Inverse Lagrangian Formulation for the Deformation of Hyperelastic Solids Wei Hong* Department of Aerospace Engineering, Iowa State University, Ames, IA 50010 Abstract Although the field equations of solid mechanics are commonly written in the undeformed configuration, neither fundamental theories nor mathematics prohibits the analysis from being carried out in a deformed configuration. In this letter, following recent developments in inverse deformation problems, the governing equations for static hyperelasticity problems are formulated in the current configuration after deformation. An inverse mapping from the deformed geometry to the original one is solved as the unknown field. Such an approach, herein referred to as the inverse Lagrangian formulation, is exemplified with several applications in addition to the inverse problem of known deformed geometry by design and determination of the original geometry. Applications are found in steady-state fluid-structure interaction problems, such as the design of microfluidic devices. More importantly, the method is also found to be useful in analyzing mechanical instability and bifurcation problems, as the inverse mapping from a buckled state to the original remains unique. Keywords inverse Lagrangian formulation; inverse design problem; inverse elastostatics; instability *Email: whong@iastate.edu 1. Introduction In continuum mechanics, the Lagrangian formulation, which traces the motion of material particles from a specific reference state, is often used for the deformation of solid structures. On the other hand, the Eulerian formulation which directly solves for the fields in the current state, is more often used for fluid dynamics problems. Although tremendous effort has been put on integrating the different algorithms in the two formulations, such as the Arbitrary-Lagrangian-Eulerian (ALE) method which bridges the two by introducing a deforming mesh (Hirt et al, 1974; Liu et al, 1988), solving soliddeformation problems directly in a deformed state is not as common. Fundamental theories never prevent the Eulerian formulation of elasticity problems. In fact, the Navier-Stokes equations are equally 1 applicable to deformable solids. The technical difficulties hindering the direct implementation of elasticity problems in Eulerian formulation are two folds: the constitutive laws of solids, especially elastic solids, usually require the knowledge of the undeformed configurations; the deformed geometry of a solid structure is often unknown. In the last two decades, various numerical methods have been developed to tackle the inverse elasto-static problem (e.g. Govindjee and Mihalic, 1996, 1998; Yamada, 1998; Koishi and Govindjee, 2001; Lu et al., 2007, 2008; Rajagopal et al., 2007; Fachinotti et al., 2008; Gee et al., 2009; Albanesi et al., 2010; Sellier, 2011): Given the deformed geometry of a body subject to prescribed loads, what is the original undeformed geometry? While differ in mathematical details, most studies formulate the problem in the deformed configuration. To obtain the geometry of the undeformed state, the mapping from the original to the current spatial location of the same material particle is often calculated as the main dependent variable, just as in the common Lagrangian formulation of hyperelasticity. However, instead of a function the original coordinates of the particle, the mapping is written as a field in the deformed frame. One direct application of these methods is the inverse design problem, in which the deformed shape of a part under the actual working condition is of practical interest and is prescribed by design. With the geometry designed in the working condition under the desired loads, the original undeformed geometry, together with the stress and strain fields, is obtained as a solution to the mathematical problem. Although the resulting undeformed geometry could be more complex and tends to have non-straight edges or non-flat surfaces, the fabrication would not be overly complicated due to the widely available additive manufacturing technologies. Practical examples of this kind include the design of high-precision forming (Sellier, 2006), turbine blades (Fachinotti et al, 2008), compliant mechanisms (Albanesi et al, 2013), and structures with elastomeric parts (Govindjee and Mihalic, 1996, 1998; Koishi and Govindjee, 2001). A related but special application is the stress analysis of biological tissues, of which the geometry of the undeformed reference state is unknown (e.g. Rajagopal et al., 2007; Lu et al., 2007, 2008; Gee et al., 2009, 2010). With the field equations for the elastic deformation written in the deformed configuration, such methods are naturally compatible with the governing equations in fluid dynamics which are usually in Eulerian formulation. Fluid-structure interaction problems, especially those in steady-state, can be solved entirely over the deformed frame without invoking ALE or moving boundaries (Sellier, 2011; Vavourakis et al., 2011). As will be shown in the current letter, such an approach is very useful in designing soft structures for the purpose of directing fluid flow, e.g. in a microfluidic device. 2 The mathematical formulation in this letter follows closely to Fachinotti et al (2008), with the extension of handling the volumetric incompressibility by using a mixed displacement-pressure formulation. This formulation differs from Govindjee and Mihalic (1996, 1997) mainly in the material laws: the same material laws in terms of the Lagrangian strain measures are used just as regular hyperelasticity. In the literature, similar approaches do not carry a common name, and are instead often named after specific applications, such as the inverse design problem (Fachinotti et al, 2008) or the inverse elastostatics problem (Govindjee and Mihalic, 1996; Lu et al, 2007). However, as will be demonstrated in this letter, the method has wider applications than just the inverse design problems. To differentiate from common Eulerian approaches which solves the history-independent velocity or acceleration field, it is referred to as the inverse Lagrangian formulation in the current letter. Besides applications in inverse design problems and fluid-structure interaction, the main focus of the current letter is the application of the inverse Lagrangian approach to the analyses of instability problems in solid structures. Not only because of the importance in engineering applications, but also due to the intriguing mechanics, instability phenomena have always been a topic of interest. As an example, the rich and controllable surface instabilities on a soft material (or in a stiff-film-soft-substrate system) have been studied extensively (Biot, 1963; Huang et al., 2005; Genzer and Groenewold, 2006; Hong et al., 2009; Hohlfeld and Mahadevan, 2011; Li et al, 2012; Wang and Zhao, 2014) and found promising applications in material characterization (Stafford et al, 2004), fabrication of flexible electronics (Choi et al., 2007; Kim and Rogers, 2008; Baca et al., 2008), surface functionalization (Chan and Crosby, 2006; Zang et al., 2013). It is non-trivial, however, to numerically predict the onset of instability or carry out post-buckling analysis. Each instability instance mathematically corresponds to a bifurcation point, beyond which the solution is non-unique, i.e. there exists more than one mapping from the undeformed to the deformed state satisfying the equilibrium and boundary conditions. Besides the singularity of the linearized coefficient matrix, numerical calculation tends to stay on a uniform (trivial) solution instead of branching out to the physically stable one with lower energy. To overcome the critical point and perform post-buckling analysis, numerical schemes usually introduce imperfections of small amplitude as well as artificial damping to ensure convergence. In some circumstances, the result is found to be sensitive to the imperfection (Cao and Hutchinson, 2012). Nevertheless, if the mathematical problem is reformulated and instead the inverse mapping from the deformed state is sought for, no bifurcation may be involved and the solution is usually unique. As will be demonstrated in this letter, without the need for artificial imperfection or damping, the inverse 3 Lagrangian approach is a robust numerical method for determining the onset of instability as well as the post-buckling behavior in soft elastic solids. 2. The Inverse Lagrangian Formulation Although the method can be applied to general elastic or inelastic solids, it is illustrated here through the deformation of hyperelastic solids as a timely example. Let us first review the mathematical description via the commonly used Lagrangian approach. Consider a material particle located at X in the undeformed configuration, and denote its position after deformation by x . The mapping X x uniquely specifies the motion of all particles in the continuum, and the field of deformation is commonly described by the deformation gradient tensor FiK xi , X K (1) which is the spatial gradient of xX with respect to the undeformed geometry. The behavior of a hyperelastic material is often written in terms of the relation between the deformation gradient and its work conjugate, the nominal stress siK siK F W , FiK (2) where W is the elastic free energy per unit undeformed volume. The material behavior may also be written in terms of various other stress and strain measures. In equilibrium, the field of nominal stress is in balance with body force b in the bulk (e.g. Bower, 2009) siK bi 0 , X K (3) and surface traction t on any surface with prescribed load siK N K t i . (4) Here, the unit normal vector N is defined on the surface before deformation. Equations (1)-(3) form a closed differential system to determine the mapping xX when proper boundary conditions such as Eq. 4 (4) or a given displacement boundary condition are prescribed. Such a framework is commonly adopted in the analyses of hyperelasticity problems, and is often referred to as the Lagrangian formulation. The Lagrangian formulation is mathematically equivalent to the following framework which is based on the inverse mapping x X , if it exists. For (locally) differentiable Xx , the inverse deformation gradient H Ki X K xi (5) captures the field of deformation. H is the inverse tensor of the deformation gradient, H F 1 . In the deformed configuration, the equilibrium condition is more conveniently written in terms of the true stress tensor σ as (e.g. Bower, 2009) ij x j bi 0 , (6) and the surface traction is balanced by ij n j t i . (7) Here, the unit normal vector n is defined on the deformed surface. By using the geometric relation between different stress measures, the material constitutive relation (2) can be written in terms of the true stress ij siK F jK det F det H W F jK . FiK (8) Just like the Lagrangian formulation, Eqs. (5), (6), and (8) also consist a differential system, through which the inverse mapping Xx can be identified when proper boundary conditions are prescribed. This formulation described above follows closely to Fachinotti et al (2008), in which it was mainly applied to inverse design problems. In the limiting case of small deformation, when the difference between the deformed and undeformed geometries is negligible, the inverse Lagrangian formulation simply reduces to the Lagrangian formulation. Only for problems involving finite deformation, the two formulations will lead to different equations. 5 The inverse Lagrangian formulation is mathematically well posed. It is less often applied, perhaps because the deformed geometry of solid objects is seldom known. Nevertheless, some unique features of the inverse Lagrangian formulation make it suitable for the special circumstances as will be demonstrated in the following sections. Due to the mathematical similarity between the two formulations, almost all numerical methods for elasticity problems in the Lagrangian formulation will equally be applicable to those in the inverse Lagrangian formulation. Here as a demonstration, a finite-element method will be adopted. Through simple mathematical operations such as integration by parts and the divergence theorem, the differential equation (6) and the traction boundary condition (7) can be written into a weak form (Fachinotti et al, 2008) ui ij dv tiui da , ui x . b u i i x j t (9) Here, the integrations are carried out over the volume of the domain under consideration, and the surfaces t on which the traction boundary condition is prescribed. Both and t are defined in the deformed configuration. The displacement vector ux x Xx . Unlike its counterpart in the Lagrangian formulation, the weak form (9) may not have a clear physical meaning or correspond to a virtual work principal. However, the lack of physical interpretation does not prohibit the numerical implementation. In fact, the procedure is almost identical due to the mathematical similarity. The only difference is that all fields, known or unknown, are expressed as functions of the coordinates x in the deformed configuration. In the Lagrangian formulation, the thermodynamic stability dictates the stiffness tensor to be positive definite. Here, it is easy to show that the tensor derivative of the true stress σ with respect to the Cauchy deformation tensor c H T H is negative definite. Although the inverse Lagrangian formulation can take any constitutive models, here the neoHookean model will be used for the purpose of demonstration. With the free-energy function given by W 2 F : F 3 , a neo-Hookean solid has true stress σ det H F F T , where modulus of the material. 6 is the initial shear Hyperelastic materials are often assumed to be incompressible. Inspired by the displacementpressure mixed formulation (e.g. Zienkiewicz et al., 2005; Govindjee and Mihalic, 1998), we enforce the volume incompressibility by adding the pressure field px as a Lagrange multiplier to the weak form: F iK F jK det H p ij ui biui dv det H 1pdv tiui da , ui x, px . (10) x j t Here, the specific weak form (10) is taken such that the combination ij FiK FjK det H ij p ij (11) represents the true stress. 3. Inverse Problems a A Fig. 1. Axisymmetric deformation of a quarter-circular wedge. Constrained by frictionless rollers on the flat surfaces, the quarter cylinder of radius A is deformed into a half cylinder of radius a A 2 . The thickness direction is constrained and the mode of deformation is plane strain. To verify the formulation and the numerical implementation, we first test the method on an example with known analytical solution. As sketched in Fig. 1, a quarter cylinder of radius A is constrained with rollers on the two flat surfaces. Subject to a displacement boundary condition, one of the two surfaces is rotated by 2 , and the sample deforms into a half cylinder. The mode of deformation is plane strain, and the material is assumed to be incompressible neo-Hookean. The problem is kinematically determinate, and the field of deformation can be readily calculated from volume conservation. In polar coordinates, the principal stretches are r 1 the stress-stretch relation of a neo-Hookean material, the true stresses are 7 2 and 2 . Using r pr , 2 2 pr (12) where pr is the field of hydrostatic pressure, which could be determined from the equilibrium equation in the radial direction, d r dr r r 0 . Using the boundary condition r a 0 , the pressure field is calculated as pr r 1 3 ln , where a A 2 a 2 is the radius in the deformed configuration. The corresponding stress field is then given by 3 r r 2 ln a . 3 r 1 ln 2 a (13) Numerically, we build the 2D model in COMSOL Multiphysics under the inverse Lagrangian frame, by implementing the weak form (11) in the deformed semi-circular domain. Given displacement boundary conditions are prescribed on the two straight edges, and a traction-free boundary condition is prescribed on the curved edge. To aid the convergence, gradually changing rotation angles on one displacement boundary is applied. A triangular mesh is created on the semi-circular domain, denser near the center ( ~ 0.03a ) and coarser near the perimeter ( ~ 0.1a ). The resulting true stress distribution along the radius of the cylinder is plotted in Fig. 2, together with the analytical solution (13). The good agreement between the numerical and analytical results validates the method. r Fig. 2. Radial distribution of dimensionless true stresses, r and . The circles and squares are numerical results by using the finite element method via the inverse Lagrangian formulation, and the continuous curves are plotted from the anylitical solution, Eq. (13). 8 As introduced in Section 1, the inverse Lagrangian formulation is naturally suitable for solving inverse problems, namely the computation of the original undeformed geometry for manufacturing, when the geometry in the deformed state under working load is known (or specified by design). Limited by conventional manufacturing techniques, structural designs commonly chose relatively simple geometric features (e.g. flat surfaces and straight edges) in the undeformed configuration (when the part is machined). Emerging techniques of additive manufacturing no longer require the original shape to be simple. One can simply design the geometry in the working state, use the inverse Lagranian method to calculate the shape before deformation, and then directly print it. m Deformed Undeformed (calculated) Fig. 3. Deformation and stress distribution of an inflated tubular structure. By symmetry, only a quarter of the structure is modeled. The computation is carried out over the deformed geometry with the inverse Lagrangian formulation. The mesh grids in the deformed and undeformed configurations represent the distortion, instead of the actual elements for calculation. The shades indicate the distribution of the normalized mean stress m kk . As an illustrative example, let us consider the following inverse problem. A hollow tube made of rubbery elastomer is designed to work in an inflated state of internal pressure 0.1 , with being the material’s initial modulus. The functional design requires the external shape of the cross section to be circular and the internal to be a rounded square, as shown by Fig. 3. With conventional finite element method, such an inverse problem will require multiple rounds of trial and error (or parametric optimization). Here by taking the inverse Lagrangian approach, we formulate and solve the problem in 9 the deformed geometry within one step. The resulting undeformed shape and stress distribution are shown in Fig. 3. The resulting undeformed state has smooth but non-flat surfaces, which should not pose difficulty to existing 3D printing technologies (e.g. Shepherd et al., 2011). With a calibrated material constitutive relation, such an approach could lead to very accurate geometries for soft structures in their working conditions. 4. Fluid-Structure Interaction Since the inverse Lagrangian method solves the unknown fields in the deformed configuration, it can be seamlessly integrated with common fluid solvers, which are formulate in Eulerian description in terms of the current coordinates. If the deformed geometry of the immersed solid structure is known, the coupling between the flow and the solid structure will be unidirectional and easy to handle: the fluid flow provides pressure and viscous stress to the solid surface as boundary conditions, with which the fields of deformation and stress in the solid can be determined. Because the deformed shape is already prescribed, the fluid solver requires no additional information from the structural solution, and no further iteration is needed. p (kPa) Fig. 4. Steady flow around an inflated soft membrane. The mesh shows the undeformed position of the membrane, calculated via the inverse Lagrangian approach. The shades in the fluid domain shows the pressure distribution, and that in the membrane shows the hoop stress, both in the unit of kPa. The curves in the fluid domain represent the streamlines of the flow. The only technical difficulty remains in the determination of the deformed geometry. However, to most engineering applications, the deformed geometry is actually of more importance. Take microfluidic devices as an example, it has long been noticed that the deformation of soft components caused by fluid pressure may be significant and should be taken into consideration in design (Gervais et 10 al., 2006). Such complexity can be totally avoided if one directly designs the deformed geometry. To illustrate this application, we calculate the following microfluid device as an example. A soft rubbery membrane (initial shear modulus 0.2 MPa) is inflated in a fluid channel to perturb the flow, as shown by Fig. 4. The fluid channel has a height of 2mm and length of 5mm, and is filled with water (viscosity 1 mPa·s). It is assumed that a semi-circular shape of radius 1mm is preferred by design. The device operates at an inlet velocity of 1 m/s and the rubber membrane is inflated with an internal pressure of 5 kPa. For simplicity, the deformed thickness of the membrane is set to be 0.1 mm uniformly. For the fluid flow, non-slip boundary conditions are applied to both the top and bottom surfaces, and to the deformed membrane surface. For the hyperelastic membrane, the bottom edges are fixed, a uniform pressure is applied to the internal surface, and the calculated fluid pressure and viscous stress are applied to the outer surface. The flow problem is solved with the built-in steady laminar fluid solver in COMSOL Multiphysics, while the deformation of the elastic membrane is solved with the inverse Lagrangian formulation, also implemented in COMSOL Multiphysics. The resulting deformation and stress fields in the membrane are plotted in Fig. 4, together with the flow field in the fluid channel. For this soft membrane, the deformed geometry differs significantly from the original shape. Clearly, to work against the flow-induced pressure gradient in the fluid, the undeformed geometry of the membrane needs to be asymmetric and leaning towards upstream. If the as-manufactured membrane is symmetric and circular instead, the flow would be very different. To achieve the same design task with conventional approaches, even with the arbitrary Lagrangian-Eulerian formulation, much more effort is needed to determine the geometry for manufacture. Just as all nonlinear problems, the solution may be non-unique and the deformed configuration may be dependent on the actual deformation history. For problems with loads given in terms of displacements, it is possible to add intermediate steps to prescribe the loading history. In general, however, it is not easy to include such information in the current framework, as the intermediate geometries are often unknown. It is even more difficult for fluid-solid interaction problems in which the steady-state flow fields will take time to establish. To overcome these difficulties, one will need to extend the inverse Lagrangian formulation to include dynamic terms, so that the entire deformation process can be predicted incrementally, and the effect of loading history can be accounted for. Such a development, however, is beyond the scope of the current letter. 11 5. Instability and Bifurcation Problems For decades, instability phenomena have been a focal point for researches in the field of soft matter mechanics. The inevitable geometric nonlinearity brings varieties to the form of instability, and at the same time complicates the analysis. For example, conventional finite-element method formulated under the Lagrangian description may encounter singularity in the coefficient matrix at the critical point. In addition, due to the non-uniqueness in solution, the numerically converged solution may not be the physical one. In the absence of perturbation, numerical solvers tend to converge to the homogeneous solution even beyond the critical point, even though the homogeneous solution may be physically unstable. To obtain an inhomogeneous solution, artificial defects or numerical perturbations will need to be added. Such an approach lowers the accuracy of computation, especially in systems that are sensitive to the mode or amplitude of perturbation. Response S1 S2 f fc O Generalized load Fig. 5. Schematic illustration of a bifurcation problem. At the critical point into two branches. Under a higher load, undeformed state O: f c , the solution bifurcates f f c , multiple solutions exist for mapping from the original O S1 and O S2 . On the other hand, the inverse mapping from one of the deformed states, e.g. S2 O , could be unique. By using the inverse Lagrangian formulation, some of these numerical difficulties could be avoided. As illustrated by Fig. 5, beyond the critical point f c , the mathematical system with the conventional Lagrangian formulation loses the solution uniqueness, and has multiple solutions in terms of the mapping from the undeformed to the deformed state: O S1 and O S2 . Nevertheless, if the same problem is formulated in the deformed configuration via the inverse Lagrangian approach, the uniqueness of the solution could be maintained even beyond the critical point. From a deformed state 12 S2 , there is only one inverse mapping S2 O which satisfies the equilibrium and boundary conditions. The other solution S1 O , formulated over a different deformed geometry, now becomes an entirely different problem. y Deflection wx Deformed geometry P P x Undeformed geometry (to be solved) Fig. 6. Sketch of a classic Euler buckling problem considered under the inverse Lagrangian frame. The deformed geometry is taken to be known, while the undeformed geometry (or the deflection) is an unknown field to be solved. To show the mathematical difference between the two approaches, let us temporarily depart from the finite-element method and consider the classic Euler buckling problem. A column of length L and bending stiffness EI , is pinned at both ends as sketched in Fig. 6. Under the assumptions of the linear beam theory, the amplitude of the post-buckling deflection is indeterministic, and the stiffness is singular at the critical point. Now consider the equilibrium in the deformed state. With a deformed geometry given by y A sin x L , the moment balance requires that M Py . Substituting in the relation between the lateral deflection wx and the moment, we arrive at EI d 2w Py . dx (14) Equation (14) can be directly integrated to obtain the solution wx L2 PA x sin . 2 EI L (15) Unlike its counterpart via the conventional Lagrangian approach, solution (15) exists and is unique for any axial load P , although the general case would represent the bending and compression from an initially curved beam, as sketched in Fig. 6. The critical state of buckling corresponds to the case when the column is originally straight, or w y , which gives the classic result of the critical axial force 13 Pc 2 EI L2 . However, it does not cause the ill conditioning of the governing equation (14) or affect the uniqueness of the solution. a b y Fig. 7. (a) Sketch of a Euler buckling problem. (b) Deformation and stress distribution in the buckled hyperelastic column, in two modes of buckling. The shades indicate the normalized true axial stress y , and the mesh shows the calculated geometry of the undeformed configuration. To further demonstrate the application of the inverse Lagrangian formulation to instability analyses, we will now use it to solve an Euler buckling problem numerically. As sketched in Fig. 7a, a column of aspect ratio 10:1 is subject to axial compression. The rotation in both ends are constrained, but the top end moves freely in the lateral direction. The deformation is plane strain. For simplicity, the material is also assumed to be neo-Hookean and incompressible, so that the formulation in Section 2 can be directly applied. As shown by Fig. 7b, the computation is carried out in the domains corresponding to the deformed states. The deformed geometries are obtained by superimposing a sinusoidal deflection onto the neutral axis of the column. As examples, the first two modes of buckling are calculated, and the representative fields of deformation and stress are plotted in Fig. 7b. To identify the critical points and illustrate the bifurcation, we further plot the amplitudes of lateral deflection as a function of the axial compression for both modes in Fig. 8. The critical strain of each mode can be easily identified from the intersection between the computed curve corresponding to the buckled states and that of the homogeneously deformed states. The resulting critical strains are very close to those given by the Euler buckling theory: c 0.0082 for the first mode, and c 0.033 for the second mode. 14 Although the uniformly compressed states can also be calculate via the inverse Lagrangian approach, such calculation is omitted as the solutions all fall on the horizontal axis in Fig. 8. Deflection amplitude 0.25 0.2 0.15 0.1 0.05 0 0 0.02 0.04 Nominal compressive strain 0.06 Fig. 8. Deflection amplitudes in the first two buckling modes of an axially compressed column, plotted as functions of the nominal compressive strain. As the homogeneous solution is just the horizontal axis, the plot indicates bifurcation at the critical points. Now let us turn to the wrinkling instability of a stiff thin film bonded to a soft thick substrate, which has been extensively studied (Cerda and Mahadevan, 2003; Chen and Hutchinson, 2004; Huang et al., 2005; Genzer and Groenewold, 2006) due to various applications (e.g. Choi et al., 2007; Stafford et al., 2004). Here as an example, we consider the film-substrate system in which the initial shear modulus of the film is 100 times that of the substrate, and the substrate thickness is 30 times film thickness h before deformation. In contrast to the conventional approach, we solve the problem in the wrinkled geometry via the inverse Lagrangian formulation. As shown by Fig. 9, a rectangular region with sinusoidal top edge is used as the computation domain. Only half of the wrinkling wavelength is computed, and symmetry boundary conditions are assumed on left and bottom edges. For simplicity, incompressible neo-Hookean material model is adopted for both the film and the substrate, and the mode of deformation is taken to be plane strain. A displacement boundary condition is applied gradually over the right boundary. When the computed top surface flattens, the undeformed state is thought to be reached. It should be noted that due to the numerical error in the prescription of the deformed geometry, the top surface may not reach a mathematically flat state. In practice, the state in which the peak and trough points have the same vertical coordinate is regarded as the flat state. A representative solution is shown in Fig. 8, in which the mesh indicates the calculated original geometry, 15 and the shades represent the horizontal component of the Eulerian-Almansi strain e 1 2 I H T H . As shown by Fig. 8, for wrinkles of relatively small amplitude, the resulting undeformed top surface is relatively flat, indicating that the prescribed sinusoidal surface is quite close to the actual geometry of the wrinkled state. e xx Fig. 9. Calculated field of deformation of a wrinkled film-substrate system by using the inverse Lagrangian formulation. The shades indicate the horizontal component exx of the Eulerian-Almansi strain, and the mesh represents the undeformed geometry from computation. The mesh is also that used for finite element calculation. For the case shown, the wrinkle wavelength is taken to be 20 times the film thickness, and the substrate thickness is 30 times film thickness before deformation. Only a half period is included for computation and symmetry conditions are assumed on corresponding boundaries. To identify the critical point of the wrinkling instability, we plot the wrinkle amplitude as a function of the applied compression in Fig. 10a. Unlike the Euler column, here the wavelength of the wrinkles is a continuous variable: wrinkles of any wavelength could be a possible deformed state. As shown by Fig. 10a, each wavelength corresponds to a critical compressive strain, shown as the intersection between the corresponding curve and the horizontal axis. In the absence of external constraints or perturbation, the system will branch into the wrinkling state of the lowest critical strain. To determine this state, we plot the critical strains as a function of the dimensionless wavelength h in Fig. 10b. The dependence of the critical strain on the wrinkle wavelength is non-monotonic, and 16 shows a clear minimum point. For this example, the wavelength of the lowest critical strain is c 20h , and the corresponding nominal compressive strain for the onset of instability is c 0.0237 . Both values are very close to the analytical solution for the case of linear elastic materials and infinite substrate, c 20.2h and c 0.0241 (Chen and Hutchinson, 2004; Huang et al., 2005). As the strain is relatively low, the small discrepancy is mainly attributed to the finite thickness effect of the substrate. b 0.02405 0.024 0.02395 0.0239 0.02385 0.0238 0.02375 0.0237 0.02365 λ = 16h 0.5 Critical nominal strain εc Wrinkle amplitude A/h a 0.6 λ = 20h 0.4 λ = 28h 0.3 0.2 0.1 0 0.022 0.024 0.026 0.028 0.03 Nominal compressive strain ε 0.032 18 19 20 21 Wavelength (λ/h) 22 Fig. 10. (a) Normalized wrinkle amplitudes plotted as functions of the applied compressive strain, for various wrinkle wavelengths. The intersection of each curve with the horizontal axis corresponds to the critical point of instability. (b) Critical compressive strain for wrinkling instability as a function of the dimensionless wavelength h. The minimum point is the strain and wavelength of the wrinkles that will spontaneously develop in a large system with no additional constraints or perturbation. As the last example, we will study the instability of a periodical porous structure under equal biaxial compression. A 2D structure contains a periodical array of circular holes, as shown by Fig. 11a, is made of incompressible neo-Hookean solid. When compressed (either uniaxially or biaxially) beyond a critical strain, the four-fold symmetry of the structure will be broken, and the circular holes will turn into oval shapes of alternating orientations. This instability phenomenon has been demonstrated experimentally and analyzed numerically (Mullin et al., 2007; Bertoldi and Boyce, 2008; Bertoldi et al., 2008). Very recently, it is also proposed that such structures can be regarded as a meta material exhibiting phase transition (Yang et al., 2016). Here, it is used as an example to show case the application of the inverse Lagrangian formulation. The computation is carried out over a representative unit cell of the structure (quarter of that shown on Fig. 11a), and symmetric boundary conditions are applied on all edges except the free surface of the holes. The ratio between the hole diameter and the 17 center-to-center distance is taken to be 9:10 in the undeformed state. To prescribe the deformed geometry, we first utilize the incompressibility and evaluate the area of each hole under a given nominal compressive strain. For simplicity, we assume the deformed shape of each hole to be an ellipse. The aspect ratio of the ellipse is then determined through a line search to achieve a symmetric undeformed structure. Using such a strategy, we compute the deformation and stress distribution in both branches of the solution, and plot the results in Fig. 11. b Nominal compressive stress σ/μ a 0.04 0.035 0.03 0.025 0.02 0.015 0.01 0.005 0 0 0.01 0.02 0.03 Nominal compressive strain 0.04 Fig. 11. (a) Stable equilibrium state of a periodic porous structure under biaxial compression of linear strain 0.035 . The shades show the mean stress normalized by the initial modulus, m , and the mesh represents the computed undeformed geometry. The mesh is also that used for the finite-element calculation. (b) Nominal compressive stress as a function of the applied compressive strain, in both the symmetrically compressed state (unstable, circles) and the state with broken symmetry (squares). The intersection at c 0.007 is the bifurcation point or the onset of instability. A representative result from computation is shown by Fig. 11a, in which the mesh indicates the calculated undeformed geometry, and the shades represent the normalized mean stress m . By breaking the four-fold symmetry of the structure and rotating the region in the midst of four holes, the mode of deformation transitions from a hydrostatic-compression-dominated state to one mainly manifested by bending of the thin ligaments, and the overall stress is greatly relieved. It can also be noticed on Fig. 11a that the computed undeformed shape of the holes is not exactly circular. Such a discrepancy is again due to the inaccuracy in prescribing the deformed geometry. In other words, the result suggests that the buckled shape of the holes is not exactly an ellipse. While the numerical results 18 could be further improved, e.g. by describing the deformed geometry with more parameters and carrying out a parametric optimization, such a process is not the main focus of the current letter. In Fig. 11b, by integrating the true stresses along the boundaries, we computed the nominal stress of the structure, and plotted it against the applied strain for both non-buckled and buckled states. The bifurcation point c 0.007 indicates the onset of the instability. Finally, it should be noted that although the inverse Lagrangian formulation changes the mathematics to avoid singularities or non-uniqueness of solutions near the critical points, it by no means alters the intrinsic physical behaviors of the structures. If the structure is sensitive to geometric imperfections, a small change in the original geometry may lead to a large difference in the deformed configuration. In these cases, the inverse Lagrangian formulation may still be used to analyze the instability and post-buckling behaviors, but is not suitable for the inverse design of structures. In fact, for better control of the target geometry, critical points of mechanical instabilities should be avoided in structural design if at all possible. 6. Conclusion In this letter, following Fachinotti et al (2008), the field equations of static hyperelastic problems are formulated in the deformed configuration, via the inverse Lagrangian approach. Different from the conventional Lagrangian formulation which solves for the undeformed-to-deformed mapping, or the Eulerian formulation of fluid dynamics which solves for the current-time velocity, the inverse Lagrangian formulation solves for the backward mapping from the deformed state to the original undeformed one. For linear elasticity problems with indistinguishable undeformed and deformed geometries, the inverse Lagrangian and Lagrangian formulations lead to identical equations. Through numerical examples, the inverse Lagrangian formulation has found some unique and interesting applications. Similar as those demonstrated in the literature, a direct application is in the inverse problems of identifying undeformed geometries from the prescribed geometries after deformation, especially in the design of soft structures which focuses more on the geometric accuracy under working conditions. With the aid of 3D printing technologies, the elevated complexity in the undeformed geometry should not become a challenge to manufacturing processes. Similarly, the method can be applied to the design of soft microfluid devices by solving the steady-state fluid-structure interaction as a decoupled problem. More importantly, the application of the inverse Lagrangian formulation to mechanical instability problems is demonstrated. By solving for the inverse mapping 19 from the deformed to the undeformed state, such an approach avoids the solution non-uniqueness in the conventional Lagrangian formulation. No artificial imperfection or numerical damping is needed for critical or post-buckling analysis. It should be admitted that the unknown geometry of the deformed configuration remains a major drawback for the inverse Lagrangian approach, especially for the postbuckling analysis at relatively large amplitudes. In future works, this method could be improved by using parametric optimization to more accurately determine the deformed geometries. References Albanesi, A.E., Fachinotti, V.D. and Cardona, A., 2010. Inverse finite element method for large‐displacement beams. International Journal for Numerical Methods in Engineering, 84(10), pp.11661182. Albanesi, A.E., Pucheta, M.A. and Fachinotti, V.D., 2013. A new method to design compliant mechanisms based on the inverse beam finite element model. Mechanism and Machine Theory, 65, pp.14-28. Baca, A.J., Ahn, J.H., Sun, Y., Meitl, M.A., Menard, E., Kim, H.S., Choi, W.M., Kim, D.H., Huang, Y. and Rogers, J.A., 2008. Semiconductor Wires and Ribbons for High‐Performance Flexible Electronics. Angewandte Chemie International Edition, 47(30), pp.5524-5542. Bertoldi, K. and Boyce, M.C., 2008. Mechanically triggered transformations of phononic band gaps in periodic elastomeric structures. Physical Review B, 77(5), p.052105. Bertoldi, K., Boyce, M.C., Deschanel, S., Prange, S.M. and Mullin, T., 2008. Mechanics of deformationtriggered pattern transformations and superelastic behavior in periodic elastomeric structures. Journal of the Mechanics and Physics of Solids, 56(8), pp.2642-2668. Biot, M.A., 1963. Surface instability of rubber in compression. Applied Scientific Research, Section A, 12(2), pp.168-182. Bower, A.F., 2009. Applied mechanics of solids. CRC press, New York. Cao, Y. and Hutchinson, J.W., 2012, January. From wrinkles to creases in elastomers: the instability and imperfection-sensitivity of wrinkling. In Proceedings of the Royal Society of London A: Mathematical, Physical and Engineering Sciences (Vol. 468, No. 2137, pp. 94-115). The Royal Society. Chan, E.P. and Crosby, A.J., 2006. Fabricating microlens arrays by surface wrinkling. ADVANCED MATERIALS-DEERFIELD BEACH THEN WEINHEIM-, 18(24), p.3238. 20 Chen, X. and Hutchinson, J.W., 2004. Herringbone buckling patterns of compressed thin films on compliant substrates. Journal of applied mechanics, 71(5), pp.597-603. Choi, W.M., Song, J., Khang, D.Y., Jiang, H., Huang, Y.Y. and Rogers, J.A., 2007. Biaxially stretchable “wavy” silicon nanomembranes. Nano Letters, 7(6), pp.1655-1663. Cerda, E. and Mahadevan, L., 2003. Geometry and physics of wrinkling. Physical review letters, 90(7), p.074302. Fachinotti, V.D., Cardona, A. and Jetteur, P., 2008. Finite element modelling of inverse design problems in large deformations anisotropic hyperelasticity. International Journal for Numerical Methods in Engineering, 74(6), pp.894-910. Gee, M.W., Förster, C. and Wall, W.A., 2010. A computational strategy for prestressing patient‐specific biomechanical problems under finite deformation. International Journal for Numerical Methods in Biomedical Engineering, 26(1), pp.52-72. Gee, M.W., Reeps, C., Eckstein, H.H. and Wall, W.A., 2009. Prestressing in finite deformation abdominal aortic aneurysm simulation. Journal of biomechanics, 42(11), pp.1732-1739. Genzer, J. and Groenewold, J., 2006. Soft matter with hard skin: From skin wrinkles to templating and material characterization. Soft Matter, 2(4), pp.310-323. Gervais, T., El-Ali, J., Günther, A. and Jensen, K.F., 2006. Flow-induced deformation of shallow microfluidic channels. Lab on a Chip, 6(4), pp.500-507. Govindjee, S. and Mihalic, P.A., 1996. Computational methods for inverse finite elastostatics. Computer Methods in Applied Mechanics and Engineering, 136(1), pp.47-57. Govindjee, S. and Mihalic, P.A., 1998. Computational methods for inverse deformations in quasi‐incompressible finite elasticity. International Journal for Numerical Methods in Engineering, 43(5), pp.821-838. Hirt, C.W., Amsden, A.A. and Cook, J.L., 1974. An arbitrary Lagrangian-Eulerian computing method for all flow speeds. Journal of Computational Physics, 14(3), pp.227-253. Hohlfeld, E. and Mahadevan, L., 2011. Unfolding the sulcus. Physical review letters, 106(10), p.105702. Hong, W., Zhao, X. and Suo, Z., 2009. Formation of creases on the surfaces of elastomers and gels. Applied Physics Letters, 95(11), p.111901. 21 Huang, Z.Y., Hong, W. and Suo, Z., 2005. Nonlinear analyses of wrinkles in a film bonded to a compliant substrate. Journal of the Mechanics and Physics of Solids, 53(9), pp.2101-2118. Koishi, M. and Govindjee, S., 2001. Inverse design methodology of a tire. Tire Science and Technology, 29(3), pp.155-170. Li, B., Cao, Y.P., Feng, X.Q. and Gao, H., 2012. Mechanics of morphological instabilities and surface wrinkling in soft materials: a review. Soft Matter, 8(21), pp.5728-5745. Liu, W.K., Herman, C., Jiun-Shyan, C. and Ted, B., 1988. Arbitrary Lagrangian-Eulerian Petrov-Galerkin finite elements for nonlinear continua. Computer methods in applied mechanics and engineering, 68(3), pp.259-310. Lu, J., Zhou, X. and Raghavan, M.L., 2007. Computational method of inverse elastostatics for anisotropic hyperelastic solids. International journal for numerical methods in engineering, 69(6), pp.1239-1261. Lu, J., Zhou, X. and Raghavan, M.L., 2008. Inverse method of stress analysis for cerebral aneurysms. Biomechanics and modeling in mechanobiology, 7(6), pp.477-486. Mullin, T., Deschanel, S., Bertoldi, K. and Boyce, M.C., 2007. Pattern transformation triggered by deformation. Physical review letters, 99(8), p.084301. Rajagopal, V., Chung, J.H., Bullivant, D., Nielsen, P.M. and Nash, M.P., 2007. Determining the finite elasticity reference state from a loaded configuration. International Journal for Numerical Methods in Engineering, 72(12), pp.1434-1451. Sellier, M., 2006. Optimal process design in high-precision glass forming. International Journal of Forming Processes, 9(1), p.61. Sellier, M., 2011. An iterative method for the inverse elasto-static problem. Journal of Fluids and Structures, 27(8), pp.1461-1470. Shepherd, R.F., Ilievski, F., Choi, W., Morin, S.A., Stokes, A.A., Mazzeo, A.D., Chen, X., Wang, M. and Whitesides, G.M., 2011. Multigait soft robot. Proceedings of the National Academy of Sciences, 108(51), pp.20400-20403. Stafford, C.M., Harrison, C., Beers, K.L., Karim, A., Amis, E.J., VanLandingham, M.R., Kim, H.C., Volksen, W., Miller, R.D. and Simonyi, E.E., 2004. A buckling-based metrology for measuring the elastic moduli of polymeric thin films. Nature materials, 3(8), pp.545-550. 22 Vavourakis, V., Papaharilaou, Y. and Ekaterinaris, J.A., 2011. Coupled fluid–structure interaction hemodynamics in a zero-pressure state corrected arterial geometry. Journal of biomechanics, 44(13), pp.2453-2460. Wang, Q. and Zhao, X., 2014. Phase diagrams of instabilities in compressed film-substrate systems. Journal of applied mechanics, 81(5), p.051004. Yamada, T., 1995. Finite element procedure of initial shape determination for hyperelasticity. In Computational Mechanics’ 95 (pp. 2456-2461). Springer Berlin Heidelberg. Yang, D., Jin, L., Martinez, R.V., Bertoldi, K., Whitesides, G.M. and Suo, Z., 2016. Phase-transforming and switchable metamaterials. Extreme Mechanics Letters, 6, pp.1-9. Zang, J., Ryu, S., Pugno, N., Wang, Q., Tu, Q., Buehler, M.J. and Zhao, X., 2013. Multifunctionality and control of the crumpling and unfolding of large-area graphene. Nature materials, 12(4), pp.321-325. Zienkiewicz, O.C., Taylor, R.L. and Zhu, J.Z., 2005. The finite element method: its basis and fundamentals. 6th ed., Elsevier, Oxford. 23