Straintronics and twistronics in bilayer graphene Federico Escudero, Dong Wang, Pierre A. Pantaleón, Shengjun Yuan, Francisco Guinea, Zhen Zhan Abstract The interplay of twist and strain in bilayer graphene enables the formation of moiré patterns and narrow bands that host correlated and topological phases. While magic-angle twisted bilayer graphene has been widely studied, strain provides an additional and realistic control knob for band engineering. In this work, we first generate a global method to construct commensurate supercells for arbitrary twist and strain. Then, using atomistic tight-binding and strain-extended continuum models to study the commensurate structures, we identify configurations that minimize the bandwidth beyond the magic angle. The results reveal a strong dependence of band narrowing and topology on strain type, magnitude, direction and lattice relaxation. Particularly, shear strain produces a stronger distortion than uniaxial strain. Including electron-electron interactions through a self-consistent Hartree potential shows that strain broadens the bare bands while reducing electrostatic renormalization. Strain also drives topological transitions as the narrow and remote bands hybridize, establishing twisted and strained bilayer graphene as a tunable platform for flat-band and topological phenomena. I INTRODUCTION The discovery of correlated phases and unconventional superconductivity in twisted bilayer graphene (TBG) has attracted significant attention over the last few years 3 ; 10 ; 11 ; 48 ; 107 ; 62 ; 85 ; 110 ; 106 ; 20 ; 12 . These phenomena are intrinsically connected to the emergence of very flat bands due to the moiré potential 59 ; 99 ; 95 ; 9 , induced by the lattice mismatch created by the twist or strain 4 ; 69 ; 52 ; 86 ; 8 ; 22 . The quenching of the kinetic energy in the flat bands promotes the appearance of the observed electronic correlations. Any approach to understand the nature of the correlated phases in TBG must then start from a solid understanding of the nature and origin of the flat bands, supported by accurate modelling methods. With only a relative twist, the flat bands are well-known to arise at an infinite set of magic angles , the largest one being around 1^{\circ} 59 ; 9 ; 97 . However, in practice the samples usually have at least some residual strain 46 ; 39 , defects 1 ; 61 ; 88 or even twist angle disorders 104 ; 87 ; 6 , typically arising during their fabrication 28 . The presence of strains, in particular, can significantly modify the geometrical and electronic properties of the system since the moiré acts as a magnifying glass 42 ; 8 ; 47 ; 67 ; 64 ; 92 ; 103 ; 22 . Although this may be seen as undesirable, one can conversely use strains as an additional tune parameter in the system 44 ; 65 ; 47 ; 46 ; 78 . Compared to the twist, which only rotates the layers, strains distort the layers and thus can lead to a plethora of moiré patterns with potentially rich properties 49 ; 22 . This has motivated recent experimental advances in developing new techniques to induce and manipulate the strain in moiré heterostructures 78 ; 36 ; 29 ; 89 , opening a path to straintronics and twistronics, whereby the electronic properties can be tuned by the combined interplay between twist and strain. Among different types of strain, uniaxial heterostrain is the most common type, and has been observed in many experimental samples 42 ; 48 ; 105 ; 67 ; 103 . Consequently, most theoretical works focus on the effects of the uniaxial heterostrain. Recently, under some new well-developed techniques, shear heterostrain is introduced to manipulate the moiré patterns 108 ; 74 ; 14 . A natural question, then, is how does the presence of arbitrary strain and twist modifies the narrow bands in twisted bilayer graphene. An initial step involves computing the electronic spectra of the system and study its evolution under different kinds of relevant strain. Since narrow bands generally arise only under small lattice mismatches 59 ; 9 , at which the supercells can contain tens of thousands of atoms, previous studies have mostly employed extensions of effective continuum models under strain 8 ; 86 ; 22 , or topological heavy fermion models 38 ; 37 . Yet, despite its importance, there is no comprehensive study of the electronic properties under strain by means of more realistic approaches, such as atomistic models. The challenge for that relies not only on the huge size of the supercell, but also on the fact that under twist and strain the system is in general incommensurate, and it is not clear at which configurations there can actually be a commensurate solution 109 . The commensurate and incommensurate structures may in fact have distinct ground-state properties 31 . In this work, we present a comprehensive analysis of the tight-binding (TB) properties of twisted and strained bilayer graphene (TSBG), focusing on the optimal conditions for the emergence of narrow bands. Two relevant types of strain are considered: uniaxial and shear. We first tackle the problem of obtaining commensurate structures under both twist and strain. We find that by generally adding a small biaxial strain, one can always find particular twist and strain values at which the system is commensurate. Using these commensurate structures, we then obtain the band structure and the density of states (DOS) by using a full atomistic TB model. We find that: (i) albeit the strain tends to increase the bandwidth at the (nonstrain) magic angle 8 ; 22 , there are yet other twist angles at which the bandwidth can be minimum, resulting in a shift of magic angle with strain; (ii) the emergence of narrow bands depends critically on the strain direction; (iii) the shear strain produces a stronger distortion of the geometry and electronic properties of TSBG; (iv) the gap between the narrow and remote bands (induced by the lattice relaxation) is mainly determined by the strain-dependent bandwidth of the narrow bands. Our atomistic results are then compared with that of the strain-extended continuum model 79 ; 101 ; 2 ; 70 , which introduces two main modifications: (i) the change in the moiré vectors by which electrons in the different layers are coupled through the moiré potential 49 ; 22 ; (ii) the introduction of strain-induced fields 96 ; 63 . We show that with just a few suitable parameter choices, the continuum model yields results in excellent agreement with the TB ones. In particular, we find that the strain-induced gauge potential, accounting for the change in the hopping energies within the Dirac approximation 96 , plays a key role in capturing the electronic behavior in TSBG. Using the strain-extended continuum model, we extend our analysis of the bandwidth evolution under twist and strain. We find that the twist angle at which the bandwidth is minimum sensitively depends on the strain direction. Yet, we remarkably see that the minimum bandwidth of the narrow bands (at the optimal twist angle) increases practically linearly with the strain magnitude. We further consider the effect of electrostatic interactions, as accounted by the Hartree potential 34 ; 18 ; 32 . Due to the increase of bandwidth under strain, the Hartree effect is weaker than that in the only twisted configurations. Consequently, as the strain increases, there is a competition between the increase of bandwidth of the bare bands, and the decrease of the Hartree potential. We show the synergy between these effects can lead to bandwidths under twist and strain that are actually comparable, if not smaller, than those with only twist angles. As any comprehensive account of correlated phases must consider such strong renormalization of the spectra by electrostatic interactions 17 ; 80 ; 77 ; 102 ; 41 , we conclude that TSBG has the potential to be a platform for new and rich correlated phenomena. Finally, we analyze the influence of strain on the topology of the narrow bands around the magic angle. By introducing a small mass term that breaks the inversion symmetry, we compute the valley Chern number C of the narrow bands for different strain magnitudes and directions. Due to the strain effect in both increasing the narrow bandwidth and reducing the remote bandgap, there are topological ( C=\pm 1 ) to trivial ( C=0 ) transitions as the strain increases, with a non-trivial dependence on the strain direction. We show that these topological transitions take place when the narrow bands close their gap with the remote bands. In the noninteracting case, we find that the sum of Chern number in top and bottom narrow band is always zero, i.e., they are both topological or both trivial. However, upon taking into account the electrostatic interactions we find asymmetrical topological transitions, whereby one narrow band can be topological while the other is trivial. We associate this behavior to the asymmetrical renormalization of the narrow bands due to the Hartree potential, which implies that they close their gap with the remote bands at different strain magnitudes. The paper is organized as follow: In Sec. II , we introduce a general formalism for moiré commensurability with any twist and strain, and discuss the Dirac point shifts due to the moiré deformation. In Sec. III , we calculate the electronic structures by using the TB and continuum methods, including the uniaxial strain, shear strain and lattice relaxations. These two models yield results in excellent agreement. Then, we investigate the narrow band modulation by both twist and strain. In Sec. IV , we study the effect of the electronic interactions. The band topology with strain is investigated in Sec. V . Our conclusions follow in Sec. VI . II Moiré commensurability with twist and strain In order to employ the TB method to obtain the band structure, we first require a commensurate structure. Although for only twist configurations the set of commensurate twist angles is well known, there is no close expression for the set of twist and strain that give commensurate structures. In what follows we will explicitly describe a global method to construct a commensurate structure of TSBG for a given twist angle and strain, taking into account different types of strain observed in experiments. II.1 Structural analysis of the moiré patterns II.1.1 Commensurate structures with twist We consider two graphene layers rotated in the plane by an angle \theta , with the rotation origin at the AA site, and assume the constructed supercell is commensurate with only one moiré pattern 66 . Each moiré pattern contains three different high-symmetry stackings, namely AA, AB and domain wall (DW) stackings, similar to the strained configuration shown in Fig. 1 (a). For the non-strain case, the rotation angle for the commensurate condition is 59 ; 100 \cos\theta=\frac{3i^{2}+3i+1/2}{3i^{2}+3i+1},\quad i=0,1,2\ldots (1) where i is an integer. The commensurate supercell vectors are \displaystyle\mathbf{L}_{1} \displaystyle=i\mathbf{a}_{1}+\left(i+1\right)\mathbf{a}_{2}, \displaystyle\mathbf{L}_{2} \displaystyle=-\left(i+1\right)\mathbf{a}_{1}+\left(2i+1\right)\mathbf{a}_{2}, (2) where \mathbf{a}_{1}=a(1,0) and \mathbf{a}_{2}=a(1/2,\sqrt{3}/2) are the lattice vectors of monolayer graphene, with a\simeq 2.46\text{\AA} being the lattice constant. Therefore, a commensurate supercell with twist angle \theta and vectors \mathbf{L}_{1,2} is exclusively identified by the integer i . When strain is introduced into the system, Eqs. ( 1 ) and ( II.1.1 ) become invalid, and a general formalism is needed to determine the twist, strain and moiré vectors of the commensurate structure. Figure 1: (a) Moiré pattern of bilayer graphene with a twist angle \theta=3.89{}^{\circ} and uniaxial strain \epsilon_{u}=3.7\% . The atoms in the top and bottom layers are plotted with blue and red dots, respectively. The AA, AB and DW stacking regions are labeled. Due to the strain the AA regions are elliptical. (b) Illustration of the moiré Brillouin zone, with the reciprocal lattice vectors labeled by \mathbf{G}_{1} and \mathbf{G}_{2} . We label six corner K points. The red dashed line is the momentum path for the band structure plots. (c), (d), (e) Schematically show the uniaxial, biaxial and shear strains, respectively. Only the top layer (blue line) is deformed. The undeformed bottom layer is plot with red lines. (f) Schematic construction of the moiré cell with the lattice vectors of the two graphene layers. (g) Schematic construction of the moiré cell in three steps by following the Eq. ( 12 ) with \mathcal{E}\to\mathcal{E}_{b}+\mathcal{E}_{u/s} : (1) an isotropic rescaling corresponding to \epsilon_{b} ; (2) an anisotropic rescaling corresponding to \epsilon_{u} in a direction given by \phi ; (3) a rotation by an angle \theta . We only plot the first two steps in (g). II.1.2 Uniaxial, shear and biaxial strains We introduce three important strains that are reported by experiments. An arbitrary two-dimensional strain tensor is given by \mathcal{E}=\begin{pmatrix}\epsilon_{xx}&\epsilon_{xy}\\ \epsilon_{xy}&\epsilon_{yy}\end{pmatrix}, (3) where \epsilon_{ij}=\left(\partial u_{i}/\partial x_{j}+\partial u_{j}/\partial x_{i}\right)/2 , with \mathbf{u} the displacement vector that accounts for the deformation induced by stress 54 ; 35 . The three relevant types of strain that we shall consider are uniaxial, shear and biaxial (Figs. 1 (c)-(e)). Their corresponding strain tensors read 79 ; 8 ; 49 ; 22 \displaystyle\mathcal{E}_{u} \displaystyle=R_{\phi}\left(\begin{array}[]{cc}\epsilon&0\\ 0&-\nu\epsilon\end{array}\right)R_{-\phi} \displaystyle\left(\mathrm{uniaxial}\right), \displaystyle\mathcal{E}_{s} \displaystyle=R_{\phi}\left(\begin{array}[]{cc}0&\epsilon\\ \epsilon&0\end{array}\right)R_{-\phi} \displaystyle\left(\mathrm{shear}\right), \displaystyle\mathcal{E}_{b} \displaystyle=\left(\begin{array}[]{cc}\epsilon&0\\ 0&\epsilon\end{array}\right) \displaystyle\left(\mathrm{biaxial}\right). Here R_{\phi}\equiv R\left(\phi\right) is the rotation matrix, \epsilon is the strain magnitude, \phi is the strain direction relative to the x axis, and \nu is the Poisson’s ratio ( \nu\simeq 0.16 in graphene). The shear strain can be written as \mathcal{E}_{s}=R_{\phi+\pi/4}\left(\begin{array}[]{cc}\epsilon&0\\ 0&-\epsilon\end{array}\right)R_{-\phi-\pi/4}. (10) Comparing with the uniaxial strain tensor we then have the relation \mathcal{E}_{s}\left(\epsilon,\phi\right)=\mathcal{E}_{u}\left(\epsilon,\phi+\pi/4,\nu\rightarrow 1\right). (11) This means that a shear strain with magnitude \epsilon and direction \phi_{s} can be thought as uniaxial strain with magnitude \epsilon and direction \phi_{u}=\phi_{s}+\pi/4 , but with Poisson’s ratio \nu\rightarrow 1 (i.e., the limit case in which the lateral contraction is equal to the applied longitudinal tension). In this paper, we restrict ourselves to TBG with a heterostrain, which refers to relative strains between two layers. Specifically, we assume that the top layer is strained and the bottom layer is rotated. Note that our definition of heterostrain differs from that in Refs. 8 ; 22 , where the two layers are strained in opposite directions with equal magnitude, i.e. a symmetric configuration. Instead, our definition is closer to the experimental cases in Ref. 42 ; 48 , where the concept of heterostrain was first introduced. These two types of heterostrain show slightly different modification of the geometrical and electronic properties of the moiré systems. However, these two definitions induce practically the same perturbation effects to the TBG around the first magic angle and are indistinguishable in both theory 23 and experiment 108 . II.1.3 Moiré geometry with twist and strain An application of strain in the top layer, and a rotation in the bottom layer, transform their lattice vectors as \displaystyle\mathbf{a}_{t,i} \displaystyle=\left(\mathbb{I}+\mathcal{E}\right)\mathbf{a}_{i}, \displaystyle\mathbf{a}_{b,i} \displaystyle=R_{\theta}\mathbf{a}_{i}, (12) where \mathbb{I} is the 2\times 2 identity matrix. The reciprocal vectors follow as \displaystyle\mathbf{b}_{t,i} \displaystyle=\left(\mathbb{I}+\mathcal{E}\right)^{-1}\mathbf{b}_{i}, \displaystyle\mathbf{b}_{b,i} \displaystyle=R_{\theta}\mathbf{b}_{i}, (13) where \mathbf{b}_{i} are the reciprocal lattice vectors of the honeycomb lattice. The reciprocal moiré vectors \mathbf{G}_{i} (see Fig. 1 (b)) can then be calculated by taking the difference between the deformed lattice vectors in each layer 5 ; 49 ; 22 ; 52 \mathbf{G}_{i}=\mathbf{b}_{t,i}-\mathbf{b}_{b,i}. (14) The real space moiré vectors \mathbf{L}_{i} are determined by the relation \mathbf{L}_{i}\cdot\mathbf{G}_{j}=2\pi\delta_{ij} . In principle, the moiré vectors \mathbf{G}_{i} and \mathbf{L}_{i} define the moiré structure 49 ; 8 . However, in practice, the definition of \mathbf{L}_{i} does not guarantee commensurability in the supercell. II.1.4 Commensurate structures with twist and strain In this part, we develop a general geometrical formalism for commensurate structures in TSBG. The analysis of the commensurate supercell can be performed by expressing the moiré lattice vectors as function of those of the two graphene layers, as shown in Fig. 1 (f) \begin{pmatrix}\mathbf{L}_{1}\\ \mathbf{L}_{2}\end{pmatrix}=\begin{pmatrix}i&j\\ k&l\end{pmatrix}\begin{pmatrix}\mathbf{a}_{t,1}\\ \mathbf{a}_{t,2}\end{pmatrix}=\begin{pmatrix}m&n\\ q&r\end{pmatrix}\begin{pmatrix}\mathbf{a}_{b,1}\\ \mathbf{a}_{b,2}\end{pmatrix}. (15) Then the top and bottom graphene lattice can be related by a Park-Madden transformation matrix \begin{aligned} \begin{pmatrix}\mathbf{a}_{t,1}\\ \mathbf{a}_{t,2}\end{pmatrix}&=\frac{1}{il-jk}\begin{pmatrix}lm-jq&ln-jr\\ -km+iq&-kn+ir\end{pmatrix}\begin{pmatrix}\mathbf{a}_{b,1}\\ \mathbf{a}_{b,2}\end{pmatrix}\\ &=\begin{pmatrix}a&b\\ c&d\end{pmatrix}\begin{pmatrix}\mathbf{a}_{b,1}\\ \mathbf{a}_{b,2}\end{pmatrix}\end{aligned}. (16) For an arbitrary transformation matrix composed of contributions of twist and strain, there are four variables ( a , b , c , d ) corresponding to a pair of eight integers ( i,j,k,l,m,n,q,r ) that define the Park-Madden matrix. The set of eight integers can be determined experimentally through atomically-resolved microscopy 42 ; 5 . The analysis of the commensurate supercell also can be expressed as a function of the elementary geometrical deformations in Eq. ( 12 ), which in general depends on four parameters: the twist angle and the three components of the strain tensor. These four parameters completely determine, in principle, the (2 \times 2) matrix in Eq. ( 16 ). If, for simplicity, one assumes that the system contains only uniaxial or shear strain, then one is left with only three parameters, namely, the twist and the strain magnitude and direction. Therefore, we introduce an additional - extremely small - biaxial strain in the top layer, so that Eq. ( 12 ) becomes \mathbf{a}_{t,i}\rightarrow\left(\mathbb{I}+\mathcal{E}_{u/s}+\mathcal{E}_{b}\right)\mathbf{a}_{i} [see Fig. 1 (g)]. Then the system has four parameters ( \theta,\epsilon_{u/s},\phi,\epsilon_{b} ) that can fully determine the set of eight integers in Eq. ( 16 ). To obtain the commensurate solutions, it is convenient to rewrite \mathbf{a}_{t,i}\simeq\left(\mathbb{I}+\mathcal{E}_{b}\right)\left(\mathbb{I}+\mathcal{E}_{u/s}\right)\mathbf{a}_{i} by taking \mathcal{E}_{b}\mathcal{E}_{u/s}\rightarrow 0 under the limit of small deformations. The transformation matrix that links \left(\mathbf{a}_{t,1},\mathbf{a}_{t,2}\right) with \left(\mathbf{a}_{b,1},\mathbf{a}_{b,2}\right) can then be written as \left(\begin{array}[]{c}\mathbf{a}_{t,1}\\ \mathbf{a}_{t,2}\end{array}\right)=P\left(1+\epsilon_{b}\right)\left(\mathbb{I}+\mathcal{E}_{u/s}\right)R_{\theta}P^{-1}\left(\begin{array}[]{c}\mathbf{a}_{b,1}\\ \mathbf{a}_{b,2}\end{array}\right), (17) where P=\left(\begin{array}[]{cc}1&0\\ 1/2&\sqrt{3}/2\end{array}\right) is the basis of the triangular lattice vectors. By combining then Eqs. ( 16 ) and ( 17 ) we can relate the physical parameters ( \theta,\epsilon_{u/s},\phi,\epsilon_{b} ) to the eight integers ( i,j,k,l,m,n,q,r ) 5 . The described procedure to obtain commensurate structures can be generalized to any arbitrary strain tensor, with the inclusion of lattice relaxation effects 53 . A detailed, step-by-step algorithm of how to obtain commensurate structures with any twist and strain, with some example solutions, can be found in the Secs. S1 and S2 . II.2 Dirac point shifts from moiré geometry deformation When the strain is small, we can consider that the length of the two moiré vectors remains, to a first approximation, practically equal. In that case, the borders of the mBZ are given by the three points (see Fig. 1 (b)) 22 : \displaystyle\mathbf{K}_{1} \displaystyle=-\frac{(1+2\chi)\mathbf{G}_{1}-\lambda\mathbf{G}_{2}}{2(1+\chi)}, \displaystyle\mathbf{K}_{3} \displaystyle=\mathbf{K}_{1}+\mathbf{G}_{1}, \displaystyle\mathbf{K}_{5} \displaystyle=\mathbf{K}_{1}+\mathbf{G}_{1}-\lambda\mathbf{G}_{2}, (18) and their negatives. Here \chi=\absolutevalue{\vb{G}_{1} \cdot\vb{G}_{2}}/\absolutevalue{\vb{G}_{1} \cdot\vb{G}_{1}} and \lambda=\mathrm{sign}(\mathbf{G}_{1}\cdot\mathbf{G}_{2})+\delta_{0,\mathbf{G}_{1}\cdot\mathbf{G}_{2}} . When the angle between \mathbf{G}_{1} and \mathbf{G}_{2} is 120{}^{\circ} , the moiré pattern is triangular and the points above collapse to yield a hexagonal mBZ. Figure 2: (a) Position of the Dirac points projected within the mBZ, for the commensurate solutions of magic angle \theta\sim 1.05^{\circ} without strain. (b) The same case as (a) for uniaxial strain \epsilon_{u}\sim 0.1\% . (c) The same case as (a) for shear strain \epsilon_{s}\sim 0.1\% . In each case, plots on the left and middle sides show the Dirac points in the top (red) and bottom (blue) layers, and the mBZ periodically translated from the origin for the K and K^{\prime} valleys, respectively. Plots on the right side show the position of the Dirac points within the mBZ, when translated by the moiré vectors. In (b) and (c), the positions of Dirac points from the top strained layer are slightly shifted from the five possible positions in the mBZ. Next, we check how the undeformed Dirac points, at the corners of the graphene BZ, are mapped into the mBZ. First we translate Eq. ( 15 ) to the reciprocal space as: \displaystyle\left(\begin{array}[]{c}\mathbf{b}_{t,1}\\ \mathbf{b}_{t,2}\end{array}\right) \displaystyle=\left(\begin{array}[]{cc}i&j\\ k&l\end{array}\right)\left(\begin{array}[]{c}\mathbf{G}_{1}\\ \mathbf{G}_{2}\end{array}\right), \displaystyle\left(\begin{array}[]{c}\mathbf{b}_{b,1}\\ \mathbf{b}_{b,2}\end{array}\right) \displaystyle=\left(\begin{array}[]{cc}m&n\\ q&r\end{array}\right)\left(\begin{array}[]{c}\mathbf{G}_{1}\\ \mathbf{G}_{2}\end{array}\right). The projection of a Dirac point at \mathbf{K}_{0}=\mathbf{b}_{1}/3+2\mathbf{b}_{2}/3 is then given by: \displaystyle\mathbf{K}_{0} \displaystyle=\left(\begin{array}[]{cc}\mathbf{G}_{1}&\mathbf{G}_{2}\end{array}\right)\left(\begin{array}[]{c}\frac{1}{3}i+\frac{2}{3}j\\ \frac{1}{3}k+\frac{2}{3}l\end{array}\right) \displaystyle=\left(\begin{array}[]{cc}\mathbf{G}_{1}&\mathbf{G}_{2}\end{array}\right)\left(\begin{array}[]{c}\frac{1}{3}m+\frac{2}{3}n\\ \frac{1}{3}q+\frac{2}{3}r\end{array}\right). For different pairs of ( i , j , k , l , m , n , q , r ), there are five different types of projections: (0,0) , (0,\frac{1}{3}) , (\frac{1}{3},0) , (\frac{1}{3},\frac{1}{3}) , (\frac{1}{3},\frac{2}{3}) . These projections will always fall into the five high-symmetry points. The situation under strain changes because the borders of the mBZ are no longer located at \mathbf{G}_{1}/3+2\mathbf{G}_{2}/3 (and translations by reciprocal moiré vectors), as in the non-strain case. Consequently, the Dirac points in the mBZ are rather located at arbitrary, strain-dependent positions. Some examples of the geometrical positions of the Dirac points in TSBG are plotted in Figs. 2 (b)-(c). The Dirac points are effectively away from the corners of the mBZ, and are no longer degenerate. It should be noted that our analysis here only accounts for the geometrical position of the Dirac points in each layer, i.e., the projection of the borders of their respective BZ. As we will discuss in the following sections, under strain and relaxation the Dirac points are also sightly shifted by strain-induced fields. Moreover, due to the broken symmetries under strain, the actual position of the moiré Dirac points is further influenced by the moiré potential that couples the two layers 92 ; 22 ; 23 . In the particular non-strain case, the commensurate condition is given by Eq. ( 1 ), and Eq. ( II.2 ) becomes \mathbf{K}_{0}=\left(\begin{array}[]{cc}\mathbf{G}_{1}&\mathbf{G}_{2}\end{array}\right)\left(\begin{array}[]{c}2/3\\ 1/3\end{array}\right), (37) which means that if there is no strain, the Dirac cones are always projected to the corners of the mBZ (Fig. 2 (a)). As discussed in the following section, the shift of the Dirac cones and the deformation of the mBZ under strain explain why, in the TB calculations, the K_{1} , K_{2} , K_{3} , K_{4} and \Gamma points appear displaced compared to the unstrained case. II.3 Numerical models We first compute the electronic structure of commensurate TSBG by using a full atomistic TB Hamiltonian. Then, we construct a general effective continuum model to describe the TB results. A key step for comparing the TB and continuum results is identifying the Dirac points in the strained mBZ, discussed in Sec. II.2 (see also Fig. 2 ). II.3.1 Tight-binding model The TB Hamiltonian of the TSBG is generated by only considering the p_{z} orbital of the carbon atom as 100 H=\sum_{i}\varepsilon_{i}c^{\dagger}_{i}c_{i}+\sum_{\langle i,j\rangle}t_{ij}c^{\dagger}_{i}c_{j} (38) where c_{i} is an annihilation operator for the i state, \varepsilon_{i} is the on-site potential, \langle i,j\rangle is the sum over index with i\neq j , t_{ij} is the hopping integral between i and j orbitals, which follows the Slater-Koster (SK) formalism. As detailed above, including a very small biaxial strain guarantees commensurate solutions for arbitrary twist and uniaxial/shear strain, from which we calculate the bands. For conciseness, in the main text we quote the approximate twist and strain values of the commensurate solutions. The exact strain and twist parameters are listed in Sec. S2 . We identify the valley character with a valley operator 83 ; 84 , and relax the moiré supercell with LAMMPS 81 . Details of the TB calculations are given in Sec. S3 . II.3.2 Strain-extended continuous model In the continuum model, the electronic properties of the system are accounted by the coupling of the Dirac points in each layer with an effective moiré-induced interlayer potential. Neglecting couplings between different valleys in each layer, the continuum model Hamiltonian for the K valley takes the form 42 ; 8 ; 76 H=\left(\begin{array}[]{cc}h_{b}\left(\mathbf{k}\right)+\mathcal{S}_{b}&U^{\dagger}\left(\mathbf{r}\right)\\ U\left(\mathbf{r}\right)&h_{t}\left(\mathbf{k}\right)+\mathcal{S}_{t}\end{array}\right), (39) where the index \ell=b,t refers to the bottom and top layers, respectively. h_{\ell}\left(\mathbf{k}\right) is the Dirac Hamiltonian relative to the twisted and strained Dirac points: h_{\ell}\left(\mathbf{k}\right)=-\hbar v\boldsymbol{\sigma}\cdot R_{\theta_{\ell}}\left(1+\mathcal{E}_{\ell}\right)\left(\mathbf{k}-\mathbf{K}_{\ell}\right), (40) where \boldsymbol{\sigma}=\left(\sigma_{x},\sigma_{y}\right) are the Dirac matrices and \mathbf{K}_{\ell}=\left(1-\mathcal{E}_{\ell}\right)R_{\ell}\left(\theta_{\ell}\right)\mathbf{K}_{0} , where \mathbf{K}_{0}=-\left(2\mathbf{b}_{1}+\mathbf{b}_{2}\right)/3 is the Dirac point of a honeycomb layer. The strain introduces an additional term \mathcal{S}_{\ell} that includes a deformation and gauge potential 96 ; 63 ; 101 ; 72 ; 73 \mathcal{S}_{\ell}=\mathbb{I}V_{\ell}-\hbar v\boldsymbol{\sigma}\cdot R_{\theta_{\ell}}\left(1+\mathcal{E}_{\ell}\right)\mathbf{A}_{\ell}, (41) where \displaystyle V_{\ell} \displaystyle=g\left(\epsilon_{xx}^{\ell}+\epsilon_{yy}^{\ell}\right), (42) \displaystyle\mathbf{A}_{\ell} \displaystyle=\frac{\sqrt{3}}{2a}\beta\left(\epsilon_{xx}^{\ell}-\epsilon_{yy}^{\ell},-2\epsilon_{xy}^{\ell}\right), (43) with g=4 eV and \beta=3.14 for graphene 101 ; 8 . The scalar potential V_{\ell} shifts the Dirac points in energy, resembling the effect of a perpendicular electric field. The vector potential \mathbf{A}_{\ell} shifts the Dirac points in momentum and accounts for the strain-induced change in the hopping energies within the Dirac approximation 96 ; 63 . The moiré-induced coupling potential U\left(\mathbf{r}\right) depends on the interplay between twist and strain through its Fourier expansion in terms of the moiré vectors 69 ; 52 ; 22 . At small deformations (i.e., low twist and strain) the Fourier expansion can be truncated to the first three leading order terms 22 : U\left(\mathbf{r}\right)=U_{1}+U_{2}e^{i\mathbf{G}_{1}\cdot\mathbf{r}}+U_{3}e^{i\left(\mathbf{G}_{1}+\mathbf{G}_{2}\right)\cdot\mathbf{r}}, (44) where U_{j}=\left(\begin{array}[]{cc}u_{0}&u_{1}e^{-\omega_{j}}\\ u_{1}e^{\omega_{j}}&u_{0}\end{array}\right), (45) with \omega_{j}=\left(j-1\right)2\pi/3 . Here u_{0} and u_{1} are the effective AA and AB/BA hopping amplitudes. The values of these hopping energies are expected to depend on the local lattice deformations. In general, a rigid configuration implies equal hopping energies u_{0}=u_{1} . This results in the remote bands always touching the flat middle bands, i.e., there is no gap between them 9 ; 69 . However, relaxation effects tend to shrink and increase the interlayer distance of the energetic AA regions, compared to the most favorable AB/BA, thus leading to an effective smaller hopping u_{0}\mathrm{eV} , u_{1}=0.096\,\mathrm{eV} and u_{0}=0.05952\,\mathrm{eV} . In line with the TB results, the continuum model reflects that the strain effectively increases the bandwidth of the narrow bands around the magic angle. Nevertheless, there is always a minimum bandwidth depending on the twist and strain. Thus, one can still identify potential twist and strain configurations at which electronic correlations could be maximized. Here, we shall particularly focus on the identification of the twist angle at which the bandwidth of the narrow bands is minimum. Figure 7 shows the bandwidth evolution as a function of the twist angle \theta and the uniaxial strain magnitude \epsilon , with different directions \phi , for the relaxed configuration of continuum model parameters. In general, we observe that the twist angle at which the bandwidth is minimum tends to shift in the presence of strain. The shift is nonuniform and depends non-trivially on the strain direction. Thus, we observe that as the strain magnitude increases, the magic angle tends to increase when \phi=0^{\circ} , but it tends to decrease when \phi=30^{\circ} . For shear strain one obtains a similar behavior as in Fig. 7 , only that the effect is stronger for the same strain magnitude, and the dependence with the strain direction is shifted [cf. Eq. ( 11 )]. Remarkably, the minimum bandwidth always seems to follow a linear dependence with the strain magnitude 8 . We have checked that this behavior persist even without the strain-fields (see Fig. S14 ). The main difference then is that the bandwidth evolution becomes almost insensitive to the strain direction, but scales linearly with the strain magnitude. Such linear dependence of the bandwidth is roughly due to the linear dependence of the moiré vector with the strain strength [see Eq. ( 14 )], which introduce the strain effect through the moiré potential given by Eq. ( 44 ). It should be noted that under strain a minimum bandwidth does not necessarily correlate to a higher DOS. This is because with strain the narrow bands are not, in general, uniformly flat over the whole mBZ. A band at a particular twist and strain configuration can have, for instance, a larger bandwidth than at other configuration, but yet be flatter over a wider region of the mBZ. Consequently, the configuration with higher bandwidth would still have higher vHs. In this sense, a minimum bandwidth should only be considered as an indicator for the appearance of strong electronic correlations. Although a nonzero strain tends to increase the bandwidth, it is crucial such an increase depends on the twist angle. Since most samples are, at least, likely to inherit some kind of random strain 20 ; 48 ; 40 ; 15 (e.g., due to their fabrication method), which can vary from sample to sample, our results highlight that the notion of magic angle is intrinsically connected to the experimental conditions of the system. IV Strain and electrostatic interactions Our analysis so far has neglected the role of electron-electron interactions in the system. However, these interactions are actually crucial in the narrow band regime due to the quenching of the kinetic energy. In this section we will address, in particular, the role of the electrostatic interactions within the continuum model, as accounted by the Hartree potential 34 ; 18 ; 32 . Our main interest will be the effect of electrostatic interactions on the bandwidth and charge density of the twisted and strained bilayer configurations. The Hartree interaction is the direct (classical) interaction of an electron with the surrounding charge density: V_{H}\left(\mathbf{r}\right)=\int d\mathbf{r}^{\prime}v_{C}\left(\mathbf{r}-\mathbf{r}^{\prime}\right)\delta\rho\left(\mathbf{r}^{\prime}\right), (49) where v_{C}\left(\mathbf{r}-\mathbf{r}^{\prime}\right) is the bare Coulomb potential and \delta\rho\left(\mathbf{r}^{\prime}\right) is the electronic charge density with respect to CNP. Replacing the plane-wave expansion of the Bloch states in TSBG leads to \displaystyle V_{H}\left(\mathbf{r}\right) \displaystyle=\sum_{\mathbf{G}\neq 0}V_{H}\left(\mathbf{G}\right)e^{-i\mathbf{G}\cdot\mathbf{r}}, (50) \displaystyle V_{H}\left(\mathbf{G}\right) \displaystyle=\frac{v_{C}\left(\mathbf{G}\right)}{A_{c}}\sum_{\mathbf{k},\mathbf{G}^{\prime}}\sum_{n,\eta,i}^{\prime}u_{n,\mathbf{k},\eta,i}^{*}\left(\mathbf{G}^{\prime}+\mathbf{G}\right)u_{n,\mathbf{k},\eta,i}\left(\mathbf{G}^{\prime}\right), (51) where u_{n,\mathbf{k},\eta,i}\left(\mathbf{G}\right) are the Fourier coefficients of the band, valley/spin and layer/sublattice indices n,\eta,i , respectively, and v_{C}\left(\mathbf{G}\right) is the Fourier transform of the bare Coulomb potential (see Sec. S12 for details). We consider a gated configuration of two metallic plates 7 , for which v_{C}\left(\mathbf{G}\right)=e^{2}\tanh\left(d\left|\mathbf{G}\right|\right)/2\varepsilon_{0}\varepsilon_{r}\left|\mathbf{G}\right| , where d is the distance between the two metallic plates, and \varepsilon_{r} is the relative primitivity of the system. For the numerical calculations we set d=40\,\mathrm{nm} . Note that the \mathbf{G}=0 term in V_{H}\left(\mathbf{r}\right) is neglected because it is canceled by the background positive charge (jellium model) 34 ; 82 . Figure 8: Evolution of the band structure and DOS as a function of the electrostatic interactions (self-consistent Hartree), from the non-interacting case at charge neutrality (filling \nu=0 ), to a filling \nu=2 with different dielectric constants \varepsilon_{r} . Panel (a) shows the results for the no strain case at \theta=1.05^{\circ} , while panel (b) shows the results for \theta=1.05^{\circ} and uniaxial strain with magnitude \epsilon=0.1\% and direction \phi=30^{\circ} . In all cases, the horizontal gray and green lines indicate the CNP and the Fermi level, respectively. In the band structures, the solid and dashed lines correspond to the K and K^{\prime} valleys. All the results are for the relaxed configuration with continuum model parameters \hbar v/a=2.13\>\mathrm{eV} , u_{1}=0.096\,\mathrm{eV} and u_{0}=0.0592\,\mathrm{eV} . Figure 9: Evolution from a nonrigid to a rigid Hartree effect as the strain increases. Panel (a) shows the continuum model density plot of the top narrow band for \theta=1.05^{\circ} and uniaxial strain with direction \phi=20^{\circ} and increasing strain magnitudes \epsilon . The top and bottom density panels show, for each strain magnitude, the bands without Hartree and with Hartree (filling \nu=2 and \varepsilon_{r}=7 ), respectively. Panel (b) shows the bandwidth in the top and bottom narrow bands, with and without Hartree, as a function of the uniaxial strain magnitude \epsilon with direction \phi=20^{0} ; other parameters as in panel (a). The colormap indicates, schematically, the nonrigid to rigid transition as the strain increases and the bandwidths become similar. In unstrained TBG, the moiré pattern is perfectly triangular, and the largest Fourier components of the Hartree potential correspond to the first six reciprocal vectors of equal magnitude 34 ; 18 . Owing to the \mathcal{C}_{3} symmetry, each reciprocal vector contributes equally to the charge density and the Hartree potential, which therefore follow the same spatial profile 18 . Under strain, however, the moiré pattern becomes distorted and loses its triangular symmetry 49 ; 22 , making the contributions from different Fourier components inequivalent. As a result, the Hartree potential is no longer proportional to the charge density. In our numerical calculations, we therefore include all Fourier components V_{H}(\mathbf{G}) within the reciprocal moiré vectors of the continuum model. The Hartree potential is diagonal in the valley/spin and sublattice/layer flavors, with matrix elements 16 \bra{\mathbf{k}+\mathbf{G}'-\mathbf{G},\eta',i'}\hat{V}_{H}\ket{\mathbf{k}+\mathbf{G}',\eta,i}=\delta_{\eta\eta^{\prime}}\delta_{ii^{\prime}}V_{H}\left(\mathbf{G}\right). (52) Since the Hartree potential depends on the occupied Bloch states from CNP, the total Hamiltonian \hat{H}=\hat{H}_{0}+\hat{V}_{H} is solved self-consistently, for different filling factors, up until convergence. Figure 8 shows the numerical results for the evolution of the band structure and density of states, from CNP ( \nu=0 ), to a filling of \nu=+2 (two electrons per moiré unit cell), for different dielectric constants \varepsilon_{r} . The results correspond to \theta=1.05^{\circ} with no strain (top panels), and uniaxial strain with magnitude \epsilon=0.1\% and direction \phi=30^{\circ} (bottom panels). In both cases, we observe that as the dielectric constant diminish and the Hartree potential increases, the band structure is strongly reshaped 34 ; 18 ; 32 ; 25 . Interestingly, by comparing the nonstrain with the strain case, we observe that the interacting band structures end up having similar bandwidths, despite having quite different single-particle bandwidth. This is because the Hartree potential diminish as the bare active bands increase their bandwidth, essentially due to the decrease in the quenching of the kinetic energy. As a result, under strain there is a competition between the increase of the non-interacting bandwidth and the decrease of the Hartree potential. The synergy between both effects determines the effective bandwidth under strain and electrostatic interactions. It may actually be that under strain the net bandwidth becomes comparable, or even smaller, than the one corresponding to the no-strain scenarios. As the strain increases, the Hartree effect not only weakens but also evolves into a nearly uniform energy shift. In other words, it changes from a non-rigid shift in the unstrained case, to an almost rigid (and small) shift at larger strains. This behavior indicates that the charge density across different momentum points becomes increasingly uniform as strain grows. Figure 9 illustrates this transition from non-rigid to rigid behavior, for the case of uniaxial strain. While the non-interacting and interacting bands differ noticeably at low strain, they become nearly identical at higher strain values, implying that the Hartree potential becomes effectively negligible. The strain threshold for this transition depends non-trivially on the strain direction (and, more generally, on the strain type). It also varies with the filling factor because of the asymmetric Hartree renormalization of the conduction and valence bands, depending on whether the system is electron- or hole-doped. Figure 10: Strain-induced topology evolution of the narrow bands, for the non-interacting case at \theta=1.05^{\circ} with (a) uniaxial heterostrain and (b) shear strain. In each case, the top left panel show the Chern number in the top and bottom narrow bands as a function of the strain direction \phi and magnitude \epsilon . For all parameters, C=+1\left(-1\right) for the top (bottom) bands. The top right panel shows the gap between the top narrow bands and the remote bands. The minimum gap, at which it closes and then re-open, corresponds to the topological transitions \left|C\right|=1\rightarrow 0 in (a); the white crosses indicate the regions where the narrow bands are trivial ( C=0 ). Figure 11: Strain-induced topology evolution of the top and bottom narrow bands with electrostatic interaction (self-consistent Hartree potential), at \theta=1.05^{\circ} with uniaxial heterostrain, for a filling of \nu=+2 electrons per moiré unit cell and dielectric constant \varepsilon_{r}=7 . The top (bottom) panel shows the Chern number in the top (bottom) narrow band, and their corresponding gap to the remote bands, as a function of the uniaxial strain direction \phi and magnitude \epsilon ; the white crosses in the remote band gap indicate where the narrow band is trivial ( C=0 ). Due to the electrostatic interactions, the topological transitions \left|C\right|=1\rightarrow 0 at which the remote band gap closes and the reopens is different for each narrow band, leading to strain configuration at which only one band topological (or trivial). V Band Topology with Strain The strain-induced reshaping of the electronic properties in TSBG is expected to influence the valley-dependent topology of the band structure. Previous studies have indeed reported a rich topological phase diagram arising from the interplay between twist and strain, both in TMDs 8 and in bilayer graphene 75 ; 76 . However, a detailed analysis of how the topology of the narrow bands in TSBG evolves with different types of strain is still lacking. Since strain preserves time-reversal symmetry, the Chern numbers of opposite valleys remain equal in magnitude and opposite in sign, yielding an overall zero Chern number, i.e., a topologically trivial system. Nevertheless, each valley can still host non-trivial topology. A clear signature of such valley topology, even without breaking time-reversal symmetry, is the nonlinear Hall effect 93 ; 60 , which can be used to probe the topological character of the narrow bands 91 . In what follows, we therefore focus on the valley-resolved topology of the TSBG band structure. We study the topology evolution of the bands by computing the Chern number \mathcal{C}_{n}=\frac{1}{2\pi}\int\boldsymbol{\Omega}_{n}\left(\mathbf{k}\right)\cdot d\mathbf{k}, (53) where \boldsymbol{\Omega}_{n}\left(\mathbf{k}\right)=i\left\langle\partial_{\mathbf{k}}\psi_{n\mathbf{k}}\right|\times\left|\partial_{\mathbf{k}}\psi_{n\mathbf{k}}\right\rangle is the Berry curvature of the n -band Bloch states \psi_{n\mathbf{k}} , d\mathbf{k} is a reciprocal-space surface vector, and the integration is over a moiré unit cell. To obtain \mathcal{C}_{n} numerically we use the Fukui-Hatsugai-Suzuki method 27 , considering different strain parameters. For comparison, we analyze separately both the noninteracting and the interacting cases with electrostatic interactions. All the results are obtained using the continuum model for the relaxed configuration. The Chern numbers are computed for the top and bottom narrow bands by introducing a small mass term \sim m\sigma_{z} in the TSBG Hamiltonian that breaks the inversion symmetry and opens a gap at the Dirac points. V.1 Non-interacting case Figure 10 shows the results for the non-interacting case at \theta=1.05^{\circ} under uniaxial and shear strain with different directions \phi and magnitudes \epsilon . When interaction effects are neglected, the two narrow bands retain opposite Chern numbers of equal magnitude, so their total Chern number remains zero. In the absence of strain, the upper and lower narrow bands are topological with valley Chern numbers of +1 and -1, respectively. Upon introducing strain, a topological transition to trivial narrow bands ( C=0 ) occurs when the gap between the narrow and remote bands closes and reopens. This behavior arises because the strain increases the bandwidth of the narrow bands and simultaneously shifts the remote bands to lower energies (see Fig. 5 ). The critical point at which the gap closes depends nontrivially on the strain magnitude and direction, and, in general, also on the twist angle. Interestingly, the non-interacting band topology is preserved up to relatively large strain magnitudes ( \epsilon\sim 2\% ). Figure 12: (a) Charge density \rho_{\mathbf{k}}\left(\mathbf{r}\right) in the top narrow band and the closest remote band, at the momentum point \mathbf{k} of minimum direct gap between them, for twist angle \theta=1.05^{\circ} and uniaxial strain at \phi=20^{\circ} , with magnitudes \epsilon=0.75\% and \epsilon=0.8\% . The transition from a topological ( C=1 ) to trivial ( C=0 ) narrow band takes place between the two strain magnitudes (cf. Figure 10 ). When the transition occurs there is a charge density transfer from the top to the remote band. Panel (b) shows the noninteracting 3D band structure for the uniaxial strain magnitude \epsilon=0.8\% . The zoom inset highlights the minimum gap between the top narrow band and the remote band, at which the charge transfer takes place. The strain-induced touching between the narrow and the remote bands that triggers the topological transition is accompanied by a transfer of charge density. Figure 12 shows this transfer of charge density \rho_{\mathbf{k}}\left(\mathbf{r}\right) for uniaxial strain (non-interacting case), at the momentum point \mathbf{k} of minimum (direct) gap between the narrow and remote bands. As clearly seen, the charge density of the topological band ( C=1 ) is directly transferred to the remote band after the gap closes and the narrow band becomes trivial ( C=0 ). This charge transfer behavior occurs, in general, for any kind of twist and strain that induce a remote gap closing. The dependence on the specific type of strain is reflected in the profile of the charge density transferred (e.g, its symmetries), and the specific twist and strain parameters at which the topological transition takes place. For biaxial strain, for instance, the system retains the C_{3} symmetry and the topological transition involves the transfer of ring-like to AA centered charge densities between the top and remote bands (Sec. S13 ). V.2 Interacting case Figure 11 shows the topology evolution of the narrow bands after including electrostatic interactions. In contrast to the noninteracting case, we now see that by increasing the strain magnitude the topology of the top and bottom narrow bands becomes asymmetrical. That is, there are now strain configurations for which the sum of the top and bottom band Chern number is not zero. This is because for any nonzero filling the Hartree potential reshapes the top and bottom bands asymmetrically, cf. Fig. 8 . In general, for positive fillings \nu (electron-doped) the bottom narrow band is more strongly renormalized, and vice versa for negative fillings \nu . This leads to different strain parameters at which the gap with the remote bands closes and then reopens, and therefore, wider strain regimes in which only one band is topological. Compared to the noninteracting case of Fig. 10 , we particularly see that the Hartree potential shrinks and increase, respectively, the regimes where the bottom and top bands are topological. Since the Hartree potential is practically symmetric with respect to charge neutrality 34 ; 18 , the regime in which only one band is topological is reversed when the system is hole-doped ( \nu<0 ). Perturbation Topology Change