Nonlinear isotropic odd elasticity Shiheng Zhao (赵世恒), Pierre A. Haas Abstract The nonconservative elastic responses of active solids have driven a recent explosion of interest in two-dimensional “odd” elasticity: small, linear deformations of these Cauchy elastic solids enable new behaviour absent from classical, passive elasticity. Here, we establish the description of large, nonlinear deformations of isotropic two-dimensional Cauchy elastic solids. We apply our framework to the Rivlin problem, perhaps the simplest problem of elasticity lacking a linear analogue: a square deforms under dead load tractions. Surprisingly, we find that oddness suppresses the bifurcations of a passive Rivlin square. By contrast, we discover that the bifurcations of a three-dimensional Rivlin cube survive oddness even though there is no isotropic, odd linear elasticity in three dimensions. Our results thus form the basis for describing large deformations of active, biological solids while revealing their unexpected nonlinear behaviour that arises even in minimal problems. The modern field theory of nonlinear elasticity is often based on the concept of hyperelasticity proposed by Green [ 1 , 2 ] , in which the mechanical stresses are derived from a conserved strain energy function. Cauchy elasticity [ 1 , 3 ] is based instead on a direct, local relation between stresses and the deformation gradient tensor. The consequent lack of a conserved energy allows the net work performed by stress along a closed path to be nonzero in a Cauchy elastic material. For this reason, Cauchy elasticity received little attention until metamaterials and biological materials provided examples of systems with local energy sources and sinks that offer a physical realisation of this central feature of Cauchy elasticity. Indeed, the advent of active matter has brought about considerable recent interest in the physics of “odd” elasticity [ 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 ] : linear Cauchy elasticity by another name would smell as sweet [ *[][;foramodernedition , see , e.g. , ]shakespeare , *rj , 3 ] . Two-dimensional (2D) linear odd elasticity adds the additional elastic moduli allowed by removing the symmetries implied by conservation of energy and angular moment to the stress-strain relation of passive linear elasticity. These odd moduli couple, for example, compression to rotation, and hence give rise to novel behaviour that is not allowed in passive elastic solids. Interestingly, this linear phenomenology is limited to 2D: there is no three-dimensional (3D) isotropic linear odd elasticity [ 26 , 4 , 5 ] . This treatment of 2D odd elasticity has however been limited to the linear regime and hence to small deformations. This cannot describe the large deformations of active biological solids. Here, we set up a constitutive framework for finite deformations of 2D odd elastic solids. We apply our framework to the classical Rivlin problem [ 27 , 28 ] , the bifurcation of an elastic square under dead load tractions that is absent from linear elasticity. Strikingly, we find that the rich behaviour of a passive Rivlin square does not appear for an odd Rivlin square, which only allows trivial deformations in general. We then construct a nonlinear theory of isotropic incompressible odd elasticity in 3D. Although there is no linear isotropic odd elasticity in 3D, we discover non-trivial behaviour in an odd Rivlin cube. 2D isotropic nonlinear odd elasticity. We consider a 2D deformation that maps each point \boldsymbol{X}\in\mathbb{R}^{2} in a body to its deformed position \boldsymbol{x}(\boldsymbol{X})\in\mathbb{R}^{2} . The deformation gradient tensor associated to this map is \mathbfsfit{F}=\partial\boldsymbol{x}/\partial\boldsymbol{X} . Cauchy elasticity postulates the relation between the Cauchy stress \mathbfsfit{T} and the deformation. For a simple material, assumed here, this relation is instantaneous and local, viz., \mathbfsfit{T}=\mathbfsfit{T}(\mathbfsfit{F}) . This relation must satisfy the fundamental requirement of objectivity , i.e., that the physics do not change under rigid-body motions [ 1 , 2 , 29 , 3 ] . Mathematically, this is expressed by the condition \mathbfsfit{T}(\mathbfsfit{QF})=\mathbfsfit{QT}(\mathbfsfit{F})\mathbfsfit{Q}^{\top} for all deformations \mathbfsfit{F} and all rotations \mathbfsfit{Q}\in\text{SO}(2) [ 2 ] . Moreover, this relation must obey the symmetries of the material. For an isotropic material, this condition reads \mathbfsfit{Q}\mathbfsfit{T}(\mathbfsfit{F})\mathbfsfit{Q}^{\top}=\mathbfsfit{T}\bigl(\mathbfsfit{Q}\mathbfsfit{F}\mathbfsfit{Q}^{\top}\bigr) for all deformations \mathbfsfit{F} and all rotations \mathbfsfit{Q}\in\text{SO}(2) [ 2 ] . Combining these conditions, {\mathbfsfit{T}(\mathbfsfit{QF})\!=\!\mathbfsfit{T}\bigl((\mathbfsfit{QF})\mathbfsfit{Q}^{\top}\bigr)} for all \mathbfsfit{QF} and all \mathbfsfit{Q}^{\top}\in\text{SO}(2) ; equivalently, {\mathbfsfit{T}(\mathbfsfit{F})=\mathbfsfit{T}(\mathbfsfit{FQ})} for all \mathbfsfit{F} and all \mathbfsfit{Q}\in\text{SO}(2) . We now introduce the polar decomposition \mathbfsfit{F}=\mathbfsfit{VR} , where \mathbfsfit{V}=\mathbfsfit{V}^{\top} and \mathbfsfit{R}\in\text{SO}(2) . Then, choosing \mathbfsfit{Q}=\mathbfsfit{R}^{\top} in the last condition, we obtain \mathbfsfit{T}(\mathbfsfit{F})=\mathbfsfit{T}(\mathbfsfit{FQ})=\mathbfsfit{T}\bigl((\mathbfsfit{VR})\mathbfsfit{R}^{\top}\bigr)=\mathbfsfit{T}(\mathbfsfit{V}) . Moreover, we have \mathbfsfit{T}\bigl(\mathbfsfit{Q}{1}\mathbfsfit{F}\mathbfsfit{Q}{2}^{\top}\bigr)=\mathbfsfit{Q}{1}\mathbfsfit{T}\bigl(\mathbfsfit{F}\mathbfsfit{Q}{2}^{\top}\bigr)\mathbfsfit{Q}{1}^{\top}=\mathbfsfit{Q}{1}\mathbfsfit{T}(\mathbfsfit{F})\mathbfsfit{Q}{1}^{\top} for all \mathbfsfit{Q}{1},\mathbfsfit{Q}{2}\in\text{SO}(2) . Taking \mathbfsfit{Q}{2}=\mathbfsfit{Q}{1}\mathbfsfit{R} , and using the previous result, this becomes {\mathbfsfit{T}\bigl(\mathbfsfit{QVQ}^{\top}\bigr)=\mathbfsfit{Q}\mathbfsfit{T}(\mathbfsfit{V})\mathbfsfit{Q}^{\top}} for all \mathbfsfit{Q}\in\text{SO}(2) . Now let \smash{\mathbfsfit{B}=\mathbfsfit{FF}^{\top}=\mathbfsfit{V}^{2}} be the left Cauchy–Green tensor [ 1 , 2 , 29 ] . Write {\mathbfsfit{T}(\mathbfsfit{V})\!=\!\mathcalbf{T}\smash{\bigl(\mathbfsfit{V}^{2}\bigr)}\!=\!\mathcalbf{T}(\mathbfsfit{B})} , so {\mathbfsfit{Q}\mathcalbf{T}(\mathbfsfit{B})\mathbfsfit{Q}^{\top}=\mathbfsfit{T}\bigl(\smash{\mathbfsfit{QVQ}}^{\top}\bigr)=\mathcalbf{T}\bigl((\smash{\mathbfsfit{QVQ}}^{\top})^{2}\bigr)=\mathcalbf{T}\bigl(\smash{\mathbfsfit{QBQ}}^{\top}\bigr)} for all \mathbfsfit{Q}\in\text{SO}(2) . This condition says that \mathbfsfit{T} is an isotropic function of \mathbfsfit{B} 1 1 1 In hyperelasticity [ 29 ] , an analogous argument establishes that the energy density W is a (scalar) isotropic function of \mathbfsfit{B} , but we are not aware of a reference for this result in Cauchy elasticity. . The irreducible representations of 2D isotropic second-rank tensors are known in general [ 31 ] . These results now imply that \displaystyle\mathbfsfit{T}=t_{1}\mathbfsfit{I}+t_{2}\mathbfsfit{B}+t_{3}[\mathbfsfit{B},\teps]+t_{4}\teps, (1a) in which \mathbfsfit{I} is the identity, \teps is the 2D alternating symbol, and [\mathbfsfit{B},\teps]=\mathbfsfit{B}\teps-\teps\mathbfsfit{B} , and where t_{1} , t_{2} , t_{3} , t_{4} are isotropic scalar functions, so [ 31 ] are functions of \mathcal{I}=\tr{\mathbfsfit{B}} and \mathcal{J}=\det{\mathbfsfit{F}} , since \det{\mathbfsfit{B}}=\mathcal{J}^{2} . Geometrically, \teps is rotation by \pi/2 . This term is allowed by objectivity and isotropy to appear in 2D Cauchy elasticity because rotations commute in 2D. They do not commute in 3D however; fundamentally, this is why there is no linear isotropic odd elasticity in 3D [ 26 ] . By contrast \mathbfsfit{T}=t_{1}\mathbfsfit{I}+t_{2}\mathbfsfit{B} in 2D hyperelasticity 2 2 2 The corresponding result in 3D hyperelasticity is well-known [ 29 , 33 ] , but we are not aware of an exact reference for the 2D result: since the energy density W is a scalar function of \mathbfsfit{B} only [ 30 ] , it is a function of \mathcal{I},\mathcal{J} only; the definition \mathbfsfit{T}=\mathcal{J}^{-1}\mathbfsfit{F}\partial W/\partial\mathbfsfit{F} [ 29 ] then yields the representation \mathbfsfit{T}=t_{1}\mathbfsfit{I}+t_{2}\mathbfsfit{B} , where t_{1}=\partial W/\partial\mathcal{J} and t_{2}=2\mathcal{J}^{-1}\partial W/\partial\mathcal{I} . Thus, unlike in Eq. ( 1a ) , t_{1},t_{2} are not arbitrary. This means that 2\partial t_{1}/\partial\mathcal{I}\neq t_{2}+\mathcal{J}\partial t_{2}/\partial\mathcal{J} introduces odd behaviour, too. We will raise a similar point in our discussion of 3D nonlinear odd elasticity. . Thus the last two terms in Eq. ( 1a ) introduce odd behaviour. The second of these additionally breaks angular momentum conservation, because \mathbfsfit{T}\neq\mathbfsfit{T}^{\top} if t_{4}\neq 0 . It is useful to express Eq. ( 1a ) in terms of the deviatoric part of \mathbfsfit{B} , \dev{\mathbfsfit{B}}=\mathbfsfit{B}-(\mathcal{I}/2)\mathbfsfit{I} . Renaming coefficients, we therefore write \displaystyle\mathbfsfit{T}=t_{1}\mathbfsfit{I}+t_{2}\dev{\mathbfsfit{B}}+t_{3}[\mathbfsfit{B},\teps]+t_{4}\teps. (1b) We also require that there be no residual stresses, i.e., \mathbfsfit{T}=\mathbfsf{0} if \mathbfsfit{F}=\mathbfsfit{I} . This implies that t_{1}\mathbfsfit{I}+t_{4}\teps=\mathbfsf{0} and hence t_{1}=t_{4}=0 if \mathcal{I}=2 , \mathcal{J}=1 . To understand the physical meaning of t_{1},t_{2},t_{3},t_{4} , we linearise Eq. ( 1b ) around \mathbfsfit{F}=\mathbfsfit{I}\Rightarrow\mathcal{I}=2,\mathcal{J}=1 by introducing a small displacement \boldsymbol{x}=\boldsymbol{X}+\boldsymbol{u} , so that \mathbfsfit{F}=\mathbfsfit{I}+\boldsymbol{\nabla u} with \|\boldsymbol{\nabla u}\|\ll 1 . In the Supplemental Material 3 3 3 See Supplemental Material at [url to be inserted], which includes Refs. [ 35 , 36 , 37 , 1 , 29 , 38 , 3 ] , for details of the calculations. , we show that, in this limit, \displaystyle\mathbfsfit{T}=\mathbfsfit{t}+\mathcal{O}\bigl(\|\boldsymbol{\nabla u}\|^{2}\bigr), (2a) with \displaystyle\mathbfsfit{t} \displaystyle=t_{1}^{(1)}(\boldsymbol{\nabla\cdot\boldsymbol{u}})\mathbfsfit{I}+t_{2}^{(0)}\dev{\mathbfsfit{S}}+t_{3}^{(0)}[\mathbfsfit{S},\teps] \displaystyle\qquad+t_{4}^{(1)}(\boldsymbol{\nabla\cdot\boldsymbol{u}})\teps, (2b) where \mathbfsfit{S}=\bigl(\boldsymbol{\nabla u}+\boldsymbol{\nabla u}^{\top}\bigr)/2 , and t_{1}^{(1)},t_{2}^{(0)},t_{3}^{(0)},t_{4}^{(1)} result from expansions of t_{1},t_{2},t_{3},t_{4} near \mathcal{I}=2,\mathcal{J}=1 . We map this result onto the constitutive equations of linear two-dimensional odd elasticity [ 35 ] , \mathbfsfit{t}=\mathbfsfit{C}:\boldsymbol{\nabla u} , with \displaystyle C_{ijk\ell} \displaystyle=\kappa\,\delta_{ij}\delta_{k\ell}+\mu\big(\delta_{ik}\delta_{j\ell}+\delta_{i\ell}\delta_{jk}-\delta_{ij}\delta_{k\ell}\big) \displaystyle\qquad+A\,\varepsilon_{ij}\delta_{k\ell}-K\big(\varepsilon_{ik}\delta_{j\ell}+\varepsilon_{j\ell}\delta_{ik}\big), (3) in which \kappa is the bulk modulus, \mu is the shear modulus, and A,K are odd moduli [ 4 ] . This determines \smash{t_{1}^{(1)}}=\kappa , \smash{t_{2}^{(0)}}=2\mu , \smash{t_{3}^{(0)}}=-K , \smash{t_{4}^{\smash{(1)}}}=-A [ 34 ] , and hence exhbits the physical meaning of t_{1},t_{2},t_{3},t_{4}(\mathcal{I},\mathcal{J}) , but does not of course determine them uniquely. In what follows, we specialise to the simplest odd constitutive relations: we require that t_{1},t_{2},t_{3},t_{4} be independent of \mathcal{I} and choose their dependence on \mathcal{J} by analogy with 2D compressible neo-Hookean elasticity [ [Theconstitutiverelationsofneo--Hookeanelasticityare , again , betterknowninthe3Dcase~\cite[cite]{{\color[rgb]{0.1328 , 0.1875 , 0.5898}[}\@@bibref{Number}{ogden , 2 , goriely}{}{}{\color[rgb]{0.1328 , 0.1875 , 0.5898}]}};forthe2Dcase , see , e.g. , ][and]Chen2017 , *Zhao2021 ] . With this, we reduce Eqs. ( 1 ) to \displaystyle\mathbfsfit{T}=\kappa(\mathcal{J}-1)\mathbfsfit{I}+\dfrac{\mu}{\mathcal{J}^{2}}\dev{\mathbfsfit{B}}-\dfrac{K}{2\mathcal{J}^{2}}[\mathbfsfit{B},\teps]-A(\mathcal{J}-1)\teps. (4a) In particular, \mathbfsfit{T}=\kappa(\mathcal{J}-1)\mathbfsfit{I}+\mu\mathcal{J}^{-2}\dev{\mathbfsfit{B}} if {A=K=0} . This is the stress in a neo-Hookean solid with strain energy {W=\bigl[\mu\bigl(\mathcal{J}^{-1}\mathcal{I}-2\bigr)+{\kappa}(\mathcal{J}-1)^{2}\bigr]/2} [ 36 , 40 ] . We extend this relation to the incompressible case \mathcal{J}=1 by introducing the pressure p . On absorbing the isotropic part of \mathbfsfit{B} into p , \displaystyle\mathbfsfit{T}=-p\mathbfsfit{I}+\mu\mathbfsfit{B}-\dfrac{K}{2}[\mathbfsfit{B},\teps]. (4b) In this model, {\mathbfsfit{T}=\mathbfsfit{T}^{\top}} , so angular momentum is conserved, but this need not be the case. Indeed, for linearised deformations, \mathcal{J}=1\Rightarrow\boldsymbol{\nabla\cdot u}=0 , so \mathbfsfit{t}=\mathbfsfit{t}^{\top} from Eq. ( 2b ) and angular momentum is conserved. However, \mathcal{J}=1 , \mathcal{I}\neq 2 and hence t_{4}\neq 0 are allowed for finite deformations. Thus angular momentum conservation can be broken in isotropic incompressible odd elasticity for finite deformations only. Figure 1: The compressible 2D Rivlin problem. (a) Bifurcation diagram, for A=K=0 , of a square \mathcal{S} under dead load tractions \tau (inset): For {\tau<0} (compression), the square shrinks past \tau=\tau_{\text{min}} towards the fully compressed state \lambda=\tau=0 . For \tau>0 (traction), the square expands. At \tau=\tau_{\ast} , the square branch bifurcates to two rectangular branches. (b) Corresponding bifurcation diagram for an odd square satisfying the parameter matching condition \mu A+\kappa K=0 . The square rotates for \tau\neq 0 . At \tau=\tau_{\ast} , the rotated square bifurcates to a parallelogram. The odd compressible Rivlin square. We apply this minimal model of 2D isotropic nonlinear odd elasticity to the Rivlin problem [ 27 , 28 , 41 , 42 , 43 , 38 , 44 ] : a square \mathcal{S} deforms under dead load tractions \tau applied to its undeformed faces 4 4 4 The Rivlin problem is a solvable paradigm of nonlinear elasticity [ 44 ] , but experimental realisations are tricky, if not impossible [ 38 ] because, with the assumed dead loads, the total force on each side of \mathcal{S} is constant, so the (Cauchy) stress on the deformed faces is not constant. . At a critical load \tau=\tau_{\ast} , a bifurcation breaks the symmetry of the square shape [Fig. 1 (a) ]. This bifurcation is absent from linear elasticity. The dead load boundary conditions are {\mathbfsfit{P}\cdot\boldsymbol{N}=\tau\boldsymbol{N}} on the boundary of \mathcal{S} , where \mathbfsfit{P}=\mathcal{J}\mathbfsfit{TF}^{-\top} is the first Piola–Kirchhoff stress [ 29 ] and \boldsymbol{N} is the (outward) normal to \mathcal{S} . For homogeneous deformations, this implies {\mathbfsfit{P}=\tau\mathbfsfit{I}} . We start by solving for isotropic deformations \mathbfsfit{F}=\lambda\mathbfsfit{R} , where \lambda>0 is an isotropic stretch and \mathbfsfit{R} is rotation by an angle \theta . In the Supplemental Material [ 34 ] , we obtain \displaystyle\tau=\bigl(k^{2}+A^{2}\bigr)^{1/2}\lambda\bigl(\lambda^{2}-1\bigr), \displaystyle\tan{\theta}=\dfrac{A}{\kappa}. (5) Thus the square shrinks ( \lambda<1 ) as the compression {\tau<0} increases in magnitude, until a minimum value \tau=\tau_{\min} [Fig. 1 (a) ]. Because the dead load tractions apply a constant force per undeformed area, the compressive force becomes smaller as the square shrinks towards the fully compressed \lambda=0 . Next, we seek anisotropic homogeneous deformations: we introduce the polar decomposition \mathbfsfit{F}=\mathbfsfit{VR} , with {\mathbfsfit{V}=\operatorname{diag}(\lambda_{1},\lambda_{2})} . We find [ 34 ] that an anisotropic solution, \lambda_{1}\neq\lambda_{2} exists only if \mu A+\kappa K=0 . This is satisfied for an even square ( A=K=0 ). In this case, \mathbfsfit{R}=\mathbfsfit{I} and these solutions define two rectangular branches bifurcating off the square branch at \tau=\tau_{\ast} [Fig. 1 (a) ], corresponding to \lambda=\lambda_{\ast} , with [ 34 ] \displaystyle\lambda_{\ast}^{2}=\dfrac{1+\sqrt{1+8\mu/\kappa}}{2}. (6) The odd case is more interesting: \mu A+\kappa K\neq 0 in general, so there is no anisotropic homogeneous deformation and the square does not bifurcate. This result for the nonlinear Rivlin problem contrasts with problems of linear elasticity, in which oddness gives rise to behaviour absent from even materials [ 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 21 , 20 , 22 , 23 ] . If however {\mu A+\kappa K=0} , then \mathbfsfit{R}\neq\mathbfsfit{I} , and the rotated square bifurcates to a parallelogram at the same \tau=\tau_{\ast} , \lambda=\lambda_{\ast} [Fig. 1 (b) ]. It is known that the even Rivlin problem only admits homogeneous deformations under the mild assumptions of Ref. [ 28 ] . This argument does not carry over to the odd case, so we analyse inhomogeneous deformations by solving the equations of incremental elasticity [ 1 ] , around the isotropic deformation \mathbfsfit{F}=\lambda\mathbfsfit{R} , for a finite-wavelength increment \delta\boldsymbol{u}=\boldsymbol{a}\,\mathrm{e}^{\mathrm{i}\boldsymbol{q}\cdot\boldsymbol{X}} , in which \boldsymbol{q} is a polarisation vector, and \boldsymbol{a} is the magnitude of the increment. This calculation, which we postpone to the Supplemental Material [ 34 ] , leads to \mathbfsfit{Q}\cdot\boldsymbol{a}=\boldsymbol{0} , in which \mathbfsfit{Q} is interpreted as an acoustic tensor [ 1 , 46 ] . There is a nontrivial solution, \boldsymbol{a}\neq\boldsymbol{0} , if and only if [ 34 ] \displaystyle 0=\det{\mathbfsfit{Q}}=\bigl(\kappa\mu-AK\bigr)+\dfrac{\mu^{2}+K^{2}}{\lambda^{4}}, (7) which requires \kappa\mu-AK<0 . Hence inhomogeneous deformations first arise at \tau=\tau_{\ast\ast} , corresponding to \displaystyle\lambda=\lambda_{\ast\ast}=\left(\dfrac{\mu^{2}+K^{2}}{AK-\kappa\mu}\right)^{1/4}. (8) Examples of these inhomogeneous deformation modes are shown in Fig. 2 . We expect odd materials to still resist compression or simple shear, i.e., \kappa,\mu>0 [ [Forevenmaterials , thermodynamicstability , $W⩾0$ , requires$κ⩾0$and$μ⩾0$:see , e.g. , ][ , butthisargumentcannotbemadeforoddmaterials.]landaulifshitz ] , so this shows that inhomogeneous solutions exist only if the material is “sufficiently” odd, AK>\kappa\mu>0 . Again, this is different from linear odd elasticity, where new behaviour arises for any A,K\neq 0 . We also note that the condition AK>\kappa\mu>0 is incompatible with \mu A+\kappa K=0 . Figure 2: Bifurcations of a very odd compressible Rivlin square. If AK>\kappa\mu , then inhomogeneous deformations bifurcate off the square branch at \tau=\tau_{\ast\ast} . Insets show possible modes [ 34 ] . We can analyse the incompressible Rivlin square similarly. We find [ 34 ] the homogeneous bifurcation condition to be (2\mu-\tau_{\ast})^{2}+4K^{2}=0 . Thus an even square ( K=0 ) bifurcates at \tau_{\ast}=2\mu , but, again, there is no bifurcation for an odd square ( K\neq 0 ). Moreover, there are no inhomogeneous solutions in the incompressible case [ 34 ] . 3D isotropic, incompressible nonlinear odd elasticity. In 3D, the Cauchy stress in an isotropic Cauchy elastic solid can be written as [ 3 ] \displaystyle\mathbfsfit{T}=-p\mathbfsfit{I}+\alpha(\mathcal{I}_{1},\mathcal{I}_{2})\mathbfsfit{B}+\beta(\mathcal{I}_{1},\mathcal{I}_{2})\mathbfsfit{B}^{2}, (9) in which p is pressure, \alpha,\beta are (arbitrary) functions of the principal invariants \mathcal{I}_{1}=\tr{\mathbfsfit{B}} , {\mathcal{I}_{2}=[(\tr{\mathbfsfit{B}})^{2}-\tr{\mathbfsfit{B}^{2}}]/2} . For a hyperelastic material with isotropic energy density W(\mathcal{I}_{1},\mathcal{I}_{2}) , the Cauchy stress is also of this form, with \alpha=2W_{,\mathcal{I}_{1}}+2\mathcal{I}_{1}{W}_{,\mathcal{I}_{2}} , \beta=-2{W}_{,\mathcal{I}_{2}} [ 29 ] , where commata denote differentiation. This implies {W_{,\mathcal{I}_{1}}=(\alpha+\mathcal{I}_{1}\beta)/2} , W_{,\mathcal{I}_{2}}=-\beta/2 . A necessary and sufficient condition for such W to exist is (W_{,\mathcal{I}_{1}})_{,\mathcal{I}_{2}}=(W_{,\mathcal{I}_{2}})_{,\mathcal{I}_{1}} [ 48 ] , which reduces to \displaystyle\frac{\partial\alpha}{\partial\mathcal{I}_{2}}+\mathcal{I}_{1}\,\frac{\partial\beta}{\partial\mathcal{I}_{2}}+\frac{\partial\beta}{\partial\mathcal{I}_{1}}=0. (10) This is not however a sufficient condition for the material described by Eq. ( 9 ) to be even, because of the additional constraint of thermodynamic stability [ 47 ] , that W\geqslant 0 for all deformations. This makes the point that even a hyperelastic solid can be odd. By analogy with polynomial hyperelastic constitutive laws [ 49 , 1 ] , we can expand \alpha,\beta as power series in {\mathcal{I}_{1}-3} , \mathcal{I}_{2}-3 [ 34 ] . In what follows, we focus on the minimal case in which this expansion is truncated at linear order, \displaystyle\alpha(\mathcal{I}_{1},\mathcal{I}_{2}) \displaystyle=\alpha_{00}+\alpha_{10}(\mathcal{I}_{1}-3)+\alpha_{01}(\mathcal{I}_{2}-3), (11a) \displaystyle\beta(\mathcal{I}_{1},\mathcal{I}_{2}) \displaystyle=\beta_{00}+\beta_{10}(\mathcal{I}_{1}-3)+\beta_{01}(\mathcal{I}_{2}-3). (11b) Clearly, condition ( 10 ) holds for all \mathcal{I}_{1},\mathcal{I}_{2} if and only if \alpha_{01}=-\beta_{10} and \beta_{01}=0 . Expansions ( 11 ) thus define a six-dimensional space of Cauchy elastic materials, with a four-dimensional subspace of hyperelastic materials. The odd Rivlin cube. We apply this minimal model of 3D incompressible odd elasticity to the Rivlin cube [ 27 , 28 , 41 , 42 , 43 , 44 ] . We seek a homogeneous deformation {\mathbfsfit{F}=\mathrm{diag}(\lambda_{1},\lambda_{2},\lambda_{3})} , with {\lambda_{1}\lambda_{2}\lambda_{3}=1} by incompressibility. Then \mathbfsfit{B}=\mathrm{diag}(\lambda_{1}^{2},\lambda_{2}^{2},\lambda_{3}^{2}) , {\mathbfsfit{B}^{2}=\mathrm{diag}(\lambda_{1}^{4},\lambda_{2}^{4},\lambda_{3}^{4})} . On substitution into Eq. ( 9 ) , this determines the first Piola–Kirchhoff stress \mathbfsfit{P}=\mathbfsfit{TF}^{-\top} . For dead loads \tau and the assumed homogeneous deformations, \mathbfsfit{P}=\tau\mathbfsfit{I} . Hence \displaystyle\tau\lambda_{i}=-p+\alpha\lambda_{i}^{2}+\beta\lambda_{i}^{4},\quad i=1,2,3. (12) Subtracting these equations from each other yields \displaystyle(\lambda_{i}-\lambda_{j})(m_{i,j}-\tau)=0,\quad i,j=1,2,3, (13) where m_{i,j}=(\lambda_{i}+\lambda_{j})\left[\alpha+\beta\left(\lambda_{i}^{2}+\lambda_{j}^{2}\right)\right] . This allows the three solution classes of the Rivlin problem [ 27 , 28 , 44 ] : (1) The trivial solution is \lambda_{1}=\lambda_{2}=\lambda_{3}=1 . (2) In a biaxial deformation, without loss of generality, \lambda_{1}=\lambda_{2}=\lambda , {\lambda_{3}=\lambda^{-2}} , with \lambda\neq 1 . Equations ( 13 ) then reduce to \tau=m_{1,2} , whence \displaystyle\tau(\lambda)=\frac{\alpha\big(\lambda^{4}-\lambda^{-2}\big)+\beta\big(\lambda^{6}-\lambda^{-6}\big)}{\lambda^{3}-1}. (14) To determine the point \tau=\tau_{\ast} at which this biaxial solution bifurcates off trivial one, we take the limit \lambda\to 1 in Eq. ( 14 ) . This yields \tau\to\tau_{\ast}=2\mu . (3) In a triaxial deformation, \lambda_{1}\neq\lambda_{2}\neq\lambda_{3}\neq\lambda_{1} , so m_{1,2}=m_{2,3}=m_{3,1}=\tau from Eq. ( 13 ) . This implies that 0=m_{1,2}-m_{3,1}=(\lambda_{2}-\lambda_{3})(\alpha+\beta r) , where r=\lambda_{1}^{2}+\lambda_{2}^{2}+\lambda_{3}^{2}+\lambda_{1}\lambda_{2}+\lambda_{2}\lambda_{3}+\lambda_{3}\lambda_{1} . Since \lambda_{2}\neq\lambda_{3} , \displaystyle\alpha+\beta r=0. (15) Conversely, if \alpha+\beta r=0 , then m_{1,2}=m_{2,3}=m_{3,1} . Now, on the trivial branch, m_{1,2}=m_{2,3}=m_{3,1}=2\mu and r=6 , so \alpha_{00}+6\beta_{00}=0 is a necessary condition for a triaxial solution to bifurcate off the trivial branch, which happens at \tau=\tau_{\ast}=2\mu , too. In the Supplemental Material [ 34 ] , we solve Eq. ( 15 ) by expansion near \lambda_{1}=\lambda_{2}=\lambda_{3}=1 to obtain necessary and sufficient conditions for such a triaxial bifurcation off the trivial branch, \alpha_{00}+6\beta_{00}=0 and 4(\alpha_{01}+\alpha_{10})+5\beta_{00}+24(\beta_{01}+\beta_{10})=0 . The classical biaxial bifurcation of an incompressible Rivlin cube [ 27 , 28 ] thus persists in the space of Cauchy elastic materials defined by Eqs. ( 11 ) . There is a subset of these materials with an additional, triaxial bifurcation of the trivial branch, but the hyperelastic space in which this additional bifurcation happens is only a two-dimensional subspace of this four-dimensional space. In other words, this additional bifurcation behaviour of the Rivlin cube requires more parameter matching conditions for an even material than for an odd one, in contrast with our results for the Rivlin square. In this Letter, we have constructed minimal models of nonlinear isotropic odd elasticity and, through the minimal example of the Rivlin problem, shown the surprising behaviour they afford: In 2D, oddness suppresses, in general, the behaviour of a passive Rivlin square, in contrast to the rich behaviour of isotropic linear odd elasticity in 2D. Somewhat conversely, there is no isotropic linear odd elasticity in 3D, but the bifurcation behaviour of a Rivlin cube survives oddness, and new bifurcations can appear. More conceptually, our work has taken some of the first steps in the programme of nonlinear Cauchy elasticity set out in Ref. [ 3 ] . We have analysed one possible, minimal extension of linear odd elasticity, which is of course not unique. For example, we have focused here on materials with a positive linear shear modulus, \mu>0 , but purely nonlinear elasticity [ 50 ] at \mu=0 , with a positive nonlinear shear modulus is allowed thermodynamically. A fuller understanding of the effect of these different nonlinearities on different mechanical bifurcations therefore remains an important question for future work. Moreover, we have analysed isotropic materials. Anisotropy can lead to linear odd moduli even in 3D [ 4 ] . What additional terms does microscopic structure, resulting, e.g., from nematic order, allow in the nonlinear relation between \mathbfsfit{T} and \mathbfsfit{F} ? These additional terms are well-studied in hyperelasticity [ 29 ] , but we are not aware of corresponding results for Cauchy elasticity. Classifying the tensor functions of \mathbfsfit{F} satisfying objectivity and the symmetries allowed by this additional structure therefore remains a challenge for future work. More abstractly, we have introduced a natural polynomial expansion of the Cauchy stress \mathbfsfit{T} in {\mathcal{I}_{1}-3} , \mathcal{I}_{2}-3 in Eq. ( 9 ) , motivated by the analogous expansion of the energy density W that defines polynomial hyperelastic laws [ 1 , 49 ] . However, even the lowest-order truncation of \mathbfsfit{T} , with \alpha(\mathcal{I}_{1},\mathcal{I}_{2})=\alpha_{00} , \beta(\mathcal{I}_{1},\mathcal{I}_{2})=\beta_{00} , is seen to imply W\propto 2(\alpha_{00}+3\beta_{00})(\mathcal{I}_{1}-3)-2\beta_{00}(\mathcal{I}_{2}-3)+\beta_{00}(\mathcal{I}_{1}-3)^{2} .By contrast, the lowest-order polynomial (Mooney–Rivlin) law [ 29 , 49 ] , {W\propto w_{10}(\mathcal{I}_{1}-3)+w_{01}(\mathcal{I}_{2}-3)} is less nonlinear. In other words, these two natural expansions do not “commute”, the consequences of which remain another open problem. A more general question remains open, too: What are biological systems or metamaterials that realise the nonlinear odd effects that we have revealed here? Acknowledgements. We thank Vincenzo Galgano for a conversation on tensor representations and gratefully acknowledge funding from the Max Planck Society. 33 35 , 36 , 37 , 1 , 29 , 38 , 3