbf{h}^{(\ell)}\in(\mathbb{R}^2)^\ell,\ |\mathbf{h}^{(\ell)}|=1,\\ \mathbf{a}\in\mathcal{D}_j^\kappa,\ x\in\Gamma_M}}\sum_{\alpha_{3-j}\in\mathcal{A}_{3-j}}\left|V_{3-j,\alpha_{3-j};\omega_j^\ell}^{(\mathrm{inter})}\left(\Delta_{3-j,j}[0](x,\alpha_{3-j})+\mathbf{a}\right)[\mathbf{h}^{(\ell)}]\right|. m ℓ , 3 − j ( inter ) ( ω j ℓ ) := h ( ℓ ) ∈ ( R 2 ) ℓ , ∣ h ( ℓ ) ∣ = 1 , a ∈ D j κ , x ∈ Γ M sup α 3 − j ∈ A 3 − j ∑ V 3 − j , α 3 − j ; ω j ℓ ( inter ) ( Δ 3 − j , j [ 0 ] ( x , α 3 − j ) + a ) [ h ( ℓ ) ] .
Assumption 2 (Decay assumption). We assume that there exists k ∈ N ≥ 5 k\in\mathbb{N}^{\geq5} k ∈ N ≥ 5 such that for all j ∈ { 1 , 2 } j\in\{1,2\} j ∈ { 1 , 2 } and ℓ ∈ { 1 , … , k } \ell\in\{1,\ldots,k\} ℓ ∈ { 1 , … , k } , we have that
M ℓ , j ( m o n o ) : = ∑ ω j ℓ ∈ Ω j ℓ m ℓ , j ( m o n o ) ( ω j ℓ ) < ∞ , M ℓ , 3 − j ( i n t e r ) : = ∑ ω j ℓ ∈ Ω j ℓ m ℓ , 3 − j ( i n t e r ) ( ω j ℓ ) < ∞ . M_{\ell,j}^{(\mathrm{mono})} := \sum_{\boldsymbol{\omega}_j^\ell\in\Omega_j^\ell}m_{\ell,j}^{(\mathrm{mono})}(\boldsymbol{\omega}_j^\ell)<\infty,\qquad M_{\ell,3-j}^{(\mathrm{inter})} := \sum_{\boldsymbol{\omega}_j^\ell\in\Omega_j^\ell}m_{\ell,3-j}^{(\mathrm{inter})}(\boldsymbol{\omega}_j^\ell)<\infty. M ℓ , j ( mono ) := ω j ℓ ∈ Ω j ℓ ∑ m ℓ , j ( mono ) ( ω j ℓ ) < ∞ , M ℓ , 3 − j ( inter ) := ω j ℓ ∈ Ω j ℓ ∑ m ℓ , 3 − j ( inter ) ( ω j ℓ ) < ∞. We also abbreviate M ℓ ( m ) : = ∑ j = 1 2 M ℓ , j ( m ) M_\ell^{(m)}:=\sum_{j=1}^{2}M_{\ell,j}^{(m)} M ℓ ( m ) := ∑ j = 1 2 M ℓ , j ( m ) for m ∈ { m o n o , i n t e r } m\in\{\mathrm{mono},\mathrm{inter}\} m ∈ { mono , inter } . In addition, we assume that
V j , α j ( m o n o ) ( 0 ) = 0 , sup x ∈ Γ M sup a ∈ D j κ ∣ V 3 − j , α 3 − j ( i n t e r ) ( Δ 3 − j , j [ 0 ] ( x , α 3 − j ) + a ) ∣ < ∞ . V_{j,\alpha_j}^{(\mathrm{mono})}(\mathbf{0})=0,\qquad\sup_{x\in\Gamma_M}\sup_{\mathbf{a}\in\mathcal{D}_j^\kappa}\left|V_{3-j,\alpha_{3-j}}^{(\mathrm{inter})}\left(\Delta_{3-j,j}[0](x,\alpha_{3-j})+\mathbf{a}\right)\right|<\infty. V j , α j ( mono ) ( 0 ) = 0 , x ∈ Γ M sup a ∈ D j κ sup V 3 − j , α 3 − j ( inter ) ( Δ 3 − j , j [ 0 ] ( x , α 3 − j ) + a ) < ∞. As we will see, this decay assumption suffices to establish the well-posedness of the relaxation problem in the Cauchy–Born approximation. In order to ensure the well-posedness for the relaxation problem associated with the general atomistic energy density, we will impose the following condition.
For p ∈ [ 1 , ∞ ] p\in[1,\infty] p ∈ [ 1 , ∞ ] , 2 ≤ ℓ ≤ m 2\leq\ell\leq m 2 ≤ ℓ ≤ m , and ω j ℓ ∈ Ω j ℓ \boldsymbol{\omega}_j^\ell\in\Omega_j^\ell ω j ℓ ∈ Ω j ℓ , where ω j ℓ = ( R j ( n ) , α j ( n ) ) ∈ Ω j \boldsymbol{\omega}_j^\ell=(R_j^{(n)},\alpha_j^{(n)})\in\Omega_j ω j ℓ = ( R j ( n ) , α j ( n ) ) ∈ Ω j , we define
m ℓ , p , j ( m o n o , a t ) ( ω j ℓ ) : = m_{\ell,p,j}^{(\mathrm{mono,at})}(\boldsymbol{\omega}_j^\ell):= m ℓ , p , j ( mono , at ) ( ω j ℓ ) := m ℓ , j ( m o n o ) ( ω j ℓ ) ( ∑ n = 1 ℓ ∣ R j ( n ) ∣ ) 2 ( ∑ n = 2 ℓ ( ∣ R j ( 1 ) × R j ( n ) ∣ + ∣ R j ( 1 ) ∣ + ∣ R j ( n ) ∣ ) ) 1 ( ℓ − 1 ) p . m_{\ell,j}^{(\mathrm{mono})}(\boldsymbol{\omega}_j^\ell)\left(\sum_{n=1}^{\ell}|R_j^{(n)}|\right)^2\left(\sum_{n=2}^{\ell}\left(|R_j^{(1)}\times R_j^{(n)}|+|R_j^{(1)}|+|R_j^{(n)}|\right)\right)^{\frac{1}{(\ell-1)p}}. m ℓ , j ( mono ) ( ω j ℓ ) ( n = 1 ∑ ℓ ∣ R j ( n ) ∣ ) 2 ( n = 2 ∑ ℓ ( ∣ R j ( 1 ) × R j ( n ) ∣ + ∣ R j ( 1 ) ∣ + ∣ R j ( n ) ∣ ) ) ( ℓ − 1 ) p 1 .
Assumption 3 (Sufficient decay for modelling error). We assume that there exists k ∈ N ≥ 3 k \in\mathbb{N}^{\ge3} k ∈ N ≥ 3 such that for all j ∈ { 1 , 2 } j \in\{1,2\} j ∈ { 1 , 2 } and ℓ ∈ { 1 , … , k } \ell\in\{1,\ldots,k\} ℓ ∈ { 1 , … , k } , we have that
M ℓ , p , j ( m o n o , a t ) : = ∑ ω j ℓ ∈ Ω j ℓ m ℓ , p , j ( m o n o , a t ) ( ω j ℓ ) < ∞ , M_{\ell,p,j}^{(\mathrm{mono},\mathrm{at})} := \sum_{\omega_j^\ell\in\Omega_j^\ell} m_{\ell,p,j}^{(\mathrm{mono},\mathrm{at})}(\omega_j^\ell) < \infty, M ℓ , p , j ( mono , at ) := ω j ℓ ∈ Ω j ℓ ∑ m ℓ , p , j ( mono , at ) ( ω j ℓ ) < ∞ , and we abbreviate M ℓ , p ( m o n o , a t ) : = ∑ j = 1 2 M ℓ , p , j ( m o n o , a t ) M_{\ell,p}^{(\mathrm{mono},\mathrm{at})} := \sum_{j=1}^{2} M_{\ell,p,j}^{(\mathrm{mono},\mathrm{at})} M ℓ , p ( mono , at ) := ∑ j = 1 2 M ℓ , p , j ( mono , at ) .
Stability As part of our assumptions, we will impose a lattice stability condition. This condition suffices to show the existence of a locally unique minimizer of e ( t o t ) e^{(\mathrm{tot})} e ( tot ) .
Assumption 4 (Stacking-based stability of lattice). We assume that
ν S t a c k : = min j = 1 , 2 inf ∥ ∇ v ∥ 2 = 1 δ 2 E j ( m o n o ) ( 0 ) [ v , v ] > 0. \nu_{\mathrm{Stack}} := \min_{j=1,2}\inf_{\lVert\nabla v\rVert_2=1}\delta^2 E_j^{(\mathrm{mono})}(0)[v,v] > 0. ν Stack := j = 1 , 2 min ∥ ∇ v ∥ 2 = 1 inf δ 2 E j ( mono ) ( 0 ) [ v , v ] > 0.
One could state an analogous assumption in terms δ 2 Δ E j ( m o n o ) ( 0 ) \delta^2\Delta E_j^{(\mathrm{mono})}(0) δ 2 Δ E j ( mono ) ( 0 ) . Crucially, such a condition in addition to smallness of δ 2 Δ E j ( i n t e r ) ( 0 ) \delta^2\Delta E_j^{(\mathrm{inter})}(0) δ 2 Δ E j ( inter ) ( 0 ) does not suffice to prove existence of a locally unique minimizer, nor its stability.
Convergence of energy functionals
We now address the convergence of the total energy density e ( t o t ) ( u ) e^{(\mathrm{tot})}(\mathbf{u}) e ( tot ) ( u ) as well as of the total offset energy Δ E u ( t o t ) ( δ u ) \Delta E_{\mathbf{u}}^{(\mathrm{tot})}(\delta\mathbf{u}) Δ E u ( tot ) ( δ u ) . Crucially, we will see that u \mathbf{u} u will be uniformly non-decaying, think L l o c 1 L^1_{\mathrm{loc}} L loc 1 , while δ u \delta\mathbf{u} δ u will be required to decay in ℓ 1 ( Ω ) \ell^1(\Omega) ℓ 1 ( Ω ) .
Energy density
We start by recalling some results established in [18 ] . A useful tool to prove the existence of the thermodynamic limit e ( t o t ) e^{(\mathrm{tot})} e ( tot ) is the following Birkhoff ergodic theorem , see [20 ] and also [18 Proposition 3.3] for this formulation.
Proposition 2.3 (Ergodic Theorem). Let f ∈ L p e r 1 ( Γ M ; L p e r 1 ( Γ j ) ) f \in L^1_{\mathrm{per}}(\Gamma_M;L^1_{\mathrm{per}}(\Gamma_j)) f ∈ L per 1 ( Γ M ; L per 1 ( Γ j )) . Then
lim N → ∞ ∑ R j ∈ R j ( N ) f ( R j + ω M , R j + ω 3 − j ) ( 2 N + 1 ) 2 = ∫ Γ M d x f ( x + ω M , ( I − A 3 − j A j − 1 ) x + ω 3 − j ) \lim_{N\to\infty}\sum_{R_j\in\mathcal{R}_j(N)}\frac{f(R_j+\omega_M,R_j+\omega_{3-j})}{(2N+1)^2}
=\int_{\Gamma_M} \mathrm{d}x\,f\left(x+\omega_M,(I-A_{3-j}A_j^{-1})x+\omega_{3-j}\right) N → ∞ lim R j ∈ R j ( N ) ∑ ( 2 N + 1 ) 2 f ( R j + ω M , R j + ω 3 − j ) = ∫ Γ M d x f ( x + ω M , ( I − A 3 − j A j − 1 ) x + ω 3 − j ) converges for almost all ( ω M , ω 3 − j ) ∈ Γ M × Γ 3 − j (\omega_M,\omega_{3-j})\in\Gamma_M\times\Gamma_{3-j} ( ω M , ω 3 − j ) ∈ Γ M × Γ 3 − j and also in L 1 ( Γ M × Γ 3 − j ) L^1(\Gamma_M\times\Gamma_{3-j}) L 1 ( Γ M × Γ 3 − j ) .
In order to employ Proposition 2.3 to ( e N , γ ( t o t ) ) N (e_{N,\gamma}^{(\mathrm{tot})})_N ( e N , γ ( tot ) ) N , we also introduce the following generalized site potential
Ψ j ( i n t e r ) [ u ] ( x , y ) : = 1 ∣ Γ j ∣ ∑ α j ∈ A j V j , α j ( i n t e r ) ( ( Y 3 − j ( x − y + R 3 − j , α 3 − j ) − Y j ( x , α j ) ) ω 3 − j ∈ Ω 3 − j ) . \Psi_j^{(\mathrm{inter})}[\mathbf{u}](x,y):=
\frac{1}{|\Gamma_j|}\sum_{\alpha_j\in\mathcal{A}_j}V_{j,\alpha_j}^{(\mathrm{inter})}\left((Y_{3-j}(x-y+R_{3-j,\alpha_{3-j}})-Y_j(x,\alpha_j))_{\omega_{3-j}\in\Omega_{3-j}}\right). Ψ j ( inter ) [ u ] ( x , y ) := ∣ Γ j ∣ 1 α j ∈ A j ∑ V j , α j ( inter ) ( ( Y 3 − j ( x − y + R 3 − j , α 3 − j ) − Y j ( x , α j ) ) ω 3 − j ∈ Ω 3 − j ) .
Theorem 2.4. Let u ∈ D κ \mathbf{u} \in\mathcal{D}_{\kappa} u ∈ D κ . Assume that ϕ j ( m o n o ) [ u j ] ∈ L 1 ( Γ M ) \phi_{j}^{(\mathrm{mono})}[u_{j}] \in L^{1}(\Gamma_{M}) ϕ j ( mono ) [ u j ] ∈ L 1 ( Γ M ) , and that Ψ j ( i n t e r ) [ u ] ∈ L 1 ( Γ M × Γ 3 − j ) \Psi_{j}^{(\mathrm{inter})}[\mathbf{u}] \in L^{1}(\Gamma_{M} \times\Gamma_{3-j}) Ψ j ( inter ) [ u ] ∈ L 1 ( Γ M × Γ 3 − j ) and for any γ = ( γ 1 , γ 2 ) ∈ Γ 1 × Γ 2 \gamma= (\gamma_{1},\gamma_{2}) \in\Gamma_{1} \times\Gamma_{2} γ = ( γ 1 , γ 2 ) ∈ Γ 1 × Γ 2 that ϕ j , γ ( i n t e r ) [ u ] ∈ L 1 ( Γ M ) \phi_{j,\gamma}^{(\mathrm{inter})}[\mathbf{u}] \in L^{1}(\Gamma_{M}) ϕ j , γ ( inter ) [ u ] ∈ L 1 ( Γ M ) , respectively. We then have for almost all γ ∈ Γ 1 × Γ 2 \gamma\in\Gamma_{1} \times\Gamma_{2} γ ∈ Γ 1 × Γ 2 that
lim N → ∞ e N , γ ( t o t ) ( u ) = e γ ( t o t ) ( u ) . \lim_{N \to\infty} e_{N,\gamma}^{(\mathrm{tot})}(\mathbf{u}) = e_{\gamma}^{(\mathrm{tot})}(\mathbf{u}). N → ∞ lim e N , γ ( tot ) ( u ) = e γ ( tot ) ( u ) .
Remark 2.5. In [18 ] , it was also proved that, under a Diophantine condition on A 1 A_{1} A 1 , A 2 A_{2} A 2 , and requiring additional regularity of u \mathbf{u} u , V ( m o n o ) V^{(\mathrm{mono})} V ( mono ) , and V ( i n t e r ) V^{(\mathrm{inter})} V ( inter ) , one obtains a rate of convergence of O ( N − 1 ) O(N^{-1}) O ( N − 1 ) . There, the result was only proved in the case γ = 0 \gamma= 0 γ = 0 , but the ideas presented there can be extended to the case γ ≠ 0 \gamma\ne0 γ = 0 . The Diophantine condition presented in [18 ] can be, more generally, stated as
dist ( A 1 T A 2 − T n , Z 2 ) ≥ K ∣ n ∣ 2 σ ∀ n ∈ Z 2 ∖ { 0 } . \operatorname{dist}(A_{1}^{T}A_{2}^{-T}n,\mathbb{Z}^{2}) \ge\frac{K}{\lvert n\rvert^{2\sigma}} \qquad\forall n \in\mathbb{Z}^{2} \setminus\{0\}. dist ( A 1 T A 2 − T n , Z 2 ) ≥ ∣ n ∣ 2 σ K ∀ n ∈ Z 2 ∖ { 0 } .
In particular, Diophantineness is a quantification of incommensurability.
We are left with proving that the assumptions on V ( m o n o ) V^{(\mathrm{mono})} V ( mono ) and V ( i n t e r ) V^{(\mathrm{inter})} V ( inter ) stated in § 2.6 imply the conditions of Theorem 2.4 .
Our assumptions immediately imply that ϕ j ( i n t e r ) ∈ L 1 ( Γ M ) \phi_{j}^{(\mathrm{inter})} \in L^{1}(\Gamma_{M}) ϕ j ( inter ) ∈ L 1 ( Γ M ) and that Ψ j ( i n t e r ) ∈ L 1 ( Γ M × Γ 3 − j ) \Psi_{j}^{(\mathrm{inter})} \in L^{1}(\Gamma_{M} \times\Gamma_{3-j}) Ψ j ( inter ) ∈ L 1 ( Γ M × Γ 3 − j ) . In the case of ϕ j ( m o n o ) \phi_{j}^{(\mathrm{mono})} ϕ j ( mono ) , observe that
∣ ϕ j ( m o n o ) [ u j ] ( x ) − ϕ j ( m o n o ) [ 0 ] ∣ ≲ 1 ∣ Γ j ∣ sup α j ∑ ω j ′ ∈ Ω j ′ m 1 , j ( m o n o ) ( ω j ′ ) ∣ u j ( ω j ′ + x ) − u j ( x , α j ) ∣ ∣ R j ′ ∣ \left\lvert\phi_{j}^{(\mathrm{mono})}[u_{j}](x) - \phi_{j}^{(\mathrm{mono})}[0] \right\rvert\lesssim\frac{1}{\lvert\Gamma_{j} \rvert} \sup_{\alpha_{j}} \sum_{\omega_{j}' \in\Omega_{j}'} m_{1,j}^{(\mathrm{mono})}(\omega_{j}') \frac{\left\lvert u_{j}(\omega_{j}' + x) - u_{j}(x,\alpha_{j}) \right\rvert}{\lvert R_{j}' \rvert} ϕ j ( mono ) [ u j ] ( x ) − ϕ j ( mono ) [ 0 ] ≲ ∣ Γ j ∣ 1 α j sup ω j ′ ∈ Ω j ′ ∑ m 1 , j ( mono ) ( ω j ′ ) ∣ R j ′ ∣ u j ( ω j ′ + x ) − u j ( x , α j )
≲ κ M 1 , j ( m o n o ) ∣ Γ j ∣ , \lesssim\frac{\kappa M_{1,j}^{(\mathrm{mono})}}{\lvert\Gamma_{j} \rvert}, ≲ ∣ Γ j ∣ κ M 1 , j ( mono ) ,
where we used the fact that u ∈ D κ \mathbf{u} \in\mathcal{D}_{\kappa} u ∈ D κ . In particular, we obtain that ϕ j ( m o n o ) ∈ L 1 ( Γ M ) \phi_{j}^{(\mathrm{mono})} \in L^{1}(\Gamma_{M}) ϕ j ( mono ) ∈ L 1 ( Γ M ) .
Offset energy
Expanding Δ E u ( t o t ) \Delta E_{\mathbf{u}}^{(\mathrm{tot})} Δ E u ( tot ) up to second order in δ u \delta\mathbf{u} δ u , we obtain
Δ E u ( t o t ) ( δ u ) = ∑ j = 1 2 ∑ ω j ∈ Ω j [ ∑ ω j ′ V j , α j ; ω j ′ ( m o n o ) ( Δ j [ u j ( e q ) ] ) ( ω j ) ⋅ Δ ω j ′ ( j ) δ u j ( ω j ) + ∫ 0 1 d t ( 1 − t ) ∑ ω j ′ , ω j ′ ′ V j , α j ; ω j ′ , ω j ′ ′ ( m o n o ) ( Δ j [ u j ( e q ) + t δ u j ] ) ( ω j ) \Delta E_{\mathbf{u}}^{(\mathrm{tot})}(\delta\mathbf{u}) =
\sum_{j=1}^{2}\sum_{\omega_{j}\in\Omega_{j}}[\sum_{\omega_{j}'} V_{j,\alpha_{j};\omega_{j}'}^{(\mathrm{mono})}(\Delta_{j}[u_{j}^{(\mathrm{eq})}])(\omega_{j})\cdot\Delta_{\omega_{j}'}^{(j)}\delta u_{j}(\omega_{j})
+
\int_{0}^{1}\mathrm{d}t\,(1-t)\sum_{\omega_{j}',\omega_{j}''}V_{j,\alpha_{j};\omega_{j}',\omega_{j}''}^{(\mathrm{mono})}(\Delta_{j}[u_{j}^{(\mathrm{eq})}+t\delta u_{j}])(\omega_{j}) Δ E u ( tot ) ( δ u ) = j = 1 ∑ 2 ω j ∈ Ω j ∑ [ ω j ′ ∑ V j , α j ; ω j ′ ( mono ) ( Δ j [ u j ( eq ) ]) ( ω j ) ⋅ Δ ω j ′ ( j ) δ u j ( ω j ) + ∫ 0 1 d t ( 1 − t ) ω j ′ , ω j ′′ ∑ V j , α j ; ω j ′ , ω j ′′ ( mono ) ( Δ j [ u j ( eq ) + t δ u j ]) ( ω j )
[ Δ ω j ( j ) δ u j ( ω j ) , Δ ω j ′ ( j ) δ u j ( ω j ) ] + ∑ ω 3 − j V j , α j ; ω 3 − j ( i n t e r ) ( Δ j , 3 − j [ u ( e q ) ] ( ω j ) ⋅ Δ ω 3 − j ( j , 3 − j ) δ u ( ω j ) ) + ∫ 0 1 d t ( 1 − t ) ∑ ω 3 − j , ω 3 − j ′ V j , α j ; ω 3 − j , ω 3 − j ′ ( i n t e r ) ( Δ j , 3 − j [ u ( e q ) + t δ u ] ( ω j ) ) [ Δ ω 3 − j ( j , 3 − j ) δ u ( ω j ) , Δ ω 3 − j ′ ( j , 3 − j ) δ u ( ω j ) ] . \begin{aligned}
\left[\Delta_{\omega_j}^{(j)}\delta u_j(\omega_j),\Delta_{\omega'_j}^{(j)}\delta u_j(\omega_j)\right]
+\sum_{\omega_{3-j}}V_{j,\alpha_j;\omega_{3-j}}^{(\mathrm{inter})}\left(\Delta_{j,3-j}[u^{(\mathrm{eq})}](\omega_j)\cdot\Delta_{\omega_{3-j}}^{(j,3-j)}\delta u(\omega_j)\right) \\
&+\int_0^1\mathrm{d}t\,(1-t)\sum_{\omega_{3-j},\omega'_{3-j}}V_{j,\alpha_j;\omega_{3-j},\omega'_{3-j}}^{(\mathrm{inter})}\left(\Delta_{j,3-j}[u^{(\mathrm{eq})}+t\delta u](\omega_j)\right)
\left[\Delta_{\omega_{3-j}}^{(j,3-j)}\delta u(\omega_j),\Delta_{\omega'_{3-j}}^{(j,3-j)}\delta u(\omega_j)\right].
\end{aligned} [ Δ ω j ( j ) δ u j ( ω j ) , Δ ω j ′ ( j ) δ u j ( ω j ) ] + ω 3 − j ∑ V j , α j ; ω 3 − j ( inter ) ( Δ j , 3 − j [ u ( eq ) ] ( ω j ) ⋅ Δ ω 3 − j ( j , 3 − j ) δ u ( ω j ) ) + ∫ 0 1 d t ( 1 − t ) ω 3 − j , ω 3 − j ′ ∑ V j , α j ; ω 3 − j , ω 3 − j ′ ( inter ) ( Δ j , 3 − j [ u ( eq ) + t δ u ] ( ω j ) ) [ Δ ω 3 − j ( j , 3 − j ) δ u ( ω j ) , Δ ω 3 − j ′ ( j , 3 − j ) δ u ( ω j ) ] .
A straight-forward calculation then yields that
∣ Δ u ( t o t ) ( δ u ) ∣ ≲ ( M 1 ( m o n o ) + M 1 ( i n t e r ) ) ∥ δ u ∥ ℓ 1 + ( M 2 ( m o n o ) + M 2 ( i n t e r ) ) ∥ δ u ∥ ℓ 2 2 . \left|\Delta_{u}^{(\mathrm{tot})}(\delta u)\right|\lesssim\left(M_1^{(\mathrm{mono})}+M_1^{(\mathrm{inter})}\right)\|\delta u\|_{\ell^1}+\left(M_2^{(\mathrm{mono})}+M_2^{(\mathrm{inter})}\right)\|\delta u\|_{\ell^2}^2. Δ u ( tot ) ( δ u ) ≲ ( M 1 ( mono ) + M 1 ( inter ) ) ∥ δ u ∥ ℓ 1 + ( M 2 ( mono ) + M 2 ( inter ) ) ∥ δ u ∥ ℓ 2 2 .
First variation
Our goal is to derive the expressions for the first Fréchet derivative of the energy density as well as of the offset energy. For now, we will assume sufficient regularity of e ( t o t ) e^{(\mathrm{tot})} e ( tot ) and Δ E u ( t o t ) \Delta E_u^{(\mathrm{tot})} Δ E u ( tot ) , and assume the existence of a critical point u ( e q ) = ( u 1 ( e q ) , u 2 ( e q ) ) u^{(\mathrm{eq})}=(u_1^{(\mathrm{eq})},u_2^{(\mathrm{eq})}) u ( eq ) = ( u 1 ( eq ) , u 2 ( eq ) ) of e ( t o t ) e^{(\mathrm{tot})} e ( tot ) . In Section 3, we will show that our assumptions on V ( m o n o ) V^{(\mathrm{mono})} V ( mono ) and V ( i n t e r ) V^{(\mathrm{inter})} V ( inter ) suffice to justify the following results.
For simplicity, let us set γ 1 = γ 2 = 0 \gamma_1=\gamma_2=0 γ 1 = γ 2 = 0 . It is useful to introduce the following notation:
Thermodynamic limit energy density
For v = ( v 1 , v 2 ) ∈ W ‾ p e r \mathbf{v}=(v_1,v_2)\in\overline{W}_{\mathrm{per}} v = ( v 1 , v 2 ) ∈ W per , we have that
δ e ( t o t ) ( u ) [ v ] = ∑ j = 1 2 1 ∣ Γ j ∣ ∑ α j ∈ A j ∫ Γ M d x ( ∑ ω j ′ V j , α j ; ω j ′ ( m o n o ) ( Δ j [ u j ] ( x , α j ) ⋅ ( v j ( ω j ′ + x ) − v j ( x , α j ) ) ) + ∑ ω 3 − j V j , α j ; ω 3 − j ( i n t e r ) ( Δ j , 3 − j [ u ] ( x , α j ) ⋅ ( v 3 − j ( ω 3 − j + A 3 − j A j − 1 x ) − v j ( x , α j ) ) ) ) . \delta e^{(\mathrm{tot})}(\mathbf{u})[\mathbf{v}]=
\sum_{j=1}^{2}\frac{1}{|\Gamma_j|}\sum_{\alpha_j\in\mathcal{A}_j}\int_{\Gamma_{\mathcal{M}}}\mathrm{d}x\left(\sum_{\omega'_j}V_{j,\alpha_j;\omega'_j}^{(\mathrm{mono})}\left(\Delta_j[u_j](x,\alpha_j)\cdot\left(v_j(\omega'_j+x)-v_j(x,\alpha_j)\right)\right)
+\sum_{\omega_{3-j}}V_{j,\alpha_j;\omega_{3-j}}^{(\mathrm{inter})}\left(\Delta_{j,3-j}[u](x,\alpha_j)\cdot\left(v_{3-j}(\omega_{3-j}+A_{3-j}A_j^{-1}x)-v_j(x,\alpha_j)\right)\right)\right). δ e ( tot ) ( u ) [ v ] = j = 1 ∑ 2 ∣ Γ j ∣ 1 α j ∈ A j ∑ ∫ Γ M d x ω j ′ ∑ V j , α j ; ω j ′ ( mono ) ( Δ j [ u j ] ( x , α j ) ⋅ ( v j ( ω j ′ + x ) − v j ( x , α j ) ) ) + ω 3 − j ∑ V j , α j ; ω 3 − j ( inter ) ( Δ j , 3 − j [ u ] ( x , α j ) ⋅ ( v 3 − j ( ω 3 − j + A 3 − j A j − 1 x ) − v j ( x , α j ) ) ) .
We now sort terms according to the coefficients v j v_j v j
δ e ( t o t ) ( u ) [ v ] = ∑ j = 1 2 1 ∣ Γ j ∣ ∣ Γ M ∣ ∑ α j ∈ A j ∫ d x [ ∑ ω j ′ ( V j , α j ; ω j ′ ( m o n o ) ( Δ j [ u j ] ( x − R j ′ , α j ′ ) ) 1 Γ M + R j ′ ( x ) − V j , α j ; ω j ′ ( m o n o ) ( Δ j [ u j ] ( x , α j ) ) 1 Γ M ( x ) ) − ∑ ω 3 − j V j , α j ; ω 3 − j ( i n t e r ) ( Δ j , 3 − j [ u ] ( x , α j ) ) 1 Γ M ( x ) + ∑ R j , α 3 − j V 3 − j , α 3 − j ; ω j ( i n t e r ) ( Δ 3 − j , j [ u ] ( A 3 − j A j − 1 ( x − R j ) , α 3 − j ) ) 1 A j A 3 − j − 1 Γ M + R j ( x ) ⋅ v j ( x , α j ) . \begin{aligned}
\delta e^{(\mathrm{tot})}(\mathbf{u})[\mathbf{v}]=
\sum_{j=1}^{2}\frac{1}{|\Gamma_j||\Gamma_{\mathcal{M}}|}\sum_{\alpha_j\in\mathcal{A}_j}\int\mathrm{d}x[\sum_{\omega'_j}(V_{j,\alpha_j;\omega'_j}^{(\mathrm{mono})}(\Delta_j[u_j](x-R'_j,\alpha'_j))\mathbf{1}_{\Gamma_{\mathcal{M}}+R'_j}(x)
-V_{j,\alpha_j;\omega'_j}^{(\mathrm{mono})}(\Delta_j[u_j](x,\alpha_j))\mathbf{1}_{\Gamma_{\mathcal{M}}}(x))
-\sum_{\omega_{3-j}}V_{j,\alpha_j;\omega_{3-j}}^{(\mathrm{inter})}(\Delta_{j,3-j}[u](x,\alpha_j))\mathbf{1}_{\Gamma_{\mathcal{M}}}(x) \\
&+ \sum_{R_j,\alpha_{3-j}} V^{(\mathrm{inter})}_{3-j,\alpha_{3-j};\omega_j}\left(\Delta_{3-j,j}[\mathbf{u}]\left(A_{3-j}A_j^{-1}(x-R_j),\alpha_{3-j}\right)\right)\mathbf{1}_{A_jA_{3-j}^{-1}\Gamma_M+R_j}(x) \\
&\cdot v_j(x,\alpha_j).
\end{aligned} δ e ( tot ) ( u ) [ v ] = j = 1 ∑ 2 ∣ Γ j ∣∣ Γ M ∣ 1 α j ∈ A j ∑ ∫ d x [ ω j ′ ∑ ( V j , α j ; ω j ′ ( mono ) ( Δ j [ u j ] ( x − R j ′ , α j ′ )) 1 Γ M + R j ′ ( x ) − V j , α j ; ω j ′ ( mono ) ( Δ j [ u j ] ( x , α j )) 1 Γ M ( x )) − ω 3 − j ∑ V j , α j ; ω 3 − j ( inter ) ( Δ j , 3 − j [ u ] ( x , α j )) 1 Γ M ( x ) + R j , α 3 − j ∑ V 3 − j , α 3 − j ; ω j ( inter ) ( Δ 3 − j , j [ u ] ( A 3 − j A j − 1 ( x − R j ) , α 3 − j ) ) 1 A j A 3 − j − 1 Γ M + R j ( x ) ⋅ v j ( x , α j ) .
Observe that x ↦ D R j V ( m o n o ) ( Δ j [ u j ] ( x ) ) x\mapsto D_{R_j}V^{(\mathrm{mono})}\left(\Delta_j[u_j](x)\right) x ↦ D R j V ( mono ) ( Δ j [ u j ] ( x ) ) and v j v_j v j are R M \mathcal{R}_M R M -periodic. That allows us to simplify the expression as
δ ε ( t o t ) ( u ) [ v ] = \delta\varepsilon^{(\mathrm{tot})}(\mathbf{u})[\mathbf{v}]= δ ε ( tot ) ( u ) [ v ] =
∑ j = 1 2 1 ∣ Γ j ∣ ∑ α j ∈ A j ∫ Γ M d x [ ∑ ω j ′ V j , α j ; ω j ′ ( m o n o ) ( Δ j [ u j ] ( x − R j ′ , α j ′ ) ) − V j , α j ; ω j ′ ( m o n o ) ( Δ j [ u j ] ( x , α j ) ) \sum_{j=1}^{2}\frac{1}{|\Gamma_j|}\sum_{\alpha_j\in A_j}\int_{\Gamma_M}\mathrm{d}x\left[\sum_{\omega'_j}V^{(\mathrm{mono})}_{j,\alpha_j;\omega'_j}\left(\Delta_j[u_j]\left(x-R'_j,\alpha'_j\right)\right)-V^{(\mathrm{mono})}_{j,\alpha_j;\omega'_j}\left(\Delta_j[u_j](x,\alpha_j)\right)\right. j = 1 ∑ 2 ∣ Γ j ∣ 1 α j ∈ A j ∑ ∫ Γ M d x ω j ′ ∑ V j , α j ; ω j ′ ( mono ) ( Δ j [ u j ] ( x − R j ′ , α j ′ ) ) − V j , α j ; ω j ′ ( mono ) ( Δ j [ u j ] ( x , α j ) )
+ ∑ R j , R M , α 3 − j V 3 − j , α 3 − j ; ω j ( i n t e r ) ( Δ 3 − j , j [ u ] ( A 3 − j A j − 1 ( x − R j − R M ) , α 3 − j ) ) 1 A j A 3 − j − 1 Γ M + R j + R M ( x ) − ∑ ω 3 − j V j , α j ; ω 3 − j ( i n t e r ) ( Δ j , 3 − j [ u ] ( x , α j ) ) ] ⋅ v j ( x , α j ) . \left.+\sum_{R_j,R_M,\alpha_{3-j}}V^{(\mathrm{inter})}_{3-j,\alpha_{3-j};\omega_j}\left(\Delta_{3-j,j}[\mathbf{u}]\left(A_{3-j}A_j^{-1}(x-R_j-R_M),\alpha_{3-j}\right)\right)\mathbf{1}_{A_jA_{3-j}^{-1}\Gamma_M+R_j+R_M}(x)-\sum_{\omega_{3-j}}V^{(\mathrm{inter})}_{j,\alpha_j;\omega_{3-j}}\left(\Delta_{j,3-j}[\mathbf{u}](x,\alpha_j)\right)\right]\cdot v_j(x,\alpha_j). + R j , R M , α 3 − j ∑ V 3 − j , α 3 − j ; ω j ( inter ) ( Δ 3 − j , j [ u ] ( A 3 − j A j − 1 ( x − R j − R M ) , α 3 − j ) ) 1 A j A 3 − j − 1 Γ M + R j + R M ( x ) − ω 3 − j ∑ V j , α j ; ω 3 − j ( inter ) ( Δ j , 3 − j [ u ] ( x , α j ) ) ⋅ v j ( x , α j ) .
We would like to simplify the the third term. For that, observe that translation-invariance of V ( i n t e r ) V^{(\mathrm{inter})} V ( inter ) implies
V 3 − j ; R j ( i n t e r ) ( ( a R j ′ , R j ′ ) ) = V 3 − j ; 0 ( i n t e r ) ( ( a R j ′ + R j ) R j ′ ) . (8) V^{(\mathrm{inter})}_{3-j;R_j}\left((aR'_j,R'_j)\right)=V^{(\mathrm{inter})}_{3-j;0}\left((aR'_{j}+R_j)R'_j\right). \tag*{(8)} V 3 − j ; R j ( inter ) ( ( a R j ′ , R j ′ ) ) = V 3 − j ; 0 ( inter ) ( ( a R j ′ + R j ) R j ′ ) . (8)
Ignoring the α \alpha α -dependence for now, we thus find that
∑ R j , R M V 3 − j ; R j ( i n t e r ) ( Δ 3 − j , j [ u ] ( A 3 − j A j − 1 ( x − R j − R M ) ) ) 1 A j A 3 − j − 1 Γ M + R j + R M ( x ) = ∑ R j , R M V 3 − j ; R j ( i n t e r ) ( ( Y 3 − j ( A 3 − j A j − 1 ( x − R j − R M ) ) − Y j ( x − R j − R M + R j ′ ) ) R j ′ ) 1 A j A 3 − j − 1 Γ M + R j + R M ( x ) = ∑ R j , R M V 3 − j ; 0 ( i n t e r ) ( ( Y 3 − j ( A 3 − j A j − 1 ( x − R j − R M ) ) − Y j ( x − R M + R j ′ ) ) R j ′ ) 1 A j A 3 − j − 1 Γ M + R j + R M ( x ) . (9) \begin{aligned}
\sum_{R_j,R_M}V^{(\mathrm{inter})}_{3-j;R_j}\left(\Delta_{3-j,j}[\mathbf{u}]\left(A_{3-j}A_j^{-1}(x-R_j-R_M)\right)\right)\mathbf{1}_{A_jA_{3-j}^{-1}\Gamma_M+R_j+R_M}(x) \\
&=\sum_{R_j,R_M}V^{(\mathrm{inter})}_{3-j;R_j}\left(\left(Y_{3-j}\left(A_{3-j}A_j^{-1}(x-R_j-R_M)\right)-Y_j\left(x-R_j-R_M+R'_j\right)\right)_{R'_j}\right)\mathbf{1}_{A_jA_{3-j}^{-1}\Gamma_M+R_j+R_M}(x) \\
&=\sum_{R_j,R_M}V^{(\mathrm{inter})}_{3-j;0}\left(\left(Y_{3-j}\left(A_{3-j}A_j^{-1}(x-R_j-R_M)\right)-Y_j\left(x-R_M+R'_j\right)\right)_{R'_j}\right)\mathbf{1}_{A_jA_{3-j}^{-1}\Gamma_M+R_j+R_M}(x). \tag*{(9)}
\end{aligned} R j , R M ∑ V 3 − j ; R j ( inter ) ( Δ 3 − j , j [ u ] ( A 3 − j A j − 1 ( x − R j − R M ) ) ) 1 A j A 3 − j − 1 Γ M + R j + R M ( x ) = R j , R M ∑ V 3 − j ; R j ( inter ) ( ( Y 3 − j ( A 3 − j A j − 1 ( x − R j − R M ) ) − Y j ( x − R j − R M + R j ′ ) ) R j ′ ) 1 A j A 3 − j − 1 Γ M + R j + R M ( x ) = R j , R M ∑ V 3 − j ; 0 ( inter ) ( ( Y 3 − j ( A 3 − j A j − 1 ( x − R j − R M ) ) − Y j ( x − R M + R j ′ ) ) R j ′ ) 1 A j A 3 − j − 1 Γ M + R j + R M ( x ) . (9)
Thanks to Lemma 2.1 , we have that
− A 3 − j A j − 1 R M + R M = D j → 3 − j R M = ( − 1 ) j A 3 − j A M − 1 R M . -A_{3-j}A_j^{-1}R_M+R_M=D_{j\to3-j}R_M=(-1)^jA_{3-j}A_M^{-1}R_M. − A 3 − j A j − 1 R M + R M = D j → 3 − j R M = ( − 1 ) j A 3 − j A M − 1 R M .
Thus we obtain
− A 3 − j A j − 1 ( R j + R M ) + R M = ( − 1 ) j + 1 A 3 − j A M − 1 ( ( − 1 ) j A M A j − 1 R j − R M ) . (10) -A_{3-j}A_j^{-1}(R_j+R_M)+R_M=(-1)^{j+1}A_{3-j}A_M^{-1}\left((-1)^jA_MA_j^{-1}R_j-R_M\right). \tag*{(10)} − A 3 − j A j − 1 ( R j + R M ) + R M = ( − 1 ) j + 1 A 3 − j A M − 1 ( ( − 1 ) j A M A j − 1 R j − R M ) . (10)
Due to R M \mathcal{R}_M R M -periodicity of u 3 − j u_{3-j} u 3 − j , we thus find that
u 3 − j ( A 3 − j A j − 1 ( x − R j − R M ) ) = u 3 − j ( A 3 − j A j − 1 ( x − R j − R M ) + R M ) = u 3 − j ( A 3 − j A j − 1 x + ( − 1 ) j + 1 A 3 − j A M − 1 ( ( − 1 ) j A M A j − 1 R j − R M ) ) . (11) \begin{aligned}
u_{3-j}\left(A_{3-j}A_j^{-1}(x-R_j-R_M)\right)
&=u_{3-j}\left(A_{3-j}A_j^{-1}(x-R_j-R_M)+R_M\right)\\
&=u_{3-j}\left(A_{3-j}A_j^{-1}x+(-1)^{j+1}A_{3-j}A_M^{-1}\left((-1)^jA_MA_j^{-1}R_j-R_M\right)\right). \tag*{(11)}
\end{aligned} u 3 − j ( A 3 − j A j − 1 ( x − R j − R M ) ) = u 3 − j ( A 3 − j A j − 1 ( x − R j − R M ) + R M ) = u 3 − j ( A 3 − j A j − 1 x + ( − 1 ) j + 1 A 3 − j A M − 1 ( ( − 1 ) j A M A j − 1 R j − R M ) ) . (11)
As a consequence of {(xref)}(2.7) and {(xref)}(2.8), we shift R M → R M + ( − 1 ) j A M A j − 1 R j R_M\to R_M+(-1)^jA_MA_j^{-1}R_j R M → R M + ( − 1 ) j A M A j − 1 R j in {(xref)}(2.6) to obtain
∑ R M , R j V 3 − j ; 0 ( i n t e r ) ( ( Y 3 − j ( A 3 − j A j − 1 ( x − R M ) ) − Y j ( x − R M + R j ′ ) ) R j ′ ) 1 A j A 3 − j − 1 Γ M + ( I + ( − 1 ) j A M A j − 1 ) R j + R M ( x ) . (12) \sum_{R_M,R_j}V_{3-j;0}^{(\mathrm{inter})}\left(\left(Y_{3-j}\left(A_{3-j}A_j^{-1}(x-R_M)\right)-Y_j\left(x-R_M+R'_j\right)\right)_{R'_j}\right)
\mathbf{1}_{A_jA_{3-j}^{-1}\Gamma_M+(I+(-1)^jA_MA_j^{-1})R_j+R_M}(x). \tag*{(12)} R M , R j ∑ V 3 − j ; 0 ( inter ) ( ( Y 3 − j ( A 3 − j A j − 1 ( x − R M ) ) − Y j ( x − R M + R j ′ ) ) R j ′ ) 1 A j A 3 − j − 1 Γ M + ( I + ( − 1 ) j A M A j − 1 ) R j + R M ( x ) . (12)
Notice the only R j R_j R j -dependence is on the characteristic function. Moreover, we have that
( I + ( − 1 ) j A M A j − 1 ) R j = A j ( A M − 1 + ( − 1 ) j A j − 1 ) A M A j − 1 R j = ( − 1 ) j + 1 A j A 3 − j − 1 A M A j − 1 R j , \begin{aligned}
\left(I+(-1)^jA_MA_j^{-1}\right)R_j
&=A_j\left(A_M^{-1}+(-1)^jA_j^{-1}\right)A_MA_j^{-1}R_j\\
&=(-1)^{j+1}A_jA_{3-j}^{-1}A_MA_j^{-1}R_j,
\end{aligned} ( I + ( − 1 ) j A M A j − 1 ) R j = A j ( A M − 1 + ( − 1 ) j A j − 1 ) A M A j − 1 R j = ( − 1 ) j + 1 A j A 3 − j − 1 A M A j − 1 R j ,
which, in turn, implies
∑ R j 1 A j A 3 − j − 1 Γ M + ( I + ( − 1 ) j A M A j − 1 ) R j + R M ( x ) = 1. (13) \sum_{R_j}\mathbf{1}_{A_jA_{3-j}^{-1}\Gamma_M+(I+(-1)^jA_MA_j^{-1})R_j+R_M}(x)=1. \tag*{(13)} R j ∑ 1 A j A 3 − j − 1 Γ M + ( I + ( − 1 ) j A M A j − 1 ) R j + R M ( x ) = 1. (13)
Combining {(xref)}(2.9) and {(xref)}(2.10), we obtain
∑ R M V 3 − j ; 0 ( i n t e r ) ( ( Y 3 − j ( A 3 − j A j − 1 ( x − R M ) ) − Y j ( x − R M + R j ′ ) ) R j ′ ) . (14) \sum_{R_M}V_{3-j;0}^{(\mathrm{inter})}\left(\left(Y_{3-j}\left(A_{3-j}A_j^{-1}(x-R_M)\right)-Y_j\left(x-R_M+R'_j\right)\right)_{R'_j}\right). \tag*{(14)} R M ∑ V 3 − j ; 0 ( inter ) ( ( Y 3 − j ( A 3 − j A j − 1 ( x − R M ) ) − Y j ( x − R M + R j ′ ) ) R j ′ ) . (14)
Using again R M \mathcal{R}_M R M -periodicity of u k u_k u k , we compute
Y 3 − j ( A 3 − j A j − 1 ( x − R M ) ) − Y j ( x − R M + R j ′ ) = ( A 3 − j A j − 1 − I ) x + ( I − A 3 − j A j − 1 ) R M − R j ′ + u 3 − j ( A 3 − j A j − 1 x + ( I − A 3 − j A j − 1 ) R M ) − u j ( x + R j ′ ) . \begin{aligned}
Y_{3-j}\left(A_{3-j}A_j^{-1}(x-R_M)\right)-Y_j(x-R_M+R'_j)
&=(A_{3-j}A_j^{-1}-I)x+(I-A_{3-j}A_j^{-1})R_M-R'_j\\
&\quad+u_{3-j}\left(A_{3-j}A_j^{-1}x+(I-A_{3-j}A_j^{-1})R_M\right)-u_j(x+R'_j).
\end{aligned} Y 3 − j ( A 3 − j A j − 1 ( x − R M ) ) − Y j ( x − R M + R j ′ ) = ( A 3 − j A j − 1 − I ) x + ( I − A 3 − j A j − 1 ) R M − R j ′ + u 3 − j ( A 3 − j A j − 1 x + ( I − A 3 − j A j − 1 ) R M ) − u j ( x + R j ′ ) .
Abbreviating R j : = − A j A 3 − j − 1 ( I − A 3 − j A j − 1 ) R M ∈ R j R_j:=-A_jA_{3-j}^{-1}(I-A_{3-j}A_j^{-1})R_M\in\mathcal{R}_j R j := − A j A 3 − j − 1 ( I − A 3 − j A j − 1 ) R M ∈ R j , due to Lemma {(xref)}2.1, we obtain
Y 3 − j ( A 3 − j A j − 1 ( x − R M ) ) − Y j ( x − R M + R j ′ ) = ( A 3 − j A j − 1 − I ) ( x − R j ) − ( R j ′ + R j ) + u 3 − j ( A 3 − j A j − 1 ( x − R j ) ) − u j ( x − R j + ( R j ′ + R j ) ) = Y 3 − j ( A 3 − j A j − 1 ( x − R j ) ) − Y j ( x − R j + ( R j ′ + R j ) ) . (15) \begin{aligned}
Y_{3-j}\left(A_{3-j}A_j^{-1}(x-R_M)\right)-Y_j(x-R_M+R'_j)
&=(A_{3-j}A_j^{-1}-I)(x-R_j)-(R'_j+R_j)\\
&\quad+u_{3-j}\left(A_{3-j}A_j^{-1}(x-R_j)\right)-u_j\left(x-R_j+(R'_j+R_j)\right)\\
&=Y_{3-j}\left(A_{3-j}A_j^{-1}(x-R_j)\right)-Y_j\left(x-R_j+(R'_j+R_j)\right). \tag*{(15)}
\end{aligned} Y 3 − j ( A 3 − j A j − 1 ( x − R M ) ) − Y j ( x − R M + R j ′ ) = ( A 3 − j A j − 1 − I ) ( x − R j ) − ( R j ′ + R j ) + u 3 − j ( A 3 − j A j − 1 ( x − R j ) ) − u j ( x − R j + ( R j ′ + R j ) ) = Y 3 − j ( A 3 − j A j − 1 ( x − R j ) ) − Y j ( x − R j + ( R j ′ + R j ) ) . (15)
Employing {(xref)}(2.5), {(xref)}(2.11) and {(xref)}(2.12) yield
∑ R j V 3 − j ; R j ( i n t e r ) ( Δ 3 − j , j [ u ] ( A 3 − j A j − 1 ( x − R j ) ) ) . \sum_{R_j}V_{3-j;R_j}^{(\mathrm{inter})}\left(\Delta_{3-j,j}[u]\left(A_{3-j}A_j^{-1}(x-R_j)\right)\right). R j ∑ V 3 − j ; R j ( inter ) ( Δ 3 − j , j [ u ] ( A 3 − j A j − 1 ( x − R j ) ) ) .
Altogether, we have thus proved that
δ e ( t o t ) ( u ) [ v ] = ∑ j = 1 2 1 ∣ Γ j ∣ ∑ α j ∈ A j ∫ M d x [ ∑ ω j ′ ( V j , α j ; ω j ′ ( m o n o ) ( Δ j [ u j ] ( x − R j ′ , α j ′ ) ) − V j , α j ; ω j ′ ( m o n o ) ( Δ j [ u j ] ( x , α j ) ) ) + ∑ R j , α 3 − j V 3 − j , α 3 − j ; ω j ( i n t e r ) ( Δ 3 − j , j [ u ] ( A 3 − j A j − 1 ( x − R j ) , α 3 − j ) ) − ∑ ω 3 − j V j , α j ; ω 3 − j ( i n t e r ) ( Δ j , 3 − j [ u ] ( x , α j ) ) ] ⋅ v j ( x , α j ) . \delta e^{(\mathrm{tot})}(\mathbf{u})[\mathbf{v}] =
\sum_{j=1}^{2} \frac{1}{|\Gamma_j|} \sum_{\alpha_j\in A_j} \int_{\mathcal{M}} \mathrm{d}x \left[
\sum_{\omega'_j} \left( V^{(\mathrm{mono})}_{j,\alpha_j;\omega'_j}\left(\Delta_j[u_j](x-R'_j,\alpha'_j)\right) - V^{(\mathrm{mono})}_{j,\alpha_j;\omega'_j}\left(\Delta_j[u_j](x,\alpha_j)\right)\right)
+ \sum_{R_j,\alpha_{3-j}} V^{(\mathrm{inter})}_{3-j,\alpha_{3-j};\omega_j}\left(\Delta_{3-j,j}[\mathbf{u}]\left(A_{3-j}A_j^{-1}(x-R_j),\alpha_{3-j}\right)\right)
- \sum_{\omega_{3-j}} V^{(\mathrm{inter})}_{j,\alpha_j;\omega_{3-j}}\left(\Delta_{j,3-j}[\mathbf{u}](x,\alpha_j)\right)
\right] \cdot v_j(x,\alpha_j). δ e ( tot ) ( u ) [ v ] = j = 1 ∑ 2 ∣ Γ j ∣ 1 α j ∈ A j ∑ ∫ M d x ω j ′ ∑ ( V j , α j ; ω j ′ ( mono ) ( Δ j [ u j ] ( x − R j ′ , α j ′ ) ) − V j , α j ; ω j ′ ( mono ) ( Δ j [ u j ] ( x , α j ) ) ) + R j , α 3 − j ∑ V 3 − j , α 3 − j ; ω j ( inter ) ( Δ 3 − j , j [ u ] ( A 3 − j A j − 1 ( x − R j ) , α 3 − j ) ) − ω 3 − j ∑ V j , α j ; ω 3 − j ( inter ) ( Δ j , 3 − j [ u ] ( x , α j ) ) ⋅ v j ( x , α j ) .
Consequently, the Euler-Lagrange equation for the minimizer u ( e q ) \mathbf{u}^{(\mathrm{eq})} u ( eq ) is given by
∑ ω j ′ ( V j , α j ; ω j ′ ( m o n o ) ( Δ j [ u j ( e q ) ] ( x − R j ′ , α j ′ ) ) − V j , α j ; ω j ′ ( m o n o ) ( Δ j [ u j ( e q ) ] ( x , α j ) ) ) + ∑ R j , α 3 − j V 3 − j , α 3 − j ; ω j ( i n t e r ) ( Δ 3 − j , j [ u ( e q ) ] ( A 3 − j A j − 1 ( x − R j ) , α 3 − j ) ) − ∑ ω 3 − j V j , α j ; ω 3 − j ( i n t e r ) ( Δ j , 3 − j [ u ( e q ) ] ( x , α j ) ) = 0. (16) \sum_{\omega'_j} \left( V^{(\mathrm{mono})}_{j,\alpha_j;\omega'_j}\left(\Delta_j[u_j^{(\mathrm{eq})}](x-R'_j,\alpha'_j)\right) - V^{(\mathrm{mono})}_{j,\alpha_j;\omega'_j}\left(\Delta_j[u_j^{(\mathrm{eq})}](x,\alpha_j)\right)\right)
+ \sum_{R_j,\alpha_{3-j}} V^{(\mathrm{inter})}_{3-j,\alpha_{3-j};\omega_j}\left(\Delta_{3-j,j}[\mathbf{u}^{(\mathrm{eq})}]\left(A_{3-j}A_j^{-1}(x-R_j),\alpha_{3-j}\right)\right)
- \sum_{\omega_{3-j}} V^{(\mathrm{inter})}_{j,\alpha_j;\omega_{3-j}}\left(\Delta_{j,3-j}[\mathbf{u}^{(\mathrm{eq})}](x,\alpha_j)\right) = 0. \tag*{(16)} ω j ′ ∑ ( V j , α j ; ω j ′ ( mono ) ( Δ j [ u j ( eq ) ] ( x − R j ′ , α j ′ ) ) − V j , α j ; ω j ′ ( mono ) ( Δ j [ u j ( eq ) ] ( x , α j ) ) ) + R j , α 3 − j ∑ V 3 − j , α 3 − j ; ω j ( inter ) ( Δ 3 − j , j [ u ( eq ) ] ( A 3 − j A j − 1 ( x − R j ) , α 3 − j ) ) − ω 3 − j ∑ V j , α j ; ω 3 − j ( inter ) ( Δ j , 3 − j [ u ( eq ) ] ( x , α j ) ) = 0. (16)
Offset energy
In order to study phonons, we now also look at the energy difference resulting from subtracting the groundstate total energy, as opposed to its energy density.
Then we define
Δ E γ , N ; u ( e q ) ( t o t ) ( δ u ) : = ∑ j = 1 2 ∑ ω j ∈ Ω j ( N ) [ ( V j , α j ( m o n o ) ( Δ j [ u j ( e q ) + δ u j ] ( ω j ) ) − V j , α j ( m o n o ) ( Δ j [ u j ( e q ) ] ( ω j ) ) ) + ( V j , α j ( i n t e r ) ( Δ j , 3 − j [ u ( e q ) + δ u ] ( ω j ) ) − V j , α j ( i n t e r ) ( Δ j , 3 − j [ u ( e q ) ] ( ω j ) ) ) ] . \Delta E^{(\mathrm{tot})}_{\gamma,N;\mathbf{u}^{(\mathrm{eq})}}(\delta\mathbf{u}) :=
\sum_{j=1}^{2} \sum_{\omega_j\in\Omega_j(N)}
\left[
\left(V^{(\mathrm{mono})}_{j,\alpha_j}\left(\Delta_j[u_j^{(\mathrm{eq})}+\delta u_j](\omega_j)\right) - V^{(\mathrm{mono})}_{j,\alpha_j}\left(\Delta_j[u_j^{(\mathrm{eq})}](\omega_j)\right)\right)
+
\left(V^{(\mathrm{inter})}_{j,\alpha_j}\left(\Delta_{j,3-j}[\mathbf{u}^{(\mathrm{eq})}+\delta\mathbf{u}](\omega_j)\right) - V^{(\mathrm{inter})}_{j,\alpha_j}\left(\Delta_{j,3-j}[\mathbf{u}^{(\mathrm{eq})}](\omega_j)\right)\right)
\right]. Δ E γ , N ; u ( eq ) ( tot ) ( δ u ) := j = 1 ∑ 2 ω j ∈ Ω j ( N ) ∑ [ ( V j , α j ( mono ) ( Δ j [ u j ( eq ) + δ u j ] ( ω j ) ) − V j , α j ( mono ) ( Δ j [ u j ( eq ) ] ( ω j ) ) ) + ( V j , α j ( inter ) ( Δ j , 3 − j [ u ( eq ) + δ u ] ( ω j ) ) − V j , α j ( inter ) ( Δ j , 3 − j [ u ( eq ) ] ( ω j ) ) ) ] .
For the limit N → ∞ N \to\infty N → ∞ to exist, we need to impose integrability of δ u \delta\mathbf{u} δ u . In particular, we consider δ u ∈ ℓ 1 ( Ω 1 × Ω 2 ) \delta\mathbf{u} \in\ell^1(\Omega_1 \times\Omega_2) δ u ∈ ℓ 1 ( Ω 1 × Ω 2 ) . We focus on the linear-in-δ u \delta\mathbf{u} δ u part and, as above, we reorder terms according to the coefficient δ u j ( ω j ) \delta u_j(\omega_j) δ u j ( ω j )
∑ j = 1 2 ∑ ω j [ ∑ ω j ′ ( V j , α j ; ω j ′ ( m o n o ) ( Δ j [ u j ( e q ) ] ( R j − R j ′ , α j ) ) − V j , α j ; ω j ′ ( m o n o ) ( Δ j [ u j ( e q ) ] ( ω j ) ) ) + ∑ ω 3 − j V j , α j ; ω 3 − j ( i n t e r ) ( Δ j , 3 − j [ u ( e q ) ] ( ω j ) ) − ∑ R j ′ , α 3 − j V 3 − j , α 3 − j ( i n t e r ) ( R j ′ , α j ) ( Δ 3 − j , j [ u ( e q ) ] ( A 3 − j A j − 1 ( R j − R j ′ ) , α 3 − j ) ) ] ⋅ δ u j ( ω j ) . \begin{aligned}
\sum_{j=1}^{2}\sum_{\omega_j}\Biggl[&\sum_{\omega'_j}\left(V^{(\mathrm{mono})}_{j,\alpha_j;\omega'_j}\left(\Delta_j[u^{(\mathrm{eq})}_j](R_j-R'_j,\alpha_j)\right)-V^{(\mathrm{mono})}_{j,\alpha_j;\omega'_j}\left(\Delta_j[u^{(\mathrm{eq})}_j](\omega_j)\right)\right)\\
&+\sum_{\omega_{3-j}}V^{(\mathrm{inter})}_{j,\alpha_j;\omega_{3-j}}\left(\Delta_{j,3-j}[u^{(\mathrm{eq})}](\omega_j)\right)\\
&-\sum_{R'_j,\alpha_{3-j}}V^{(\mathrm{inter})}_{3-j,\alpha_{3-j}}(R'_j,\alpha_j)\left(\Delta_{3-j,j}[u^{(\mathrm{eq})}]\left(A_{3-j}A_j^{-1}(R_j-R'_j),\alpha_{3-j}\right)\right)\Biggr]\cdot\delta u_j(\omega_j).
\end{aligned} j = 1 ∑ 2 ω j ∑ [ ω j ′ ∑ ( V j , α j ; ω j ′ ( mono ) ( Δ j [ u j ( eq ) ] ( R j − R j ′ , α j ) ) − V j , α j ; ω j ′ ( mono ) ( Δ j [ u j ( eq ) ] ( ω j ) ) ) + ω 3 − j ∑ V j , α j ; ω 3 − j ( inter ) ( Δ j , 3 − j [ u ( eq ) ] ( ω j ) ) − R j ′ , α 3 − j ∑ V 3 − j , α 3 − j ( inter ) ( R j ′ , α j ) ( Δ 3 − j , j [ u ( eq ) ] ( A 3 − j A j − 1 ( R j − R j ′ ) , α 3 − j ) ) ] ⋅ δ u j ( ω j ) .
We recover the Euler-Lagrange equation (16) , evaluated at the lattice positions R j R_j R j . In particular, the first variation δ Δ ∞ E ( 0 ) = 0 \delta\Delta_\infty E(0)=0 δ Δ ∞ E ( 0 ) = 0 vanishes, which means the solution u ( e q ) \mathbf{u}^{(\mathrm{eq})} u ( eq ) of (16) is a critical point under ℓ 1 \ell^1 ℓ 1 -perturbations.
Models
Cauchy-Born approximation
A common further simplification to the Cauchy-Born approximation is the linear elasticity approximation
W j ( M ) : = M : C j : M , C j , k l m n : = λ j δ k l δ m n + μ j ( δ k n δ l m + δ k m δ l n ) , W_j(M):=M:C_j:M,
\qquad
C_{j,klmn}:=\lambda_j\delta_{kl}\delta_{mn}+\mu_j(\delta_{kn}\delta_{lm}+\delta_{km}\delta_{ln}), W j ( M ) := M : C j : M , C j , k l mn := λ j δ k l δ mn + μ j ( δ k n δ l m + δ k m δ l n ) ,
where λ j , μ j > 0 \lambda_j,\mu_j>0 λ j , μ j > 0 denote the Lamé parameters of an individual layer. Then the monolayer energy density is given by
1 2 ∫ Γ M d x ∇ u j ( x ) : C j : ∇ u j ( x ) . \frac{1}{2}\int_{\Gamma_M}\mathrm{d}x\,\nabla u_j(x):C_j:\nabla u_j(x). 2 1 ∫ Γ M d x ∇ u j ( x ) : C j : ∇ u j ( x ) .
In order to allow for out-of-plane relaxation, we need to extend the monolayer energy model beyond linear elasticity, e.g., to
∫ Γ M d x 1 2 ( λ j ∑ k = 1 2 ε k , k ( j ) ( x ) 2 + 2 μ j ∑ k , ℓ = 1 2 ε k , ℓ ( j ) ( x ) 2 + κ j ∣ D 2 u j , 3 ( x ) ∣ 2 ) , \int_{\Gamma_M}\mathrm{d}x\,\frac{1}{2}\left(\lambda_j\sum_{k=1}^{2}\varepsilon^{(j)}_{k,k}(x)^2+2\mu_j\sum_{k,\ell=1}^{2}\varepsilon^{(j)}_{k,\ell}(x)^2+\kappa_j\left|D^2u_{j,3}(x)\right|^2\right), ∫ Γ M d x 2 1 λ j k = 1 ∑ 2 ε k , k ( j ) ( x ) 2 + 2 μ j k , ℓ = 1 ∑ 2 ε k , ℓ ( j ) ( x ) 2 + κ j D 2 u j , 3 ( x ) 2 ,
where ε k , ℓ ( j ) : = 1 2 ( ∂ k u j , ℓ + ∂ ℓ u j , k + ∂ k u j , 3 ∂ ℓ u j , 3 ) \varepsilon^{(j)}_{k,\ell}:=\frac{1}{2}(\partial_k u_{j,\ell}+\partial_\ell u_{j,k}+\partial_k u_{j,3}\partial_\ell u_{j,3}) ε k , ℓ ( j ) := 2 1 ( ∂ k u j , ℓ + ∂ ℓ u j , k + ∂ k u j , 3 ∂ ℓ u j , 3 ) is the strain tensor, λ \lambda λ , μ \mu μ are the Lamé parameters as before, and κ \kappa κ is the bending rigidity, see, e.g., [34 ] .
Pair potentials
For the interlayer energy and in order to model the monolayer offset energy before taking a continuum limit, we may consider pair potential interactions. Common choices for such pair potentials include
Lennard-Jones potentials v L J ( X ) = 4 ε 0 [ ( σ ∣ X ∣ ) 12 − ( σ ∣ X ∣ ) 6 ] v_{\mathrm{LJ}}(X)=4\varepsilon_0\left[\left(\frac{\sigma}{|X|}\right)^{12}-\left(\frac{\sigma}{|X|}\right)^6\right] v LJ ( X ) = 4 ε 0 [ ( ∣ X ∣ σ ) 12 − ( ∣ X ∣ σ ) 6 ] for X ∈ R 3 X\in\mathbb{R}^3 X ∈ R 3 , with equilibrium distance 2 1 6 σ 2^{\frac{1}{6}}\sigma 2 6 1 σ ,
Morse potentials v M o r s e ( X ) = E 0 ( e − 2 a ( ∣ X ∣ − r 0 ) − 2 e − a ( ∣ X ∣ − r 0 ) ) v_{\mathrm{Morse}}(X)=E_0\left(e^{-2a(|X|-r_0)}-2e^{-a(|X|-r_0)}\right) v Morse ( X ) = E 0 ( e − 2 a ( ∣ X ∣ − r 0 ) − 2 e − a ( ∣ X ∣ − r 0 ) ) for X ∈ R 3 X\in\mathbb{R}^3 X ∈ R 3 , with equilibrium distance r 0 r_0 r 0 ,
Kolmogorov-Crespi potentials Introduced in [21 , 29 ] , we define
v K C ( X j − X k , n j , n k ) : = e − λ ( r j k − z 0 ) ( C + f ( ρ j k ) + f ( ρ k j ) ) − A 0 ( z 0 r j k ) 6 v_{\mathrm{KC}}(X_j-X_k,n_j,n_k):=e^{-\lambda(r_{jk}-z_0)}\left(C+f(\rho_{jk})+f(\rho_{kj})\right)-A_0\left(\frac{z_0}{r_{jk}}\right)^6 v KC ( X j − X k , n j , n k ) := e − λ ( r j k − z 0 ) ( C + f ( ρ j k ) + f ( ρ k j ) ) − A 0 ( r j k z 0 ) 6
for X j , X k ∈ R 3 X_j,X_k\in\mathbb{R}^3 X j , X k ∈ R 3 , where r j k : = ∣ X j − X k ∣ r_{jk}:=|X_j-X_k| r j k := ∣ X j − X k ∣ , n ℓ n_\ell n ℓ denotes the normal vector to the s p 2 sp^2 s p 2 -plane4 in the vicinity of atom ℓ \ell ℓ , ρ ℓ m : = r ℓ m 2 − ( n ℓ ⋅ ( X ℓ − X m ) ) 2 \rho_{\ell m}:=r_{\ell m}^2-\left(n_\ell\cdot(X_\ell-X_m)\right)^2 ρ ℓ m := r ℓ m 2 − ( n ℓ ⋅ ( X ℓ − X m ) ) 2 and
f ( ρ ) : = e − ( ρ / δ ) 2 ∑ n = 0 2 C 2 n ( ρ / δ ) 2 n . f(\rho):=e^{-(\rho/\delta)^2}\sum_{n=0}^{2}C_{2n}(\rho/\delta)^{2n}. f ( ρ ) := e − ( ρ / δ ) 2 n = 0 ∑ 2 C 2 n ( ρ / δ ) 2 n .
For simplicity, one could choose n ℓ = e z n_\ell=e_z n ℓ = e z . In this case, the Kolmogorov potential simplifies to
v K C ( r , ρ ) = e − λ ( r − z 0 ) ( C + 2 f ( ρ ) ) − A 0 ( z 0 r ) 6 , v_{\mathrm{KC}}(r,\rho)=e^{-\lambda(r-z_0)}\left(C+2f(\rho)\right)-A_0\left(\frac{z_0}{r}\right)^6, v KC ( r , ρ ) = e − λ ( r − z 0 ) ( C + 2 f ( ρ ) ) − A 0 ( r z 0 ) 6 ,
see, e.g., [14 ] . In case of bilayer graphene, the parameters are given by ( C , C 0 , C 2 , C 4 ) = ( 3.030 , 15.71 , 12.29 , 4.933 ) m e V (C,C_0,C_2,C_4)=(3.030,15.71,12.29,4.933)\,\mathrm{meV} ( C , C 0 , C 2 , C 4 ) = ( 3.030 , 15.71 , 12.29 , 4.933 ) meV , δ = 0.578 A ˚ \delta=0.578\,\text{\AA} δ = 0.578 A ˚ , λ = 3.629 A ˚ − 1 \lambda=3.629\,\text{\AA}^{-1} λ = 3.629 A ˚ − 1 , A 0 = 10.238 m e V A_0=10.238\,\mathrm{meV} A 0 = 10.238 meV , z 0 = 3.34 A ˚ z_0=3.34\,\text{\AA} z 0 = 3.34 A ˚ .
Combined potentials, e.g., v ( x , z ) = v M o r s e ( x ) v L J ( z ) v(x,z)=v_{\mathrm{Morse}}(x)v_{\mathrm{LJ}}(z) v ( x , z ) = v Morse ( x ) v LJ ( z ) for x ∈ R 2 x\in\mathbb{R}^2 x ∈ R 2 , z ∈ R z\in\mathbb{R} z ∈ R .
Well-definedness of energy functionals
We now show that the previous calculations are meaningful by showing that the energy functionals are well-defined and sufficiently regular.
Atomistic model
The following statements can be proved analogously to [33 Theorem 1] .
Lemma 3.1. Let j ∈ { 1 , 2 } j\in\{1,2\} j ∈ { 1 , 2 } , m ∈ N m\in\mathbb{N} m ∈ N . Assume that u = ( u 1 , u 2 ) ∈ K κ \mathbf{u}=(u_1,u_2)\in\mathcal{K}_\kappa u = ( u 1 , u 2 ) ∈ K κ and v ℓ ∈ W ‾ p e r 1 , 2 ( Γ M ; ( R 2 ) A j ) v_\ell\in\overline{W}^{1,2}_{\mathrm{per}}(\Gamma_M;(\mathbb{R}^2)^{\mathcal{A}_j}) v ℓ ∈ W per 1 , 2 ( Γ M ; ( R 2 ) A j ) , ℓ = 1 , … , m \ell=1,\ldots,m ℓ = 1 , … , m . For any 1 ≤ p 1 , … , p k ≤ ∞ 1\leq p_1,\ldots,p_k\leq\infty 1 ≤ p 1 , … , p k ≤ ∞ such that ∑ ℓ = 1 m 1 p ℓ = 1 \sum_{\ell=1}^{m}\frac{1}{p_\ell}=1 ∑ ℓ = 1 m p ℓ 1 = 1 , we have that
∣ δ m e j ( m o n o ) ( u j ) [ v 1 , … , v m ] ∣ ≲ M m , j ( m o n o ) ∏ ℓ = 1 m ∥ ∇ v ℓ ∥ L p ℓ ( Γ M ) . \left|\delta^m e_j^{(\mathrm{mono})}(u_j)[v_1,\ldots,v_m]\right|\lesssim M_{m,j}^{(\mathrm{mono})}\prod_{\ell=1}^{m}\|\nabla v_\ell\|_{L^{p_\ell}(\Gamma_M)}. δ m e j ( mono ) ( u j ) [ v 1 , … , v m ] ≲ M m , j ( mono ) ℓ = 1 ∏ m ∥∇ v ℓ ∥ L p ℓ ( Γ M ) .
For the next result, we recall the first-order interpolants defined in (5) .
Lemma 3.2. Let j ∈ { 1 , 2 } j \in\{1,2\} j ∈ { 1 , 2 } , m ∈ N m \in\mathbb{N} m ∈ N . Assume that u = ( u 1 , u 2 ) ∈ K κ \mathbf{u}=(u_1,u_2)\in\mathcal{K}_{\kappa} u = ( u 1 , u 2 ) ∈ K κ and that δ u , δ v ℓ ∈ ℓ 1 ( Ω j ) \delta u,\delta v_{\ell}\in\ell^1(\Omega_j) δ u , δ v ℓ ∈ ℓ 1 ( Ω j ) , ℓ = 1 , … , m \ell=1,\ldots,m ℓ = 1 , … , m . For any 1 ≤ p 1 , … , p k ≤ ∞ 1\leq p_1,\ldots,p_k\leq\infty 1 ≤ p 1 , … , p k ≤ ∞ such that ∑ ℓ = 1 m 1 p ℓ = 1 \sum_{\ell=1}^{m}\frac{1}{p_{\ell}}=1 ∑ ℓ = 1 m p ℓ 1 = 1 , we have that
∣ δ m Δ E j ; u j ( m o n o ) ( δ u ) [ δ v 1 , … , δ v m ] ∣ ≲ M m , j ( m o n o ) ∏ ℓ = 1 m ∥ ∇ ζ ∥ ℓ 1 ∥ δ v ℓ ∥ ℓ 1 ( A j ; ℓ p ℓ ( R 2 ; R 3 ) ) . \left|\delta^m\Delta E_{j;u_j}^{(\mathrm{mono})}(\delta u)[\delta v_1,\ldots,\delta v_m]\right|
\lesssim M_{m,j}^{(\mathrm{mono})}\prod_{\ell=1}^{m}\|\nabla\zeta\|_{\ell^{1}}\|\delta v_{\ell}\|_{\ell^{1}(A_j;\ell^{p_{\ell}}(\mathbb{R}^{2};\mathbb{R}^{3}))}. δ m Δ E j ; u j ( mono ) ( δ u ) [ δ v 1 , … , δ v m ] ≲ M m , j ( mono ) ℓ = 1 ∏ m ∥∇ ζ ∥ ℓ 1 ∥ δ v ℓ ∥ ℓ 1 ( A j ; ℓ p ℓ ( R 2 ; R 3 )) .
Proof. In order to employ [33 Lemma 6] , we shall map the deformations over R j \mathcal{R}_j R j to deformations over Z 2 \mathbb{Z}^{2} Z 2 . In particular, we define
δ w j ( n ) : = δ v j ( A j n ) , δ w j ( x ) : = ∑ n ∈ Z 2 ζ ( x − n ) δ w j ( n ) = δ v j ( A j x ) , \delta w_j(n):=\delta v_j(A_jn),\qquad\delta w_j(x):=\sum_{n\in\mathbb{Z}^{2}}\zeta(x-n)\delta w_j(n)=\delta v_j(A_jx), δ w j ( n ) := δ v j ( A j n ) , δ w j ( x ) := n ∈ Z 2 ∑ ζ ( x − n ) δ w j ( n ) = δ v j ( A j x ) ,
see (5) . Defining
F ( m o n o ) ( δ w 1 , … , δ w m ) : = δ m Δ E j ; u j ( m o n o ) ( δ u ) [ δ v 1 , … , δ v m ] , F^{(\mathrm{mono})}(\delta w_1,\ldots,\delta w_m):=\delta^m\Delta E_{j;u_j}^{(\mathrm{mono})}(\delta u)[\delta v_1,\ldots,\delta v_m], F ( mono ) ( δ w 1 , … , δ w m ) := δ m Δ E j ; u j ( mono ) ( δ u ) [ δ v 1 , … , δ v m ] ,
[33 Lemma 6] then implies that
∣ F ( m o n o ) ( δ w 1 , … , δ w m ) ∣ ≲ M m , j ( m o n o ) ∏ ℓ = 1 m ∥ ∇ δ w j ∥ ℓ 1 ( A j ; L p ℓ ( R 2 ; R 3 ) ) \left|F^{(\mathrm{mono})}(\delta w_1,\ldots,\delta w_m)\right|
\lesssim M_{m,j}^{(\mathrm{mono})}\prod_{\ell=1}^{m}\|\nabla\delta w_j\|_{\ell^1(A_j;L^{p_\ell}(\mathbb{R}^{2};\mathbb{R}^{3}))} F ( mono ) ( δ w 1 , … , δ w m ) ≲ M m , j ( mono ) ℓ = 1 ∏ m ∥∇ δ w j ∥ ℓ 1 ( A j ; L p ℓ ( R 2 ; R 3 ))
≲ M m , j ( m o n o ) ∏ ℓ = 1 m ∥ A j ∇ δ v ℓ ∥ ℓ 1 ( A j ; L p ℓ ( R 2 ; R 3 ) ) ∣ Γ j ∣ 1 p ℓ . (17) \lesssim M_{m,j}^{(\mathrm{mono})}\prod_{\ell=1}^{m}\frac{\|A_j\nabla\delta v_\ell\|_{\ell^1(A_j;L^{p_\ell}(\mathbb{R}^{2};\mathbb{R}^{3}))}}{|\Gamma_j|^{\frac{1}{p_\ell}}}. \tag*{(17)} ≲ M m , j ( mono ) ℓ = 1 ∏ m ∣ Γ j ∣ p ℓ 1 ∥ A j ∇ δ v ℓ ∥ ℓ 1 ( A j ; L p ℓ ( R 2 ; R 3 )) . (17)
Next, we compute
∥ A j ∇ δ v ℓ ∥ L p = ∑ R j ∈ R j ∫ Γ j d x ∣ ∑ R j ′ ∈ R j A j ∇ ζ j ( x + R j − R j ′ ) δ v ℓ ( R j ) ∣ p . (18) \|A_j\nabla\delta v_\ell\|_{L^p}
=\sum_{R_j\in\mathcal{R}_j}\int_{\Gamma_j}\mathrm{d}x\left|\sum_{R'_j\in\mathcal{R}_j}A_j\nabla\zeta_j(x+R_j-R'_j)\delta v_\ell(R_j)\right|^p. \tag*{(18)} ∥ A j ∇ δ v ℓ ∥ L p = R j ∈ R j ∑ ∫ Γ j d x R j ′ ∈ R j ∑ A j ∇ ζ j ( x + R j − R j ′ ) δ v ℓ ( R j ) p . (18)
Using the fact that ζ j \zeta_j ζ j has compact support, there exists a universal constant r 0 > 0 r_0>0 r 0 > 0 such that ⋃ x ∈ Γ j supp ζ j ( x − ⋅ ) ⊆ B r 0 ( 0 ) ∩ R j \bigcup_{x\in\Gamma_j}\operatorname{supp}\zeta_j(x-\cdot)\subseteq B_{r_0}(0)\cap\mathcal{R}_j ⋃ x ∈ Γ j supp ζ j ( x − ⋅ ) ⊆ B r 0 ( 0 ) ∩ R j . With that, we obtain that
∑ R j ∈ R j ∫ Γ j d x ∣ ∑ R j ′ ∈ R j A j ∇ ζ j ( x + R j − R j ′ ) δ v ℓ ( R j ) ∣ p \sum_{R_j\in\mathcal{R}_j}\int_{\Gamma_j}\mathrm{d}x\left|\sum_{R'_j\in\mathcal{R}_j}A_j\nabla\zeta_j(x+R_j-R'_j)\delta v_\ell(R_j)\right|^p R j ∈ R j ∑ ∫ Γ j d x R j ′ ∈ R j ∑ A j ∇ ζ j ( x + R j − R j ′ ) δ v ℓ ( R j ) p
≲ ∑ R j ∈ R j ∫ Γ j d x ∑ R j ′ ∈ R j ∩ B r 0 ( R j ) ∣ δ v ℓ ( R j ′ ) ∣ p ∑ R j ′ ∈ B r 0 ( 0 ) ∩ R j ∣ A j ∇ ζ j ( x − R j ′ ) ∣ p \lesssim\sum_{R_j\in\mathcal{R}_j}\int_{\Gamma_j}\mathrm{d}x\sum_{R'_j\in\mathcal{R}_j\cap B_{r_0}(R_j)}|\delta v_\ell(R'_j)|^p\sum_{R'_j\in B_{r_0}(0)\cap\mathcal{R}_j}|A_j\nabla\zeta_j(x-R'_j)|^p ≲ R j ∈ R j ∑ ∫ Γ j d x R j ′ ∈ R j ∩ B r 0 ( R j ) ∑ ∣ δ v ℓ ( R j ′ ) ∣ p R j ′ ∈ B r 0 ( 0 ) ∩ R j ∑ ∣ A j ∇ ζ j ( x − R j ′ ) ∣ p
≲ ∥ δ v ℓ ∥ ℓ p p ∑ R j ∈ B r 0 ( 0 ) ∩ R j ∫ Γ j + R j d x ∣ A j ∇ ζ j ( x ) ∣ p . (19) \lesssim\|\delta v_\ell\|_{\ell^p}^p\sum_{R_j\in B_{r_0}(0)\cap\mathcal{R}_j}\int_{\Gamma_j+R_j}\mathrm{d}x\,|A_j\nabla\zeta_j(x)|^p. \tag*{(19)} ≲ ∥ δ v ℓ ∥ ℓ p p R j ∈ B r 0 ( 0 ) ∩ R j ∑ ∫ Γ j + R j d x ∣ A j ∇ ζ j ( x ) ∣ p . (19)
where the constant is universal and independent of p p p . Recalling ζ j ( x ) = ζ ( A j − 1 x ) \zeta_j(x)=\zeta(A_j^{-1}x) ζ j ( x ) = ζ ( A j − 1 x ) , we further estimate the previous expression by
∣ Γ j ∣ ∥ δ v ℓ ∥ p p ∥ ∇ ζ ∥ p p . (20) |\Gamma_j|\|\delta v_\ell\|_p^p\|\nabla\zeta\|_p^p . \tag*{(20)} ∣ Γ j ∣∥ δ v ℓ ∥ p p ∥∇ ζ ∥ p p . (20)
up to a universal constant. Collecting (17) , (18) , (19) , (20) , we conclude the proof.
For the next result, we abbreviate for v = ( v 1 , v 2 ) \mathbf{v}=(v_1,v_2) v = ( v 1 , v 2 ) , v j ∈ L 1 ( Γ M , ( R 2 ) A j ) v_j\in L^1(\Gamma_{\mathcal M},(\mathbb{R}^2)^{A_j}) v j ∈ L 1 ( Γ M , ( R 2 ) A j ) , p ∈ [ 1 , ∞ ] p\in[1,\infty] p ∈ [ 1 , ∞ ]
∥ v ∥ p : = sup α ∈ A 1 × A 2 ( ∥ v 1 ( ⋅ , α 1 ) ∥ p + ∥ v 2 ( ⋅ , α 2 ) ∥ p ) . \|\mathbf{v}\|_p:=\sup_{\alpha\in A_1\times A_2}\left(\|v_1(\cdot,\alpha_1)\|_p+\|v_2(\cdot,\alpha_2)\|_p\right). ∥ v ∥ p := α ∈ A 1 × A 2 sup ( ∥ v 1 ( ⋅ , α 1 ) ∥ p + ∥ v 2 ( ⋅ , α 2 ) ∥ p ) .
Lemma 3.3. Let j ∈ { 1 , 2 } j\in\{1,2\} j ∈ { 1 , 2 } , m ∈ N m\in\mathbb{N} m ∈ N . Assume that u ∈ K κ \mathbf{u}\in\mathcal{K}_\kappa u ∈ K κ and v ( ℓ ) = ( v 1 ( ℓ ) , v 2 ( ℓ ) ) ∈ W ‾ p e r 1 , 2 \mathbf{v}^{(\ell)}=(v_1^{(\ell)},v_2^{(\ell)})\in\overline{W}^{1,2}_{\mathrm{per}} v ( ℓ ) = ( v 1 ( ℓ ) , v 2 ( ℓ ) ) ∈ W per 1 , 2 , ℓ = 1 , … , m \ell=1,\ldots,m ℓ = 1 , … , m . For any 1 ≤ p 1 , … , p m ≤ ∞ 1\le p_1,\ldots,p_m\le\infty 1 ≤ p 1 , … , p m ≤ ∞ such that ∑ ℓ = 1 m 1 p ℓ = 1 \sum_{\ell=1}^{m}\frac{1}{p_\ell}=1 ∑ ℓ = 1 m p ℓ 1 = 1 , we have that
∣ δ e j m ( i n t e r ) ( u ) [ v ( 1 ) , … , v ( m ) ] ∣ ≲ max { 1 , ∣ Γ j ∣ ∣ Γ 3 − j ∣ } M m , 3 − j ( i n t e r ) ∏ ℓ = 1 m ∥ v ( ℓ ) ∥ L p ℓ ( Γ M ) . \left|\delta_{e_j}^{m\,(\mathrm{inter})}(\mathbf{u})[\mathbf{v}^{(1)},\ldots,\mathbf{v}^{(m)}]\right|
\lesssim
\max\left\{1,\frac{|\Gamma_j|}{|\Gamma_{3-j}|}\right\}
M_{m,3-j}^{(\mathrm{inter})}\prod_{\ell=1}^{m}\|\mathbf{v}^{(\ell)}\|_{L^{p_\ell}(\Gamma_{\mathcal M})}. δ e j m ( inter ) ( u ) [ v ( 1 ) , … , v ( m ) ] ≲ max { 1 , ∣ Γ 3 − j ∣ ∣ Γ j ∣ } M m , 3 − j ( inter ) ℓ = 1 ∏ m ∥ v ( ℓ ) ∥ L p ℓ ( Γ M ) .
Proof. We follow the proof of [33 Lemma 6] . Then we find that
V j , α j ; ω 3 − j m ( i n t e r ) ( Δ j , 3 − j [ u ] ( x , α 3 − j ) ) [ ( Δ ω 3 − j m v m ) ( x , α j ) ] ≲ m m , 3 − j ( i n t e r ) ( ω j ℓ ) ∏ ℓ = 1 m ∣ ( Δ ω 3 − j ( ℓ ) v ( ℓ ) ) ( x , α j ) ∣ . V_{j,\alpha_j;\omega_{3-j}^{m}}^{(\mathrm{inter})}\left(\Delta_{j,3-j}[\mathbf{u}](x,\alpha_{3-j})\right)\left[(\Delta_{\omega_{3-j}^{m}}\mathbf{v}^{m})(x,\alpha_j)\right]
\lesssim
m_{m,3-j}^{(\mathrm{inter})}(\omega_j^{\ell})\prod_{\ell=1}^{m}\left|(\Delta_{\omega_{3-j}^{(\ell)}}\mathbf{v}^{(\ell)})(x,\alpha_j)\right|. V j , α j ; ω 3 − j m ( inter ) ( Δ j , 3 − j [ u ] ( x , α 3 − j ) ) [ ( Δ ω 3 − j m v m ) ( x , α j ) ] ≲ m m , 3 − j ( inter ) ( ω j ℓ ) ℓ = 1 ∏ m ( Δ ω 3 − j ( ℓ ) v ( ℓ ) ) ( x , α j ) .
Consequently, and using R M \mathcal{R}_{\mathcal M} R M -periodicity of v j v_j v j we obtain that
∣ δ e j m ( i n t e r ) ( u ) [ v m ] ∣ ≲ ∑ ω 3 − j ℓ ∈ Ω 3 − j ℓ m m , 3 − j ( i n t e r ) ( ω j ℓ ) sup α j ∫ Γ M d x ∏ ℓ = 1 m ∣ ( Δ ω 3 − j ( ℓ ) v ( ℓ ) ) ( x , α j ) ∣ \left|\delta_{e_j}^{m\,(\mathrm{inter})}(\mathbf{u})[\mathbf{v}^{m}]\right|
\lesssim
\sum_{\omega_{3-j}^{\ell}\in\Omega_{3-j}^{\ell}}
m_{m,3-j}^{(\mathrm{inter})}(\omega_j^\ell)
\sup_{\alpha_j}\int_{\Gamma_{\mathcal M}}\mathrm{d}x\prod_{\ell=1}^{m}\left|(\Delta_{\omega_{3-j}^{(\ell)}}\mathbf{v}^{(\ell)})(x,\alpha_j)\right| δ e j m ( inter ) ( u ) [ v m ] ≲ ω 3 − j ℓ ∈ Ω 3 − j ℓ ∑ m m , 3 − j ( inter ) ( ω j ℓ ) α j sup ∫ Γ M d x ℓ = 1 ∏ m ( Δ ω 3 − j ( ℓ ) v ( ℓ ) ) ( x , α j )
≲ M m , 3 − j ( i n t e r ) sup α 3 − j ℓ ∏ ℓ = 1 m ( ∥ v 3 − j ( ℓ ) ( ω 3 − j ℓ + A 3 − j A j − 1 ⋅ ) ∥ p ℓ + ∥ v j ( ℓ ) ( ⋅ , α j ) ∥ p ℓ ) . \lesssim
M_{m,3-j}^{(\mathrm{inter})}
\sup_{\alpha_{3-j}^{\ell}}
\prod_{\ell=1}^{m}
\left(
\|v_{3-j}^{(\ell)}(\omega_{3-j}^{\ell}+A_{3-j}A_j^{-1}\cdot)\|_{p_\ell}
+\|v_j^{(\ell)}(\cdot,\alpha_j)\|_{p_\ell}
\right). ≲ M m , 3 − j ( inter ) α 3 − j ℓ sup ℓ = 1 ∏ m ( ∥ v 3 − j ( ℓ ) ( ω 3 − j ℓ + A 3 − j A j − 1 ⋅ ) ∥ p ℓ + ∥ v j ( ℓ ) ( ⋅ , α j ) ∥ p ℓ ) .
Substituting A 3 − j A j − 1 x → x A_{3-j}A_j^{-1}x\to x A 3 − j A j − 1 x → x , we obtain due to R M \mathcal{R}_{\mathcal M} R M -periodicity of v 3 − j ( ℓ ) v_{3-j}^{(\ell)} v 3 − j ( ℓ )
∣ δ e j m ( i n t e r ) ( u ) [ v m ] ∣ ≲ max { 1 , ∣ Γ j ∣ ∣ Γ 3 − j ∣ } M m , 3 − j ( i n t e r ) ∏ ℓ = 1 m ∥ v ( ℓ ) ∥ p ℓ , \left|\delta_{e_j}^{m\,(\mathrm{inter})}(\mathbf{u})[\mathbf{v}^{m}]\right|
\lesssim
\max\left\{1,\frac{|\Gamma_j|}{|\Gamma_{3-j}|}\right\}
M_{m,3-j}^{(\mathrm{inter})}
\prod_{\ell=1}^{m}\|\mathbf{v}^{(\ell)}\|_{p_\ell}, δ e j m ( inter ) ( u ) [ v m ] ≲ max { 1 , ∣ Γ 3 − j ∣ ∣ Γ j ∣ } M m , 3 − j ( inter ) ℓ = 1 ∏ m ∥ v ( ℓ ) ∥ p ℓ ,
concluding the proof.
With analogous steps, we obtain the following result
Lemma 3.4. Let j ∈ { 1 , 2 } j \in\{1,2\} j ∈ { 1 , 2 } , m ∈ N m \in\mathbb{N} m ∈ N . Assume that u ∈ D κ \mathbf{u} \in\mathcal{D}_{\kappa} u ∈ D κ , and δ u , δ v ( ℓ ) = ( δ v 1 ( ℓ ) , δ v 2 ( ℓ ) ) ∈ ℓ 1 ( Ω ) \delta\mathbf{u},\delta\mathbf{v}^{(\ell)} = (\delta v_{1}^{(\ell)},\delta v_{2}^{(\ell)}) \in\ell^{1}(\Omega) δ u , δ v ( ℓ ) = ( δ v 1 ( ℓ ) , δ v 2 ( ℓ ) ) ∈ ℓ 1 ( Ω ) , ℓ = 1 , … , m \ell=1,\ldots,m ℓ = 1 , … , m . For any 1 ≤ p 1 , … , p m ≤ ∞ 1 \le p_{1},\ldots,p_{m} \le\infty 1 ≤ p 1 , … , p m ≤ ∞ such that ∑ ℓ = 1 m 1 p ℓ = 1 \sum_{\ell=1}^{m} \frac{1}{p_{\ell}}=1 ∑ ℓ = 1 m p ℓ 1 = 1 , we have that
∣ δ m Δ E j ( i n t e r ) ( u ) [ δ v ( 1 ) , … , δ v ( m ) ] ∣ ≲ max { 1 , ∣ Γ j ∣ ∣ Γ 3 − j ∣ } M m , 3 − j ( i n t e r ) ∏ ℓ = 1 m ∥ δ v ( ℓ ) ∥ ℓ p ℓ ( Ω ) . \left|\delta^{m}\Delta E_{j}^{(\mathrm{inter})}(\mathbf{u})[\delta\mathbf{v}^{(1)},\ldots,\delta\mathbf{v}^{(m)}]\right| \lesssim\max\left\{1,\frac{|\Gamma_{j}|}{|\Gamma_{3-j}|}\right\}M_{m,3-j}^{(\mathrm{inter})}\prod_{\ell=1}^{m}\|\delta\mathbf{v}^{(\ell)}\|_{\ell^{p_{\ell}}(\Omega)}. δ m Δ E j ( inter ) ( u ) [ δ v ( 1 ) , … , δ v ( m ) ] ≲ max { 1 , ∣ Γ 3 − j ∣ ∣ Γ j ∣ } M m , 3 − j ( inter ) ℓ = 1 ∏ m ∥ δ v ( ℓ ) ∥ ℓ p ℓ ( Ω ) .
Proposition 3.5. Assume that V j , α j ( m o n o ) V_{j,\alpha_{j}}^{(\mathrm{mono})} V j , α j ( mono ) and V j , α j ( i n t e r ) V_{j,\alpha_{j}}^{(\mathrm{inter})} V j , α j ( inter ) satisfy the decay conditions 5 . Assume that u ∈ K κ \mathbf{u} \in\mathcal{K}_{\kappa} u ∈ K κ and δ u j ∈ c 0 ( Ω j ) \delta u_{j} \in c_{0}(\Omega_{j}) δ u j ∈ c 0 ( Ω j ) , j = 1 , 2 j=1,2 j = 1 , 2 .
(i) ϕ j ( m o n o ) [ u j ] , ϕ j ( i n t e r ) [ u ] ∈ L 1 ( Γ M ) \phi_{j}^{(\mathrm{mono})}[u_{j}],\phi_{j}^{(\mathrm{inter})}[\mathbf{u}] \in L^{1}(\Gamma_{M}) ϕ j ( mono ) [ u j ] , ϕ j ( inter ) [ u ] ∈ L 1 ( Γ M ) .
(ii) e ( t o t ) ∈ C k ( K κ ) e^{(\mathrm{tot})} \in C^{k}(\mathcal{K}_{\kappa}) e ( tot ) ∈ C k ( K κ ) .
(i) Δ Φ j ( m o n o ) [ δ u j ; u j ] , Δ Φ j ( i n t e r ) [ δ u ; u ] ∈ ℓ 1 ( R j ) \Delta\Phi_{j}^{(\mathrm{mono})}[\delta u_{j};u_{j}],\Delta\Phi_{j}^{(\mathrm{inter})}[\delta\mathbf{u};\mathbf{u}] \in\ell^{1}(\mathcal{R}_{j}) Δ Φ j ( mono ) [ δ u j ; u j ] , Δ Φ j ( inter ) [ δ u ; u ] ∈ ℓ 1 ( R j ) .
(ii) There exists a unique continuous extension Δ E u ( t o t ) ( ⋅ ) : ℓ 1 ( Ω 1 × Ω 2 ) → R \Delta E_{\mathbf{u}}^{(\mathrm{tot})}(\cdot):\ell^{1}(\Omega_{1}\times\Omega_{2})\to\mathbb{R} Δ E u ( tot ) ( ⋅ ) : ℓ 1 ( Ω 1 × Ω 2 ) → R .
(iii) Δ E u ( t o t ) ( ⋅ ) ∈ C k ( ℓ 1 ( Ω 1 × Ω 2 ) ) \Delta E_{\mathbf{u}}^{(\mathrm{tot})}(\cdot)\in C^{k}(\ell^{1}(\Omega_{1}\times\Omega_{2})) Δ E u ( tot ) ( ⋅ ) ∈ C k ( ℓ 1 ( Ω 1 × Ω 2 )) .
Proof. The proof essentially follows that of [33 Section 2.4, Theorem 1] .
Next, we recall from [18 Theorem 2.17] the ergodic theorem resulting in e ( t o t ) e^{(\mathrm{tot})} e ( tot ) .
Proposition 3.6 (Ergodic theorem). Let u ∈ K κ \mathbf{u} \in\mathcal{K}_{\kappa} u ∈ K κ . Then we have that
lim N → ∞ ∑ j = 1 2 1 ( 2 N + 1 ) 2 ∣ Γ j ∣ ∑ R j ∈ R j ( N ) ( ϕ j ( m o n o ) [ u j ] ( R j ) + ϕ j ( i n t e r ) [ u ] ( R j ) ) = e ( t o t ) ( u ) . \lim_{N\to\infty}\sum_{j=1}^{2}\frac{1}{(2N+1)^{2}|\Gamma_{j}|}\sum_{R_{j}\in\mathcal{R}_{j}(N)}\left(\phi_{j}^{(\mathrm{mono})}[u_{j}](R_{j})+\phi_{j}^{(\mathrm{inter})}[\mathbf{u}](R_{j})\right)=e^{(\mathrm{tot})}(\mathbf{u}). N → ∞ lim j = 1 ∑ 2 ( 2 N + 1 ) 2 ∣ Γ j ∣ 1 R j ∈ R j ( N ) ∑ ( ϕ j ( mono ) [ u j ] ( R j ) + ϕ j ( inter ) [ u ] ( R j ) ) = e ( tot ) ( u ) .
Cauchy-Born approximation
In addition to the many-body potentials, we also introduce the Cauchy-Born approximation. For simplicity, we restrict to a single lattice degree of freedom per layer. For an analysis of the continuum/Cauchy-Born approximation in the case of multilattices, we refer the interested reader to [32 ] .
e j ( C B ) ( u j ) : = ∫ Γ M d x V j ( m o n o ) ( ( ∇ u j ( x ) ⋅ R j ) R j ∈ R j ) ∣ Γ j ∣ . e_{j}^{(\mathrm{CB})}(u_{j}):=\int_{\Gamma_{M}} \mathrm{d}x\,\frac{V_{j}^{(\mathrm{mono})}\left((\nabla u_{j}(x)\cdot R_{j})_{R_{j}\in\mathcal{R}_{j}}\right)}{|\Gamma_{j}|}. e j ( CB ) ( u j ) := ∫ Γ M d x ∣ Γ j ∣ V j ( mono ) ( ( ∇ u j ( x ) ⋅ R j ) R j ∈ R j ) .
Lemma 3.7. Let j ∈ { 1 , 2 } j\in\{1,2\} j ∈ { 1 , 2 } , m ∈ N m\in\mathbb{N} m ∈ N . Assume that u = ( u 1 , u 2 ) ∈ K κ \mathbf{u}=(u_{1},u_{2})\in\mathcal{K}_{\kappa} u = ( u 1 , u 2 ) ∈ K κ and v ℓ ∈ W ‾ p e r 1 , 2 ( Γ M ; ( R 2 ) A j ) v_{\ell}\in\overline{W}^{1,2}_{\mathrm{per}}(\Gamma_{M};(\mathbb{R}^{2})^{\mathcal{A}_{j}}) v ℓ ∈ W per 1 , 2 ( Γ M ; ( R 2 ) A j ) , ℓ = 1 , … , m \ell=1,\ldots,m ℓ = 1 , … , m . For any 1 ≤ p 1 , … , p m ≤ ∞ 1\le p_{1},\ldots,p_{m}\le\infty 1 ≤ p 1 , … , p m ≤ ∞ such that ∑ ℓ = 1 m 1 p ℓ = 1 \sum_{\ell=1}^{m}\frac{1}{p_{\ell}}=1 ∑ ℓ = 1 m p ℓ 1 = 1 , we have that
∣ δ m e j ( C B ) ( u j ) [ v 1 , … , v m ] ∣ ≲ M m , j ( m o n o ) ∏ ℓ = 1 m ∥ ∇ v j ∥ L p ℓ ( Γ M ) . \left|\delta^{m}e_{j}^{(\mathrm{CB})}(u_{j})[v_{1},\ldots,v_{m}]\right|\lesssim M_{m,j}^{(\mathrm{mono})}\prod_{\ell=1}^{m}\|\nabla v_{j}\|_{L^{p_{\ell}}(\Gamma_{M})}. δ m e j ( CB ) ( u j ) [ v 1 , … , v m ] ≲ M m , j ( mono ) ℓ = 1 ∏ m ∥∇ v j ∥ L p ℓ ( Γ M ) .
Stacking-based relaxation
Instead of studying relaxation in its full generality, we provide sufficient conditions for the existence of a locally unique minimizer. We will only sketch the proof in the case of a single sublattice degree per layer. We believe that the analysis presented in [32 ] can be adapted to the present context to generalize our results to the case of multilattices. However, for the purpose of clarity and to better emphasize the main ideas, namely the difference between stacking- and phonon-based relaxation, we shall omit this generalization.
Relaxation in Cauchy-Born approximation
We start by proving the existence of energy minimizers in the Cauchy-Born approximation. The reason for that is that we can employ elliptic regularity theory to prove sufficient regularity for minimizers, in order to employ perturbation theory after that. Here, we treat the interlayer energy contribution as a small perturbation. The following statement about δ 2 e j ( C B ) \delta^2 e_j^{(\mathrm{CB})} δ 2 e j ( CB ) is given by [9 Proposition 4.1] . The statement about δ 2 e j ( m o n o ) \delta^2 e_j^{(\mathrm{mono})} δ 2 e j ( mono ) can be proved analogously, see, e.g., [33 Proposition 4] .
Lemma 4.1. For any v ∈ W p e r 1 , 2 ( Γ M ; ( R 2 ) A j ) v \in W_{\mathrm{per}}^{1,2}(\Gamma_M;(\mathbb{R}^2)^{A_j}) v ∈ W per 1 , 2 ( Γ M ; ( R 2 ) A j ) , we have that
δ 2 e j ( C B ) ( 0 ) [ v , v ] ≥ ν S t a c k ∥ ∇ v ∥ 2 2 . \delta^2 e_j^{(\mathrm{CB})}(0)[v,v] \ge\nu_{\mathrm{Stack}}\lVert\nabla v\rVert_2^2. δ 2 e j ( CB ) ( 0 ) [ v , v ] ≥ ν Stack ∥ ∇ v ∥ 2 2 . Moreover, there exists κ 1 > 0 \kappa_1 > 0 κ 1 > 0 such that for all κ ∈ ( 0 , κ 1 ) \kappa\in(0,\kappa_1) κ ∈ ( 0 , κ 1 ) , u ∈ K κ \mathbf{u} \in\mathcal{K}_\kappa u ∈ K κ , j = 1 , 2 j=1,2 j = 1 , 2 , we find that
δ 2 e j ( C B ) ( u j ) [ v , v ] , δ 2 e j ( m o n o ) ( u j ) [ v , v ] ≥ ν S t a c k 2 ∥ ∇ v ∥ 2 2 . \delta^2 e_j^{(\mathrm{CB})}(\mathbf{u}_j)[v,v],\ \delta^2 e_j^{(\mathrm{mono})}(\mathbf{u}_j)[v,v] \ge\frac{\nu_{\mathrm{Stack}}}{2}\lVert\nabla v\rVert_2^2. δ 2 e j ( CB ) ( u j ) [ v , v ] , δ 2 e j ( mono ) ( u j ) [ v , v ] ≥ 2 ν Stack ∥ ∇ v ∥ 2 2 .
Proposition 4.2 (Stacking-based stability of displaced lattices). Let κ 1 \kappa_1 κ 1 be as in Lemma 4.1 . Then there exists a constant δ 0 > 0 \delta_0 > 0 δ 0 > 0 dependent on ∣ Γ 1 ∣ \lvert\Gamma_1\rvert ∣ Γ 1 ∣ , ∣ Γ 2 ∣ \lvert\Gamma_2\rvert ∣ Γ 2 ∣ such that
M 2 ( i n t e r ) ≤ ν S t a c k max { ∣ Γ 1 ∣ ∣ Γ 2 ∣ , ∣ Γ 2 ∣ ∣ Γ 1 ∣ } ∣ Γ M ∣ δ 0 M_2^{(\mathrm{inter})} \le\frac{\nu_{\mathrm{Stack}}}{\max\left\{\frac{\lvert\Gamma_1\rvert}{\lvert\Gamma_2\rvert},\frac{\lvert\Gamma_2\rvert}{\lvert\Gamma_1\rvert}\right\}\lvert\Gamma_M\rvert}\delta_0 M 2 ( inter ) ≤ max { ∣ Γ 2 ∣ ∣ Γ 1 ∣ , ∣ Γ 1 ∣ ∣ Γ 2 ∣ } ∣ Γ M ∣ ν Stack δ 0 implies that, for all κ ∈ ( 0 , κ 1 ) \kappa\in(0,\kappa_1) κ ∈ ( 0 , κ 1 ) , u ∈ K κ \mathbf{u} \in\mathcal{K}_\kappa u ∈ K κ and v ∈ W ‾ p e r 1 , 2 \mathbf{v} \in\overline{W}_{\mathrm{per}}^{1,2} v ∈ W per 1 , 2 , we have that
δ 2 e ( t o t ) ( u ) [ v , v ] ≥ ν S t a c k 2 ∥ ∇ v ∥ 2 2 . \delta^2 e^{(\mathrm{tot})}(\mathbf{u})[\mathbf{v},\mathbf{v}] \ge\frac{\nu_{\mathrm{Stack}}}{2}\lVert\nabla\mathbf{v}\rVert_2^2. δ 2 e ( tot ) ( u ) [ v , v ] ≥ 2 ν Stack ∥ ∇ v ∥ 2 2 .
Proof. By Lemma 3.3 , we have that
δ 2 e j ( i n t e r ) ( u ) [ v , v ] ≲ max { 1 , ∣ Γ j ∣ ∣ Γ 3 − j ∣ } M 2 , 3 − j ( i n t e r ) ∥ v ∥ 2 2 \delta^2 e_j^{(\mathrm{inter})}(\mathbf{u})[\mathbf{v},\mathbf{v}]
\lesssim\max\left\{1,\frac{\lvert\Gamma_j\rvert}{\lvert\Gamma_{3-j}\rvert}\right\}M_{2,3-j}^{(\mathrm{inter})}\lVert\mathbf{v}\rVert_2^2 δ 2 e j ( inter ) ( u ) [ v , v ] ≲ max { 1 , ∣ Γ 3 − j ∣ ∣ Γ j ∣ } M 2 , 3 − j ( inter ) ∥ v ∥ 2 2
≲ max { 1 , ∣ Γ j ∣ ∣ Γ 3 − j ∣ } ∣ Γ M ∣ M 2 , 3 − j ( i n t e r ) ∥ ∇ v ∥ 2 2 , \lesssim\max\left\{1,\frac{\lvert\Gamma_j\rvert}{\lvert\Gamma_{3-j}\rvert}\right\}\lvert\Gamma_M\rvert M_{2,3-j}^{(\mathrm{inter})}\lVert\nabla\mathbf{v}\rVert_2^2, ≲ max { 1 , ∣ Γ 3 − j ∣ ∣ Γ j ∣ } ∣ Γ M ∣ M 2 , 3 − j ( inter ) ∥ ∇ v ∥ 2 2 ,
where, in the second step, we employed Poincaré’s inequality and the fact that ∫ Γ M v ( x ) d x = 0 \int_{\Gamma_M} v(x)\,\mathrm{d}x = 0 ∫ Γ M v ( x ) d x = 0 . In particular, there exists a constant δ 0 > 0 \delta_0 > 0 δ 0 > 0 dependent on ∣ Γ 1 ∣ |\Gamma_1| ∣ Γ 1 ∣ , ∣ Γ 2 ∣ |\Gamma_2| ∣ Γ 2 ∣ such that if
M 2 ( i n t e r ) ≤ ν S t a c k max { ∣ Γ 1 ∣ ∣ Γ 2 ∣ , ∣ Γ 2 ∣ ∣ Γ 1 ∣ } ∣ Γ M ∣ δ 0 M_2^{(\mathrm{inter})} \le\frac{\nu_{\mathrm{Stack}}}{\max\left\{\frac{|\Gamma_1|}{|\Gamma_2|},\frac{|\Gamma_2|}{|\Gamma_1|}\right\}|\Gamma_M|}\delta_0 M 2 ( inter ) ≤ max { ∣ Γ 2 ∣ ∣ Γ 1 ∣ , ∣ Γ 1 ∣ ∣ Γ 2 ∣ } ∣ Γ M ∣ ν Stack δ 0
then
δ 2 e ( t o t ) ( u ) [ v , v ] ≥ ν S t a c k 2 ∥ ∇ v ∥ 2 2 . \delta^2 e^{(\mathrm{tot})}(\mathbf{u})[\mathbf{v},\mathbf{v}] \ge\frac{\nu_{\mathrm{Stack}}}{2}\|\nabla\mathbf{v}\|_2^2. δ 2 e ( tot ) ( u ) [ v , v ] ≥ 2 ν Stack ∥∇ v ∥ 2 2 .
This concludes the proof.
Next, we want to prove the existence of a local minimizer. For that, we employ the following generalized inverse function theorem as stated, e.g., in [33 Lemma 14] .
Lemma 4.3. Let X , Y X,Y X , Y be Banach spaces, O O O an open subset of X X X , and let T : O → Y T:O\to Y T : O → Y be Fréchet differentiable. Assume that x 0 ∈ O x_0\in O x 0 ∈ O satisfies
∥ T ( x 0 ) ∥ Y ≤ η , ∥ δ T ( x 0 ) − 1 ∥ L ( Y , X ) ≤ σ , B ‾ X ( x 0 , 2 η σ ) ⊆ O . \|T(x_0)\|_Y \le\eta,\qquad\|\delta T(x_0)^{-1}\|_{L(Y,X)} \le\sigma,\qquad\overline{B}_X(x_0,2\eta\sigma)\subseteq O. ∥ T ( x 0 ) ∥ Y ≤ η , ∥ δ T ( x 0 ) − 1 ∥ L ( Y , X ) ≤ σ , B X ( x 0 , 2 η σ ) ⊆ O . Assume that δ T \delta T δ T satisfies the Lipschitz bound
∥ δ T ( x 1 ) − δ T ( x 2 ) ∥ L ( X , Y ) ≤ L ∥ x 1 − x 2 ∥ X ∀ x 1 , x 2 ∈ B ‾ X ( x 0 , 2 η σ ) , \|\delta T(x_1)-\delta T(x_2)\|_{L(X,Y)} \le L\|x_1-x_2\|_X\qquad\forall x_1,x_2\in\overline{B}_X(x_0,2\eta\sigma), ∥ δ T ( x 1 ) − δ T ( x 2 ) ∥ L ( X , Y ) ≤ L ∥ x 1 − x 2 ∥ X ∀ x 1 , x 2 ∈ B X ( x 0 , 2 η σ ) , and that 2 L σ 2 η < 1 2L\sigma^2\eta<1 2 L σ 2 η < 1 . Then there exists a unique x ∈ B X ( x 0 , 2 η σ ) x\in B_X(x_0,2\eta\sigma) x ∈ B X ( x 0 , 2 η σ ) such that T ( x ) = 0 T(x)=0 T ( x ) = 0 .
Next, we first prove the existence of local minimizers for when the monolayer energy is approximated by the Cauchy-Born energy. This in turn will allow us to prove the existence of a local minimizers for the non-approximated monolayer energy in proximity of the Cauchy-Born minimizer. This strategy was also pursued in [33 ] . In particular, there are cases when the solution to the problem in the Cauchy-Born approximation is stable while the atomistic model is unstable, see [12 , 19 , 33 ] .
For this purpose, we introduce the Cauchy-Born approximation of the total energy
e C B ( t o t ) ( u ) : = ∑ j = 1 2 ( e j ( C B ) ( u j ) + e j ( i n t e r ) ( u ) ) . e_{\mathrm{CB}}^{(\mathrm{tot})}(\mathbf{u}) := \sum_{j=1}^{2}\left(e_j^{(\mathrm{CB})}(u_j)+e_j^{(\mathrm{inter})}(\mathbf{u})\right). e CB ( tot ) ( u ) := j = 1 ∑ 2 ( e j ( CB ) ( u j ) + e j ( inter ) ( u ) ) .
Theorem 4.4. Let δ 0 > 0 \delta_0>0 δ 0 > 0 be as in Proposition 4.2 . Then there exists a constant δ ( C B ) ∈ ( 0 , δ 0 ) \delta^{(\mathrm{CB})}\in(0,\delta_0) δ ( CB ) ∈ ( 0 , δ 0 ) dependent on ∣ Γ 1 ∣ |\Gamma_1| ∣ Γ 1 ∣ , ∣ Γ 2 ∣ |\Gamma_2| ∣ Γ 2 ∣ such that if
M 2 ( i n t e r ) ≤ ν S t a c k max { ∣ Γ 1 ∣ ∣ Γ 2 ∣ , ∣ Γ 2 ∣ ∣ Γ 1 ∣ } ∣ Γ M ∣ δ ( C B ) , M_2^{(\mathrm{inter})} \le\frac{\nu_{\mathrm{Stack}}}{\max\left\{\frac{|\Gamma_1|}{|\Gamma_2|},\frac{|\Gamma_2|}{|\Gamma_1|}\right\}|\Gamma_M|}\delta^{(\mathrm{CB})}, M 2 ( inter ) ≤ max { ∣ Γ 2 ∣ ∣ Γ 1 ∣ , ∣ Γ 1 ∣ ∣ Γ 2 ∣ } ∣ Γ M ∣ ν Stack δ ( CB ) ,
and that
max j = 1 , 2 ∥ D j → 3 − j ∥ 2 M 2 ( i n t e r ) ( 1 + ν S t a c k − 1 max j = 1 , 2 ∥ D j → 3 − j ∥ 2 2 ∑ ℓ = 1 3 M ℓ ( i n t e r ) 2 ) ( ∑ ℓ = 3 5 M ℓ ( m o n o ) + ∑ ℓ = 3 4 M ℓ ( i n t e r ) ) ≤ δ ( C B ) , \max_{j=1,2}\lVert D_{j\to3-j}\rVert_{2}M_{2}^{(\mathrm{inter})}\left(1+\nu_{\mathrm{Stack}}^{-1}\max_{j=1,2}\lVert D_{j\to3-j}\rVert_{2}^{2}\sum_{\ell=1}^{3}M_{\ell}^{(\mathrm{inter})\,2}\right)\left(\sum_{\ell=3}^{5}M_{\ell}^{(\mathrm{mono})}+\sum_{\ell=3}^{4}M_{\ell}^{(\mathrm{inter})}\right)\leq\delta^{(\mathrm{CB})}, j = 1 , 2 max ∥ D j → 3 − j ∥ 2 M 2 ( inter ) ( 1 + ν Stack − 1 j = 1 , 2 max ∥ D j → 3 − j ∥ 2 2 ℓ = 1 ∑ 3 M ℓ ( inter ) 2 ) ( ℓ = 3 ∑ 5 M ℓ ( mono ) + ℓ = 3 ∑ 4 M ℓ ( inter ) ) ≤ δ ( CB ) ,
then there exists a solution u ( C B ) ∈ W ‾ p e r 3 , 2 \mathbf{u}^{(\mathrm{CB})}\in\overline{W}^{3,2}_{\mathrm{per}} u ( CB ) ∈ W per 3 , 2 of
δ C B e ( t o t ) ( u ( C B ) ) [ v ] = 0 ∀ v ∈ W ‾ p e r 3 , 2 . \delta_{\mathrm{CB}}^{e\,(\mathrm{tot})}\left(\mathbf{u}^{(\mathrm{CB})}\right)[\mathbf{v}]=0\quad\forall\mathbf{v}\in\overline{W}^{3,2}_{\mathrm{per}}. δ CB e ( tot ) ( u ( CB ) ) [ v ] = 0 ∀ v ∈ W per 3 , 2 .
This solution is stable in the sense that
δ C B 2 ( t o t ) ( u ( C B ) ) [ v , v ] ≳ ∥ ∇ v ∥ 2 2 ∀ v ∈ W ‾ p e r 1 , 2 . \delta_{\mathrm{CB}}^{2\,(\mathrm{tot})}\left(\mathbf{u}^{(\mathrm{CB})}\right)[\mathbf{v},\mathbf{v}]\gtrsim\lVert\nabla\mathbf{v}\rVert_{2}^{2}\quad\forall\mathbf{v}\in\overline{W}^{1,2}_{\mathrm{per}}. δ CB 2 ( tot ) ( u ( CB ) ) [ v , v ] ≳ ∥ ∇ v ∥ 2 2 ∀ v ∈ W per 1 , 2 .
It also satisfies the a priori bound
∥ u ( C B ) ∥ W 3 , 2 ≲ max j = 1 , 2 ∥ D j → 3 − j ∥ 2 M 2 ( i n t e r ) ( 1 + ν S t a c k − 1 max j = 1 , 2 ∥ D j → 3 − j ∥ 2 2 ∑ ℓ = 1 3 M ℓ ( i n t e r ) ) . \lVert\mathbf{u}^{(\mathrm{CB})}\rVert_{W^{3,2}}\lesssim\max_{j=1,2}\lVert D_{j\to3-j}\rVert_{2}M_{2}^{(\mathrm{inter})}\left(1+\nu_{\mathrm{Stack}}^{-1}\max_{j=1,2}\lVert D_{j\to3-j}\rVert_{2}^{2}\sum_{\ell=1}^{3}M_{\ell}^{(\mathrm{inter})}\right). ∥ u ( CB ) ∥ W 3 , 2 ≲ j = 1 , 2 max ∥ D j → 3 − j ∥ 2 M 2 ( inter ) ( 1 + ν Stack − 1 j = 1 , 2 max ∥ D j → 3 − j ∥ 2 2 ℓ = 1 ∑ 3 M ℓ ( inter ) ) .
Proof. The proof is standard and is only a slight modification of the arguments presented in the proof of [9 Theorem 4.3] . We want to apply Lemma 4.3 to X = W ‾ p e r 3 , 2 X=\overline{W}^{3,2}_{\mathrm{per}} X = W per 3 , 2 , Y = W ‾ p e r 1 , 2 Y=\overline{W}^{1,2}_{\mathrm{per}} Y = W per 1 , 2 , x 0 = 0 x_{0}=0 x 0 = 0 , O = B X ( 0 , κ ′ ) O=B_{X}(0,\kappa') O = B X ( 0 , κ ′ ) , where κ ′ > 0 \kappa'>0 κ ′ > 0 is chosen small enough such that, due to the Sobolev embedding W ‾ p e r 3 , 2 ⊆ W ‾ p e r 1 , ∞ \overline{W}^{3,2}_{\mathrm{per}}\subseteq\overline{W}^{1,\infty}_{\mathrm{per}} W per 3 , 2 ⊆ W per 1 , ∞ , B X ( 0 , κ ′ ) ⊆ K κ B_{X}(0,\kappa')\subseteq K_{\kappa} B X ( 0 , κ ′ ) ⊆ K κ for κ ∈ ( 0 , κ 1 ) \kappa\in(0,\kappa_{1}) κ ∈ ( 0 , κ 1 ) and κ 1 > 0 \kappa_{1}>0 κ 1 > 0 chosen as in Lemma 4.1. We define the mapping
T : O → Y , u ↦ δ C B e ( t o t ) ( u ) . T:O\to Y,\quad\mathbf{u}\mapsto\delta_{\mathrm{CB}}^{e\,(\mathrm{tot})}(\mathbf{u}). T : O → Y , u ↦ δ CB e ( tot ) ( u ) .
Residual bound: Observe that, due to
δ e j ( C B ) ( 0 ) [ v ] ∝ ∫ Γ μ d x v ( x ) = 0 \delta_{e_{j}}^{(\mathrm{CB})}(0)[\mathbf{v}]\propto\int_{\Gamma_{\mu}}\mathrm{d}x\,v(x)=0 δ e j ( CB ) ( 0 ) [ v ] ∝ ∫ Γ μ d x v ( x ) = 0
for v ∈ W ‾ p e r 1 , 2 \mathbf{v}\in\overline{W}^{1,2}_{\mathrm{per}} v ∈ W per 1 , 2 , we have that
T ( 0 ) ( x ) = ∑ j = 1 2 δ e j ( i n t e r ) ( 0 ) ( x ) = ∑ ℓ , j = 1 2 [ ∑ R 3 − j ( V j ; R 3 − j ( i n t e r ) ) ℓ ( Δ j , 3 − j [ 0 ] ( x ) ) − ∑ R j ( V 3 − j ; R j ( i n t e r ) ) ℓ ( Δ 3 − j , j [ 0 ] ( A 3 − j A j − 1 ( x − R j ) ) ) ] e ℓ + j . (21) \begin{aligned}
T(0)(x)=\sum_{j=1}^{2}\delta_{e_{j}}^{(\mathrm{inter})}(0)(x) \\
&=\sum_{\ell,j=1}^{2}\left[\sum_{R_{3-j}}\left(V_{j;R_{3-j}}^{(\mathrm{inter})}\right)_{\ell}\left(\Delta_{j,3-j}[0](x)\right)-\sum_{R_{j}}\left(V_{3-j;R_{j}}^{(\mathrm{inter})}\right)_{\ell}\left(\Delta_{3-j,j}[0]\left(A_{3-j}A_{j}^{-1}(x-R_{j})\right)\right)\right]e_{\ell+j}. \tag*{(21)}
\end{aligned} T ( 0 ) ( x ) = j = 1 ∑ 2 δ e j ( inter ) ( 0 ) ( x ) = ℓ , j = 1 ∑ 2 R 3 − j ∑ ( V j ; R 3 − j ( inter ) ) ℓ ( Δ j , 3 − j [ 0 ] ( x ) ) − R j ∑ ( V 3 − j ; R j ( inter ) ) ℓ ( Δ 3 − j , j [ 0 ] ( A 3 − j A j − 1 ( x − R j ) ) ) e ℓ + j . (21)
where e k e_k e k is the standard unit vector in the k k k -th direction, and where we recall (16) . Consequently, we obtain that
∥ T ( 0 ) ∥ Y ≤ ∑ j = 1 2 ∥ δ e j ( i n t e r ) ( 0 ) ∥ W 1 , 2 ≲ ∑ j = 1 2 ∥ ∇ δ e j ( i n t e r ) ( 0 ) ∥ 2 . \|T(0)\|_Y \le\sum_{j=1}^{2}\|\delta e_j^{(\mathrm{inter})}(0)\|_{W^{1,2}} \lesssim\sum_{j=1}^{2}\|\nabla\delta e_j^{(\mathrm{inter})}(0)\|_2 . ∥ T ( 0 ) ∥ Y ≤ j = 1 ∑ 2 ∥ δ e j ( inter ) ( 0 ) ∥ W 1 , 2 ≲ j = 1 ∑ 2 ∥∇ δ e j ( inter ) ( 0 ) ∥ 2 .
where, in the last step, we used Poincaré’s inequality. Plugging in the expression (21) , we thus obtain
∥ T ( 0 ) ∥ Y ≲ max j = 1 , 2 ∥ D j → 3 − j ∥ 2 M 2 ( i n t e r ) . (22) \|T(0)\|_Y \lesssim\max_{j=1,2}\|D_{j\to3-j}\|_2 M_2^{(\mathrm{inter})}. \tag*{(22)} ∥ T ( 0 ) ∥ Y ≲ j = 1 , 2 max ∥ D j → 3 − j ∥ 2 M 2 ( inter ) . (22)
By assumption, we thus can choose ∥ T ( 0 ) ∥ Y \|T(0)\|_Y ∥ T ( 0 ) ∥ Y small enough.
Stability estimate: Proposition 4.2 implies that δ T ( 0 ) − 1 ∈ L ( Y , Y ) \delta T(0)^{-1}\in\mathcal{L}(Y,Y) δ T ( 0 ) − 1 ∈ L ( Y , Y ) , e.g., by the Lax-Milgrams theorem. We adapt the proof of [9 Theorem 4.3] to the present case to prove that even δ T ( 0 ) − 1 ∈ L ( Y , X ) \delta T(0)^{-1}\in\mathcal{L}(Y,X) δ T ( 0 ) − 1 ∈ L ( Y , X ) .
For that, let v ∈ Y \mathbf{v}\in Y v ∈ Y . Due to the stability estimate described above, there then exists a unique solution u ∈ Y \mathbf{u}\in Y u ∈ Y to δ T ( 0 ) [ u ] = v \delta T(0)[\mathbf{u}]=\mathbf{v} δ T ( 0 ) [ u ] = v with ∥ u ∥ Y ≤ 2 ν S t a c k ∥ v ∥ Y \|\mathbf{u}\|_Y\le\frac{2}{\nu_{\mathrm{Stack}}}\|\mathbf{v}\|_Y ∥ u ∥ Y ≤ ν Stack 2 ∥ v ∥ Y .
Let w : = − ∑ j = 1 2 δ 2 e j ( i n t e r ) ( 0 ) [ u , ⋅ ] \mathbf{w}:=-\sum_{j=1}^{2}\delta^2 e_j^{(\mathrm{inter})}(0)[\mathbf{u},\cdot] w := − ∑ j = 1 2 δ 2 e j ( inter ) ( 0 ) [ u , ⋅ ] . A straight-forward but lengthy computation, analogously to that leading to (22) yields
∥ w ∥ W 1 , 2 ≲ max j = 1 , 2 ∥ D j → 3 − j ∥ 2 2 ∑ ℓ = 1 3 M ℓ ( i n t e r ) ∥ u ∥ W 1 , 2 . \|\mathbf{w}\|_{W^{1,2}}\lesssim\max_{j=1,2}\|D_{j\to3-j}\|_2^2\sum_{\ell=1}^{3}M_\ell^{(\mathrm{inter})}\|\mathbf{u}\|_{W^{1,2}}. ∥ w ∥ W 1 , 2 ≲ j = 1 , 2 max ∥ D j → 3 − j ∥ 2 2 ℓ = 1 ∑ 3 M ℓ ( inter ) ∥ u ∥ W 1 , 2 .
Then u \mathbf{u} u solves the elliptic problem with constant coefficients
∑ j = 1 2 δ 2 e j ( C B ) ( 0 ) [ u , ⋅ ] = v + w . \sum_{j=1}^{2}\delta^2 e_j^{(\mathrm{CB})}(0)[\mathbf{u},\cdot]=\mathbf{v}+\mathbf{w}. j = 1 ∑ 2 δ 2 e j ( CB ) ( 0 ) [ u , ⋅ ] = v + w .
Elliptic regularity theory as explained, e.g., in [15 , 16 ] , then yields that u ∈ X \mathbf{u}\in X u ∈ X , with the a priori estimate ∥ u ∥ W 3 , 2 ≲ ∥ v + w ∥ W 1 , 2 \|\mathbf{u}\|_{W^{3,2}}\lesssim\|\mathbf{v}+\mathbf{w}\|_{W^{1,2}} ∥ u ∥ W 3 , 2 ≲ ∥ v + w ∥ W 1 , 2 . In particular, δ T ( 0 ) : X → Y \delta T(0):X\to Y δ T ( 0 ) : X → Y is an isomorphism with the stability estimate
∥ δ T ( 0 ) − 1 ∥ L ( Y , X ) ≲ ( 1 + ν S t a c k − 1 max j = 1 , 2 ∥ D j → 3 − j ∥ 2 2 ∑ ℓ = 1 3 M ℓ ( i n t e r ) ) . \|\delta T(0)^{-1}\|_{\mathcal{L}(Y,X)}\lesssim\left(1+\nu_{\mathrm{Stack}}^{-1}\max_{j=1,2}\|D_{j\to3-j}\|_2^2\sum_{\ell=1}^{3}M_\ell^{(\mathrm{inter})}\right). ∥ δ T ( 0 ) − 1 ∥ L ( Y , X ) ≲ ( 1 + ν Stack − 1 j = 1 , 2 max ∥ D j → 3 − j ∥ 2 2 ℓ = 1 ∑ 3 M ℓ ( inter ) ) .
Lipschitz bound: A lengthy but straight-forward calculation, employing the Sobolev embedding ∥ u ∥ 1 , ∞ ≲ ∥ u ∥ 2 , 4 ≲ ∥ u ∥ 3 , 2 \|\mathbf{u}\|_{1,\infty}\lesssim\|\mathbf{u}\|_{2,4}\lesssim\|\mathbf{u}\|_{3,2} ∥ u ∥ 1 , ∞ ≲ ∥ u ∥ 2 , 4 ≲ ∥ u ∥ 3 , 2 implies that
∥ δ 2 e j ( C B ) ( u ) [ v , ⋅ ] − δ 2 e j ( C B ) ( u ′ ) [ v , ⋅ ] ∥ W 1 , 2 (23) \|\delta^2 e_j^{(\mathrm{CB})}(\mathbf{u})[\mathbf{v},\cdot]-\delta^2 e_j^{(\mathrm{CB})}(\mathbf{u}')[\mathbf{v},\cdot]\|_{W^{1,2}} \tag*{(23)} ∥ δ 2 e j ( CB ) ( u ) [ v , ⋅ ] − δ 2 e j ( CB ) ( u ′ ) [ v , ⋅ ] ∥ W 1 , 2 (23)
≲ ∑ ℓ = 3 5 M ℓ , j ( m o n o ) ∥ u − u ′ ∥ W 3 , 2 ∥ v ∥ W 3 , 2 . \lesssim\sum_{\ell=3}^{5}M_{\ell,j}^{(\mathrm{mono})}\|\mathbf{u}-\mathbf{u}'\|_{W^{3,2}}\|\mathbf{v}\|_{W^{3,2}}. ≲ ℓ = 3 ∑ 5 M ℓ , j ( mono ) ∥ u − u ′ ∥ W 3 , 2 ∥ v ∥ W 3 , 2 .
Analogously, one proves that
∥ δ 2 e j ( i n t e r ) ( u ) [ v , ⋅ ] − δ 2 e j ( i n t e r ) ( u ′ ) [ v , ⋅ ] ∥ W 1 , 2 ≲ ∑ ℓ = 3 4 M ℓ , j ( i n t e r ) ∥ u − u ′ ∥ W 3 , 2 ∥ v ∥ W 3 , 2 . (24) \left\|\delta^{2}e_{j}^{(\mathrm{inter})}(\mathbf{u})[\mathbf{v},\cdot]-\delta^{2}e_{j}^{(\mathrm{inter})}(\mathbf{u}^{\prime})[\mathbf{v},\cdot]\right\|_{W^{1,2}}
\lesssim\sum_{\ell=3}^{4}M_{\ell,j}^{(\mathrm{inter})}\left\|\mathbf{u}-\mathbf{u}^{\prime}\right\|_{W^{3,2}}\left\|\mathbf{v}\right\|_{W^{3,2}} . \tag*{(24)} δ 2 e j ( inter ) ( u ) [ v , ⋅ ] − δ 2 e j ( inter ) ( u ′ ) [ v , ⋅ ] W 1 , 2 ≲ ℓ = 3 ∑ 4 M ℓ , j ( inter ) ∥ u − u ′ ∥ W 3 , 2 ∥ v ∥ W 3 , 2 . (24)
With a computation analogous to [33 Proof of Theorem 3] and similar to those proving (23) , (24) , one can also prove that
∥ T ( u + v ) − T ( u ) − δ 2 e ( t o t ) ( u ) [ v , ⋅ ] ∥ W 1 , 2 ≲ ( ∑ ℓ = 3 5 M ℓ ( m o n o ) + ∑ ℓ = 3 4 M ℓ ( i n t e r ) ) ∥ v ∥ W 3 , 2 2 . \left\|T(\mathbf{u}+\mathbf{v})-T(\mathbf{u})-\delta^{2}e^{(\mathrm{tot})}(\mathbf{u})[\mathbf{v},\cdot]\right\|_{W^{1,2}}
\lesssim\left(\sum_{\ell=3}^{5}M_{\ell}^{(\mathrm{mono})}+\sum_{\ell=3}^{4}M_{\ell}^{(\mathrm{inter})}\right)\left\|\mathbf{v}\right\|_{W^{3,2}}^{2}. T ( u + v ) − T ( u ) − δ 2 e ( tot ) ( u ) [ v , ⋅ ] W 1 , 2 ≲ ( ℓ = 3 ∑ 5 M ℓ ( mono ) + ℓ = 3 ∑ 4 M ℓ ( inter ) ) ∥ v ∥ W 3 , 2 2 .
In particular, T : X → Y T:X\to Y T : X → Y is Fréchet differentiable and its derivative δ 2 e ( t o t ) \delta^{2}e^{(\mathrm{tot})} δ 2 e ( tot ) satisfies the Lipschitz bound
∥ δ T ( u ) − δ T ( u ′ ) ∥ L ( X , Y ) ≲ ( ∑ ℓ = 3 5 M ℓ ( m o n o ) + ∑ ℓ = 3 4 M ℓ ( i n t e r ) ) ∥ u − u ′ ∥ W 3 , 2 . \left\|\delta T(\mathbf{u})-\delta T(\mathbf{u}^{\prime})\right\|_{\mathcal{L}(X,Y)}
\lesssim\left(\sum_{\ell=3}^{5}M_{\ell}^{(\mathrm{mono})}+\sum_{\ell=3}^{4}M_{\ell}^{(\mathrm{inter})}\right)\left\|\mathbf{u}-\mathbf{u}^{\prime}\right\|_{W^{3,2}}. ∥ δ T ( u ) − δ T ( u ′ ) ∥ L ( X , Y ) ≲ ( ℓ = 3 ∑ 5 M ℓ ( mono ) + ℓ = 3 ∑ 4 M ℓ ( inter ) ) ∥ u − u ′ ∥ W 3 , 2 .
Consequently, Lemma 4.3 implies the existence of a local minimizer u ( C B ) \mathbf{u}^{(\mathrm{CB})} u ( CB ) and its a priori bound. The stability bound follows from Proposition 4.2.
Relaxation in atomistic model
Our goal is now to show that existence of a minimizer for the full atomistic model which is a neighborhood of a minimizer for the Cauchy-Born approximation. This will require to impose additional decay assumptions on V ( m o n o ) V^{(\mathrm{mono})} V ( mono ) . In particular, these assumptions are sufficient to justify the existence of a minimizer for the atomistic model.
The next statement employs [33 Lemma 12] . However, stating it properly would require us to introduce some notation that is similar to that in the present work. In order to avoid confusion and to keep the presentation brief, we thus omit an extensive introduction, and refer the interested reader to the original reference instead.
Theorem 4.5. Let δ ( C B ) > 0 \delta^{(\mathrm{CB})}>0 δ ( CB ) > 0 be given as in Theorem 4.4 . Assume that U ( C B ) \mathbf{U}^{(\mathrm{CB})} U ( CB ) is a local minimizer of e C B ( t o t ) e_{\mathrm{CB}}^{(\mathrm{tot})} e CB ( tot ) and, for any a 0 > 0 a_{0}>0 a 0 > 0 , define
u 0 ( C B ) ( x ) : = a 0 − 1 U ( C B ) ( a 0 x ) ∀ x ∈ Γ M ( 0 ) : = 1 a 0 Γ M . \mathbf{u}_{0}^{(\mathrm{CB})}(x):=a_{0}^{-1}\mathbf{U}^{(\mathrm{CB})}(a_{0}x)\quad\forall x\in\Gamma_{\mathcal{M}}^{(0)}:=\frac{1}{a_{0}}\Gamma_{\mathcal{M}}. u 0 ( CB ) ( x ) := a 0 − 1 U ( CB ) ( a 0 x ) ∀ x ∈ Γ M ( 0 ) := a 0 1 Γ M .
Then there exists there exists a constant δ ( m o n o ) ∈ ( 0 , δ ( C B ) ) \delta^{(\mathrm{mono})} \in(0,\delta^{(\mathrm{CB})}) δ ( mono ) ∈ ( 0 , δ ( CB ) ) dependent on ∣ Γ 1 ∣ |\Gamma_1| ∣ Γ 1 ∣ , ∣ Γ 2 ∣ |\Gamma_2| ∣ Γ 2 ∣ such that if
a 0 M 2 ( i n t e r ) max j = 1 , 2 ∥ D j → 3 − j ∥ 2 ν S t a c k 2 ( 1 + max j = 1 , 2 ∥ D j → 3 − j ∥ 2 ν S t a c k ∑ ℓ = 1 3 M ℓ ( i n t e r ) ) a_0 M_2^{(\mathrm{inter})} \frac{\max_{j=1,2}\|D_{j\to3-j}\|^2}{\nu_{\mathrm{Stack}}^2}\left(1+\frac{\max_{j=1,2}\|D_{j\to3-j}\|^2}{\nu_{\mathrm{Stack}}}\sum_{\ell=1}^{3}M_\ell^{(\mathrm{inter})}\right) a 0 M 2 ( inter ) ν Stack 2 max j = 1 , 2 ∥ D j → 3 − j ∥ 2 ( 1 + ν Stack max j = 1 , 2 ∥ D j → 3 − j ∥ 2 ℓ = 1 ∑ 3 M ℓ ( inter ) )
∑ ℓ = 2 3 M ℓ , 2 ( m o n o , a t ) ( M 3 ( m o n o ) + max { ∣ Γ 1 ∣ ∣ Γ 2 ∣ , ∣ Γ 2 ∣ ∣ Γ 1 ∣ } M 3 ( i n t e r ) ) ≤ δ ( m o n o ) \sum_{\ell=2}^{3}M_{\ell,2}^{(\mathrm{mono,at})}\left(M_3^{(\mathrm{mono})}+\max\left\{\frac{|\Gamma_1|}{|\Gamma_2|},\frac{|\Gamma_2|}{|\Gamma_1|}\right\}M_3^{(\mathrm{inter})}\right)\leq\delta^{(\mathrm{mono})} ℓ = 2 ∑ 3 M ℓ , 2 ( mono , at ) ( M 3 ( mono ) + max { ∣ Γ 2 ∣ ∣ Γ 1 ∣ , ∣ Γ 1 ∣ ∣ Γ 2 ∣ } M 3 ( inter ) ) ≤ δ ( mono )
and
M 2 ( i n t e r ) ≤ ν S t a c k max { ∣ Γ 1 ∣ ∣ Γ 2 ∣ , ∣ Γ 2 ∣ ∣ Γ 1 ∣ } ∣ Γ M ∣ δ ( C B ) M_2^{(\mathrm{inter})}\leq\frac{\nu_{\mathrm{Stack}}}{\max\left\{\frac{|\Gamma_1|}{|\Gamma_2|},\frac{|\Gamma_2|}{|\Gamma_1|}\right\}|\Gamma_M|}\delta^{(\mathrm{CB})} M 2 ( inter ) ≤ max { ∣ Γ 2 ∣ ∣ Γ 1 ∣ , ∣ Γ 1 ∣ ∣ Γ 2 ∣ } ∣ Γ M ∣ ν Stack δ ( CB )
and
max j = 1 , 2 ∥ D j → 3 − j ∥ 2 M 2 ( i n t e r ) ( 1 + max j = 1 , 2 ∥ D j → 3 − j ∥ 2 ν S t a c k ∑ ℓ = 1 3 M ℓ ( i n t e r ) ) 2 \max_{j=1,2}\|D_{j\to3-j}\|^2M_2^{(\mathrm{inter})}\left(1+\frac{\max_{j=1,2}\|D_{j\to3-j}\|^2}{\nu_{\mathrm{Stack}}}\sum_{\ell=1}^{3}M_\ell^{(\mathrm{inter})}\right)^2 j = 1 , 2 max ∥ D j → 3 − j ∥ 2 M 2 ( inter ) ( 1 + ν Stack max j = 1 , 2 ∥ D j → 3 − j ∥ 2 ℓ = 1 ∑ 3 M ℓ ( inter ) ) 2
( ∑ ℓ = 3 5 M ℓ ( m o n o ) + ∑ ℓ = 3 4 M ℓ ( i n t e r ) ) ≤ δ ( C B ) , \left(\sum_{\ell=3}^{5}M_\ell^{(\mathrm{mono})}+\sum_{\ell=3}^{4}M_\ell^{(\mathrm{inter})}\right)\leq\delta^{(\mathrm{CB})}, ( ℓ = 3 ∑ 5 M ℓ ( mono ) + ℓ = 3 ∑ 4 M ℓ ( inter ) ) ≤ δ ( CB ) ,
there exists a stable local minimizer u ( e q ) \mathbf{u}^{(\mathrm{eq})} u ( eq ) of e ( t o t ) e^{(\mathrm{tot})} e ( tot ) which satisfies
a 0 ∥ ∇ u ( e q ) − ∇ u ( C B ) ∥ L 2 ( Γ M ( 0 ) ) ≲ a 0 M 2 ( i n t e r ) max j = 1 , 2 ∥ D j → 3 − j ∥ 2 ν S t a c k ∑ ℓ = 2 3 M ℓ , 2 ( m o n o , a t ) a_0\|\nabla\mathbf{u}^{(\mathrm{eq})}-\nabla\mathbf{u}^{(\mathrm{CB})}\|_{L^2(\Gamma_M^{(0)})}\lesssim a_0M_2^{(\mathrm{inter})}\frac{\max_{j=1,2}\|D_{j\to3-j}\|^2}{\nu_{\mathrm{Stack}}}\sum_{\ell=2}^{3}M_{\ell,2}^{(\mathrm{mono,at})} a 0 ∥∇ u ( eq ) − ∇ u ( CB ) ∥ L 2 ( Γ M ( 0 ) ) ≲ a 0 M 2 ( inter ) ν Stack max j = 1 , 2 ∥ D j → 3 − j ∥ 2 ℓ = 2 ∑ 3 M ℓ , 2 ( mono , at )
( 1 + max j = 1 , 2 ∥ D j → 3 − j ∥ 2 ν S t a c k ∑ ℓ = 1 3 M ℓ ( i n t e r ) ) . \left(1+\frac{\max_{j=1,2}\|D_{j\to3-j}\|^2}{\nu_{\mathrm{Stack}}}\sum_{\ell=1}^{3}M_\ell^{(\mathrm{inter})}\right). ( 1 + ν Stack max j = 1 , 2 ∥ D j → 3 − j ∥ 2 ℓ = 1 ∑ 3 M ℓ ( inter ) ) .
Proof. We want to apply Lemma 4.3 for the case x 0 = u ( C B ) x_0=\mathbf{u}^{(\mathrm{CB})} x 0 = u ( CB ) , X = W ‾ p e r 1 , 2 X=\overline{W}_{\mathrm{per}}^{1,2} X = W per 1 , 2 , endowed with the W ˙ 1 , 2 \dot{W}^{1,2} W ˙ 1 , 2 norm, Y = W ‾ p e r − 1 , 2 Y=\overline{W}_{\mathrm{per}}^{-1,2} Y = W per − 1 , 2 , endowed with the W ˙ − 1 , 2 \dot{W}^{-1,2} W ˙ − 1 , 2 norm, O : = B W ‾ p e r 1 , 2 ( 0 , η 0 ) O:=B_{\overline{W}_{\mathrm{per}}^{1,2}}(0,\eta_0) O := B W per 1 , 2 ( 0 , η 0 ) . Let κ 1 > 0 \kappa_1>0 κ 1 > 0 be as in Lemma 4.1 . We choose η 0 \eta_0 η 0 small enough s.t. B W ‾ p e r 1 , 2 ( x 0 , η 0 ) ⊆ W ‾ p e r 1 , 2 ∩ D κ 1 B_{\overline{W}_{\mathrm{per}}^{1,2}}(x_0,\eta_0)\subseteq\overline{W}_{\mathrm{per}}^{1,2}\cap D_{\kappa_1} B W per 1 , 2 ( x 0 , η 0 ) ⊆ W per 1 , 2 ∩ D κ 1 . Then we define
T ( w ) : = δ e ( t o t ) ( u ( C B ) + w ) : W ‾ p e r → R . T(\mathbf{w}):=\delta e^{(\mathrm{tot})}(\mathbf{u}^{(\mathrm{CB})}+\mathbf{w}):\overline{W}_{\mathrm{per}}\to\mathbb{R}. T ( w ) := δ e ( tot ) ( u ( CB ) + w ) : W per → R .
Lemmata 3.1 and 3.3 imply that T T T is Fréchet differentiable on O O O and that δ T \delta T δ T is Lipschitz continuous on O O O with Lipschitz constant
C ( M 3 ( m o n o ) + max { ∣ Γ 1 ∣ ∣ Γ 2 ∣ , ∣ Γ 2 ∣ ∣ Γ 1 ∣ } M 3 ( i n t e r ) ) C\left(M_3^{(\mathrm{mono})}+\max\left\{\frac{|\Gamma_1|}{|\Gamma_2|},\frac{|\Gamma_2|}{|\Gamma_1|}\right\}M_3^{(\mathrm{inter})}\right) C ( M 3 ( mono ) + max { ∣ Γ 2 ∣ ∣ Γ 1 ∣ , ∣ Γ 1 ∣ ∣ Γ 2 ∣ } M 3 ( inter ) )
for some universal constant C > 0 C>0 C > 0 .
The stability estimate follows immediately from Proposition 4.2 . In particular, we have that
∥ δ T ( 0 ) − 1 ∥ L ( Y , X ) ≤ 2 ν S t a c k . \lVert\delta T(0)^{-1} \rVert_{\mathcal{L}(Y,X)} \le\frac{2}{\nu_{\mathrm{Stack}}}. ∥ δ T ( 0 ) − 1 ∥ L ( Y , X ) ≤ ν Stack 2 .
Next, we want to establish a residual bound. Given v ∈ W ‾ p e r \mathbf{v} \in\overline{W}_{\mathrm{per}} v ∈ W per and recalling (6) , we abbreviate
v ~ : = ( v ~ j ) j = 1 2 , v ~ j ( x ) : = ζ ∗ v j ( x ) = ∫ Γ M ( 0 ) d y ζ ( x − y ) v j ( y ) , \widetilde{\mathbf{v}} := (\widetilde{v}_j)_{j=1}^{2}, \qquad\widetilde{v}_j(x) := \zeta* v_j(x) = \int_{\Gamma_{\mathcal{M}}^{(0)}} \mathrm{d}y\,\zeta(x-y)v_j(y), v := ( v j ) j = 1 2 , v j ( x ) := ζ ∗ v j ( x ) = ∫ Γ M ( 0 ) d y ζ ( x − y ) v j ( y ) ,
where we extended v j v_j v j trivially by 0 0 0 outside of Γ M \Gamma_{\mathcal{M}} Γ M .
Observe that
T ( 0 ) [ v ~ ] = δ e ( t o t ) ( u ( C B ) ) [ v ~ ] = ∑ j = 1 2 ( δ e j ( m o n o ) ( u j ( C B ) ) [ v ~ j ] − δ e j ( C B ) ( u j ( C B ) ) [ v j ] ) . (25) T(0)[\widetilde{\mathbf{v}}] = \delta e^{(\mathrm{tot})}(\mathbf{u}^{(\mathrm{CB})})[\widetilde{\mathbf{v}}] = \sum_{j=1}^{2} \left( \delta e_j^{(\mathrm{mono})}(u_j^{(\mathrm{CB})})[\widetilde{v}_j] - \delta e_j^{(\mathrm{CB})}(u_j^{(\mathrm{CB})})[v_j] \right). \tag*{(25)} T ( 0 ) [ v ] = δ e ( tot ) ( u ( CB ) ) [ v ] = j = 1 ∑ 2 ( δ e j ( mono ) ( u j ( CB ) ) [ v j ] − δ e j ( CB ) ( u j ( CB ) ) [ v j ] ) . (25)
where we used the fact that e C B ( t o t ) ( u ( C B ) ) = 0 e_{\mathrm{CB}}^{(\mathrm{tot})}(\mathbf{u}^{(\mathrm{CB})}) = 0 e CB ( tot ) ( u ( CB ) ) = 0 . Recalling the localization formula (7) , we have that
Δ R j ( j ) v ~ j = ∫ Γ M ( 0 ) d y 1 ∣ Γ j ∣ ∫ 0 1 d t ζ j ( x + t R j − y ) ∇ R j v ( y ) ⏟ = : ζ j , x , R j ( y ) . \Delta_{R_j}^{(j)}\widetilde{v}_j = \int_{\Gamma_{\mathcal{M}}^{(0)}} \mathrm{d}y\,\underbrace{\frac{1}{|\Gamma_j|}\int_{0}^{1}\mathrm{d}t\,\zeta_j(x+tR_j-y)\nabla_{R_j}v(y)}_{=:\zeta_{j,x,R_j}(y)}. Δ R j ( j ) v j = ∫ Γ M ( 0 ) d y =: ζ j , x , R j ( y ) ∣ Γ j ∣ 1 ∫ 0 1 d t ζ j ( x + t R j − y ) ∇ R j v ( y ) .
we define, as in [33 ] , the atomistic stress density tensor
S j ( m o n o ) ( u ; y ) : = 1 ∣ Γ j ∣ ∫ Γ M ( 0 ) d x ∑ R j ∈ R j V j ; R j ( m o n o ) ( Δ j [ u ] ( x ) ) ⊗ R j ζ j , x , R j ( y ) , \mathcal{S}_j^{(\mathrm{mono})}(u;y) := \frac{1}{|\Gamma_j|}\int_{\Gamma_{\mathcal{M}}^{(0)}}\mathrm{d}x\sum_{R_j\in\mathcal{R}_j}V_{j;R_j}^{(\mathrm{mono})}\left(\Delta_j[u](x)\right)\otimes R_j\,\zeta_{j,x,R_j}(y), S j ( mono ) ( u ; y ) := ∣ Γ j ∣ 1 ∫ Γ M ( 0 ) d x R j ∈ R j ∑ V j ; R j ( mono ) ( Δ j [ u ] ( x ) ) ⊗ R j ζ j , x , R j ( y ) ,
which allows us to compute that
δ e j ( m o n o ) ( u j ( C B ) ) [ ζ j ∗ v j ] = 1 ∣ Γ j ∣ ∫ Γ M ( 0 ) d x ∑ R j ∈ R j V j ; R j ( m o n o ) ( Δ j [ u j ( C B ) ] ( x ) ) ⋅ Δ R j ( j ) ( ζ j ∗ v j ) ( x ) = ∫ Γ M ( 0 ) d y S j ( m o n o ) ( u j ( C B ) ; y ) : ∇ v j ( y ) . \begin{aligned}
\delta e_j^{(\mathrm{mono})}(u_j^{(\mathrm{CB})})[\zeta_j*v_j] = \frac{1}{|\Gamma_j|}\int_{\Gamma_{\mathcal{M}}^{(0)}}\mathrm{d}x\sum_{R_j\in\mathcal{R}_j}V_{j;R_j}^{(\mathrm{mono})}\left(\Delta_j[u_j^{(\mathrm{CB})}](x)\right)\cdot\Delta_{R_j}^{(j)}(\zeta_j*v_j)(x) \\
&= \int_{\Gamma_{\mathcal{M}}^{(0)}}\mathrm{d}y\,\mathcal{S}_j^{(\mathrm{mono})}(u_j^{(\mathrm{CB})};y):\nabla v_j(y).
\end{aligned} δ e j ( mono ) ( u j ( CB ) ) [ ζ j ∗ v j ] = ∣ Γ j ∣ 1 ∫ Γ M ( 0 ) d x R j ∈ R j ∑ V j ; R j ( mono ) ( Δ j [ u j ( CB ) ] ( x ) ) ⋅ Δ R j ( j ) ( ζ j ∗ v j ) ( x ) = ∫ Γ M ( 0 ) d y S j ( mono ) ( u j ( CB ) ; y ) : ∇ v j ( y ) .
Similarly, abbreviating the first Piola–Kirchhoff stress density tensor
S j ( P K ) ( u ; y ) : = 1 ∣ Γ j ∣ ∑ R j ∈ R j V j ; R j ( m o n o ) ( ( ∇ u ( x ) ⋅ R j ) R j ∈ R j ) ⊗ R j , \mathcal{S}_j^{(\mathrm{PK})}(u;y) := \frac{1}{|\Gamma_j|}\sum_{R_j\in\mathcal{R}_j}V_{j;R_j}^{(\mathrm{mono})}\left((\nabla u(x)\cdot R_j)_{R_j\in\mathcal{R}_j}\right)\otimes R_j, S j ( PK ) ( u ; y ) := ∣ Γ j ∣ 1 R j ∈ R j ∑ V j ; R j ( mono ) ( ( ∇ u ( x ) ⋅ R j ) R j ∈ R j ) ⊗ R j ,
we can rewrite
δ e j ( C B ) ( u j ( C B ) ) [ v j ] = ∫ Γ M d y S j ( P K ) ( u j ( C B ) ; y ) : ∇ v j ( y ) . \delta e_j^{(\mathrm{CB})}(u_j^{(\mathrm{CB})})[v_j] = \int_{\Gamma_{\mathcal{M}}}\mathrm{d}y\,\mathcal{S}_j^{(\mathrm{PK})}(u_j^{(\mathrm{CB})};y):\nabla v_j(y). δ e j ( CB ) ( u j ( CB ) ) [ v j ] = ∫ Γ M d y S j ( PK ) ( u j ( CB ) ; y ) : ∇ v j ( y ) .
Consequently, (25) implies the estimate
∣ T ( 0 ) [ v ~ ] ∣ ≤ ∑ j = 1 2 ∥ S j ( m o n o ) ( u j ( C B ) ) − S j ( P K ) ( u j ( C B ) ) ∥ L 2 ( Γ M ( 0 ) ) ∥ ∇ v j ∥ L 2 ( Γ M ( 0 ) ) . |T(0)[\widetilde{\mathbf{v}}]| \le\sum_{j=1}^{2} \left\| \mathcal{S}_{j}^{(\mathrm{mono})}\left(u_{j}^{(\mathrm{CB})}\right)-\mathcal{S}_{j}^{(\mathrm{PK})}\left(u_{j}^{(\mathrm{CB})}\right)\right\|_{L^{2}(\Gamma_{\mathcal{M}}^{(0)})}\|\nabla v_{j}\|_{L^{2}(\Gamma_{\mathcal{M}}^{(0)})}. ∣ T ( 0 ) [ v ] ∣ ≤ j = 1 ∑ 2 S j ( mono ) ( u j ( CB ) ) − S j ( PK ) ( u j ( CB ) ) L 2 ( Γ M ( 0 ) ) ∥∇ v j ∥ L 2 ( Γ M ( 0 ) ) .
An analogous computation to [33 Proposition 3] now implies that, given v ∈ W ‾ p e r \mathbf{v}\in\overline{W}_{\mathrm{per}} v ∈ W per with ∥ ∇ v ∥ L 2 ( Γ M ( 0 ) ) = 1 \|\nabla\mathbf{v}\|_{L^{2}(\Gamma_{\mathcal{M}}^{(0)})}=1 ∥∇ v ∥ L 2 ( Γ M ( 0 ) ) = 1 ,
∣ T ( 0 ) [ v ~ ] ∣ ≲ ( M 2 , 2 ( m o n o , a t ) ∥ ∇ 3 u ( C B ) ∥ L 2 ( Γ M ( 0 ) ) + M 3 , 2 ( m o n o , a t ) ∥ ∇ 2 u ( C B ) ∥ L 4 ( Γ M ( 0 ) ) 2 ) ≲ a 0 ( M 2 , 2 ( m o n o , a t ) ∥ ∇ 3 U ( C B ) ∥ L 2 ( Γ M ) + M 3 , 2 ( m o n o , a t ) ∥ ∇ 2 U ( C B ) ∥ L 4 ( Γ M ) 2 ) , \begin{aligned}
|T(0)[\widetilde{\mathbf{v}}]| &\lesssim\left(M_{2,2}^{(\mathrm{mono,at})}\|\nabla^{3}\mathbf{u}^{(\mathrm{CB})}\|_{L^{2}(\Gamma_{\mathcal{M}}^{(0)})}+M_{3,2}^{(\mathrm{mono,at})}\|\nabla^{2}\mathbf{u}^{(\mathrm{CB})}\|_{L^{4}(\Gamma_{\mathcal{M}}^{(0)})}^{2}\right)\\
&\lesssim a_{0}\left(M_{2,2}^{(\mathrm{mono,at})}\|\nabla^{3}\mathbf{U}^{(\mathrm{CB})}\|_{L^{2}(\Gamma_{\mathcal{M}})}+M_{3,2}^{(\mathrm{mono,at})}\|\nabla^{2}\mathbf{U}^{(\mathrm{CB})}\|_{L^{4}(\Gamma_{\mathcal{M}})}^{2}\right),
\end{aligned} ∣ T ( 0 ) [ v ] ∣ ≲ ( M 2 , 2 ( mono , at ) ∥ ∇ 3 u ( CB ) ∥ L 2 ( Γ M ( 0 ) ) + M 3 , 2 ( mono , at ) ∥ ∇ 2 u ( CB ) ∥ L 4 ( Γ M ( 0 ) ) 2 ) ≲ a 0 ( M 2 , 2 ( mono , at ) ∥ ∇ 3 U ( CB ) ∥ L 2 ( Γ M ) + M 3 , 2 ( mono , at ) ∥ ∇ 2 U ( CB ) ∥ L 4 ( Γ M ) 2 ) ,
where we used the definition of u ( C B ) \mathbf{u}^{(\mathrm{CB})} u ( CB ) and the scaling properties of the involved norms. Recalling the a priori bound in Theorem 4.4 and employing the fact that ∥ ∇ v ~ j ∥ 2 ∼ ∥ ∇ v j ∥ 2 \|\nabla\widetilde{v}_{j}\|_{2}\sim\|\nabla v_{j}\|_{2} ∥∇ v j ∥ 2 ∼ ∥∇ v j ∥ 2 , we thus obtain the upper bound
∥ T ( 0 ) ∥ Y ≲ a 0 ∑ ℓ = 2 3 M ℓ , 2 ( m o n o , a t ) max j = 1 , 2 ∥ D j → 3 − j ∥ 2 M 2 ( i n t e r ) ( 1 + max j = 1 , 2 ∥ D j → 3 − j ∥ 2 2 ν S t a c k ∑ ℓ = 1 3 M ℓ ( i n t e r ) ) . \|T(0)\|_{Y}\lesssim a_{0}\sum_{\ell=2}^{3}M_{\ell,2}^{(\mathrm{mono,at})}\max_{j=1,2}\|D_{j\to3-j}\|_{2}M_{2}^{(\mathrm{inter})}\left(1+\frac{\max_{j=1,2}\|D_{j\to3-j}\|_{2}^{2}}{\nu_{\mathrm{Stack}}}\sum_{\ell=1}^{3}M_{\ell}^{(\mathrm{inter})}\right). ∥ T ( 0 ) ∥ Y ≲ a 0 ℓ = 2 ∑ 3 M ℓ , 2 ( mono , at ) j = 1 , 2 max ∥ D j → 3 − j ∥ 2 M 2 ( inter ) ( 1 + ν Stack max j = 1 , 2 ∥ D j → 3 − j ∥ 2 2 ℓ = 1 ∑ 3 M ℓ ( inter ) ) .
In particular, there exists a universal constant δ ( m o n o ) > 0 \delta^{(\mathrm{mono})}>0 δ ( mono ) > 0 such that if we thus choose a 0 > 0 a_{0}>0 a 0 > 0 such that
a 0 M 2 ( i n t e r ) max j = 1 , 2 ∥ D j → 3 − j ∥ 2 ν S t a c k 2 ( 1 + max j = 1 , 2 ∥ D j → 3 − j ∥ 2 2 ν S t a c k ∑ ℓ = 1 3 M ℓ ( i n t e r ) ) ∑ ℓ = 2 3 M ℓ , 2 ( m o n o , a t ) ( M 3 ( m o n o ) + max { ∣ Γ 1 ∣ ∣ Γ 2 ∣ , ∣ Γ 2 ∣ ∣ Γ 1 ∣ } M 3 ( i n t e r ) ) ≤ δ ( m o n o ) , a_{0}M_{2}^{(\mathrm{inter})}\frac{\max_{j=1,2}\|D_{j\to3-j}\|_{2}}{\nu_{\mathrm{Stack}}^{2}}\left(1+\frac{\max_{j=1,2}\|D_{j\to3-j}\|_{2}^{2}}{\nu_{\mathrm{Stack}}}\sum_{\ell=1}^{3}M_{\ell}^{(\mathrm{inter})}\right)
\sum_{\ell=2}^{3}M_{\ell,2}^{(\mathrm{mono,at})}\left(M_{3}^{(\mathrm{mono})}+\max\left\{\frac{|\Gamma_{1}|}{|\Gamma_{2}|},\frac{|\Gamma_{2}|}{|\Gamma_{1}|}\right\}M_{3}^{(\mathrm{inter})}\right)\le\delta^{(\mathrm{mono})}, a 0 M 2 ( inter ) ν Stack 2 max j = 1 , 2 ∥ D j → 3 − j ∥ 2 ( 1 + ν Stack max j = 1 , 2 ∥ D j → 3 − j ∥ 2 2 ℓ = 1 ∑ 3 M ℓ ( inter ) ) ℓ = 2 ∑ 3 M ℓ , 2 ( mono , at ) ( M 3 ( mono ) + max { ∣ Γ 2 ∣ ∣ Γ 1 ∣ , ∣ Γ 1 ∣ ∣ Γ 2 ∣ } M 3 ( inter ) ) ≤ δ ( mono ) ,
Lemma 4.3 implies that there exists a locally unique solution v \mathbf{v} v of T [ v ] = 0 T[\mathbf{v}]=0 T [ v ] = 0 . We define u ( e q ) : = u ( C B ) + v \mathbf{u}^{(\mathrm{eq})}:=\mathbf{u}^{(\mathrm{CB})}+\mathbf{v} u ( eq ) := u ( CB ) + v . Lemma 4.3 yields
∥ ∇ u ( e q ) − ∇ u ( C B ) ∥ 2 = ∥ v ∥ 2 ≲ a 0 M 2 ( i n t e r ) max j = 1 , 2 ∥ D j → 3 − j ∥ 2 ν S t a c k ∑ ℓ = 2 3 M ℓ , 2 ( m o n o , a t ) ( 1 + max j = 1 , 2 ∥ D j → 3 − j ∥ 2 2 ν S t a c k ∑ ℓ = 1 3 M ℓ ( i n t e r ) ) . \|\nabla\mathbf{u}^{(\mathrm{eq})}-\nabla\mathbf{u}^{(\mathrm{CB})}\|_{2}=\|\mathbf{v}\|_{2}\lesssim a_{0}M_{2}^{(\mathrm{inter})}\frac{\max_{j=1,2}\|D_{j\to3-j}\|_{2}}{\nu_{\mathrm{Stack}}}\sum_{\ell=2}^{3}M_{\ell,2}^{(\mathrm{mono,at})}\left(1+\frac{\max_{j=1,2}\|D_{j\to3-j}\|_{2}^{2}}{\nu_{\mathrm{Stack}}}\sum_{\ell=1}^{3}M_{\ell}^{(\mathrm{inter})}\right). ∥∇ u ( eq ) − ∇ u ( CB ) ∥ 2 = ∥ v ∥ 2 ≲ a 0 M 2 ( inter ) ν Stack max j = 1 , 2 ∥ D j → 3 − j ∥ 2 ℓ = 2 ∑ 3 M ℓ , 2 ( mono , at ) ( 1 + ν Stack max j = 1 , 2 ∥ D j → 3 − j ∥ 2 2 ℓ = 1 ∑ 3 M ℓ ( inter ) ) .
This finishes the proof.
Phonon stability
As we have previously seen, a critical point for the stacking-based energy density e ( t o t ) e^{(\mathrm{tot})} e ( tot ) yields a critical point for the offset energy Δ E ( t o t ) \Delta E^{(\mathrm{tot})} Δ E ( tot ) . A natural question that we then need to answer is whether stability is also inherited. Recall that an important property of e ( t o t ) e^{(\mathrm{tot})} e ( tot ) is that it is stable in a neighborhood of zero displacement u = 0 \mathbf{u}=\mathbf{0} u = 0 , which allowed us to obtain a locally unique minimizer.
As we shall see, stability is not necessarily inherited in the following sense: First, we will see that δ 2 Δ E 0 ( t o t ) ( 0 ) < 0 \delta^{2}\Delta E_{\mathbf{0}}^{(\mathrm{tot})}(0)<0 δ 2 Δ E 0 ( tot ) ( 0 ) < 0 despite δ 2 e ( t o t ) ( 0 ) > 0 \delta^{2}e^{(\mathrm{tot})}(0)>0 δ 2 e ( tot ) ( 0 ) > 0 . After developing a stability criterion, we then show that for purely twisted bilayer systems, at twist angle tending to 0, the minimizer u θ → 0 ( e q ) \mathbf{u}_{\theta\to0}^{(\mathrm{eq})} u θ → 0 ( eq ) does indeed satisfy δ 2 Δ E u θ → 0 ( e q ) ( t o t ) ( 0 ) > 0 \delta^{2}\Delta E_{\mathbf{u}_{\theta\to0}^{(\mathrm{eq})}}^{(\mathrm{tot})}(0)>0 δ 2 Δ E u θ → 0 ( eq ) ( tot ) ( 0 ) > 0 .
For the next assumption, recall the interpolants (5) –(6) of functions defined on the lattices R j \mathcal{R}_{j} R j .
Assumption 5 (Phonon stability of monolayer). We assume that
ν P h o n o n : = inf ∥ ∇ v ∥ 2 = 1 δ 2 Δ E j ( m o n o ) [ 0 ] [ v , v ] > 0. \nu_{\mathrm{Phonon}} := \inf_{\lVert\nabla v\rVert_{2}=1}\delta^{2}\Delta E_{j}^{(\mathrm{mono})}[0][v,v] > 0. ν Phonon := ∥ ∇ v ∥ 2 = 1 inf δ 2 Δ E j ( mono ) [ 0 ] [ v , v ] > 0. In particular, we have that
δ 2 Δ E j ; δ u j ( m o n o ) ( δ u j ) ≥ ν ∥ ∇ δ u j ∥ 2 2 . \delta^{2}\Delta E_{j;\delta u_{j}}^{(\mathrm{mono})}(\delta u_{j}) \ge\nu\lVert\nabla\delta u_{j}\rVert_{2}^{2}. δ 2 Δ E j ; δ u j ( mono ) ( δ u j ) ≥ ν ∥ ∇ δ u j ∥ 2 2 .
For the interlayer potential, we consider the simplified problem for interatomic pair potentials v v v , and with only one sublattice degree of freedom. In this case, we have
δ 2 Δ E j ; u ( e q ) ( i n t e r ) [ 0 ] [ δ u , δ v ] = 1 2 ∑ R 1 , R 2 ( D 2 v ) ( R 1 + u 1 ( e q ) ( R 1 ) − R 2 − u 2 ( e q ) ( R 2 ) ) [ δ u 1 ( R 1 ) − δ u 2 ( R 2 ) , δ v 1 ( R 1 ) − δ v 2 ( R 2 ) ] . \delta^{2}\Delta E_{j;\mathbf{u}^{(\mathrm{eq})}}^{(\mathrm{inter})}[0][\delta\mathbf{u},\delta\mathbf{v}]
=
\frac{1}{2}\sum_{R_{1},R_{2}}
(D^{2}v)\left(R_{1}+u_{1}^{(\mathrm{eq})}(R_{1})-R_{2}-u_{2}^{(\mathrm{eq})}(R_{2})\right)
\left[\delta u_{1}(R_{1})-\delta u_{2}(R_{2}),\delta v_{1}(R_{1})-\delta v_{2}(R_{2})\right]. δ 2 Δ E j ; u ( eq ) ( inter ) [ 0 ] [ δ u , δ v ] = 2 1 R 1 , R 2 ∑ ( D 2 v ) ( R 1 + u 1 ( eq ) ( R 1 ) − R 2 − u 2 ( eq ) ( R 2 ) ) [ δ u 1 ( R 1 ) − δ u 2 ( R 2 ) , δ v 1 ( R 1 ) − δ v 2 ( R 2 ) ] .
Continuum approximation
We now want to study the continuum approximation, in order to find an appropriate test function to approximate the equilibrium energy.
We start by excluding the relaxation and fixing the the equilibrium distance d d d . We focus on the interlayer energy in the case of pair potentials
∑ R 1 , R 2 ( v 1 ( R 1 ) − v 2 ( R 2 ) ) T D 2 v ( R 1 − R 2 , d ) ( v 1 ( R 1 ) − v 2 ( R 2 ) ) . \sum_{R_{1},R_{2}}\left(v_{1}(R_{1})-v_{2}(R_{2})\right)^{T}D^{2}v(R_{1}-R_{2},d)\left(v_{1}(R_{1})-v_{2}(R_{2})\right). R 1 , R 2 ∑ ( v 1 ( R 1 ) − v 2 ( R 2 ) ) T D 2 v ( R 1 − R 2 , d ) ( v 1 ( R 1 ) − v 2 ( R 2 ) ) .
Observe that we can rewrite
∑ R 1 , R 2 h ( R 1 , R 2 ) = ∫ d x d y ∑ R 1 δ ( x − R 1 ) ∑ R 2 δ ( y − R 2 ) h ( x , y ) . \sum_{R_{1},R_{2}}h(R_{1},R_{2})
=
\int\mathrm{d}x\mathrm{d}y
\sum_{R_{1}}\delta(x-R_{1})
\sum_{R_{2}}\delta(y-R_{2})h(x,y). R 1 , R 2 ∑ h ( R 1 , R 2 ) = ∫ d x d y R 1 ∑ δ ( x − R 1 ) R 2 ∑ δ ( y − R 2 ) h ( x , y ) .
Notice that we can approximate δ ( x − R j ) \delta(x-R_j) δ ( x − R j ) by ζ j ( x − R j ) \zeta_j(x-R_j) ζ j ( x − R j ) . Abbreviating the approximate lattice restrictions
χ R j ( x ) : = ∑ R j ∈ R j ζ j ( x − R j ) , \chi_{\mathcal{R}_j}(x) := \sum_{R_j\in\mathcal{R}_j} \zeta_j(x-R_j), χ R j ( x ) := R j ∈ R j ∑ ζ j ( x − R j ) ,
and the approximate interaction potential
A ( x , y ) : = χ R 1 ( x ) χ R 2 ( y ) ( D 2 v ) ( x − y , d ) A(x,y) := \chi_{\mathcal{R}_1}(x)\chi_{\mathcal{R}_2}(y)(D^2v)(x-y,d) A ( x , y ) := χ R 1 ( x ) χ R 2 ( y ) ( D 2 v ) ( x − y , d )
we thus approximate δ 2 Δ E j ; u ( e q ) ( i n t e r ) [ 0 ] [ δ u , δ u ] \delta^2\Delta E^{(\mathrm{inter})}_{j;u(\mathrm{eq})}[0][\delta u,\delta u] δ 2 Δ E j ; u ( eq ) ( inter ) [ 0 ] [ δ u , δ u ] by
∫ d x d y ( δ u 1 ( x ) − δ u 2 ( y ) ) T A ( x , y ) ( δ u 1 ( x ) − δ u 2 ( y ) ) . \int\mathrm{d}x\mathrm{d}y\,(\delta u_1(x)-\delta u_2(y))^T A(x,y)(\delta u_1(x)-\delta u_2(y)). ∫ d x d y ( δ u 1 ( x ) − δ u 2 ( y ) ) T A ( x , y ) ( δ u 1 ( x ) − δ u 2 ( y )) .
For simplicity, we even consider
A ( x , y ) ← D 2 v ( x − y , d ) . A(x,y) \leftarrow D^2v(x-y,d). A ( x , y ) ← D 2 v ( x − y , d ) .
Formally, we can justify this simplification by rescaling the interlayer energy by the unit cell volume. In particular, we rescale and take the continuum limit
χ R j ( x ) ⟶ a 0 2 ∑ R j ∈ a 0 R j ζ j ( x − R j ) → a 0 → 0 ∫ d y ζ j ( x − y ) = 1 , \chi_{\mathcal{R}_j}(x) \longrightarrow a_0^2\sum_{R_j\in a_0\mathcal{R}_j}\zeta_j(x-R_j)\xrightarrow[a_0\to0]{}\int\mathrm{d}y\,\zeta_j(x-y)=1, χ R j ( x ) ⟶ a 0 2 R j ∈ a 0 R j ∑ ζ j ( x − R j ) a 0 → 0 ∫ d y ζ j ( x − y ) = 1 ,
by the normalization condition on ζ j \zeta_j ζ j .
We expand the interlayer energy according to
∫ d x d y ( v 1 ( x ) − v 2 ( y ) ) T D 2 v ( x − y , d ) ( v 1 ( x ) − v 2 ( y ) ) = (26) \begin{aligned}
\int\mathrm{d}x\mathrm{d}y\,(v_1(x)-v_2(y))^T D^2v(x-y,d)(v_1(x)-v_2(y)) &= \tag*{(26)}
\end{aligned} ∫ d x d y ( v 1 ( x ) − v 2 ( y ) ) T D 2 v ( x − y , d ) ( v 1 ( x ) − v 2 ( y )) = (26)
∫ d x d y ( v 1 ( x ) − v 1 ( y ) ) T D 2 v ( x − y , d ) ( v 1 ( x ) − v 1 ( y ) ) (27) \int\mathrm{d}x\mathrm{d}y\,(v_1(x)-v_1(y))^T D^2v(x-y,d)(v_1(x)-v_1(y)) \tag*{(27)} ∫ d x d y ( v 1 ( x ) − v 1 ( y ) ) T D 2 v ( x − y , d ) ( v 1 ( x ) − v 1 ( y )) (27)
+ ∫ d x d y ( v 1 ( y ) − v 2 ( y ) ) T D 2 v ( x − y , d ) ( v 1 ( y ) − v 2 ( y ) ) (28) +\int\mathrm{d}x\mathrm{d}y\,(v_1(y)-v_2(y))^T D^2v(x-y,d)(v_1(y)-v_2(y)) \tag*{(28)} + ∫ d x d y ( v 1 ( y ) − v 2 ( y ) ) T D 2 v ( x − y , d ) ( v 1 ( y ) − v 2 ( y )) (28)
+ 2 ∫ d x d y ( v 1 ( x ) − v 1 ( y ) ) T D 2 v ( x − y , d ) ( v 1 ( y ) − v 2 ( y ) ) . (29) +2\int\mathrm{d}x\mathrm{d}y\,(v_1(x)-v_1(y))^T D^2v(x-y,d)(v_1(y)-v_2(y)). \tag*{(29)} + 2 ∫ d x d y ( v 1 ( x ) − v 1 ( y ) ) T D 2 v ( x − y , d ) ( v 1 ( y ) − v 2 ( y )) . (29)
This allows us to make the following observations:
(27) corresponds to a monolayer contribution in layer 1. In particular, if v v v satisfies some relative compactness property w.r.t. to the monolayer energy, we can treat this contribution perturbatively.
For (28) , we notice that v 1 v_1 v 1 and v 2 v_2 v 2 are evaluated at the same position , which allows us to average over D 2 v D^2v D 2 v . Negative eigenvalues of this average could lead to negative total energies, dependent on the eigenfunctions.
(29) can be controlled via Cauchy-Schwarz’ inequality by distributing it as corrections to the two previous contributions coming from (27) and (28) .
As an interlayer potential, we fix a potential that is radial in the horizontal direction, i.e.,
v ( x , z ) = w ( ∣ x ∣ , z ) , v(x,z)=w(|x|,z), v ( x , z ) = w ( ∣ x ∣ , z ) ,
where x ∈ R 2 x\in\mathbb{R}^{2} x ∈ R 2 . A straight-forward calculation yields
D 2 v ( x , z ) = ( ∂ ρ w ( ρ , z ) ρ Q x + ∂ z 2 w ( ρ , z ) P x ∂ z 2 w ( ρ , z ) x ρ ∂ z 2 w ( ρ , z ) x T ρ ∂ z 2 w ( ρ , z ) ) , D^{2}v(x,z)=\begin{pmatrix}
\frac{\partial_{\rho}w(\rho,z)}{\rho}Q_x+\partial_{z}^{2}w(\rho,z)P_x & \partial_{z}^{2}w(\rho,z)\frac{x}{\rho}\\
\partial_{z}^{2}w(\rho,z)\frac{x^{\mathsf{T}}}{\rho} & \partial_{z}^{2}w(\rho,z)
\end{pmatrix}, D 2 v ( x , z ) = ( ρ ∂ ρ w ( ρ , z ) Q x + ∂ z 2 w ( ρ , z ) P x ∂ z 2 w ( ρ , z ) ρ x T ∂ z 2 w ( ρ , z ) ρ x ∂ z 2 w ( ρ , z ) ) ,
where P x ( y T , z ) T = ⟨ x , y ⟩ ∣ x ∣ P_x(y^{\mathsf{T}},z)^{\mathsf{T}}=\frac{\langle x,y\rangle}{|x|} P x ( y T , z ) T = ∣ x ∣ ⟨ x , y ⟩ and Q x = I − P x Q_x=I-P_x Q x = I − P x . Another straight-forward computation implies that
∫ R 2 d x D 2 v ( x , z ) = π ∫ 0 ∞ d ρ ( ( ρ ∂ ρ 2 w ( ρ , z ) + ∂ ρ w ( ρ , z ) ) I 2 0 0 2 ρ ∂ z 2 w ( ρ , z ) ) = π ( ρ ∂ ρ w ( ρ , z ) ∣ ρ = 0 ρ = ∞ 0 0 2 ∫ 0 ∞ d ρ ρ ∂ z 2 w ( ρ , z ) ) = : diag ( 0 , 0 , m ( z ) ) . (30) \begin{aligned}
\int_{\mathbb{R}^{2}}\mathrm{d}x\,D^{2}v(x,z)
=
\pi\int_{0}^{\infty}\mathrm{d}\rho
\begin{pmatrix}
(\rho\partial_{\rho}^{2}w(\rho,z)+\partial_{\rho}w(\rho,z))I_{2} & 0\\
0 & 2\rho\partial_{z}^{2}w(\rho,z)
\end{pmatrix} \\
&=
\pi\begin{pmatrix}
\left.\rho\partial_{\rho}w(\rho,z)\right|_{\rho=0}^{\rho=\infty} & 0\\
0 & 2\int_{0}^{\infty}\mathrm{d}\rho\,\rho\partial_{z}^{2}w(\rho,z)
\end{pmatrix} \\
&=: \operatorname{diag}(0,0,m(z)). \tag*{(30)}
\end{aligned} ∫ R 2 d x D 2 v ( x , z ) = π ∫ 0 ∞ d ρ ( ( ρ ∂ ρ 2 w ( ρ , z ) + ∂ ρ w ( ρ , z )) I 2 0 0 2 ρ ∂ z 2 w ( ρ , z ) ) = π ( ρ ∂ ρ w ( ρ , z ) ∣ ρ = 0 ρ = ∞ 0 0 2 ∫ 0 ∞ d ρ ρ ∂ z 2 w ( ρ , z ) ) =: diag ( 0 , 0 , m ( z )) . (30)
where, in the second line, we integrated by parts and used the fact that w ( ρ , ⋅ ) ≲ 1 ρ 2 w(\rho,\cdot)\lesssim\frac{1}{\rho^{2}} w ( ρ , ⋅ ) ≲ ρ 2 1 for ρ → ∞ \rho\to\infty ρ → ∞ and ∥ ∂ ρ w ( ⋅ , z ) ∥ ∞ < ∞ \|\partial_{\rho}w(\cdot,z)\|_{\infty}<\infty ∥ ∂ ρ w ( ⋅ , z ) ∥ ∞ < ∞ for ∣ z ∣ ≠ 0 |z|\ne0 ∣ z ∣ = 0 . We now consider two special cases.
w ( ρ , z ) = f 1 ( ρ ) f 2 ( z ) w(\rho,z)=f_{1}(\rho)f_{2}(z) w ( ρ , z ) = f 1 ( ρ ) f 2 ( z ) : Then
m ( z ) = 2 π ( ∫ 0 ∞ d ρ ρ f 1 ( ρ ) ) f 2 ′ ′ ( z ) . m(z)=2\pi\left(\int_{0}^{\infty}\mathrm{d}\rho\,\rho f_{1}(\rho)\right)f_{2}^{\prime\prime}(z). m ( z ) = 2 π ( ∫ 0 ∞ d ρ ρ f 1 ( ρ ) ) f 2 ′′ ( z ) .
w ( ρ , z ) = g ( r ) w(\rho,z)=g(r) w ( ρ , z ) = g ( r ) , where r = ρ 2 + z 2 r=\sqrt{\rho^{2}+z^{2}} r = ρ 2 + z 2 : A straight-forward calculation yields
∂ z 2 w ( ρ , z ) = ρ 2 r 3 g ′ ( r ) + ( z r ) 2 g ′ ′ ( r ) = g ′ ( r ) r + g ′ ′ ( r ) − ∂ ρ 2 w ( ρ , z ) , \partial_{z}^{2}w(\rho,z)=\frac{\rho^{2}}{r^{3}}g^{\prime}(r)+\left(\frac{z}{r}\right)^{2}g^{\prime\prime}(r)=\frac{g^{\prime}(r)}{r}+g^{\prime\prime}(r)-\partial_{\rho}^{2}w(\rho,z), ∂ z 2 w ( ρ , z ) = r 3 ρ 2 g ′ ( r ) + ( r z ) 2 g ′′ ( r ) = r g ′ ( r ) + g ′′ ( r ) − ∂ ρ 2 w ( ρ , z ) ,
ρ r g ′ ( r ) = ∂ ρ w ( ρ , z ) , ρ g ′ ′ ( r ) = z 2 ρ ∂ ρ 2 w ( ρ , z ) − z 2 ρ 2 ∂ ρ w ( ρ , z ) . \frac{\rho}{r}g^{\prime}(r)=\partial_{\rho}w(\rho,z),\qquad
\rho g^{\prime\prime}(r)=\frac{z^{2}}{\rho}\partial_{\rho}^{2}w(\rho,z)-\frac{z^{2}}{\rho^{2}}\partial_{\rho}w(\rho,z). r ρ g ′ ( r ) = ∂ ρ w ( ρ , z ) , ρ g ′′ ( r ) = ρ z 2 ∂ ρ 2 w ( ρ , z ) − ρ 2 z 2 ∂ ρ w ( ρ , z ) .
Consequently, we obtain that
m ( z ) = 2 π ∫ 0 ∞ ( ∂ ρ w ( ρ , z ) + z 2 ρ ∂ ρ 2 w ( ρ , z ) − z 2 ρ 2 ∂ ρ w ( ρ , z ) − ρ ∂ ρ 2 w ( ρ , z ) ) = 2 π [ ( z 2 ρ ∂ ρ w ( ρ , z ) − ρ ∂ ρ w ( ρ , z ) ) ∣ ρ = 0 ρ = ∞ + 2 ∫ 0 ∞ d ρ ∂ ρ w ( ρ , z ) ] . \begin{aligned}
m(z)=2\pi\int_{0}^{\infty}\left(\partial_{\rho}w(\rho,z)+\frac{z^{2}}{\rho}\partial_{\rho}^{2}w(\rho,z)-\frac{z^{2}}{\rho^{2}}\partial_{\rho}w(\rho,z)-\rho\partial_{\rho}^{2}w(\rho,z)\right) \\
&=2\pi\left[\left(\frac{z^{2}}{\rho}\partial_{\rho}w(\rho,z)-\rho\partial_{\rho}w(\rho,z)\right)\bigg|_{\rho=0}^{\rho=\infty}+2\int_{0}^{\infty}\mathrm{d}\rho\,\partial_{\rho}w(\rho,z)\right].
\end{aligned} m ( z ) = 2 π ∫ 0 ∞ ( ∂ ρ w ( ρ , z ) + ρ z 2 ∂ ρ 2 w ( ρ , z ) − ρ 2 z 2 ∂ ρ w ( ρ , z ) − ρ ∂ ρ 2 w ( ρ , z ) ) = 2 π [ ( ρ z 2 ∂ ρ w ( ρ , z ) − ρ ∂ ρ w ( ρ , z ) ) ρ = 0 ρ = ∞ + 2 ∫ 0 ∞ d ρ ∂ ρ w ( ρ , z ) ] .
Employing ∂ ρ w ( ρ , z ) = ρ r g ′ ( r ) \partial_{\rho}w(\rho,z)=\frac{\rho}{r}g^{\prime}(r) ∂ ρ w ( ρ , z ) = r ρ g ′ ( r ) and the fundamental theorem of calculus, we thus obtain
m ( z ) = 2 π ( ∣ z ∣ g ′ ( ∣ z ∣ ) + 2 g ( ∣ z ∣ ) ) . m(z)=2\pi\left(|z|g^{\prime}(|z|)+2g(|z|)\right). m ( z ) = 2 π ( ∣ z ∣ g ′ ( ∣ z ∣ ) + 2 g ( ∣ z ∣ ) ) .
Example 5.1 (3D Lennard-Jones). We have that
g L J ( r ) = 4 ε 0 [ ( σ r ) 12 − ( σ r ) 6 ] g_{\mathrm{LJ}}(r)=4\varepsilon_0\left[\left(\frac{\sigma}{r}\right)^{12}-\left(\frac{\sigma}{r}\right)^6\right] g LJ ( r ) = 4 ε 0 [ ( r σ ) 12 − ( r σ ) 6 ] for ε 0 , σ > 0 \varepsilon_0,\sigma>0 ε 0 , σ > 0 . 2 1 6 σ ≈ 1.12 σ 2^{\frac{1}{6}}\sigma\approx1.12\sigma 2 6 1 σ ≈ 1.12 σ is the equilibrium distance. Then we obtain
m L J ( z ) = 8 π ε 0 [ 5 ( σ ∣ z ∣ ) 6 − 11 ( σ ∣ z ∣ ) 12 ] . m_{\mathrm{LJ}}(z)=8\pi\varepsilon_0\left[5\left(\frac{\sigma}{|z|}\right)^6-11\left(\frac{\sigma}{|z|}\right)^{12}\right]. m LJ ( z ) = 8 π ε 0 [ 5 ( ∣ z ∣ σ ) 6 − 11 ( ∣ z ∣ σ ) 12 ] . m L J m_{\mathrm{LJ}} m LJ changes its sign from negative to positive at ∣ z ∣ = ( 11 5 ) 1 6 σ ≈ 1.14 σ |z|=\left(\frac{11}{5}\right)^{\frac{1}{6}}\sigma\approx1.14\sigma ∣ z ∣ = ( 5 11 ) 6 1 σ ≈ 1.14 σ .
In the case of bilayer graphene, we have that the equilibrium distance is given by z = d B G = 3.35 z=d_{\mathrm{BG}}=3.35 z = d BG = 3.35 Å. The parameters are given by ε 0 = 2.39 \varepsilon_0=2.39 ε 0 = 2.39 meV, σ = 3.41 \sigma=3.41 σ = 3.41 Å, see [21 ] . We compute
m L J ( B G ) = − 483.5 m e V < 0. m_{\mathrm{LJ}}^{(\mathrm{BG})}=-483.5\ \mathrm{meV}<0. m LJ ( BG ) = − 483.5 meV < 0.
Example 5.2 (3D Morse potential). We have that
g M o r s e ( r ) = E 0 ( e − 2 κ ( r − r 0 ) − 2 e − κ ( r − r 0 ) ) g_{\mathrm{Morse}}(r)=E_0\left(e^{-2\kappa(r-r_0)}-2e^{-\kappa(r-r_0)}\right) g Morse ( r ) = E 0 ( e − 2 κ ( r − r 0 ) − 2 e − κ ( r − r 0 ) ) g M o r s e ( r ) = E 0 ( e − 2 κ ( r − r 0 ) − 2 e − κ ( r − r 0 ) ) g_{\mathrm{Morse}}(r)=E_0\left(e^{-2\kappa(r-r_0)}-2e^{-\kappa(r-r_0)}\right) g Morse ( r ) = E 0 ( e − 2 κ ( r − r 0 ) − 2 e − κ ( r − r 0 ) ) for some dissociation energy E 0 > 0 E_0>0 E 0 > 0 and decay rate κ > 0 \kappa>0 κ > 0 . Then we obtain
m M o r s e ( z ) = − 4 π E 0 ( ( κ ∣ z ∣ − 1 ) e − 2 κ ( ∣ z ∣ − r 0 ) − ( κ ∣ z ∣ − 2 ) e − κ ( ∣ z ∣ − r 0 ) ) . m_{\mathrm{Morse}}(z)=-4\pi E_0\left((\kappa|z|-1)e^{-2\kappa(|z|-r_0)}-(\kappa|z|-2)e^{-\kappa(|z|-r_0)}\right). m Morse ( z ) = − 4 π E 0 ( ( κ ∣ z ∣ − 1 ) e − 2 κ ( ∣ z ∣ − r 0 ) − ( κ ∣ z ∣ − 2 ) e − κ ( ∣ z ∣ − r 0 ) ) . m M o r s e m_{\mathrm{Morse}} m Morse changes its sign from negative to positive once in the interval [ 0 , min { 2 κ , r 0 } ] \left[0,\min\left\{\frac{2}{\kappa},r_0\right\}\right] [ 0 , min { κ 2 , r 0 } ] and from positive to negative once in [ max { 2 κ , r 0 } , ∞ ) \left[\max\left\{\frac{2}{\kappa},r_0\right\},\infty\right) [ max { κ 2 , r 0 } , ∞ ) , see Figure 1 . The transition points are given by the solutions of
e κ ( ∣ z ∣ − r 0 ) − 1 = 1 κ ∣ z ∣ − 2 . e^{\kappa(|z|-r_0)}-1=\frac{1}{\kappa|z|-2}. e κ ( ∣ z ∣ − r 0 ) − 1 = κ ∣ z ∣ − 2 1 . In case of bilayer graphene, we have that, see [30 ] , κ = 1.8168 \kappa=1.8168 κ = 1.8168 Å− 1 ^{-1} − 1 , E 0 = 2.8437 E_0=2.8437 E 0 = 2.8437 meV, r 0 = 3.6891 r_0=3.6891 r 0 = 3.6891 Å and again z = d B G = 3.35 z=d_{\mathrm{BG}}=3.35 z = d BG = 3.35 Å as above. Consequently, we obtain
m M o r s e ( B G ) = − 352.8 m e V < 0. m_{\mathrm{Morse}}^{(\mathrm{BG})}=-352.8\ \mathrm{meV}<0. m Morse ( BG ) = − 352.8 meV < 0.
Figure 1. Transition distances for the Morse potential are given by the intersection of the hyperbola with the respective exponential function
Example 5.3 (Combined potential). Let
v M + L J ( x , z ) = v M o r s e ( ∣ x ∣ ) v L J ( z ) , v_{\mathrm{M+LJ}}(x,z)=v_{\mathrm{Morse}}(|x|)v_{\mathrm{LJ}}(z), v M + LJ ( x , z ) = v Morse ( ∣ x ∣ ) v LJ ( z ) , where v M o r s e v_{\mathrm{Morse}} v Morse denotes a Morse potential with parameters ( E 0 , κ , r 0 ) (E_0,\kappa,r_0) ( E 0 , κ , r 0 ) , and v L J v_{\mathrm{LJ}} v LJ denotes a Lennard-Jones potential with parameters ( E 0 ′ , σ ) = ( 1 , σ ) (E'_0,\sigma)=(1,\sigma) ( E 0 ′ , σ ) = ( 1 , σ ) . Then we obtain
m M + L J ( z ) = 3 π E 0 κ 2 ∣ z ∣ 2 ( e 2 κ r 0 − 8 e κ r 0 ) [ − 7 ( σ ∣ z ∣ ) 6 + 26 ( σ ∣ z ∣ ) 12 ] . m_{\mathrm{M+LJ}}(z)=\frac{3\pi E_0}{\kappa^2|z|^2}\left(e^{2\kappa r_0}-8e^{\kappa r_0}\right)\left[-7\left(\frac{\sigma}{|z|}\right)^6+26\left(\frac{\sigma}{|z|}\right)^{12}\right]. m M + LJ ( z ) = κ 2 ∣ z ∣ 2 3 π E 0 ( e 2 κ r 0 − 8 e κ r 0 ) [ − 7 ( ∣ z ∣ σ ) 6 + 26 ( ∣ z ∣ σ ) 12 ] .
Now there is a transition parameter r 0 = log 8 κ − 1 r_0=\log8\kappa^{-1} r 0 = log 8 κ − 1 above which m C o m b m_{\mathrm{Comb}} m Comb transitions from positive to negative at ∣ z ∣ = ( 26 7 ) 1 6 σ ≈ 1.24 σ |z|=\left(\frac{26}{7}\right)^{\frac{1}{6}}\sigma\approx1.24\sigma ∣ z ∣ = ( 7 26 ) 6 1 σ ≈ 1.24 σ , and below which the converse transition occurs. With the above parametrization, we obtain, in the case of bilayer graphene,
m M + L J ( B G ) = 11.6 k e V > 0. m_{\mathrm{M+LJ}}^{(\mathrm{BG})}=11.6\ \mathrm{keV}>0. m M + LJ ( BG ) = 11.6 keV > 0.
In each of the examples, we determine regions where phonon instabilities are possible. In particular, we now assume that m ( d ) < 0 m(d)<0 m ( d ) < 0 .
Decomposing the interlayer energy according to (26) and employing (30) , we obtain
E ( a p p r o x ) [ v ] : = ∑ j = 1 2 ( δ 2 Δ E j ( m o n o ) ( v j ) + λ 2 ∫ d x d y D 2 v ( x − y , d ) [ ( v j ( x ) − v j ( y ) ) ⊗ 2 ] ) + λ 2 ( − 1 ) j ∫ d x d y ( v j ( x ) − v j ( y ) ) T D 2 2 v ( x − y , d ) ( v 2 ( y ) − v 1 ( y ) ) + λ m ( d ) ∫ d x ∣ ( v 1 ( x ) − v 2 ( x ) ) z ∣ 2 . \begin{aligned}
E^{(\mathrm{approx})}[\mathbf{v}] := {} & \sum_{j=1}^{2}\left(\delta^{2}\Delta E_{j}^{(\mathrm{mono})}(v_{j})+\frac{\lambda}{2}\int\mathrm{d}x\mathrm{d}y\ D^{2}v(x-y,d)\left[(v_{j}(x)-v_{j}(y))^{\otimes2}\right]\right) \\
&+\frac{\lambda}{2}(-1)^{j}\int\mathrm{d}x\mathrm{d}y\ (v_{j}(x)-v_{j}(y))^{T}D_{2}^{2}v(x-y,d)(v_{2}(y)-v_{1}(y)) \\
&+\lambda m(d)\int\mathrm{d}x\ \left|(v_{1}(x)-v_{2}(x))_{z}\right|^{2}.
\end{aligned} E ( approx ) [ v ] := j = 1 ∑ 2 ( δ 2 Δ E j ( mono ) ( v j ) + 2 λ ∫ d x d y D 2 v ( x − y , d ) [ ( v j ( x ) − v j ( y ) ) ⊗ 2 ] ) + 2 λ ( − 1 ) j ∫ d x d y ( v j ( x ) − v j ( y ) ) T D 2 2 v ( x − y , d ) ( v 2 ( y ) − v 1 ( y )) + λm ( d ) ∫ d x ∣ ( v 1 ( x ) − v 2 ( x ) ) z ∣ 2 .
In order to minimize the first line, we choose v j v_j v j in each layer aligned. With that choice, the second line also vanishes in the limit. In order to minimize the last line, we choose v 1 v_1 v 1 and v 2 v_2 v 2 anti-parallel. In particular, choosing
v R ( x ) : = ϕ R ( x ) ( 2 π ) 1 2 R ( e z , − e z ) , \mathbf{v}_{R}(x):=\frac{\phi_{R}(x)}{(2\pi)^{\frac{1}{2}}R}(e_z,-e_z), v R ( x ) := ( 2 π ) 2 1 R ϕ R ( x ) ( e z , − e z ) ,
where we choose ϕ R \phi_R ϕ R as an approximate normalized characteristic function of B R ( 0 ) B_R(0) B R ( 0 ) , i.e.
ϕ R ∣ B R ( 0 ) = 1 + O ( R − 1 ) , ϕ R ∣ B R + 1 ( 0 ) c = 0 , ∥ ∇ ϕ R ∥ ∞ ≲ 1 , \left.\phi_R\right|_{B_R(0)}=1+O(R^{-1}),\qquad\left.\phi_R\right|_{B_{R+1}(0)^c}=0,\qquad\|\nabla\phi_R\|_{\infty}\lesssim1, ϕ R ∣ B R ( 0 ) = 1 + O ( R − 1 ) , ϕ R ∣ B R + 1 ( 0 ) c = 0 , ∥∇ ϕ R ∥ ∞ ≲ 1 ,
we obtain
inf ∥ v j ∥ 2 = 1 E ( a p p r o x ) [ v ] ≤ lim sup R → ∞ E ( a p p r o x ) [ v R ] = 4 λ m ( d ) < 0 , \inf_{\lVert v_j\rVert_2=1} E^{(\mathrm{approx})}[v] \le\limsup_{R\to\infty} E^{(\mathrm{approx})}[v_R] = 4\lambda m(d) < 0, ∥ v j ∥ 2 = 1 inf E ( approx ) [ v ] ≤ R → ∞ lim sup E ( approx ) [ v R ] = 4 λm ( d ) < 0 ,
for we chose m ( d ) < 0 m(d)<0 m ( d ) < 0 .
Discrete model
Analogously to (26) , we decompose the interlayer energy as
∑ R 1 , R 2 ( v 1 ( R 1 ) − v 2 ( R 2 ) ) T D 2 v ( R 1 − R 2 ) ( v 1 ( R 1 ) − v 2 ( R 2 ) ) = ∑ R 1 , R 2 ( v 1 ( R 1 ) − v 1 ( ⌊ R 2 ⌋ 1 ) ) T D 2 v ( R 1 − R 2 ) ( v 1 ( R 1 ) − v 1 ( ⌊ R 2 ⌋ 1 ) ) + ( v 1 ( ⌊ R 2 ⌋ 1 ) − v 2 ( R 2 ) ) T D 2 v ( R 1 − R 2 ) ( v 1 ( ⌊ R 2 ⌋ 1 ) − v 2 ( R 2 ) ) + 2 ( v 1 ( R 1 ) − v 1 ( ⌊ R 2 ⌋ 1 ) ) T D 2 v ( R 1 − R 2 ) ( v 1 ( ⌊ R 2 ⌋ 1 ) − v 2 ( R 2 ) ) . (31) \begin{aligned}
\sum_{R_1,R_2} (v_1(R_1)-v_2(R_2))^T D^2v(R_1-R_2)(v_1(R_1)-v_2(R_2)) ={}&
\sum_{R_1,R_2} (v_1(R_1)-v_1(\lfloor R_2\rfloor_1))^T D^2v(R_1-R_2)(v_1(R_1)-v_1(\lfloor R_2\rfloor_1)) \\
&+ (v_1(\lfloor R_2\rfloor_1)-v_2(R_2))^T D^2v(R_1-R_2)(v_1(\lfloor R_2\rfloor_1)-v_2(R_2)) \\
&+ 2(v_1(R_1)-v_1(\lfloor R_2\rfloor_1))^T D^2v(R_1-R_2)(v_1(\lfloor R_2\rfloor_1)-v_2(R_2)). \tag*{(31)}
\end{aligned} R 1 , R 2 ∑ ( v 1 ( R 1 ) − v 2 ( R 2 ) ) T D 2 v ( R 1 − R 2 ) ( v 1 ( R 1 ) − v 2 ( R 2 )) = R 1 , R 2 ∑ ( v 1 ( R 1 ) − v 1 (⌊ R 2 ⌋ 1 ) ) T D 2 v ( R 1 − R 2 ) ( v 1 ( R 1 ) − v 1 (⌊ R 2 ⌋ 1 )) + ( v 1 (⌊ R 2 ⌋ 1 ) − v 2 ( R 2 ) ) T D 2 v ( R 1 − R 2 ) ( v 1 (⌊ R 2 ⌋ 1 ) − v 2 ( R 2 )) + 2 ( v 1 ( R 1 ) − v 1 (⌊ R 2 ⌋ 1 ) ) T D 2 v ( R 1 − R 2 ) ( v 1 (⌊ R 2 ⌋ 1 ) − v 2 ( R 2 )) . (31)
After finding a sequence of phonon modes with a limiting negative energy in the continuum approximation, we would like to study the functional in the true discrete case. Analogously to above, we now set
v 1 ( R 1 ) : = ϕ R ( R 1 ) ∣ Γ 1 ∣ 1 2 ( 2 π ) 1 2 R e z , v 2 ( R 2 ) : = − ϕ R ( R 2 ) ∣ Γ 2 ∣ 1 2 ( 2 π ) 1 2 R e z . v_1(R_1):=\frac{\phi_R(R_1)|\Gamma_1|^{\frac12}}{(2\pi)^{\frac12}R}e_z,\qquad v_2(R_2):=-\frac{\phi_R(R_2)|\Gamma_2|^{\frac12}}{(2\pi)^{\frac12}R}e_z. v 1 ( R 1 ) := ( 2 π ) 2 1 R ϕ R ( R 1 ) ∣ Γ 1 ∣ 2 1 e z , v 2 ( R 2 ) := − ( 2 π ) 2 1 R ϕ R ( R 2 ) ∣ Γ 2 ∣ 2 1 e z .
Plugging this ansatz into (31) , we obtain
∑ R 1 , R 2 ( v 1 ( R 1 ) − v 2 ( R 2 ) ) T D 2 v ( R 1 − R 2 ) ( v 1 ( R 1 ) − v 2 ( R 2 ) ) = ( ∣ Γ 1 ∣ 1 2 + ∣ Γ 2 ∣ 1 2 ) 2 2 π R 2 ∑ R 2 ∈ R 2 ∩ B R ( 0 ) R 1 ∈ R 1 ∂ z 2 v ( R 1 − R 2 ) + O ( R − 1 ) , \sum_{R_1,R_2}(v_1(R_1)-v_2(R_2))^T D^2v(R_1-R_2)(v_1(R_1)-v_2(R_2))
=
\frac{(|\Gamma_1|^{\frac12}+|\Gamma_2|^{\frac12})^2}{2\pi R^2}
\sum_{\substack{R_2\in\mathcal{R}_2\cap B_R(0)\\R_1\in\mathcal{R}_1}}
\partial_z^2v(R_1-R_2)+O(R^{-1}), R 1 , R 2 ∑ ( v 1 ( R 1 ) − v 2 ( R 2 ) ) T D 2 v ( R 1 − R 2 ) ( v 1 ( R 1 ) − v 2 ( R 2 )) = 2 π R 2 ( ∣ Γ 1 ∣ 2 1 + ∣ Γ 2 ∣ 2 1 ) 2 R 2 ∈ R 2 ∩ B R ( 0 ) R 1 ∈ R 1 ∑ ∂ z 2 v ( R 1 − R 2 ) + O ( R − 1 ) ,
where C ϕ C_\phi C ϕ is the normalization constant for ϕ \phi ϕ . Observe that we have that
∑ R 2 ∈ R 2 ∩ B R ( 0 ) R 1 ∈ R 1 D 2 v ( R 1 − R 2 ) = ∑ R 2 ∈ R 2 ∩ B R ( 0 ) R 1 ∈ R 1 D 2 v ( R 1 − { R 2 } 1 ) . \sum_{\substack{R_2\in\mathcal{R}_2\cap B_R(0)\\R_1\in\mathcal{R}_1}}D^2v(R_1-R_2)
=
\sum_{\substack{R_2\in\mathcal{R}_2\cap B_R(0)\\R_1\in\mathcal{R}_1}}D^2v(R_1-\{R_2\}_1). R 2 ∈ R 2 ∩ B R ( 0 ) R 1 ∈ R 1 ∑ D 2 v ( R 1 − R 2 ) = R 2 ∈ R 2 ∩ B R ( 0 ) R 1 ∈ R 1 ∑ D 2 v ( R 1 − { R 2 } 1 ) .
In particular, employing the Birkhoff ergodic theorem, we obtain
∑ R 1 , R 2 ( v 1 ( R 1 ) − v 2 ( R 2 ) ) T D 2 v ( R 1 − R 2 ) ( v 1 ( R 1 ) − v 2 ( R 2 ) ) = ( ∣ Γ 1 ∣ 1 2 + ∣ Γ 2 ∣ 1 2 ) 2 ∣ Γ 2 ∣ 2 π ∣ Γ 1 ∣ ∫ Γ 1 d x ∑ R 1 ∈ R 1 ∂ z 2 v ( R 1 − x ) + O ( R − 1 ) = ( ∣ Γ 1 ∣ 1 2 + ∣ Γ 2 ∣ 1 2 ) 2 2 π ∣ Γ 1 ∣ / ∣ Γ 2 ∣ ∫ R 2 d x ∂ z 2 v ( x ) + O ( R − 1 ) = ( ∣ Γ 1 ∣ 1 2 + ∣ Γ 2 ∣ 1 2 ) 2 2 π ∣ Γ 1 ∣ / ∣ Γ 2 ∣ m ( d ) + O ( R − 1 ) , \begin{aligned}
\sum_{R_1,R_2} \left(v_1(R_1)-v_2(R_2)\right)^T D^2v(R_1-R_2)\left(v_1(R_1)-v_2(R_2)\right)
&= \frac{\left(\lvert\Gamma_1\rvert^{\frac{1}{2}}+\lvert\Gamma_2\rvert^{\frac{1}{2}}\right)^2\lvert\Gamma_2\rvert}{2\pi\lvert\Gamma_1\rvert}\int_{\Gamma_1} \mathrm{d}x \sum_{R_1\in\mathcal{R}_1}\partial_z^2v(R_1-x)+O(R^{-1})\\
&= \frac{\left(\lvert\Gamma_1\rvert^{\frac{1}{2}}+\lvert\Gamma_2\rvert^{\frac{1}{2}}\right)^2}{2\pi\lvert\Gamma_1\rvert/\lvert\Gamma_2\rvert}\int_{\mathbb{R}^2}\mathrm{d}x\,\partial_z^2v(x)+O(R^{-1})\\
&= \frac{\left(\lvert\Gamma_1\rvert^{\frac{1}{2}}+\lvert\Gamma_2\rvert^{\frac{1}{2}}\right)^2}{2\pi\lvert\Gamma_1\rvert/\lvert\Gamma_2\rvert}m(d)+O(R^{-1}),
\end{aligned} R 1 , R 2 ∑ ( v 1 ( R 1 ) − v 2 ( R 2 ) ) T D 2 v ( R 1 − R 2 ) ( v 1 ( R 1 ) − v 2 ( R 2 ) ) = 2 π ∣ Γ 1 ∣ ( ∣ Γ 1 ∣ 2 1 + ∣ Γ 2 ∣ 2 1 ) 2 ∣ Γ 2 ∣ ∫ Γ 1 d x R 1 ∈ R 1 ∑ ∂ z 2 v ( R 1 − x ) + O ( R − 1 ) = 2 π ∣ Γ 1 ∣ / ∣ Γ 2 ∣ ( ∣ Γ 1 ∣ 2 1 + ∣ Γ 2 ∣ 2 1 ) 2 ∫ R 2 d x ∂ z 2 v ( x ) + O ( R − 1 ) = 2 π ∣ Γ 1 ∣ / ∣ Γ 2 ∣ ( ∣ Γ 1 ∣ 2 1 + ∣ Γ 2 ∣ 2 1 ) 2 m ( d ) + O ( R − 1 ) ,
where we used (30) . Observe that the monolayer energy contributions are each of size O ( R − 1 ) O(R^{-1}) O ( R − 1 ) . As a consequence and bounding the crossing term as above via Cauchy–Schwarz, we obtain
inf ∥ δ u j ∥ 2 = 1 Δ E ( t o t ) ( 0 ) ( δ u , δ u ) ≤ lim R → ∞ Δ E ∞ ( 0 ) ( v R , v R ) ≤ λ ( ∣ Γ 1 ∣ 1 2 + ∣ Γ 2 ∣ 1 2 ) 2 2 π ∣ Γ 1 ∣ / ∣ Γ 2 ∣ m ( d ) < 0 , \inf_{\lVert\delta u_j\rVert^2=1}\Delta E^{(\mathrm{tot})}(0)(\delta u,\delta u)\leq\lim_{R\to\infty}\Delta E_\infty(0)(v_R,v_R)\leq\lambda\frac{\left(\lvert\Gamma_1\rvert^{\frac{1}{2}}+\lvert\Gamma_2\rvert^{\frac{1}{2}}\right)^2}{2\pi\lvert\Gamma_1\rvert/\lvert\Gamma_2\rvert}m(d)<0, ∥ δ u j ∥ 2 = 1 inf Δ E ( tot ) ( 0 ) ( δ u , δ u ) ≤ R → ∞ lim Δ E ∞ ( 0 ) ( v R , v R ) ≤ λ 2 π ∣ Γ 1 ∣ / ∣ Γ 2 ∣ ( ∣ Γ 1 ∣ 2 1 + ∣ Γ 2 ∣ 2 1 ) 2 m ( d ) < 0 ,
provided that m ( d ) < 0 m(d)<0 m ( d ) < 0 .
With analogous steps and following the ideas of [18 Lemma 3.6] , we obtain
inf ∥ δ u j ∥ 2 = 1 j = 1 , 2 Δ E ( t o t ) ( u ( e q ) ) ( δ u , δ u ) ≤ ( ∣ Γ 1 ∣ 1 2 + ∣ Γ 2 ∣ 1 2 ) 2 2 π ∣ Γ 1 ∣ / ∣ Γ 2 ∣ ∫ R 2 d x ∂ a 2 v ( x + u 2 ( e q ) ( A M A 1 − 1 x ) − u 1 ( e q ) ( A M A 2 − 1 x ) ) \inf_{\substack{\lVert\delta u_j\rVert^2=1\\j=1,2}}\Delta E^{(\mathrm{tot})}(\mathbf{u}^{(\mathrm{eq})})(\delta u,\delta u)
\leq\frac{\left(\lvert\Gamma_1\rvert^{\frac{1}{2}}+\lvert\Gamma_2\rvert^{\frac{1}{2}}\right)^2}{2\pi\lvert\Gamma_1\rvert/\lvert\Gamma_2\rvert}\int_{\mathbb{R}^2}\mathrm{d}x\,\partial_a^2v\left(x+u_2^{(\mathrm{eq})}\left(A_MA_1^{-1}x\right)-u_1^{(\mathrm{eq})}\left(A_MA_2^{-1}x\right)\right) ∥ δ u j ∥ 2 = 1 j = 1 , 2 inf Δ E ( tot ) ( u ( eq ) ) ( δ u , δ u ) ≤ 2 π ∣ Γ 1 ∣ / ∣ Γ 2 ∣ ( ∣ Γ 1 ∣ 2 1 + ∣ Γ 2 ∣ 2 1 ) 2 ∫ R 2 d x ∂ a 2 v ( x + u 2 ( eq ) ( A M A 1 − 1 x ) − u 1 ( eq ) ( A M A 2 − 1 x ) )
for any unit vector a ∈ S 2 a\in\mathbb{S}^2 a ∈ S 2 . With that, we obtain the following condition:
Definition 5.4 (Phonon instability). Let u ( e q ) \mathbf{u}^{(\mathrm{eq})} u ( eq ) denote the equilibrium displacements. We say the bilayer system is phonon-instable iff there exists a direction a ∈ S 2 a\in\mathbb{S}^2 a ∈ S 2
∫ R 2 d x ∂ a 2 v ( x + u 2 ( e q ) ( A M A 1 − 1 x ) − u 1 ( e q ) ( A M A 2 − 1 x ) ) < 0. \int_{\mathbb{R}^2}\mathrm{d}x\,\partial_a^2v\left(x+u_2^{(\mathrm{eq})}\left(A_MA_1^{-1}x\right)-u_1^{(\mathrm{eq})}\left(A_MA_2^{-1}x\right)\right)<0. ∫ R 2 d x ∂ a 2 v ( x + u 2 ( eq ) ( A M A 1 − 1 x ) − u 1 ( eq ) ( A M A 2 − 1 x ) ) < 0.
Sufficient condition for stability
Above, we showed a sufficient condition for instability. Conversely, we would like to establish a sufficient condition for stability. Using (31) , we obtain positive (semi-) definiteness of δ 2 Δ E ( t o t ) ( 0 ) \delta^2\Delta E^{(\mathrm{tot})}(0) δ 2 Δ E ( tot ) ( 0 ) if
∑ R 1 ∈ R 1 D 2 v ( x − R 1 ) ≥ 0 \sum_{R_1\in\mathcal{R}_1}D^2v(x-R_1)\geq0 R 1 ∈ R 1 ∑ D 2 v ( x − R 1 ) ≥ 0
is positive definite for all x ∈ Γ 1 x\in\Gamma_1 x ∈ Γ 1 . Including relaxation, this condition becomes
∑ R 1 ∈ R 1 D 2 v ( x − R 1 + u 2 ( e q ) ( A M A 1 − 1 x ) − u 1 ( e q ) ( R 1 + A M A 2 − 1 x ) ) ≥ 0. \sum_{R_1\in\mathcal{R}_1}D^2v\left(x-R_1+u_2^{(\mathrm{eq})}\left(A_MA_1^{-1}x\right)-u_1^{(\mathrm{eq})}\left(R_1+A_MA_2^{-1}x\right)\right)\geq0. R 1 ∈ R 1 ∑ D 2 v ( x − R 1 + u 2 ( eq ) ( A M A 1 − 1 x ) − u 1 ( eq ) ( R 1 + A M A 2 − 1 x ) ) ≥ 0.
Since u j ( e q ) u_j^{(\mathrm{eq})} u j ( eq ) is R M \mathcal{R}_M R M -periodic and
R 1 = − A M ( A 2 − 1 − A 1 − 1 ) R 1 = − A M A 2 − 1 + A M A 1 − 1 R 1 , R_1=-A_M\left(A_2^{-1}-A_1^{-1}\right)R_1=-A_MA_2^{-1}+A_MA_1^{-1}R_1, R 1 = − A M ( A 2 − 1 − A 1 − 1 ) R 1 = − A M A 2 − 1 + A M A 1 − 1 R 1 ,
where A M A 1 − 1 R 1 ∈ R M A_MA_1^{-1}R_1\in\mathcal{R}_M A M A 1 − 1 R 1 ∈ R M , we have that
u 2 ( e q ) ( A M A 1 − 1 x ) − u 1 ( e q ) ( R 1 + A M A 2 − 1 x ) = u 2 ( e q ) ( A M A 1 − 1 ( x − R 1 ) ) − u 1 ( e q ) ( A M A 2 − 1 ( x − R 1 ) ) = : u ( e q ) ( x − R 1 ) . (32) \begin{aligned}
u_2^{(\mathrm{eq})}\left(A_MA_1^{-1}x\right)-u_1^{(\mathrm{eq})}\left(R_1+A_MA_2^{-1}x\right)
&=u_2^{(\mathrm{eq})}\left(A_MA_1^{-1}(x-R_1)\right)-u_1^{(\mathrm{eq})}\left(A_MA_2^{-1}(x-R_1)\right)\\
&=:u^{(\mathrm{eq})}(x-R_1). \tag*{(32)}
\end{aligned} u 2 ( eq ) ( A M A 1 − 1 x ) − u 1 ( eq ) ( R 1 + A M A 2 − 1 x ) = u 2 ( eq ) ( A M A 1 − 1 ( x − R 1 ) ) − u 1 ( eq ) ( A M A 2 − 1 ( x − R 1 ) ) =: u ( eq ) ( x − R 1 ) . (32)
With this, we obtain the following condition.
Definition 5.5 (Phonon stability). Let u ( e q ) \mathbf{u}^{(\mathrm{eq})} u ( eq ) denote the equilibrium displacements, and u ( e q ) u^{(\mathrm{eq})} u ( eq ) be defined as in (32) . We say the bilayer system is phonon-stable iff
∑ R 1 ∈ R 1 D 2 v ( x − R 1 + u ( e q ) ( x − R 1 ) ) ≥ 0 \sum_{R_1\in\mathcal{R}_1}D^2v\left(x-R_1+u^{(\mathrm{eq})}(x-R_1)\right)\geq0 R 1 ∈ R 1 ∑ D 2 v ( x − R 1 + u ( eq ) ( x − R 1 ) ) ≥ 0 for all x ∈ Γ 1 x\in\Gamma_1 x ∈ Γ 1 , in the sense of quadratic forms.
GSFE based offset energy In order to find a model based on the generalized stacking fault energy determined in [9 ] , the interlayer energy needs to be replaced by
∫ R M d x Φ ( G S F E ) ( D 1 → 2 x + u 1 ( x ) − u 2 ( x ) ) . \int_{\mathcal{R}_M}\mathrm{d}x\,\Phi^{(\mathrm{GSFE})}\left(D_{1\to2}x+u_1(x)-u_2(x)\right). ∫ R M d x Φ ( GSFE ) ( D 1 → 2 x + u 1 ( x ) − u 2 ( x ) ) .
The corresponding interlayer phonon energy term is given by
∑ R 1 ∈ R 1 D 2 Φ ( G S F E ) ( D 1 → 2 R 1 + u 1 ( e q ) ( R 1 ) − u 2 ( e q ) ( R 1 ) ) [ ( δ u 1 ( R 1 ) − δ u 2 ( R 1 ) ) ⊗ 2 ] . \sum_{R_1\in\mathcal{R}_1}D^2\Phi^{(\mathrm{GSFE})}\left(D_{1\to2}R_1+u_1^{(\mathrm{eq})}(R_1)-u_2^{(\mathrm{eq})}(R_1)\right)\left[(\delta u_1(R_1)-\delta u_2(R_1))^{\otimes2}\right]. R 1 ∈ R 1 ∑ D 2 Φ ( GSFE ) ( D 1 → 2 R 1 + u 1 ( eq ) ( R 1 ) − u 2 ( eq ) ( R 1 ) ) [ ( δ u 1 ( R 1 ) − δ u 2 ( R 1 ) ) ⊗ 2 ] .
Analogously, one could replace the summation over R 1 \mathcal{R}_1 R 1 by summation over R 2 \mathcal{R}_2 R 2 , or a combination of these. Notice that here, we would need to construct a smooth interpolant of δ u 2 \delta u_2 δ u 2 . This could be avoided, if one instead approximated the energy by
∑ R 1 ∈ R 1 D 2 Φ ( G S F E ) ( D 1 → 2 R 1 + u 1 ( e q ) ( R 1 ) − u 2 ( e q ) ( R 1 ) ) [ ( δ u 1 ( R 1 ) − δ u 2 ( ⌊ R 1 ⌋ 2 ) ) ⊗ 2 ] , \sum_{R_1\in\mathcal{R}_1}D^2\Phi^{(\mathrm{GSFE})}\left(D_{1\to2}R_1+u_1^{(\mathrm{eq})}(R_1)-u_2^{(\mathrm{eq})}(R_1)\right)\left[(\delta u_1(R_1)-\delta u_2(\lfloor R_1\rfloor_2))^{\otimes2}\right], R 1 ∈ R 1 ∑ D 2 Φ ( GSFE ) ( D 1 → 2 R 1 + u 1 ( eq ) ( R 1 ) − u 2 ( eq ) ( R 1 ) ) [ ( δ u 1 ( R 1 ) − δ u 2 (⌊ R 1 ⌋ 2 ) ) ⊗ 2 ] ,
or, analogously, evaluating δ u 2 \delta u_2 δ u 2 at the nearest lattice point in R 2 \mathcal{R}_2 R 2 instead. Φ ( G S F E ) \Phi^{(\mathrm{GSFE})} Φ ( GSFE ) takes the form
Φ ( G S F E ) ( v , w ) : = Φ 0 ( G S F E ) ( 2 π v , 2 π w ) , \Phi^{(\mathrm{GSFE})}(v,w):=\Phi_0^{(\mathrm{GSFE})}(2\pi v,2\pi w), Φ ( GSFE ) ( v , w ) := Φ 0 ( GSFE ) ( 2 π v , 2 π w ) ,
Φ 0 ( G S F E ) ( v , w ) : = c 0 + c 1 [ cos ( v ) + cos ( w ) + cos ( v + w ) ] + c 2 [ cos ( v + 2 w ) + cos ( v − w ) + cos ( 2 v + w ) ] + c 3 [ cos ( 2 v ) + cos ( 2 w ) + cos ( 2 v + 2 w ) ] , \begin{aligned}
\Phi_0^{(\mathrm{GSFE})}(v,w)&:=c_0+c_1[\cos(v)+\cos(w)+\cos(v+w)]\\
&\quad+c_2[\cos(v+2w)+\cos(v-w)+\cos(2v+w)]\\
&\quad+c_3[\cos(2v)+\cos(2w)+\cos(2v+2w)],
\end{aligned} Φ 0 ( GSFE ) ( v , w ) := c 0 + c 1 [ cos ( v ) + cos ( w ) + cos ( v + w )] + c 2 [ cos ( v + 2 w ) + cos ( v − w ) + cos ( 2 v + w )] + c 3 [ cos ( 2 v ) + cos ( 2 w ) + cos ( 2 v + 2 w )] ,
Figure 2. Profile of Φ ( G S F E ) \Phi^{(\mathrm{GSFE})} Φ ( GSFE ) , at 30° rotated position, with potential minima AB/BA.
where c 0 = 6.832 c_0 = 6.832 c 0 = 6.832 , c 1 = 4.064 c_1 = 4.064 c 1 = 4.064 , c 2 = − 0.374 c_2 = -0.374 c 2 = − 0.374 , and c 3 = − 0.095 c_3 = -0.095 c 3 = − 0.095 (all in meV/unit area), see 2. In particular, the stability criterion 5.5 translates to
D 2 Φ ( G S F E ) ( D 1 → 2 x + u 1 ( e q ) ( x ) − u 2 ( e q ) ( x ) ) ≥ 0. D^2\Phi^{(\mathrm{GSFE})}\left(D_{1\to2}x+u_1^{(\mathrm{eq})}(x)-u_2^{(\mathrm{eq})}(x)\right)\ge0. D 2 Φ ( GSFE ) ( D 1 → 2 x + u 1 ( eq ) ( x ) − u 2 ( eq ) ( x ) ) ≥ 0.
Asymptotic analysis
To test our criterion, we shall restrict to purely twisted systems, i.e., we assume
A 1 = R θ / 2 A , A 2 = R − θ / 2 A . A_1=R_{\theta/2}A,\qquad A_2=R_{-\theta/2}A. A 1 = R θ /2 A , A 2 = R − θ /2 A .
Moreover, to obtain an analytical result, we study the energy in the limit θ → 0 \theta\to0 θ → 0 . As we will see, one can reduce the functional to an Allen-Cahn functional.
Reduction to Allen-Cahn functional
For that, let us consider the energy in the Cauchy-Born and linear elasticity approximation and with interlayer pair potentials. In this case, the total energy reads, see [18 ] ,
1 2 ∑ j = 1 2 ∫ Γ M d x ∇ u j ( x ) : C j : ∇ u j ( x ) + ∫ R 2 d ξ ∣ Γ M ∣ v ( ( A 1 − A 2 ) ξ + u 1 ( A 1 ξ ) − u 2 ( A 2 ξ ) ) . \frac{1}{2}\sum_{j=1}^{2}\int_{\Gamma_M}\mathrm{d}x\,\nabla u_j(x):C_j:\nabla u_j(x)+\int_{\mathbb{R}^2}\frac{\mathrm{d}\xi}{|\Gamma_M|}v\left((A_1-A_2)\xi+u_1(A_1\xi)-u_2(A_2\xi)\right). 2 1 j = 1 ∑ 2 ∫ Γ M d x ∇ u j ( x ) : C j : ∇ u j ( x ) + ∫ R 2 ∣ Γ M ∣ d ξ v ( ( A 1 − A 2 ) ξ + u 1 ( A 1 ξ ) − u 2 ( A 2 ξ ) ) .
We rescale x = A M y x=A_My x = A M y , ξ = ( A 1 − A 2 ) − 1 A y \xi=(A_1-A_2)^{-1}Ay ξ = ( A 1 − A 2 ) − 1 A y and abbreviate
U j ( y ) : = A − 1 u j ( A M y ) , ε : = 2 sin ( θ / 2 ) , I ε : = A − 1 R θ / 2 A , V ( x ) : = v ( A x ) ∣ det A ∣ , U_j(y):=A^{-1}u_j(A_My),\qquad\varepsilon:=2\sin(\theta/2),\qquad I_\varepsilon:=A^{-1}R_{\theta/2}A,\qquad V(x):=\frac{v(Ax)}{|\det A|}, U j ( y ) := A − 1 u j ( A M y ) , ε := 2 sin ( θ /2 ) , I ε := A − 1 R θ /2 A , V ( x ) := ∣ det A ∣ v ( A x ) ,
E j , a b c d : = ( A − 1 R π / 2 ) a ′ a A b ′ b ( A − 1 R π / 2 ) c ′ c A d ′ d C j , a b c d = λ ( R − π / 2 A − T A ) a b ( R − π / 2 A − T A ) c d + μ [ ( R − π / 2 A − T A ) a d ( A T A − 1 R π / 2 ) b c + ( R − π / 2 A − T A − 1 R π / 2 ) a c ( A T A ) b d ] . \mathcal{E}_{j,abcd}:=(A^{-1}R_{\pi/2})_{a'a}A_{b'b}(A^{-1}R_{\pi/2})_{c'c}A_{d'd}C_{j,abcd}
=\lambda(R_{-\pi/2}A^{-T}A)_{ab}(R_{-\pi/2}A^{-T}A)_{cd}
+\mu\left[(R_{-\pi/2}A^{-T}A)_{ad}(A^TA^{-1}R_{\pi/2})_{bc}
+(R_{-\pi/2}A^{-T}A^{-1}R_{\pi/2})_{ac}(A^TA)_{bd}\right]. E j , ab c d := ( A − 1 R π /2 ) a ′ a A b ′ b ( A − 1 R π /2 ) c ′ c A d ′ d C j , ab c d = λ ( R − π /2 A − T A ) ab ( R − π /2 A − T A ) c d + μ [ ( R − π /2 A − T A ) a d ( A T A − 1 R π /2 ) b c + ( R − π /2 A − T A − 1 R π /2 ) a c ( A T A ) b d ] .
Observe, that U j U_j U j is Z 2 \mathbb{Z}^2 Z 2 -periodic, j = 1 , 2 j=1,2 j = 1 , 2 . After dividing by ε \varepsilon ε , we rewrite the energy functional as
ε 2 ∑ j = 1 2 ∫ T 2 d y ∇ U j ( y ) : E j : ∇ U j ( y ) + 1 ε ∫ R 2 d y V ( y + U 1 ( I ε − 1 y ) − U 2 ( I ε y ) ) = ∫ T 2 d y ( ε 2 ∑ j = 1 2 ∇ U j ( y ) : E j : ∇ U j ( y ) + 1 ε ∑ n ∈ Z 2 V ( y − n + U 1 ( I ε − 1 ( y − n ) ) − U 2 ( I ε ( y − n ) ) ) ) . \begin{aligned}
\frac{\varepsilon}{2}\sum_{j=1}^{2}\int_{\mathbb{T}^2} \mathrm{d}y\, \nabla U_j(y):\mathcal{E}_j:\nabla U_j(y)+\frac{1}{\varepsilon}\int_{\mathbb{R}^2}\mathrm{d}y\,V\bigl(y+U_1(I_\varepsilon^{-1}y)-U_2(I_\varepsilon y)\bigr) \\
&=\int_{\mathbb{T}^2}\mathrm{d}y\,\left(\frac{\varepsilon}{2}\sum_{j=1}^{2}\nabla U_j(y):\mathcal{E}_j:\nabla U_j(y)
+\frac{1}{\varepsilon}\sum_{n\in\mathbb{Z}^2}V\bigl(y-n+U_1(I_\varepsilon^{-1}(y-n))-U_2(I_\varepsilon(y-n))\bigr)\right).
\end{aligned} 2 ε j = 1 ∑ 2 ∫ T 2 d y ∇ U j ( y ) : E j : ∇ U j ( y ) + ε 1 ∫ R 2 d y V ( y + U 1 ( I ε − 1 y ) − U 2 ( I ε y ) ) = ∫ T 2 d y ( 2 ε j = 1 ∑ 2 ∇ U j ( y ) : E j : ∇ U j ( y ) + ε 1 n ∈ Z 2 ∑ V ( y − n + U 1 ( I ε − 1 ( y − n )) − U 2 ( I ε ( y − n )) ) ) .
If we formally take the limit ε → 0 \varepsilon\to0 ε → 0 , we obtain that the interlayer misfit energy, multiplied by ε \varepsilon ε , converges to
∫ T 2 d y ∑ n ∈ Z 2 V ( y − n + U 1 ( y − n ) − U 2 ( y − n ) ) = ∫ T 2 d y Φ 0 ( y + U 1 ( y ) − U 2 ( y ) ) , \int_{\mathbb{T}^2}\mathrm{d}y\,\sum_{n\in\mathbb{Z}^2}V\bigl(y-n+U_1(y-n)-U_2(y-n)\bigr)=\int_{\mathbb{T}^2}\mathrm{d}y\,\Phi_0\bigl(y+U_1(y)-U_2(y)\bigr), ∫ T 2 d y n ∈ Z 2 ∑ V ( y − n + U 1 ( y − n ) − U 2 ( y − n ) ) = ∫ T 2 d y Φ 0 ( y + U 1 ( y ) − U 2 ( y ) ) ,
where we used the fact that U i U_i U i is Z 2 \mathbb{Z}^2 Z 2 -periodic and where Φ 0 ( y ) : = ∑ n ∈ Z 2 V ( y − n ) \Phi_0(y):=\sum_{n\in\mathbb{Z}^2}V(y-n) Φ 0 ( y ) := ∑ n ∈ Z 2 V ( y − n ) is the misfit energy , akin to the generalized stacking fault energy (GSFE) obtained in [9 ] . It turns out that Φ 0 \Phi_0 Φ 0 is a two-well potential with minima exactly at AB and BA stacking, corresponding to the positions x A B = 1 3 ( 1 , 1 ) T x_{AB}=\frac{1}{3}(1,1)^T x A B = 3 1 ( 1 , 1 ) T and x B A = 2 x A B x_{BA}=2x_{AB} x B A = 2 x A B . For simplicity, let us assume homobilayers, i.e., E 1 = E 2 \mathcal{E}_1=\mathcal{E}_2 E 1 = E 2 . Using the fact that
1 2 ∑ j = 1 2 ∇ U j : E : ∇ U j = ∇ u : E : ∇ u + ∇ u ˉ : E : ∇ u ˉ , \frac{1}{2}\sum_{j=1}^{2}\nabla U_j:\mathcal{E}:\nabla U_j=\nabla u:\mathcal{E}:\nabla u+\nabla\bar{u}:\mathcal{E}:\nabla\bar{u}, 2 1 j = 1 ∑ 2 ∇ U j : E : ∇ U j = ∇ u : E : ∇ u + ∇ u ˉ : E : ∇ u ˉ ,
where u : = U 1 − U 2 2 u:=\frac{U_1-U_2}{2} u := 2 U 1 − U 2 and u ˉ : = U 1 + U 2 2 \bar{u}:=\frac{U_1+U_2}{2} u ˉ := 2 U 1 + U 2 . In particular, we can rewrite the energy as
∫ T 2 d x ( ε ∇ u : E : ∇ u + 1 ε Φ 0 ( x + 2 u ) + ε ∇ u ˉ : E : ∇ u ˉ ) . \int_{\mathbb{T}^2}\mathrm{d}x\,\left(\varepsilon\nabla u:\mathcal{E}:\nabla u+\frac{1}{\varepsilon}\Phi_0(x+2u)+\varepsilon\nabla\bar{u}:\mathcal{E}:\nabla\bar{u}\right). ∫ T 2 d x ( ε ∇ u : E : ∇ u + ε 1 Φ 0 ( x + 2 u ) + ε ∇ u ˉ : E : ∇ u ˉ ) .
Consequently, the minimum is attained at u ˉ = C \bar{u}=C u ˉ = C for some constant C C C . Due to the constraint ∫ T 2 d x U j = 0 \int_{\mathbb{T}^2}\mathrm{d}x\,U_j=0 ∫ T 2 d x U j = 0 , we have that C = 0 C=0 C = 0 , i.e., u = U 1 = − U 2 u=U_1=-U_2 u = U 1 = − U 2 . Then the remaining energy is given by
∫ T 2 d x ( ε ∇ u ε : E : ∇ u ε + 1 ε Φ 0 ( x + 2 u ε ) ) . \int_{\mathbb{T}^2}\mathrm{d}x\,\left(\varepsilon\nabla u_\varepsilon:\mathcal{E}:\nabla u_\varepsilon+\frac{1}{\varepsilon}\Phi_0(x+2u_\varepsilon)\right). ∫ T 2 d x ( ε ∇ u ε : E : ∇ u ε + ε 1 Φ 0 ( x + 2 u ε ) ) .
From the Allen-Cahn theory, see, [37 ] , we find that, as ε → 0 \varepsilon\to0 ε → 0 , the sequence of minimizers x + 2 u ε x+2u_\varepsilon x + 2 u ε needs to converge to
x A B 1 E + x B A 1 E c (33) x_{AB}\mathbf{1}_E+x_{BA}\mathbf{1}_{E^c} \tag*{(33)} x A B 1 E + x B A 1 E c (33)
for some appropriate set E ⊆ T 2 E \subseteq\mathbb{T}^2 E ⊆ T 2 . Here, we set the minima of Φ 0 \Phi_0 Φ 0 to zero. A crucial assumption for the theory to be applicable is that the Hessian of Φ 0 \Phi_0 Φ 0 is positive at its wells. In particular, we require the assumption
( D 2 Φ 0 ) ( x A B ) , ( D 2 Φ 0 ) ( x B A ) > 0. (34) (D^2\Phi_0)(x_{AB}), (D^2\Phi_0)(x_{BA}) > 0. \tag*{(34)} ( D 2 Φ 0 ) ( x A B ) , ( D 2 Φ 0 ) ( x B A ) > 0. (34)
In particular, we have in this case that
( D 2 Φ 0 ) ( x A B 1 E + x B A 1 E c ) > 0 , (D^2\Phi_0)(x_{AB}\mathbf{1}_E + x_{BA}\mathbf{1}_{E^c}) > 0, ( D 2 Φ 0 ) ( x A B 1 E + x B A 1 E c ) > 0 ,
i.e., by Definition 5.5 , in the limit θ → 0 \theta\to0 θ → 0 , the bilayer system is phonon stable .
Model evaluation
We now want to evaluate the asymptotic stability criterion (34) in the models presented in § 2.10.2 . For simplicity, we will restrict to two-dimensional interlayer coupling pair-potentials, where we keep the vertical distance fixed. In the case of bilayer graphene, the geometry is given by
A = a 0 ( 3 2 3 2 − 1 2 1 2 ) , a 0 = 3 ⋅ 1.42 n m ≈ 2.46 n m , d B G ≈ 3.35 A ˚ , A = a_0 \begin{pmatrix} \frac{\sqrt{3}}{2} & \frac{\sqrt{3}}{2} \\ -\frac{1}{2} & \frac{1}{2} \end{pmatrix}, \qquad a_0 = \sqrt{3} \cdot1.42\ \mathrm{nm} \approx2.46\ \mathrm{nm}, \qquad d_{\mathrm{BG}} \approx3.35\ \text{\AA}, A = a 0 ( 2 3 − 2 1 2 3 2 1 ) , a 0 = 3 ⋅ 1.42 nm ≈ 2.46 nm , d BG ≈ 3.35 A ˚ ,
where d B G d_{\mathrm{BG}} d BG is the equilibrium distance and a 0 / 3 a_0/\sqrt{3} a 0 / 3 is the lattice constant. For a radial potential v ( 3 D ) ( x , z ) = f ( ∣ ( x , z ) ∣ ) v^{(3D)}(x,z) = f(\lvert(x,z)\rvert) v ( 3 D ) ( x , z ) = f (∣( x , z )∣) , the restricted potential is then given by
v ( 2 D ) ( x ) = f ( ∣ x ∣ 2 + d B G 2 ) = : g ( ∣ x ∣ ) ⇒ V ( x ) = v ( 2 D ) ( A x ) ∣ det A ∣ . v^{(2D)}(x) = f\left(\sqrt{\lvert x\rvert^2 + d_{\mathrm{BG}}^2}\right) =: g(\lvert x\rvert) \quad\Rightarrow\quad V(x) = \frac{v^{(2D)}(Ax)}{\lvert\det A\rvert}. v ( 2 D ) ( x ) = f ( ∣ x ∣ 2 + d BG 2 ) =: g (∣ x ∣) ⇒ V ( x ) = ∣ det A ∣ v ( 2 D ) ( A x ) .
Observe that we have that
D 2 v ( 2 D ) ( x ) = g ′ ′ ( r ) e r ⊗ e r + g ′ ( r ) r e φ ⊗ e φ , D 2 V ( x ) = A T D 2 v ( 2 D ) ( A x ) A ∣ det A ∣ , D^2v^{(2D)}(x) = g^{\prime\prime}(r)e_r \otimes e_r + \frac{g^\prime(r)}{r}e_\varphi\otimes e_\varphi,\qquad D^2V(x) = \frac{A^T D^2v^{(2D)}(Ax)A}{\lvert\det A\rvert}, D 2 v ( 2 D ) ( x ) = g ′′ ( r ) e r ⊗ e r + r g ′ ( r ) e φ ⊗ e φ , D 2 V ( x ) = ∣ det A ∣ A T D 2 v ( 2 D ) ( A x ) A ,
where x = r e r x = re_r x = r e r , e r = ( cos φ , sin φ ) T e_r = (\cos\varphi,\sin\varphi)^T e r = ( cos φ , sin φ ) T , e φ = ( − sin φ , cos φ ) T e_\varphi= (-\sin\varphi,\cos\varphi)^T e φ = ( − sin φ , cos φ ) T . To approximate Φ 0 \Phi_0 Φ 0 , we truncate the summation at some N N N , i.e., we compute
( D 2 Φ 0 ) ( A B ) ≈ ∑ n ∈ Z 2 ∥ n ∥ ∞ ≤ N D 2 V ( A B ) . (D^2\Phi_0)(AB) \approx\sum_{\substack{n\in\mathbb{Z}^2 \\ \lVert n\rVert_\infty\le N}} D^2V(AB). ( D 2 Φ 0 ) ( A B ) ≈ n ∈ Z 2 ∥ n ∥ ∞ ≤ N ∑ D 2 V ( A B ) .
We provided the profiles of the GSFEs corresponding to the respective models, and the evaluation of the Hessians at the potential wells in Figure 3 . We obtain that each of the considered models exhibits phonon stability.
Figure 3. Misfit energy for bilayer graphene evaluated at 30 ∘ 30^\circ 3 0 ∘ rotated positions with potential wells AB/BA. The truncation is chosen to be N = 10 N = 10 N = 10 . Parametrization given in § 2.10 and § 5.1 . Right: Convergence of minimal eigenvalue of Hessian at AB/BA point for growing potential sum truncation N N N . From top to bottom: Lennard-Jones, Morse, and Kolmogorov-Crespi potential
In case of Φ ( G S F E ) \Phi^{(\mathrm{GSFE})} Φ ( GSFE ) , an analogous argument to above shows that the minimizer needs to satisfy u 2 = − u 1 u_2=-u_1 u 2 = − u 1 . With analogous notation to above, the argument of Φ ( G S F E ) \Phi^{(\mathrm{GSFE})} Φ ( GSFE ) is given by x + 2 u ε x+2u_\varepsilon x + 2 u ε , which, again, needs to converge to (33) . In particular, we compute
D 2 Φ ( G S F E ) ( A B ) ≥ 117 m e V , D^2\Phi^{(\mathrm{GSFE})}(AB) \ge117\ \mathrm{meV}, D 2 Φ ( GSFE ) ( A B ) ≥ 117 meV ,
i.e., the GSFE-based model obtained in [9 ] is phonon-stable in the limit θ → 0 \theta\to0 θ → 0 .