Orbital piezomagnetic polarizability of pure insulating altermagnets in two dimensions Beryl Bell, J. W. F. Venderbos Abstract The distinctive symmetry properties of pure altermagnets make them natural candidates for piezomagnetism. Previous work motivated by the piezomagnetic properties of altermagnets has primarily focused on the spin magnetization response to applied strain. In this paper we study orbital piezomagnetic effects—the orbital magnetization response to applied strain—in minimal lattice models of pure insulating altermagnets in two dimensions. We obtain general microscopic expressions for the orbital magnetization in the presence of strain, as well as the orbital piezomagnetic polarizability, i.e., the defining response characteristic of pure altermagnets. We apply these expressions to three specific tetragonal lattice models, two corresponding to d d -wave altermagnets and one describing a g g -wave altermagnet. Whereas the d d -wave altermagnets are associated with a linear piezomagnetic polarizability, the g g -wave altermagnet exhibits a nonlinear piezomagnetic effect. Our analysis reveals how the polarizabilities are related to and determined by the Berry curvature of the occupied bands. Connections to materials of current interest are discussed. I Introduction Altermagnets are a category of magnetic materials that have attracted increasing attention in recent years Šmejkal et al. 2022a ; Šmejkal et al. 2022b ; Mazin and Editors 2022 ; Turek 2022 ; Mazin et al. 2021 ; Song et al. 2025 ; Jungwirth et al. 2025 ; Jungwirth et al. 2026 . Altermagnets have fully compensated collinear magnetic order with no net magnetization, which is similar to antiferromagnets, but nonetheless exhibit a Zeeman spin splitting of the energy bands reminiscent of ferromagnets Hayami et al. 2019 ; Bai et al. 2023 ; Yuan et al. 2021 ; Karube et al. 2022 ; Egorov and Evarestov 2021 ; Guo et al. 2023 ; Naka et al. 2019 ; Yuan and Zunger 2023 ; Hayami et al. 2020 ; González-Hernández et al. 2021 ; Naka et al. 2021 ; Shao et al. 2021 ; Ma et al. 2021 ; Bose et al. 2022 . This spin splitting is of non-relativistic origin and is caused by the particular crystallographic arrangement of the magnetic ions. The latter lacks the symmetries that enforce spin-degeneracy. In sharp contrast to ferromagnets, the energy band spin splitting in altermagnets is non-uniform and has a d -, g -, or i -wave nodal structure, which is directly linked to the symmetry and hence response properties of altermagnets Šmejkal et al. 2022a ; Šmejkal et al. 2022b . These defining characteristics can be viewed as intermediate between ferromagnetism and antiferromagnets, and have fueled renewed interest in the classification of magnets and their fundamental properties. They have further put altermagnets at the forefront of the spintronics frontier Song et al. 2025 . While the formal classification of altermagnetism is predicated on the non-relativistic limit Šmejkal et al. 2022a ; Liu et al. 2022 ; Jiang et al. 2024 ; Xiao et al. 2024 ; Chen et al. 2024 ; Schiff et al. 2025 , including the effect of spin-orbit coupling gives rise to a variety of interesting phenomena associated with altermagnetic order Fernandes et al. 2024 ; Šmejkal et al. 2022c . For instance, depending on the orientation of the Néel vector the symmetries of the magnetic space group can allow for weak ferromagnetism and an anomalous Hall effect Šmejkal et al. 2022c . Materials of this kind have been referred to as ‘mixed’ altermagnets, since the ordered moments are not strictly but only nearly compensated, giving rise to a net magnetization and a spin polarization of s -wave type Fernandes et al. 2024 . Instead, in ‘pure’ altermagnets the magnetic space group forbids both a net magnetization and an anomalous Hall effect (since the Hall vector transforms as the magnetization), and moments remain strictly compensated Fernandes et al. 2024 . Although an anomalous Hall effect is forbidden by symmetry, pure altermagnets can still show signatures of nontrivial topology, such as mirror Chern bands or topological nodal band crossings Antonenko et al. 2025 ; Fernandes et al. 2024 , and this raises the question whether the Berry curvature associated with these features can be unlocked and exploited by applying external fields. One way to achieve this is the application of external strain. The response of pure altermagnets to applied strain is directly tied to the nature of the magnetic order parameter, which, in the case of d -wave altermagnets, has the symmetry of a magnetic octupole McClarty and Rau 2024 ; Bhowal and Spaldin 2024 . As a result, the reduction of symmetry caused by strain can enable both a net magnetization and an anomalous Hall effect, which makes pure altermagnets natural candidates for harnessing piezomagnetism and for observing elasto-Hall effects Ma et al. 2021 ; McClarty and Rau 2024 ; Aoyama and Ohgushi 2024 ; Steward et al. 2023 ; Fernandes et al. 2024 ; Takahashi et al. 2025 . The piezomagnetic and elasto-Hall responses of pure altermagnets are a particularly appealing feature, and have inspired a number of proposals for manipulating and controlling the properties of altermagnets through strain Li et al. 2024 ; Chakraborty et al. 2024 ; Karetta et al. 2025 ; Belashchenko 2025 ; Naka et al. 2025 ; Zhang et al. 2025a ; Chen et al. 2025 ; Xun et al. 2025 ; Jiang et al. 2025a ; Khodas et al. 2025 ; Zhai et al. 2025 ; Xun et al. 2025 ; Smolenski et al. 2025 . In the non-relativistic limit, the strain-induced magnetization can only be of spin origin McClarty and Rau 2024 ; Khodas et al. 2025 , but in general, the induced magnetization has both a spin and an orbital magnetization component Sor 2025 ; Ye et al. 2026 . In this paper we focus on the latter and study the strain-induced orbital magnetization of pure altermagnets in two dimensions (2D). Previous work has addressed the spin magnetization of a strained 2D altermagnet (i.e., the Lieb lattice model Brekke et al. 2023 ; Antonenko et al. 2025 ) and found that it is only nonzero in a metal, when the spin-polarized bands are filled unequally Takahashi et al. 2025 ; Khodas et al. 2025 . In contrast, the orbital magnetization can be nonzero the insulating regime, and in this paper we exclusively consider altermagnetic insulators for which the spin magnetization response vanishes. We specifically consider three microscopic models for tetragonal altermagnets in 2D. These models are minimal models for altermagnets with distinct symmetry Roig et al. 2024 , in particular for altermagnets with d_{x^{2}-y^{2}} -wave Brekke et al. 2023 ; Antonenko et al. 2025 , d_{xy} -wave Maier and Okamoto 2023 , and g -wave symmetry. As such, these representative minimal models serve as an appropriate foundation for a general survey of orbital piezomagnetism in 2D. In particular, whereas the two d -wave altermagnets exhibit a linear piezomagnetic effect, i.e., a linear dependence of the orbital magnetization on applied strain, here we show that the g -wave altermagnet exhibits a non-linear orbital piezomagnetic effect Khodas et al. 2025 . To examine the orbital piezomagnetic response, we rely on the modern theory of orbital magnetization, which provides the basis for computing the orbital magnetization of periodic systems in terms of band energies and eigenstates Xiao et al. 2005 ; Thonhauser et al. 2005 ; Ceresoli et al. 2006 . For the general class of pure altermagnets considered here, we obtain a simple formula for the orbital magnetization, which is directly expressed in terms of the parametrization of the Hamiltonian Mitscherling et al. 2025 . We furthermore derive microscopic expressions for orbital piezomagnetic polarizability, i.e., the piezomagnetic response tensor. Based on these expressions we compute the polarizability for each of the three minimal models and analyze the result in terms of relevant band structure properties, specifically the Berry curvature distribution of the occupied bands. Together, this constitutes a general theory of orbital piezomagnetism in two-dimensional altermagnets. II Model altermagnets in two dimensions In this section we introduce and examine lattice models of pure altermagnets in 2D, with Néel vector oriented along the perpendicular \hat{z} direction. We first present a general formulation of the models of interest, and then discuss specific lattice realizations with tetragonal symmetry. II.1 General Hamiltonian of pure 2D altermagnets The altermagnets we consider have two magnetic sites in the unit cell, labeled A and B , and these sites define the sublattice degree of freedom. To construct the tight-binding Hamiltonian, we introduce the electron destruction operators c_{{\bf k}\alpha\sigma} , where \alpha=A,B and \sigma=\uparrow,\downarrow corresponds to spin. The Hamiltonian is then expressed as H=\sum_{{\bf k}}c^{\dagger}_{\bf k}H_{\bf k}c_{\bf k},\;\;c_{{\bf k}}=(c_{{\bf k}A\uparrow},c_{{\bf k}A\downarrow},c_{{\bf k}B\uparrow},c_{{\bf k}B\downarrow})^{T}, (1) where H_{\bf k} , the momentum-dependent Hamiltonian matrix, is the sum of two terms: H_{\bf k}=H_{0,{\bf k}}+\phi W_{\bf k}. (2) The first term is the Hamiltonian of the altermagnet without strain and the second term describes the effect of strain. Strain is parametrized by a strain field \phi . Following notation introduced in Ref. Roig et al. 2024 , the unstrained Hamiltonian H_{0,{\bf k}} takes the general form H_{0,{\bf k}}=\varepsilon_{\bf k}+t_{x,{\bf k}}\tau^{x}+t_{z,{\bf k}}\tau^{z}+\lambda_{z,{\bf k}}\tau^{y}\sigma^{z}+N_{z}\tau^{z}\sigma^{z}, (3) where we have made use of two sets of Pauli matrices, \bm{\tau}=(\tau^{x},\tau^{y},\tau^{z}) and \bm{\sigma}=(\sigma^{x},\sigma^{y},\sigma^{z}) , corresponding to the sublattice and spin degree of freedom, respectively. The functions t_{x,{\bf k}} and t_{z,{\bf k}} describe intersublattice and intrasublattice hopping processes, respectively, and \lambda_{z,{\bf k}} describes an allowed spin-orbit coupling term Roig et al. 2024 . As mentioned above, the collinear Néel order is always chosen along the \hat{z} direction, such that, in the absence of strain, a pure altermagnetic state without net magnetization is realized. Magnetic order enters the Hamiltonian via the Néel order parameter N_{z} . A key symmetry constraint is imposed by the mirror reflection \mathcal{M}_{z} , which remains a symmetry of the magnetic state when the Néel vector points along the \hat{z} direction. Since \mathcal{M}_{z} does not affect {\bf k}=(k_{x},k_{y}) , it must commute with H_{0,{\bf k}} and therefore imposes the form of Eq. ( 3 ). Specifically, the mirror symmetry \mathcal{M}_{z} implies that H_{0,{\bf k}} is diagonal in spin space (i.e., spin-orbit terms featuring \sigma^{x} and \sigma^{y} are forbidden) and can be considered in each spin sector separately 1 1 1 Technically, the Hamiltonian decomposes into two diagonal mirror sectors. The spin label \sigma may be used here to label these mirror sectors. . A convenient way to write the Hamiltonian in each spin sector is H^{\sigma}_{0,{\bf k}}=\varepsilon_{\bf k}+{\bf n}^{\sigma}_{\bf k}\cdot\bm{\tau}, (4) where {\bf n}^{\sigma}_{\bf k} is given by {\bf n}^{\sigma}_{\bf k}=(t_{x},\sigma\lambda_{z},t_{z}+\sigma N_{z}), (5) and \sigma should be read as \pm 1 for \uparrow / \downarrow . Next, consider strain. The effect of strain is described by a term W_{\bf k} in the Hamiltonian, which we further parametrize as W_{\bf k}=\chi_{\bf k}+{\bf w}_{\bf k}\cdot\bm{\tau}. (6) Since strain preserves the mirror symmetry \mathcal{M}_{z} , the full Hamiltonian can still be decomposed into separate spin sectors. We express the full Hamiltonian as H^{\sigma}_{{\bf k}}=\xi_{\bf k}+{\bf h}^{\sigma}_{\bf k}\cdot\bm{\tau}, (7) where \xi_{\bf k} and {\bf h}^{\sigma}_{\bf k} are given by \xi_{\bf k}=\varepsilon_{\bf k}+\phi\chi_{\bf k},\quad{\bf h}^{\sigma}_{\bf k}={\bf n}^{\sigma}_{\bf k}+\phi{\bf w}_{\bf k}. (8) This general formulation of the Hamiltonian provides a convenient framework for our analysis in the remainder of this paper. Since the Hamiltonian in each spin sector is a two-band Hamiltonian, it is straightforward to obtain the energy bands. We denote the energy bands E^{\sigma}_{{\bf k}n} , where n=1,2 is the band label. Explicitly, the band energies are given by E^{\sigma}_{{\bf k}1,2}=\xi_{\bf k}\mp|{\bf h}^{\sigma}_{\bf k}|. (9) It is similarly straightforward to determine the projectors onto the eigenspaces associated with the energy bands E^{\sigma}_{{\bf k}n} . These projectors, denoted P^{\sigma}_{{\bf k}n} , are given by P^{\sigma}_{{\bf k}1,2}=\frac{1}{2}(\mathbb{1}\mp{\bf h}^{\sigma}_{\bf k}\cdot\bm{\tau}/|{\bf h}^{\sigma}_{\bf k}|), (10) and will be used to obtain a general formula for the orbital magnetization [see Eqs. ( 23 ) and ( 24 )]. We conclude this subsection with a general remark regarding crystal symmetries and their implications for the energy spectrum. A key characteristic of pure altermagnets is that certain point group symmetries of the crystal lattice exchange the magnetic sublattices but do not flip the spin. In the case of a tetragonal d -wave altermagnet an example is fourfold rotation symmetry \mathcal{C}_{4z} . When combined with time-reversal ( \mathcal{T} ) such point group symmetries leave the magnetic state invariant. Let g be such a point group symmetry. Then \mathcal{T}g is a symmetry of the pure (unstrained) altermagnet described by Hamiltonian ( 4 ), and the functions \varepsilon_{{\bf k}} and {\bf n}^{\sigma}_{{\bf k}} satisfy \varepsilon_{-{\bf k}}=\varepsilon_{g{\bf k}},\quad{\bf n}^{\uparrow}_{-{\bf k}}=R{\bf n}^{\downarrow}_{g{\bf k}}, (11) where R=\text{diag}(1,1,-1) is a diagonal matrix. Since the Hamiltonian of the pure altermagnet is also manifestly even under inversion, we further have \varepsilon_{-{\bf k}}=\varepsilon_{{\bf k}} and {\bf n}^{\sigma}_{-{\bf k}}={\bf n}^{\sigma}_{{\bf k}} . These relations can be used to show that the orbital magnetization must vanish in the absence of strain (see Sec. III.1 ). II.2 Model realizations with tetragonal symmetry We now consider three different realizations of the general model introduced above, all with tetragonal symmetry and point group D_{4h} . In each of the three models the site symmetry group of the magnetic sublattices is different, however, such that they give rise to altermagnets with d_{x^{2}-y^{2}} - d_{xy} -, and g_{yx^{3}-xy^{3}} -wave symmetry. Lieb lattice model. The first model is the Lieb lattice altermagnet shown in Fig. 1 (a), which is currently one of the primary minimal models for the study of altermagnetism Brekke et al. 2023 ; Antonenko et al. 2025 ; Yershov et al. 2024 ; Kaushal and Franz 2025 ; Dürrnagel et al. 2025 ; Fu et al. 2025 . The Lieb lattice has two magnetically active sites in the unit cell (i.e., the A and B sites), shown in blue and red in Fig. 1 (a). The two colors indicate the anti-alignment of spins in the Néel state. The additional nonmagnetic site, shown in white, gives rise to the anisotropic local crystal environment seen by the A and B sites. The Lieb lattice has (symmorphic) space group P4/mmm , and with the origin chosen at the nonmagnetic site the magnetic sites sit at Wyckoff positions (\tfrac{1}{2},0) and (0,\tfrac{1}{2}) . Importantly, the two magnetic sublattices are not exchanged by the inversion \mathcal{I} , but are exchanged by the fourfold rotation \mathcal{C}_{4z} and the twofold rotation about the [11] direction. This makes the Lieb lattice an altermagnet with d_{x^{2}-y^{2}} symmetry. The uniform dispersion \varepsilon_{\bf k} and the intersublattice hopping t_{x,{\bf k}} , both defined in Eq. ( 3 ), are given by \varepsilon_{\bf k}=-2t_{0}(c_{x}+c_{y}),\quad t_{x,{\bf k}}=-4t_{1}c_{x/2}c_{y/2}, (12) where we have defined the abbreviations c_{i}\equiv\cos k_{i} and c_{i/2}\equiv\cos(k_{i}/2) , with i=x,y . Importantly, since the form of \varepsilon_{\bf k} and t_{x,{\bf k}} is not dependent on the site symmetry group of the A and B sublattices, it is the same for all three lattice models. The form of t_{z,{\bf k}} and \lambda_{z,{\bf k}} does depend on the sublattice site symmetry groups, and in the case of the Lieb lattice these functions are given by t_{z,{\bf k}}=-2t_{d}(c_{x}-c_{y}),\quad\lambda_{z,{\bf k}}=4\lambda s_{x/2}s_{y/2}. (13) Here we have used the abbreviation s_{i/2}\equiv\sin(k_{i}/2) . To introduce strain effects in the Lieb lattice model, we follow Ref. Takahashi et al. 2025 and choose \chi_{\bf k}=-2t_{0}(c_{x}-c_{y}),\quad w_{z,{\bf k}}=-2t_{d}(c_{x}+c_{y}). (14) This form of the strain field coupling is mandated by symmetry, since the applied strain must be of a particular symmetry type in order to give rise to a nonzero magnetization. In the case of d -wave altermagnets, strain must have the same symmetry as the altermagnet McClarty and Rau 2024 ; Bhowal and Spaldin 2024 ; Takahashi et al. 2025 ; Khodas et al. 2025 . Here, the applied strain must have d_{x^{2}-y^{2}} symmetry, which means the nonzero strain tensor component is \varepsilon_{xx}-\varepsilon_{yy} . Figure 1: Lieb lattice model. (a) Schematic of the two-dimensional Lieb lattice. The magnetic A and B sites are shown in blue and red, and the nonmagnetic site is shown in white. The ordered moments of the collinear altermagnetic Néel state point out of the plane, as indicated. (b) Energy spectrum with and without strain (red/blue and solid/dashed black, respectively) of the Lieb lattice model specified by Eqs. ( 12 )–( 14 ) with parameters (t_{0},t_{1},\lambda,N_{z})=(0.25t_{d},0.5t_{d},0.25t_{d},2.0t_{d}) and \phi=0.1 for the case with strain. Blue and red bands correspond to \sigma=\uparrow and \sigma=\downarrow , respectively. The energy bands E^{\sigma}_{{\bf k}n} of the Lieb lattice model, as given by Eq. ( 9 ), are shown in Fig. 1 (b) for a set of parameters that give rise to a fully gapped spectrum. The blue ( \uparrow ) and red ( \downarrow ) bands include the effect of strain, whereas the black solid ( \uparrow ) and dashed ( \downarrow ) bands correspond to the unstrained system. Note that we have chosen the unrealistically large strain value of \phi=0.1 ( 10\% ) to make the effect of strain visible. As is clear, in the presence of strain the energy bands are no longer related by \mathcal{T}\mathcal{C}_{4z} symmetry [see Eq. ( 11 )] or the mirror symmetry \mathcal{M}_{[11]} (i.e., the mirror reflection in a plane normal to the [11] direction). In particular, the manifest degeneracy of \sigma=\uparrow and \sigma=\downarrow bands on the \Gamma M line, which is characteristic of d_{x^{2}-y^{2}} -wave altermagnets, is lifted. An essential feature of the Lieb altermagnet model is the nontrivial topology of the energy bands. Previous work pointed out that the two valence bands with energies E^{\uparrow}_{{\bf k}1} and E^{\downarrow}_{{\bf k}1} have an equal but opposite Chern number, giving rise to mirror Chern bands Antonenko et al. 2025 . The nontrivial topology of the Lieb lattice bands can be intuitively understood from the Dirac points that exist on the Brillouin zone (BZ) boundary in the limit of vanishing spin-orbit coupling (i.e., \lambda=0 ) Sun et al. 2009 ; Weeks and Franz 2010 ; Antonenko et al. 2025 . Spin-orbit coupling acts as a mass term for the two Dirac points in each spin sector and the sign structure of the mass terms is such that the spin sectors have nonzero but opposite Chern numbers. The location of these Dirac points on the k_{x}=\pi and k_{y}=\pi lines is determined from the condition that the \hat{z} -component of {\bf n}^{\sigma}_{\bf k} in ( 5 ) vanishes: t_{z,{\bf k}}+\sigma N_{z}=0 Antonenko et al. 2025 . This condition defines a critical value of N_{z} (denoted N_{z,c} ) beyond which no solutions exist and the bands cannot be understood from a gapped Dirac point picture. In the Lieb lattice model this critical value is given by N_{z,c}=4t_{d}, (15) and marks a topological gap closing transition at which the mirror Chern number C_{M} changes from \pm 1 to zero Antonenko et al. 2025 . Below, in Sec. III , we will show how the topology of the Lieb lattice band structure is reflected in the piezomagnetic response. Figure 2: 2D rutile lattice model. (a) Schematic of the 2D rutile lattice with layer group P4/mbm . The magnetic sites (red and blue) have an anisotropic crystallographic environment. (b) Energy spectrum with and without strain (red/blue and solid/dashed black, respectively) of the model defined by Eq. ( 12 ) and Eqs. ( 16 )–( 17 ) with parameters (t_{0},t_{1},\lambda,N_{z})=(0.25t_{d},0.5t_{d},0.25t_{d},2.0t_{d}) and \phi=0.1 for the strained case. As in Fig. 1 (b), blue and red bands correspond to \sigma=\uparrow and \sigma=\downarrow , respectively. 2D rutile lattice model. The second model we consider was introduced in Ref. Maier and Okamoto 2023 and is shown in Fig. 2 (a). As in the Lieb lattice model, the magnetic sites experience different local crystal environments as a result of the arrangement of nonmagnetic sites, which is represented by alternatingly oriented rhombuses in Fig. 2 (a). A crystal lattice of this type can be viewed as the 2D variant of the 3D rutile lattice, the lattice of candidate altermagnets RuO 2 Sun et al. 2017 ; Berlijn et al. 2016 ; Ahn et al. 2019 ; Šmejkal et al. 2020 ; Feng et al. 2022 ; Šmejkal et al. 2022a ; Šmejkal et al. 2022b and MnF 2 Yuan et al. 2020 ; Bhowal and Spaldin 2024 . For this reason, we refer to this model as the “2D rutile” lattice model. The layer group of the 2D rutile lattice is P4/mbm . With the origin chosen at one of the magnetic sites, the space group generators are \{\mathcal{C}_{4z}|\tfrac{1}{2}\tfrac{1}{2}\} , \{\mathcal{C}_{2x}|\tfrac{1}{2}\tfrac{1}{2}\} , and \{\mathcal{I}|00\} . The fourfold and the twofold rotations exchange the magnetic sublattices and as a result, this model realizes a d_{xy} altermagnet. As mentioned above, \varepsilon_{\bf k} and t_{x,{\bf k}} are the same as in Lieb lattice model and are given by Eq. ( 12 ). The functions t_{z,{\bf k}} and \lambda_{z,{\bf k}} reflect the specific symmetries of the crystal lattice, in particular the site symmetry group of the A and B sites, and are therefore different for a d_{xy} altermagnet. In the 2D rutile lattice model these functions are given by t_{z,{\bf k}}=-2t_{d}s_{x}s_{y},\quad\lambda_{z,{\bf k}}=4\lambda c_{x/2}c_{y/2}(c_{x}-c_{y}), (16) where the same abbreviation convention as above are used. This specifies the microscopic model without strain. To obtain a piezomagnetic response in a d_{xy} -wave altermagnet, the applied strain must be of the corresponding symmetry, and therefore we include an {\varepsilon}_{xy} strain component. The functions \chi_{\bf k} and w_{z,{\bf k}} , which parametrize the strain microscopically via Eq. ( 6 ), have the form \chi_{\bf k}=-2t_{d}s_{x}s_{y},\quad w_{z,{\bf k}}=-2t_{0}(c_{x}+c_{y}). (17) The energy bands of the 2D rutile lattice model are shown in Fig. 2 (b), both with and without strain, using the same convention as in Fig. 1 (b). The symmetry breaking effects of strain on the energy bands are clearly visible. In the absence of strain the valence bands of opposite spin are degenerate both on the \Gamma X and on the XM lines (the same is true for the conduction bands). The degeneracy on the \Gamma X line is expected for an altermagnet with d_{xy} -wave spin splitting, as is the large spin splitting on the diagonal \Gamma M line. The degeneracy on the BZ boundary XM line is a consequence of non-symmorphic space group symmetries. In fact, in the nonmagnetic \mathcal{T} -symmetric state, the symmetries of the space group P4/mbm require that all energy levels on the BZ boundary are fourfold degenerate (see Appendix A ). This leads to the important observation that the 2D rutile lattice model cannot maintain an energy gap, and thus cannot describe an insulator, in the N_{z}\rightarrow 0 limit. When examining the dependence of the orbital piezomagnetic polarizability on the magnetic order parameter N_{z} one must account for this, since the computation of M_{z} tacitly assumes an insulating ground state. In sharp contrast to the Lieb lattice model, the two valence bands of the rutile model are topologically trivial and do not have a Chern number. As in the Lieb lattice model it is possible to understand the topology of the bands—in this case the trivial topology of the bands—from a continuum Dirac model. A detailed construction of such a Dirac model is presented in Appendix B . Here we summarize the features necessary for a discussion of the strain-induced orbital magnetization. In the case of the 2D rutile lattice, the (gapped) Dirac points are located on the body diagonals of the BZ, i.e., the k_{x}=\pm k_{y} lines, and their location on the diagonals is again determined from the condition t_{z,{\bf k}}+\sigma N_{z}=0 . As shown in Appendix B , as long as 0<|N_{z}|<2t_{d} this equation has four solutions in each spin sector (i.e., on each body diagonal). These solutions define two pairs of Dirac points with opposite topological charge, which implies that the Chern number in each spin sector vanishes. We again find a critical value of N_{z} beyond which a Dirac model construction is no longer possible and this value is given by N_{z,c}=2t_{d}. (18) In the 2D rutile model no gap closing transition occurs at this value of N_{z} . Two important differences with the Dirac model of the Lieb lattice altermagnet should be emphasized. First, in the case of the rutile lattice, the Dirac model is only meaningful when spin-orbit coupling is included. Recall that in the case of the Lieb lattice model the Dirac points already exist in the absence of spin-orbit coupling and are gapless, and acquire a gap when spin-orbit coupling is included. Instead, here the spin-orbit coupling provides the linear dispersion in the direction perpendicular to the body diagonal, rather than the Dirac mass (see Appendix B ). The second key difference is the number of (gapped) Dirac points in each spin sector. Whereas the Lieb lattice model has one pair (per spin sector), the rutile lattice model has two such pairs. The members of each pair have the same topological charge, but the overall topological charge vanishes. Fleshing out these differences provides a basis for the discussion of the orbital magnetization response to strain in the next section. Figure 3: Octahedral rotation lattice. (a) A lattice of magnetic sites (shown in blue and red) surrounded by rotated square cages of nonmagnetic sites (shown in white). (b) Momentum-space resolved plot of the energy difference (E^{\uparrow}_{{\bf k}1}-E^{\downarrow}_{{\bf k}1})/8t_{g} between the valence bands of opposite spin in the absence of strain. The g -wave structure of the spin splitting is evident. Energies are calculated for the model defined in Eqs. ( 12 ) and ( 19 ) with parameters (t_{0},t_{1},\lambda_{2},N_{z})=(0.25t_{g},0.5t_{g},0.5t_{g},2.0t_{g}) . Octahedral rotation model. The third and final model we consider is shown in Fig. 3 (a). The unit cell has two magnetic sites, again shown in red and blue, as well as nonmagnetic sites, shown in white. The magnetic sites are surrounded by square plaquettes of nonmagnetic sites, which are rotated in opposite directions for the two magnetic sublattices. It is this rotation of the nonmagnetic squares that is responsible for the different crystal environment seen by the magnetic sites. Rotations of surrounding ligand cages are common in magnetic oxides, where they are referred to as octahedral rotations (of the octahedral oxygen cages) Glazer 1972 , and the square lattice model considered here can therefore be viewed as a 2D variant of altermagnetism induced by octahedral rotations Schmitz et al. 2005 ; Bousquet and Spaldin 2011 ; Wang et al. 2022 ; Fernandes et al. 2024 ; Rooj et al. 2025 . For this reason, we refer to this third model as the octahedral rotation model. The layer group of this crystal structure is the same as that of the 2D rutile model, P4/mbm . Crucially, the site symmetry group is different, however. Choosing the origin at one of the magnetic sites, the space group generators are \{\mathcal{C}_{4z}|00\} , \{\mathcal{C}_{2x}|\tfrac{1}{2}\tfrac{1}{2}\} , and \{\mathcal{I}|00\} . It follows that the fourfold rotation does not exchange the magnetic sublattices, but all twofold rotations with in-plane rotation axes do. The octahedral rotation model therefore realizes a g_{yx^{3}-xy^{3}} -wave altermagnet. The functions \varepsilon_{\bf k} and t_{x,{\bf k}} are still given by Eq. ( 12 ), while the form of t_{z,{\bf k}} and \lambda_{z,{\bf k}} is determined the site symmetry group of the magnetic sites. For the octahedral rotation model we find \begin{split}t_{z,{\bf k}}&=-2t_{g}s_{x}s_{y}(c_{x}-c_{y}),\\ \lambda_{z,{\bf k}}&=4\lambda_{1}c_{x/2}c_{y/2}+4\lambda_{2}c_{x/2}c_{y/2}(c_{x}+c_{y}).\end{split} (19) Here t_{g} describes an intrasublattice fourth-nearest neighbor hopping. Note further that here we have introduced two spin-orbit terms, with parameters \lambda_{1} and \lambda_{2} , respectively. This is necessary to ensure that \lambda_{z,{\bf k}} and t_{x,{\bf k}} are not simply linearly proportional to each other; in what follows we will set \lambda_{1}=0 , unless otherwise specified. Introducing the effect of strain in this g -wave model requires generalizing Eq. ( 6 ) to include two symmetry-distinct strain components. The full Hamiltonian of the octahedral rotation lattice model therefore takes the form H_{\bf k}=H_{0,{\bf k}}+\phi_{1}W_{1,{\bf k}}+\phi_{2}W_{2,{\bf k}}, (20) where \phi_{1} and \phi_{2} are strain fields with \varepsilon_{xx}-\varepsilon_{yy} and \varepsilon_{xy} symmetry, respectively. The strain terms W_{1,{\bf k}} and W_{2,{\bf k}} are expanded as in Eq. ( 6 ), and the nonzero expansion coefficients are chosen as \begin{split}\chi_{1,{\bf k}}&=w_{2,z,{\bf k}}=-2t_{0}(c_{x}-c_{y}),\\ w_{1,z,{\bf k}}&=\chi_{2,{\bf k}}=-2t_{d}s_{x}s_{y}.\end{split} (21) Introducing two symmetry-distinct strain components \phi_{1} and \phi_{2} is necessary to obtain a magnetization response in a g -wave altermagnet. We will show in Sec. IV that the nonlinear piezomagnetic effect is proportional to \phi_{1}\phi_{2} . In Fig. 3 (b) we show a density plot of the energy difference between the two valence bands E^{\uparrow}_{{\bf k}1} and E^{\downarrow}_{{\bf k}1} . As expected for a square lattice g -wave altermagnet, the energy difference vanishes both on the principal axes and on the body diagonals. It is furthermore clear from Fig. 3 (b) that the energy difference respects fourfold rotation symmetry and breaks all vertical mirror symmetries. The energy difference also vanishes on the BZ boundary, which, as in the case of the 2D rutile lattice, is a consequence of space group symmetry (as detailed in Appendix A ). As for the 2D rutile lattice, these space group symmetries also imply that the energy bands must be fourfold degenerate on the BZ boundary in the nonmagnetic state with N_{z}=0 . As a result, the model cannot maintain an insulating gap as N_{z}\rightarrow 0 either. III Orbital magnetization and linear piezomagnetic polarizability We now turn to the main objective of this paper: the study of orbital piezomagnetism in altermagnets. As mentioned in the introduction, d -wave altermagnets are naturally piezomagnetic (i.e., exhibit a linear piezomagnetic effect) and we therefore focus first on the two d -wave models introduced in the previous section. We begin by computing the orbital magnetization in the presence of strain based on the modern theory Thonhauser et al. 2005 ; Xiao et al. 2005 ; Ceresoli et al. 2006 , and then compute the corresponding linear response coefficient, the linear orbital piezomagnetic polarizability. The piezomagnetic polarizability describes how strong the orbital magnetization response to strain is and is a property of pure altermagnets. III.1 Strain-induced orbital magnetization According to the modern theory, the orbital magnetization M_{z} of a general 2D system is given by Thonhauser et al. 2005 ; Xiao et al. 2005 ; Ceresoli et al. 2006 M_{z}=\frac{e}{\hbar}\int\frac{d^{2}{\bf k}}{(2\pi)^{2}}\sum_{n}\text{Im}\langle\partial_{x}u_{{\bf k}n}|(H_{\bf k}+E_{{\bf k}n})|\partial_{y}u_{{\bf k}n}\rangle. (22) Here H_{\bf k} is the Bloch Hamiltonian and |u_{{\bf k}n}\rangle and E_{{\bf k}n} are the Bloch eigenstates and band energies of H_{\bf k} , respectively. The derivatives \partial_{j}\equiv\partial/\partial k_{j} are momentum derivatives; n is the band label, and the sum is over all occupied bands. This expression for the orbital magnetization tacitly assumes it is calculated for an insulator with vanishing Chern number Ceresoli et al. 2006 , which is applicable to the case at hand. It is useful to rewrite the expression for the orbital magnetization in terms of derivatives of the Hamiltonian H_{\bf k} , rather than of eigenstates |u_{{\bf k}n}\rangle . The orbital magnetization can be recast as M_{z}=\frac{e}{2\hbar}\int\frac{d^{2}{\bf k}}{(2\pi)^{2}}\sum_{n,m}\frac{E_{{\bf k}n}+E_{{\bf k}m}}{(E_{{\bf k}n}-E_{{\bf k}m})^{2}}\\ \times\text{Im}\text{Tr}[P_{{\bf k}n}(\partial_{x}H_{\bf k})P_{{\bf k}m}(\partial_{y}H_{\bf k})], (23) where P_{{\bf k}n} are the projectors onto the eigenspace of energy band E_{{\bf k}n} , as given by Eq. ( 10 ). In Eq. ( 23 ) the sum on n is over all occupied bands and the sum on m is over unoccupied bands. Figure 4: Orbital magnetization. (a) Orbital magnetization M_{z} as a function of the dimensionless strain field \phi , calculated for the Lieb lattice model (see inset) using Eq. ( 24 ). We have used the parameters (t_{0},t_{d},\lambda)=(0.5t_{1},2t_{1},0.5t_{1}) and different curves correspond to N_{z}/t_{1}=1.0,2.0,3.0,4.0,5.0 (from blue to magenta). Here a is a length scale on the order of the lattice spacing defined via \hbar^{2}/2m_{e}t_{1}\equiv a^{2} . Note further that the value of the M_{z} has been multiplied by a factor 10 . (b) Same as in (a) but for the 2D rutile lattice model (see inset). For the rutile lattice model we have used the same model parameters (t_{0},t_{d},\lambda) but different curves correspond to N_{z}/t_{1}=2.0,\ldots,6.0 (from blue to magenta). The expression for the magnetization given by Eq. ( 23 ) is valid for a general 2D Hamiltonian. Here we apply it to the 2D altermagnets defined by Eq. ( 7 ). The contributions from the two spin sectors can be considered separately and then added. Each spin sector has an occupied band, given by E^{\sigma}_{{\bf k}1} , and an unoccupied band, given by E^{\sigma}_{{\bf k}2} [see Eq. ( 9 )], with associated projectors given by Eq. ( 10 ). Substituting these expressions into Eq. ( 23 ) yields the appealing result M_{z}=-\frac{e}{\hbar}\sum_{\sigma}\int\frac{d^{2}{\bf k}}{(2\pi)^{2}}\xi_{\bf k}\frac{{\bf h}^{\sigma}_{\bf k}\cdot\partial_{x}{\bf h}^{\sigma}_{\bf k}\times\partial_{y}{\bf h}^{\sigma}_{\bf k}}{2|{\bf h}^{\sigma}_{\bf k}|^{3}}, (24) which shows that M_{z} can be expressed directly in terms of the parametrization of the Hamiltonian (i.e., \xi_{\bf k} and {\bf h}^{\sigma}_{\bf k} ). We further observe that the integrand is the product of \xi_{\bf k} with the Berry curvature of the occupied valence bands. (See also Ref. Mitscherling et al. 2025 .) It is straightforward to deduce from Eq. ( 24 ) that the orbital magnetization must be zero in the absence of strain. Without strain one has \xi_{\bf k}\rightarrow\varepsilon_{\bf k} and {\bf h}^{\sigma}_{\bf k}\rightarrow{\bf n}^{\sigma}_{\bf k} , as is clear from Eq. ( 8 ). We can then exploit the fact that {\bf n}^{\uparrow}_{\bf k} and {\bf n}^{\downarrow}_{\bf k} are related via Eq. ( 11 ) to show that \frac{{\bf n}^{\uparrow}_{\bf k}\cdot\partial_{x}{\bf n}^{\uparrow}_{\bf k}\times\partial_{y}{\bf n}^{\uparrow}_{\bf k}}{2|{\bf n}^{\uparrow}_{\bf k}|^{3}}=-\frac{{\bf n}^{\downarrow}_{g{\bf k}}\cdot\partial_{x}{\bf n}^{\downarrow}_{g{\bf k}}\times\partial_{y}{\bf n}^{\downarrow}_{g{\bf k}}}{2|{\bf n}^{\downarrow}_{g{\bf k}}|^{3}}, (25) where we have used that \text{Det}(R)=-1 . In the case of d -wave altermagnets it is natural to take g as the fourfold rotation \mathcal{C}_{4z} . Performing a change of variables in the integral from {\bf k} to {\bf k}^{\prime}=g{\bf k} shows that the two terms in ( 24 ) cancel and that M_{z} must vanish in the absence of strain. Figure 5: Berry curvature distribution. (Left) Plot of the Berry curvature of the Lieb lattice model without strain. The upper and lower panels correspond to the \sigma=\uparrow and \sigma=\downarrow valence bands, respectively. The Berry curvature is concentrated around the gapped Dirac crossings, which exist on the k_{y}=\pi and k_{x}=\pi lines (respectively) and are indicated by black markers. Here we have used (t_{d},\lambda,N_{z})=(2t_{1},0.3t_{1},4.0t_{1}) . (Right) Plot of the Berry curvature of the (unstrained) 2D rutile lattice model, with upper and lower panels corresponding to \sigma=\uparrow and \sigma=\downarrow , respectively. For the rutile model we have used (t_{d},\lambda,N_{z})=(2t_{1},0.5t_{1},2.0t_{1}) . The location of the Dirac points on the body diagonal, obtained as described in Appendix B , is indicated by black dots. In Fig. 4 we show the orbital magnetization M_{z} as a function of strain, calculated using Eq. ( 24 ), for the Lieb lattice model in panel (a) and the 2D rutile lattice model in panel (b). Different curves (color coded) correspond to different values of N_{z} , and thus shed light on the dependence on the strength of (alter)magnetic order. As expected, M_{z} vanishes in the absence of strain. We further find that, for both models and the chosen parameter sets, the dependence of M_{z} on strain is essentially linear for reasonably realistic values of strain. We have verified that deviations from linearity occur at larger (and thus more unrealistic) values of strain. Two differences between the two d -wave models can be observed. The first is that the slope of the magnetization curves increases with N_{z} for the Lieb lattice model, whereas it decreases for the rutile lattice beyond N_{z}\approx 3.0t_{1} . This will be discussed in more detail below, when we compute the linear polarizability (see also Fig. 7 ) and examine its dependence on N_{z} . The second is that the magnitude of M_{z} at the same strain is significantly larger for the Lieb lattice model. This can be traced back to the specific band topology of the Lieb lattice model in the unstrained limit. As mentioned above (see also Ref. Antonenko et al. 2025 ), the energy bands of the Lieb lattice model have a nonzero mirror Chern number in the regime |N_{z}|0 . Hence, when N_{z}>0 solutions only exist in the \sigma=\uparrow sector. Denoting the momenta which solve ( 34 ) as k_{D} , we then find that \sin k_{D}=\pm\sqrt{\sigma N_{z}/2t_{d}} . From this, in turn, it follows that there are four solutions on the k_{x}=k_{y} line, as long as 0<|N_{z}|/2t_{d}<1 , which we denote k_{D,j} ( j=1,2,3,4 ). These four solutions give rise to four Dirac points, shown schematically in Fig. 9 , which form two pairs. These pairs are (k_{D,1},k_{D,4}) and (k_{D,2},k_{D,3}) , with Dirac point momenta related as k_{D,4}=-k_{D,1} and k_{D,3}=-k_{D,2} . The equation \sin k_{D}=\pm(\sigma N_{z}/2t_{d})^{1/2} also yields the critical value of N_{z} , which has been defined in main text as N_{z,c}=2t_{d} , since no solutions are obtained for N_{z}>N_{z,c} . Figure 9: Sketch of the four Dirac points which are solutions of ( 34 ). The Dirac point momenta are labeled k_{D,j} ( j=1,2,3,4 ), and are related as k_{D,4}=-k_{D,1} and k_{D,3}=-k_{D,2} . The next step is to expand the lattice Hamiltonian around each of the four Dirac points. Expanding the \hat{z} -component of {\bf n}^{\sigma}_{\bf k} amounts to an expansion of the form factor \sin k_{x}\sin k_{y} , for which we find to linear order in small momentum {\bf q}=(q_{x},q_{y}) \displaystyle\sin k_{x}\sin k_{y} \displaystyle=\sin(k_{D}+q_{x})\sin(k_{D}+q_{y}) (35) \displaystyle\approx\sin^{2}k_{D}+\sin k_{D}\cos k_{D}(q_{x}+q_{y}). (36) Note that the sign of \sin k_{D}\cos k_{D} depends on the Dirac point in question. Given the expansion of the form factor, we find for the \hat{z} -component of {\bf n}^{\sigma}_{\bf k} : n^{z,\sigma}_{{\bf q}}\approx-2t_{d}\sin k_{D}\cos k_{D}(q_{x}+q_{y}). (37) In case of the \hat{x} -component, we find that it does not vanish when evaluated at the Dirac points. We therefore only need to expand to zeroth order in {\bf q} . Specifically, we obtain n^{x,\sigma}_{{\bf q}}\approx-4t_{1}\cos^{2}(k_{D}/2)=-2t_{1}(1+\cos k_{D}). (38) Since \cos k_{D} has a different sign for the two pairs of Dirac points, their Dirac mass is of different magnitude [see Eq. ( 44 ) below]. Finally, we obtain for the \hat{y} -component: n^{y,\sigma}_{{\bf q}}\approx 2\sigma\lambda(1+\cos k_{D})\sin k_{D}(-q_{x}+q_{y}). (39) It follows from this expression that the Dirac velocity perpendicular to the body diagonal k_{x}=k_{y} is different in magnitude for the two pairs of Dirac points [see Eq. ( 43 ) below]. Putting these results together we obtain the following for the Dirac points on the k_{x}=k_{y} line. For the Dirac point pair (k_{D,1},k_{D,4})=(k_{D,1},-k_{D,1}) we find the Hamiltonian \mathcal{H}^{\pm}_{{\bf q}}=\pm v_{1}(q_{x}+q_{y})\tau^{z}\pm v_{2}(q_{y}-q_{x})\tau^{y}-m\tau^{x}, (40) and for the Dirac point pair (k_{D,2},k_{D,3})=(k_{D,2},-k_{D,2}) we find \tilde{\mathcal{H}}^{\pm}_{{\bf q}}=\mp v_{1}(q_{x}+q_{y})\tau^{z}\pm\tilde{v}_{2}(q_{y}-q_{x})\tau^{y}-\tilde{m}\tau^{x}. (41) Here v_{1} , v_{2} , and \tilde{v}_{2} are Dirac velocities defined as v_{1}=2t_{d}\sqrt{\frac{\sigma N_{z}}{2t_{d}}\left(1-\frac{\sigma N_{z}}{2t_{d}}\right)}, (42) and v_{2},\tilde{v}_{2}=2\sigma\lambda\left(1\mp\sqrt{1-\frac{\sigma N_{z}}{2t_{d}}}\right)\sqrt{\frac{\sigma N_{z}}{2t_{d}}}, (43) whereas m and \tilde{m} are Dirac masses given by m,\tilde{m}=2t_{1}\left(1\mp\sqrt{1-\frac{\sigma N_{z}}{2t_{d}}}\right). (44) Note that throughout this derivation we have set \hbar=1 and also set the lattice constant to unity ( a=1 ). Given the Dirac Hamiltonians of Eqs. ( 40 ) and ( 41 ), it is straightforward to compute the Berry curvatures of the valence bands for each Dirac point. We find \Omega^{\pm}_{xy}=\frac{mv_{1}v_{2}}{\sqrt{m^{2}+v_{1}^{2}(q_{x}+q_{y})^{2}+v_{2}^{2}(q_{x}-q_{y})^{2}}}, (45) and \tilde{\Omega}^{\pm}_{xy}=-\frac{\tilde{m}v_{1}\tilde{v}_{2}}{\sqrt{\tilde{m}^{2}+v_{1}^{2}(q_{x}+q_{y})^{2}+\tilde{v}_{2}^{2}(q_{x}-q_{y})^{2}}}, (46) respectively, and it is these expressions that have been used to construct the approximate Berry curvature shown in Fig. 6 in orange. Appendix C Non-linear orbital piezomagnetic polarizability The nonlinear piezomagnetic polarizability defined by the response equation M_{z}=\Lambda^{(2)}\phi_{1}\phi_{2} can be obtained from Eq. ( 24 ) of the main text as follows. One simply makes the substitutions \xi_{\bf k}={\varepsilon}_{\bf k}+\phi_{1}\chi_{1,{\bf k}}+\phi_{2}\chi_{2,{\bf k}}, (47) and {\bf h}^{\sigma}={\bf n}^{\sigma}_{\bf k}+\phi_{1}{\bf w}_{1,{\bf k}}+\phi_{2}{\bf w}_{2,{\bf k}}, (48) and then expands the integrand in \phi_{1} and \phi_{2} , keeping only terms of order \phi_{1}\phi_{2} . This is essentially trivial for the numerator and is also straightforward in case of the denominator. Collecting all terms yields \Lambda^{(2)}=-\frac{e}{\hbar}\sum_{\sigma}\int\frac{d^{2}{\bf k}}{(2\pi)^{2}}\Bigg\{\frac{\varepsilon}{|\mathbf{n}^{\sigma}|^{3}}\big[\mathbf{n}^{\sigma}\cdot(\partial_{x}\mathbf{w}_{1}\times\partial_{y}\mathbf{w}_{2})+\mathbf{w}_{1}\cdot(\partial_{x}\mathbf{w}_{2}\times\partial_{y}\mathbf{n}^{\sigma})+\mathbf{w}_{1}\cdot(\partial_{x}\mathbf{n}^{\sigma}\times\partial_{y}\mathbf{w}_{2})+(1\leftrightarrow 2)\big]\\ +3\frac{\mathbf{n}^{\sigma}\cdot(\partial_{x}\mathbf{n}^{\sigma}\times\partial_{y}\mathbf{n}^{\sigma})}{|\mathbf{n}^{\sigma}|^{5}}\Big[5{\varepsilon}(\hat{\mathbf{n}}^{\sigma}\cdot\mathbf{w}_{1})(\hat{\mathbf{n}}^{\sigma}\cdot\mathbf{w}_{2})-{\varepsilon}\,\mathbf{w}_{1}\cdot\mathbf{w}_{2}-\chi_{1}\,\mathbf{n}^{\sigma}\cdot\mathbf{w}_{2}-\chi_{2}\,\mathbf{n}^{\sigma}\cdot\mathbf{w}_{1}\Big]\\ -\varepsilon\frac{3}{|\mathbf{n}^{\sigma}|^{5}}\ \Big[(\mathbf{n}^{\sigma}\cdot\mathbf{w}_{1})[\mathbf{n}^{\sigma}\cdot(\partial_{x}\mathbf{n}^{\sigma}\times\partial_{y}\mathbf{w}_{2})+\mathbf{n}^{\sigma}\cdot(\partial_{x}\mathbf{w}_{2}\times\partial_{y}\mathbf{n}^{\sigma})+\mathbf{w}_{2}\cdot(\partial_{x}\mathbf{n}^{\sigma}\times\partial_{y}\mathbf{n}^{\sigma})]+(1\leftrightarrow 2)\Big]\\ +\frac{1}{|\mathbf{n}^{\sigma}|^{3}}\Big(\chi_{1}\ [\mathbf{n}^{\sigma}\cdot(\partial_{x}\mathbf{n}^{\sigma}\times\partial_{y}\mathbf{w}_{2})+\mathbf{n}^{\sigma}\cdot(\partial_{x}\mathbf{w}_{2}\times\partial_{y}\mathbf{n}^{\sigma})+\mathbf{w}_{2}\cdot(\partial_{x}\mathbf{n}^{\sigma}\times\partial_{y}\mathbf{n}^{\sigma})]+(1\leftrightarrow 2)\Big)\Bigg\} (49) which has been used to calculate \Lambda^{(2)} in Fig. 8 .