Arxiv:2103.12683V1 [Cond-Mat.Soft] 23 Mar 2021

Conformal Elasticity of Mechanism-Based Metamaterials

Michael Czajkowski,1 Corentin Coulais,2 Martin van Hecke,3, 4 and D. Zeb Rocklin*1 1School of Physics, Georgia Institute of Technology, Atlanta, Georgia 30332, USA 2Institute of Physics, Universiteit van Amsterdam, Science Park 904, 1098 XH Amsterdam, The Netherlands 3AMOLF, Science Park 104, 1098 XG Amsterdam, The Netherlands 4Huygens-Kamerlingh Onnes Lab, Universiteit Leiden, PObox 9504, 2300 RA Leiden, The Netherlands (Dated: March 24, 2021) Deformations of conventional solids are described via elasticity, a classical field theory whose form is constrained by translational and rotational symmetries. However, flexible metamaterials often contain an additional approximate symmetry due to the presence of a designer soft strain pathway. Here we show that low energy deformations of designer dilational metamaterials will be governed by a novel field theory, conformal elasticity, in which the nonuniform, nonlinear deformations observed under generic loads correspond with the well-studied—conformal—maps. We validate this approach using experiments and finite element simulations and further show that such systems obey a holographic bulk-boundary principle, which enables an unprecedented analytic method to predict and control nonuniform, nonlinear deformations. This work both presents a novel method of precise deformation control and demonstrates a general principle in which mechanisms can generate special classes of soft deformations.

I. MAIN tional strains only. These maps therefore provide a recipe to compose nonuniform soft modes from slow spatial vari- Mechanical metamaterials use patterns of cuts, pores ation of e.g. the RS mechanism (Fig.1c). We therefore and folds to achieve nonlinear [1–3], programmable [4,5], formulate our conformal hypothesis: under a generic and polar [6–8] and other exotic behavior [9–14] in response broad set of loading conditions, dilational metamaterials to external forcing. Often these features rely on the care- will respond with an angle-preserving conformal defor- ful arrangement of the cuts, pores and folds to emulate mation (Fig.1d), energetically much softer than than a mechanism, which is a special pathway of deformation conventional elastic strains. that enables the metamaterial to change shape at very small (ideally zero) energy cost. For example, the canon- ical rotating square (RS) mechanism, which consists of A. Testing the Conformal Hypothesis perfectly hinged rigid squares, enables a uniform dilational motion (Fig.1a) and has inspired the design of a To test the hypothesis of conformal deformation, we in- range of metamaterials which collapse inward rather than vestigate the elastic response of RS-based metamaterials bulge outward when compressed from the lateral direc- at a range of hinge thicknesses using finite element sim- tion [2, 15, 16]. While the dilational mechanism becomes ulations which preserve the intricate pore structure (see a true zero energy motion in the limit of vanishing hinge Methods). While the detailed microscopic strains may size, for realistic metamaterials with finite hinges this contain significant shear, we use the displacement of the uniform dilational motion is only observed for the partic- square centers to extract coarse-grained shear strain and ular case of a completely homogeneous loading condition. dilational strain fields, with the expectation that the lat- For generic, i.e. inhomogenous, loading conditions the re- ter should dominate as the hinges become small. Even sponse is more complex [17], and yet a general framework under nonuniform strain, dilation indeed dominates over to describe the general nonuniform soft deformations of shear in simulations of three-point “bending” (Fig2a), a mechanism-based metamaterials is lacking. local “dipole” dilation (Fig2b) and even when the system Here, we aim to decode the nonuniform deformations is subject to global “pure shear” via compression along of metamaterials based on a dilational mechanism. While one axis and expansion along the other (Fig2c). Even for arXiv:2103.12683v1 [cond-mat.soft] 23 Mar 2021 elastic response of ordinary materials will locally be com- dilations that vary dramatically through space (Fig.2d), posed of some finite portion of shear (Fig.1b), a dila- the average amount of local shear compared to dilation tional material will strive to expel shear everywhere in is captured in the shear fraction, defined in the SI, and favor of the dramatically softer dilational strains. Hence, varies from 10−1.5 to 10−3, in contrast with values on one may wonder whether the nonuniform deformations the order of one in conventional structures (Fig2 f,g). of dilational metamaterials may be locally composed of The low amounts of shear also imply nearly preserved pure dilation with no shear? While not all spatial pat- angles, which is the defining characteristic of conformal terns of strain are realizable due to basic geometric re- deformations. As confirmed in Fig.2e,f, the average an- strictions from compatibility relations, conformal maps gle change and the shear fraction both decrease as the are well-known to constitute the full set of compatible hinge thickness is reduced and the system approaches the smooth deformations which locally are composed of dila- ideal mechanism limit. Despite finite-size effects from the 2

b. c.

FIG. 1. Dilational Mechanism and Conformal Deformations. (a) In the ideal rotating square mechanism, a structure of rigid squares (blue) connected by frictionless hinges (pink), may be dilated and contracted at zero energy cost when the squares are rotated opposite to their neighbors. (b) Applying fixed deformations (red arrows) at the boundary of a conventional elastic material leads to a spatially varying strain field that includes shear components that change the local angles between pieces of material, such as the initially perpendicular grid lines (blue, yellow). (c) In contrast, a pure dilational metamaterial designed around the RS mechanism, may accommodate the loading without shearing of the unit cells, so that even as grid lines rotate, the angles between them remain fixed. This angle-preserving behavior arises due to a local version of the RS mechanism, in which square elastic chunks of side length a rotate about small, flexible hinges of thickness ` to open and close pores according to α(x, y), the local linear dilation factor. (d), A fabricated sample of checkerboard material approximately preserves right angles (shown in yellow) under a generic nonlinear “foot” load, suggesting conformal deformation behavior.

16 × 16 unit cell lattice, the strong preservation of angles takes the convenient form indicates that the response is locally conformal. ∞ X n The observation of approximately conformal local be- f(z) = Cnz , (1) havior suggests that deformations of the elastic RS meta- n=0 material may also correspond globally to a conformal map. This motivates an elegant formulation expressing where f(z) is the deformed position of the material ele- positions in the plane as complex numbers r = (x, y) ↔ ment initially located at z and the complex coeﬃcients x + iy = z, such that any exact conformal deformation {Cn} deﬁne the map. The ability to express such a map 3

a. “bending” b.“dipole” c. “pure shear”

d. e. decreasing f. dilation (d)

- bin fraction

- - - - -

FIG. 2. Deformations of dilational metamaterials are angle preserving. (a) A Finite-Element Simulation of a three- point bending test (the “bending” load) locates force balanced (relaxed) states of the RS-based elastic metamaterial subject to displacements on the boundaries that are incompatible with uniform dilation. (b) The “dipole” load consists of closing a single central pore by displacement of points on opposite sides of the pore, thereby generating a localized dilation in the bulk of the material. (c) The “pure shear” load consists of compression along one axis and extension along the other, which is nevertheless compatible with a spatially varying pure dilation field. (d) The resulting dilation field under the “bending” load varies widely across the sample, including nonlinear compression and expansion. (e) Nevertheless, the angle changes |∆ψ| remain small, even in proportion to the average strain magnitude ||. Here, the histogram is shown on log-linear scale, with each square marker giving the fraction of unit cells with that angle, which tend to decrease as the hinge size decreases. The preservation of angles is the defining feature of conformal maps. (f) For all loads (colored in correspondence with the box outlines in (a), (b), (c)) the dilation fraction (solid lines), the complement of the shear fraction, is nearly unity: the systems are predominantly dilational. Both the fraction of strain that is shear (dot dashed lines) and the normalized average angle change (dashed lines) decrease with decreasing hinge size. The plateau in the bending and pure shear loads can be attributed to non-continuum lattice effects.

in this reduced series form, turns the search for the con- B. Conformal Elasticity formal map closest to our data into a linear algebra problem which is readily solved. Using only the first 20 Cn Given this evidence that the RS metamaterial reponds coefficients, we use this procedure in Fig.3a,b to identify to loading with a near-conformal deformation, we present conformal maps that are able to capture all but a small an elastic energy functional, fraction ∆ (about 1%) of the observed displacements conf Z 1h (see SI for precise definition of ∆ ). Indeed, despite 2 2 2 conf E = d r G1(α)s1 + G2(α)s2+ (2) the nonzero local angle deviations, we find that RS-based 2 metamaterials will respond to loading with a deformation `2 i M(α) + a2M˜ (α)|∇α|2 that is closely matched with a conformal map even under a2 nonlinear strains. This is in contrast with generic defor- which the system minimizes subject to the boundary con- mations, which fit poorly to Eq. (1) as shown in Fig.3b ditions. Here, α is a (nonlinear) spatially varying field and in general require terms with complex conjugation of describing the dilation factor of the structure relative to z such as z2z¯5 in the expansion to be captured accurately. its equilibrium, s1 and s2 are the (linear) coarse pure shear and simple shear respectively, ` is the width of a hinge and a the width of a square piece. G1,2(α) are shear moduli, M(α) the dilation energy density and M˜ (α) the modulus associated with spatial variations in the dilation. As discussed in the SI, additional terms such as 4

a. b.

- - -

best fit conformal deformation predicted conformal deformation from boundary dilation data

c. d. e.

- - -

FIG. 3. Deformations are conformal and permit nonlinear prediction. (a) The nonlinear deformation of a three-point “bridge” load can be well fit to a conformal map (magenta grid) chosen to minimize the average squared deviation between 2 the mapped and observed displacements. (b) The fraction of variance unexplained ∆conf between observed deformations and conformal fits in both simulations and experiments is very small, demonstrating that the soft deformations correspond globally to conformal maps as well as locally to pure dilations. Experimentally, the fit is especially good in the unloading step, in which stresses arising from frictional interactions at the material boundary are able to relax. In contrast, the data for a conventional elastic material (pink) cannot be described as a conformal map. (c) According to the holographic property of conformal maps, the dilation field on the boundary of an arbitrarily chosen (gerrymandered) region, such as the one depicted in yellow for a simulation configuration, should uniquely determine the deformation within the region. (d) Indeed, the displacements inferred from the boundary dilations (orange arrows) match closely with those observed directly in simulations (blue arrows). (e) The observed displacements are now compared to the conformal maps inferred in this way, and the fraction of variance unexplained 2 plotted as ∆inference. This demonstrates that the deformations are not merely conformal, but the precise conformal map is encoded on the boundary. Here, the predictions of experimental deformations are inferred from the full rectangular boundary, while predictions of FEM data are all inferred from the gerrymandered boundary points shown in (c) to show robustness to the choice of boundary.

couplings between shear and dilation are excluded by space of ground states, we recover a nonlinear notion of scaling arguments and by the symmetries of the lattice. conformally invariant elastic theory, which was suggested Although other continuum elasticity theories for the RS by Sun. et. al. [18] previously for the linear deformations metamaterial have been developed in one [17] and two of the Kagome lattice, but has otherwise been viewed as dimensions [11, 12], and another theory for a bistable di- “unphysical” [19] until now. Then, to distinguish which lational material [13], this nonlinear analytic form, which of the many conformal deformations will arise in response centers the dilational mechanism and its gradients, has to a particular loading, we turn to the perturbative en- not been proposed previously. While this energy function ergy functional holds a rare level of insight into nonlinear deformation, it is still quite difficult to solve analytically. Z 1 `2 ∆E ≡ d2z M(|f 0(z)|) + a2M˜ (|f 0(z)|)|f 00(z)|2 , To gain more analytical traction, we employ pertur- 2 a2 bation theory in the limit in which the hinges are tiny (3) relative to the unit cell size and the material sample is expressed in terms of the conformal map f : z → f(z) de- composed of very many unit cells (i.e., the continuum scribing the local dilation and rotation, with f 0(z) ≡ αeiφ limit). In this limit, deformations that include shear (first (Eq. (1)). The energy in Eq. (3) arises from the last two terms of Eq. (2)) become prohibitively stiff and, by two terms in Eq. (2), which simultaneously breaks this comparison, dilations begin to act as a local symmetry. ground state degeneracy and the conformal invariance. Therefore, the conformal maps will constitute a degener- With this, we have reduced the difficult tensorial prob- ate space of ground states. Restricting our focus to this lem of conventional nonlinear elasticity of material with 5 pores to a scalar theory which is much more analytically deformation that will be activated in the interior. tractable. For small loading, minimizing the energy and thus predicting the deformation is reduced to a linear algebra problem. As shown in the SI, these predictions closely match the observed finite-element displacements D. Discussion with correlation coefficient R2 ≈ 0.99 − 0.9999 for hinges with `/a = 0.005. This result showcases both the accu- The intuitive concept of soft modes that are locally racy of conformal elasticity and the mathematical conve- composed of a spatially varying mechanism is not con- nience of conformal maps. fined to the RS mechanism explored here. The conformal modes explored here arise because they are the Nambu-Goldstone modes associated with the local di- C. Bulk-Boundary Correspondence Yields New lational symmetry, rather than from the details of the Analytic Insight and Control Methods specific microstructure. We therefore propose the anal- ogy: as isometries are to thin elastic sheets [20], so Solving Eq. (3) analytically is only possible in the lin- are conformal deformations to dilational metamaterials. ear regime and relies on prior knowledge of the effective Consequently, dilational materials derived from a va- stiffnesses M and M˜ . However, as shown in Fig.3a,b, the riety of different metastructures [18, 21–24] including conformal property itself extends to nonlinear deforma- fractal RS structures [23, 24], disordered “pruned” net- tions. We therefore devise a scheme to extend analytic works [25], nanoscale [26], and three dimensional gen- predictions to the nonlinear regime, relying on a mathe- eralizations [3, 27, 28] will similarly have mechanical matical property of conformal maps: bulk-boundary cor- response controlled by a set of conformal soft modes. respondence. Within our formalism, this bulk-boundary These alternate architectures provide the possibility for correspondence works as follows: if we know the amount greater ranges of dilation [22], opening the door for even of dilation along the entire boundary of a section of RS more dramatic soft deformations such as pictured for the material, we can predict analytically the local dilation kagome lattice in Fig4d. We suggest that a broad class everywhere in the material interior from the unique con- of generic mechanisms not confined to pure dilation will formal map derivative f 0(z) consistent with the boundary also generate families of soft modes that govern mate- conditions and can further integrate to infer the displace- rial response, and that this may become a fundamental ments. We illustrate the principle in Fig.3c,d,e. We re- principle for mechanism-based metamaterials. Exploring trieve the dilation on an arbitrarily chosen red domain how these may obey, e.g., a generalized bulk-boundary in Fig.3c from the FEM data, use the bulk-boundary correspondence may yield fruitful connections to exotic correspondence to calculate the deformation inside this field theories [29] as well as black hole and string theories domain (See displacement field depicted by blue arrows where holographic principles already play a vital role. in Fig.3d), and show that these are in excellent agreement with the observed displacement data (See displacement field depicted by orange arrows in Fig.3d). This method is robust to the arbitrary choice of boundary II. METHODS shape and captures > 99% of block displacements even at large strains (Fig3e). A. Generating deformations analytically The bulk-boundary correspondence can also be used for prescribing on-demand deformations. Given a target deformation field of some section of material, we can de- We analytically generate the deformations of two- termine the suitable actuation pattern by simply examin- dimensional metamaterials that locally resemble the uni- ing the local target dilation at the material boundary. We form mechanism and a rotation and hence eliminate illustrate this with three different actuation patterns in the dominant contributions to the elastic energy, taking Fig.4a,b,c. We actuate the material at its boundary (the advantage of a map between the plane of deformation red bars represent actuators) and observe that the defor- and the complex plane. In general, a deformation map ∂Xi x → X(x) has deformation tensor Fij ≡ which de- mation is in good agreement with the target deformation: ∂xj the center of the square coincides very well with the yel- forms infinitesimal material vectors as dxi → Fijdxj. To low nodes. Of course in addition to being conformal, this correspond locally to a rotation of the mechanism field, procedure is only able to actuate deformations that do the deformation tensor must be proportionate to a ro- not stretch or compress the material beyond the physical tation matrix. We seek solutions in which the rotation bounds of the mechanism itself. This task is aided by the angle φ and the dilation factor α vary over space, so that maximum modulus principle of conformal maps, in which F = αR(φ). However, this generates nontrivial restric- both the maximum and minimum dilations must occur tions because of the requirement that the deformed state on the material boundary. We can therefore guarantee not tear, i.e. that H dX = 0 along any path. By applying that for any choice of actuation that varies slowly along Stokes’ theorem, this leads to the standard geometric or the boundary, there is a physically valid soft conformal kinematic compatibility conditions: 6

a. b. c.

FIG. 4. Bulk-Boundary correspondence allows for precise control of metamaterial deformation. (a, b, c) The bulk-boundary correspondence shown in Fig.3 enables on-demand actuation of material deformation via the addition of boundary springs (red). Here, the locations of square centers observed in simulation match well with target positions (yellow dots) for force-balanced conﬁgurations of a simpliﬁed ball-and-spring model. (d) Because the principles presented here extend to any dilational metamaterial, the kagome lattice may likewise be placed in a low-energy conformal deformation. The kagome lattice is capable of undergoing a broader range of deformations without self-intersection, permitting a greater range of controllable deformations such as pictured here.

tion and whose argument is the local orientation. Dis- placements can be obtained via complex integration: ∂1F12 − ∂2F11 = 0, (4) f(z) = R dzf 0(z). This method hints at a more general ∂1F22 − ∂2F21 = 0. (5) class of spatially varying deformations in unimode materials, but the results in this case follow from the fact that This generates two independent conditions on the two the shear-free deformations of the structure are precisely fields, suggesting a unique solution subject to certain the angle-preserving ones well-known to be described by boundary conditions. However, it becomes convenient complex analytic functions. to define new variables z = x + iy, z¯ = x − iy, with i the imaginary unit. Enforcing the conditions ∂zz = ∂z¯z¯ = 1, ∂zz¯ = ∂z¯z = 0 then requires that ∂z = (1/2)(∂x − i∂y), B. Finite Element Simulation Protocol ∂z¯ = (1/2)(∂x + i∂y). By summing the second compatibility condition with (−i) times the first, we find, following algebra, that For the 2D finite-elements simulations, we use the compatibility is equivalent to the complex condition commercial software Abaqus/Standard. We use a neo- ∂z¯ [α exp(iφ)] = 0. Thus, the locally dilational defor- Hookean energy density as a material model, a shear mations of the metamaterial are exactly the conformal modulus µ = 0.333 MPa*m, bulk modulus K0 = maps of the plane, which may be expressed as ana- 16.7 GPa*m and plane strain conditions with hybrid lytic functions f 0(z) whose magnitude is the local dila- quadratic triangular elements (Abaqus type CPE6H). We 7 construct the mesh so that the thinnest parts of the sam- boundary conditions, (i) by a foot-shaped indenter, which ples are at least two elements across. We perform two was 3D printed in ABS Materials using a Dimension 3D types of simulations (i) on the metamaterials unit cells printer (Stratasys); (ii) a three points bending test, using with periodic boundary conditions, on which we perform a universal uniaxial mechanical testing machine (Instron a low strain static test (5 × 10−6)) and (ii) on the full 5943). In parallel to the mechanical loading, we recorded structure, on which we perform static nonlinear analysis. frames using a high-resolution camera (6000×4000 pixels In the simulations (i), to implement periodic bound- (Nikon D5600) with a telephoto lens (Nikon 200mm, F4), ary conditions, we define constraints on the displace- positioned at 4 meters from the sample. To ensure uni- ments of all of the nodes at the horizontal and verti- form lighting, the sample was illuminated by two led pan- cal boundaries of the unit cell and impose displacement els (Bresser, LG 900 54W) and white paper was used to using virtual nodes [30]. We perform four types of sim- ensure a uniform bright background. The 3D printed ma- ulations to extract the macroscopic linear bulk modulus terial tends to highly frictional and even adhesive, there- `2M 00(α=1) fore, we used fine powder to lubricate the interaction with K = 2 , and shear moduli G1(α = 1),G2(α = 1). 4a the bottom surface and the foot for (i). For (i), we per- In order to extract K and G1, we apply biaxial compres- comp comp −6 formed a compression-decompression cycle at a rate of sion ε11 = ε22 = 5 × 10 , ε12 = 0 and pure shear shear shear −6 0.1 mm/s up to a maximum 20 mm stroke, during which ε11 = −ε22 = 5 × 10 , ε12 = 0 to a unit cell with 400 frames were recorded (1 frame / sec). For (ii), we per- periodic boundary conditions. K and G1 are each ex- comp comp formed a compression-decompression cycle at a rate of 0.2 tracted as a slope in linear fitting σ11 = 2Kε11 and shear shear mm/s up to a maximum 40 mm stroke, during which 400 σ11 = 2G1ε11 . G2 is extracted using the identical frames were recorded (1 frame / sec). The images were procedure as for G1 applied to a unit cell that is rotated by π/4. We examine a unit cell with a = 12mm, processed using standard image-tesselation and tracking techniques to extract the positions of the squares with pretwist θ0 = 0.39 and hinge thickness ` = 0.1mm, find- 4 4 subpixel detection accuracy of 0.02 pix (1 µm). We at- ing (K,G1,G2) ∼ (35, 5.2 ∗ 10 , 2 ∗ 10 )Pa*m. These modulus measurements are essential for obtaining the tribute the strongly different behavior observed between perturbative energy minimizing prediction of deforma- compression and decompression, to a combination of fric- tion, summarized briefly in the main text and given in tional, viscoelastic and self-adhesion effects. more detail in the SI. In the simulations (ii), we use three kinds of boundary conditions, dipole, bending and pure shear, by imposing displacement of specific nodes, as described in Fig.2. We extract the coordinates (position, angle) of all the squares and use these to compute the dilation and shear field (method described in the S.I.) and to calculate the relative amount of shear, as well as the accuracy of conformal map fits and of the analytic bulk-boundary correspondence (Fig.3). Simulations are performed for a series of strains at identical material geometry as (i) and separately at linearly small strain across a range of hinge thicknesses ` = {0.02, 0.033, 0.055, 0.092, 0.155, 0.258, 0.431, 0.719, 1.199, 2.0} mm.

C. Experimental Protocol

We 3D printed a elongated meta-beam of length 306mm, width 64mm and height 40mm, consisting of a checkerboard of lattice of 48 × 10 squares of side III. ACKNOWLEDGEMENTS a = 4.8mm connected by hinges of thickness ` = 0.2mm using a Connex 500 stratasys 3D printed and the proprietary Stratatys Agilus 30 material (30 Shore A). This material is viscoelastic, its Young’s modulus at short times is E= 3.3 MPa and E= 0.6 MPa at long times (see, e.g., Dykstra et. al. JAM 2019 [31] for a calibration). For We thank S. Matsumoto and M. Dimitriyev for insight- imaging purposes, we 3D printed black square-shaped ful discussions and suggestions, S. Koot and D. Giesen pads (3.6mm) on one edge of the sample (proprietary for technical assistance. C.C. acknowledges funding from Stratasys material: Vero Black). the European Research Council via the Grant ERC-StG- The sample was tested under two sets of non-uniform Coulais-852587-Extr3Me. 8

Appendix A: Appendix linear shear. Importantly, this shear may be rewritten in terms of invariants of U as s2 = Tr[U]2 − 4Det[U], and 1. Quantifying Local Dilations and Shears is therefore also a rotationally invariant quantity. Since the stretch tensor is generally less convenient to To our knowledge, a parameter which quantifies what extract numerically, we show that this local dilation and fraction of a nonlinear deformation is composed of local shear may also be written in terms of the trace and de- dilation, as opposed to local shear, has not been intro- terminant of the metric (equivalently the “right Cauchy- duced previously. We therefore present a more detailed Green deformation tensor”). The metric is defined as derivation in the context of Finite Strain Theory. This ∂Xk ∂Xk T T is an essential ingredient for supporting our hypothesis Cij = = (F F)ij = (U U)ij . (A3) ∂x ∂x in the main text that nonuniform deformations of a dila- i j tional metamaterial are nonetheless composed primarily The coordinate system which diagonalizes U also diago- of local dilations. nalizes C, and it follows that the eigenvalues of the metric The method described here applies to data describing are the squared eigenvalues of U. Using this locally diag- deformation from a reference space x to a target space onalized frame, we may extract general relations between X(x), which are smooth enough to take derivatives. In rotationally invariant quantities the case of the RS metamaterial, the square centers provide a discrete approximation to a smooth deformation, Det[U]2 = Det[C], which may then be used to define local material deriva- (A4) Tr[U]2 = Tr[C] + 2pDet[C] . tives. More generally, one may use the lattice vectors of a material with repeating metastructure as material Using these relations, we may write the more useful for- vectors to take these material derivatives. From these mulas derivatives we assemble the deformation tensor d = pDet[C] − 1, ∂Xi (A5) Fij(x) = |x . (A1) s2 = Tr[C] − 2pDet[C] , ∂xj

For a smooth deformation, this tensor contains all re- completing our recipe to extract the local dilation and quired information to quantify dilation and shear at each shear magnitudes from the metric tensor. point in space. That these relations in Eq. (A5) discriminate confor- Using the polar decomposition theorem, we may write mal from non-conformal deformation is not entirely ob- F = R·U, where R is a rotation and U is the symmetric vious. We therefore present them in alternative form by ”right stretch tensor”. From this form, it is clear that connecting to the complex formulation where the vector an infinitesimal material vector in the reference space mapping x → X(x) is captured by the complex function dx will deform first via multiplication with U, then will z → f(z, z¯). In this case, the components of the metric be rotated via R to reach a final state dX = R · U · may be conveniently rewritten as dx. Since U is symmetric, there exists a local choice of 2 2 Cxx = |∂zf| + |∂z¯f| + 2Re ∂zf∂z¯f coordinate system where U is diagonal and acts on the 2 2 Cyy = |∂zf| + |∂z¯f| − 2Re ∂zf∂z¯f (A6) local elements dx as a combination of just dilation and Cxy = Cyx = 2Im ∂zf∂z¯f , pure shear. The eigenvalues (λ1, λ2) of U are called the principal stretches and the local material dilation and where ∂ = 1/2(∂ − i∂ ) and ∂ = 1/2(∂ + i∂ ) are shear magnitudes may be written intuitively in terms of z x y z¯ x y standard complex derivatives [32]. With these relations, them. An infinitesimal square sample of side lengths (, ) our metrics of conformal and non-conformal become oriented in the diagonal frame of U will transform into a 2 rectangle with side lengths (λ1, λ2) after deformation. 2 2 2 d = |∂zf| − |∂z¯f| − 1 , Using this, the local dilation may be defined intuitively 2 2 in terms of the area s = 4|∂z¯f| , ∆A where, remembering that the conformal condition is d ≡ = Det[U] − 1 , (A2) ∂ f = 0, the shear part now more transparently quanti- A0 z¯ fies the non-conformal part of the deformation. In addi- where ∆A is the change in area of our infinitesimal el- tion, rewriting 2 ement and A0 = is the initial area in the reference 2 2 2 space. For a similarly intuitive notion of shear, we turn to s = (∂xux − ∂yuy) + (∂xuy + ∂yux) (A7) the anisotropy of the tensor U. We define the nonlinear 2 2 = s1 + s2 shear magnitude in terms of the difference of the princi- 2 2 pal stretches as s = (λ1 − λ2) . We will see that this in terms of the components of the displacement u = X−x definition matches with conventional intuition for stan- reveals that our definition of the shear matches with the dard simple shear nonlinearly as well as general mixed combined magnitude of pure and simple shears (s1 and 9

dipole bending pure shear

FIG. 5. Elastic RS-based structures deform with local dilation over shear. For each of the applied displacement loads (columns here) from the main text, we display the spatial distributions of coarse dilation ( row 2, d) and coarse shear magnitude (row 3, |s|) displaying a strong tendency of the RS-based metamaterial toward deformations that are locally dilation dominated.

s2) from linear strain theory. Finally, for a conventional identical points in neighboring unit cells. For the RS- simple shear u = γyxˆ, this formula reassuringly recovers based metamaterial, we can choose the centers of the s2 = γ2 to nonlinear order. square elements. As shown in Fig.5, these formulae reveal that the typi- The method consists of first noting that a general con- cal magnitude of local dilation is much larger than that of formal map f(z) on a simple finite planar domain with no local shear. Nonuniform, nonlinear deformations of the holes (i.e. simply-connected) may be expressed precisely RS-based mechanism in force-balance may then indeed as an analytic function with a series expansion be composed of local dilations with very little shear. ∞ X n f(z) = Cnz . (A8) 2. Least-squares method to extract the nearest n=0 conformal map from data In contrast, a generic non-conformal deformation must be We now identify a simple fitting method of locating expressed as a double sum over all products of z and the the closest conformal map describing a discrete set of complex conjugatez ¯. The complex expansion coefficients displacement data points. Per Section.A1, the discrete Cn provide a complete description of the map and solving data should be chosen carefully to approximate a smooth for them is therefore the goal of our analysis. Finding deformation. For lattice materials, this means tracking the nearest conformal map is a matter of minimizing the 10 square error function

np X 2 Err ≡ |fi − f(zi)| , (A9) - i=1 - where the sum runs over the np discrete data points zi → fi. Inserting the form of f(z) (Eq. (A8)) and minimizing - with respect to the Cn, we ﬁnd - nc np X X m AmnCn = fiz¯i , (A10) - n=1 i=1 where we have deﬁned the matrix

np X m n Amn = z¯i zi (A11) FIG. 6. The accuracy of the conformal map with the increas- i=1 ing number of coefficients allowed in the expansion Eq. (A8). Solid lines correspond to fits with FEM data at the small- and the coefficients are now cut off at a maximum nc. est strains, while the dotted lines are for the largest strains This effectively reduces the least-squares error analysis to explored in the main text analysis. In comparison to the a straightforward linear algebra problem, which is readily conventional elastic material in the absence of pores, the fit- 2 solved. We note that the cutoff nc should be chosen to be ting quickly reaches very small errors ∆conf which is the func- 2 much less than the np data points, to avoid over-fitting. tion ∆ [ ] evaluated for the particular candidate fit u(z). For To evaluate the accuracy of the fit, we define a function loads with discrete rotational symmetry, the errors go down in steps, since only every few coefficients may be non-zero. Pnp |u − u(z )|2 ∆2[u(z)] ≡ i=1 i i , (A12) Pnp 2 i=1 |ui| 3. Energetic Estimates of Hinge Deformations where ui = fi − zi is the observed displacement data, u(z) ≡ f(z) − z are the displacements proposed to fit the data. The functional ∆2[ ], known as the fraction of un- We now aim to derive a continuum elasticity theory for explained variance, is then a general method to quantify the RS metamaterial. However, to consider the RS elastic the amount of the deformation captured by a candidate material at full detail requires allowance of all possible de- conformal function, which we use in the main text and formations of the porous elastic structure. Such a tensor in other sections of this SI. The method evaluates error field theory for deformations would be complicated and using the displacements rather than positions in order unwieldy, leaving very little room for analytic progress. to avoid a false inflation of the results at small defor- Fortunately, due to the careful design of the pores, we mation. Inserting the numerically solved closest fit into may think of the material as composed of square ele- Eq. (A12), we show in Fig.6 that only around 20 co- ments of side length a connected by comparatively small efficients are needed to fit the deformations of the RS “hinge” elements of thickness `. This enables an effec- metamaterial before the improvement starts to become tive description of the material as a collection of hinges, negligible. This is good, as the more coefficients that are written in terms of simplified variables which prove to included, the more computationally expensive the fitting be very convenient for coarse-graining. We establish this problem becomes. effective “hinge-based” energetic formalism below. We remark that, in practice, application of this method In the limit that ` a, the hinge will become very works best if the coordinate system is first centered and flexible compared to the stiff square pieces. We therefore rescaled so that the data points reside closer to unity follow successful previous examples [11, 17] in which the away from the origin. This helps to avoid large and small large square elements are approximated as rigid bodies number issues in the numerics, which includes higher and and all strain deformation is assumed to take place at the higher exponents of the positions as more coefficients are hinges. This assumption is readily verified in the Finite included. Element deformation data, where the stress is generically This constitutes a fast and accurate method to extract found to be dramatically localized to the hinge region, as the nearest conformal map from position data. Note that, shown in Fig.7. while the method is reduced to linear algebra, no assump- To quantify the energetics in this effective descriptions of linearity of the strains in either the deformation, tion, we consider a single hinge connecting two elastic or the fitting map, have been made. Indeed, this method blocks as shown in Fig.8-a. Knowledge of the defor- shows the nonlinear deformations of the RS metamaterial mation tensor Fij(x) throughout the hinge completely are well-approximated as a conformal map to nonlinear determines the system configuration, as stress and strain order. are already assumed to be zero elsewhere. Given the 11 complete constitutive relation for the elastic material, Alternatively, we may explore this mechanism by fixing the tensor field Fij(x) provides sufficient information to the orientation ψ = π/2 − 2T and allowing the positions compute the energy density at each point in space. For to relax to a square center spacing c(T ) which can be generality we write this simply as an unknown function expressed as e(x) = φ(Fij(x)). √ We now confine our focus to hinges with aspect ra- c(T ) = 2aCos(T ) + `δc(T ) . (A14) tio ∼ 1, which is the case for the experiments and Again, the second term on the right-hand side is a correc- simulations in this manuscript. This condition gener- tion term dependent on an unknown function δc which ally removes opportunities for bistability and other path- will eventually be neglected in the small hinge limit. dependent elastic behavior at the hinge. Therefore, pre- We now use these mechanism twisting functions to describing the orientations and positions of the two squares fine two parametrizations of the hinge energy. In both attached to a particular hinge will leave only one possi- cases, we first choose to move to a point along the mech- ble force-balanced deformation Fij(x) in the hinge. The anism pathway, setting the distance c and twisting angle energy of this hinge may therefore be rewritten as a func- to ψ = π/2 − 2T (c) as shown in Fig.8 (left). We may tion of the two square orientations and two 2d vector then deviate from this configuration in two ways: In the positions, six numbers altogether. Accounting for trans- first, we may choose to fix the square orientations and lational and rotational symmetry leaves only three scalar allow a small vector deviation δc in the relative positions parameters necessary to describe the configuration of the as shown in the top right of Fig.8. We will refer to this hinge. We will focus on two distinct, yet convenient as the “vector parametrization”. Alternatively, we could choices for these three parameters as described below. keep the square centers fixed in place and allow each Importantly, we note that arbitrary relative positions to have a small reorientation away from the mechanistic and orientations of a pair of squares will generally cost twisting orientation as shown in Fig.8. This defines de- a great deal of energy and may stretch the hinge to the viation angles δθA and δθB and we will refer to this as point of fracture. We wish to focus on low-energy config- the “angular parametrization”. In the end, the choices urations that do not stretch the hinge too much, which will span the same configurations of the hinge structure. includes a particular set of nonlinear rearrangements due Both parametrizations above are useful for isolating to the design based on an ideal-hinge mechanism. There- the “non-mechanistic” part of the hinge motion into small fore, in order to find an appropriate parametrization for variables. It is clear for small hinges that comparatively these low-energy configurations, we must first quantify large δc will put the hinge under large strains which scale the finite-hinge (finite-energy) analogue of this mecha- |δc| roughly as ∼ ` and will put conventional materials at nism. As shown in Fig.8a,b, fixing the distance be- risk of tearing and fracture. We therefore work in the tween the square centers (c) and allowing the square |δc| limit ` 1. Equivalently, (δθA, δθB) 1 are the ap- orientations to relax defines our twisting angle T (c) ≡ propriate conditions in the angular parametrization. We π/4 − ψ(c)/2. Matching with the common nomenclature will see below that the vector parametrization is useful for for twisting in kagome and RS lattices, this represents deriving scaling relations, while the angular parametriza- the amount the squares must each be rotated away from tion is useful for the coarse-graining. Given some config- the fully extended state to balance torque at fixed dis- uration of the hinge we may now write the hinge energy tance. To twist the squares oppositely according to T (c) in terms of the vector parameters while moving the square centers to distance c therefore constitutes our generalized mechanism. In the limit that EH = gv(T, δc, `, a, T0) , (A15) the hinge size ` goes toward zero, and the strain is in- or equivalently in terms of the angular parameters creasingly localized to an infinitesimal hinge, this motion must approach the ideal frictionless hinge version of the EH = ga(c, δθA, δθB, `, a, T0) . (A16) mechanism. We may therefore write the twisting func- where T0 is the “pretwist” present in the reference con- tion as figuration of the hinge ensemble. Finally, we note that it is useful to be able to convert between these descrip- c ` 2 T (c) = ArcCos √ + δT (c) + O(`/a) (A13) tions. Neglecting second and higher order terms in the 2a a deviations δc and dimensionless hinge size `/a, we may where δT (c) is an unknown dimensionless function cap- write turing the first-order corrections to the twisting due to (δc)x (δc)y δθA = −√ + the finite size and stiffness of the hinge. Eventually we 2a2 − c2 c will proceed to ignore this first order correction, but must (A17) (δc)x (δc)y acknowledge that the function δT (c) may diverge in the δθB = −√ − . √ 2 2 vicinity of c = 2a (the untwisted state). Both our hinge 2a − c c theory and subsequent coarse-graining of the checker- Even with perfect knowledge of the hinge geometry and board mechanics will break down near this untwisted nonlinear material elasticity, finding the form of the func- state and we therefore leave analysis of this anomalous tions gv, ga would require solving a difficult nonlinear me- point in the configuration space for future work. chanics problem. However, a lot can be discerned without 12

Stress

FIG. 7. Sample deformations from Finite Element Simulations displaying the detail of the elastic mesh method as well the characteristic localization of the stress (and therefore also strain) to a small region in the vicinity of the hinge.

a. c.

b. d.

FIG. 8. In the absence of additional forces applied near the hinge, the elastic hinge energetics may be parametrized with three numbers describing the relative orientations and positions of the squares. First, the “mechanism” portion may be traversed by ﬁxing the center-to-center distances of the squares and allowing the square orientations to relax. Then small deviations around this state are allowed by either (bottom-right) preserving the square distance and allowing small orientation changes or (top-right) preserving the square orientations and allowing a small change in relative square positions.

2 knowing the exact form of the energy and instead relying over a rescaled domain, and we find E˜H = γ EH . How- only on scaling arguments. In the vector parametriza- ever, this rescaling also impacts the relative locations of tion, our hinge deformation tensor is a function not only the large square elements. Consider the internal vector of the location within the hinge x, but also this tensor e which points from the projected corner of square A to field must depend on the inputs T and δc. It is instructive the projected corner of square B as shown in the zoom to consider rescaling the hinge reference and final states in Fig.8-c. Due to the preserved shape of the hinge, by a factor γ while preserving shape. This means that this internal vector must also scale directly with γ. This points in the hinge x are mapped to points x˜ = γx and internal vector also has an analytic form the deformation tensor is preserved via F˜ij(x˜) = Fij(x). Because the deformation tensor is preserved, so is the e = `δc(T )ˆx + δc , (A18) energy density. The only difference in computing the total energy of the rescaled hinge comes from integration where δc(T ) is again the correction to the mechanism from Eq. (A14). Considering the configurations where 13

δc = 0 in Eq. (A18), we note that because e scales with wherec ˜ = √c is the dimensionless rescaling of the square 2a γ, and because ` also scales directly with γ, then δT must center distance and the moduli are defined by not scale with γ. Noting that δ is independent of δd, T √ it will remain scale-free when we also apply such a dis- g(˜c) = gT (T ( 2ac˜)) placement δc. Because each side of Eq. (A18) must scale = gT (ArcCos(˜cCos(T0))) the same way with γ, we have now established the re- √ sult that the displacement vector δc indeed must scale Sin(T ( 2ac˜))2 √ k (˜c) = k (T ( 2ac˜)) linearly with γ. 1 s 2 √ Rescaling the hinge has no impact on the square rela- 2 Cos(T ( 2ac˜)) √ (A23) tive orientations ψ nor on the “pre-twist” T . This rescal- + kµ(T ( 2ac˜)) 0 2 ing impacts only ` and δc and the hinge total energy E √ H Sin(T ( 2ac˜))2 √ as described above (in the vector parametrization). We k2(˜c) = ks(T ( 2ac˜)) may therefore rewrite the hinge energy function 2 √ Cos(T ( 2ac˜))2 √ δc − kµ(T ( 2ac˜)) . E = g (T, δc, `, a, T ) = `2g (T, , a, T ) , (A19) 2 H v 0 1 ` 0 While the function g1 remains unknown, and the ener- where g1 is another unknown function, whose value may gies in Eqs. A16& A15 may not be evaluated directly, be thought of as a mean energy density in the hinge. these expressions provide useful insight into the nonlin- Another way of arriving at Eq. (A19) is the follow- ear macroscopic energetics of the RS metamaterial while ing: given we had started with a smaller hinge of the excluding extremal configurations, as shown in Sec.A4. same initial shape, we must arrive at the same final force- Importantly, as the function g1 defined in Eq. (A19) is balanced hinge shape by simply rescaling the applied dis- scale-free, there is no remaining “hidden” dependance on placement vector δc by the same factor as `. Therefore, the hinge size ` nor the square size a in the energy func- we must be able to write the hinge energy as in Eq. (A19) tion and the scaling is directly apparent. as a function of the current twist, the pretwist, and the We have therefore avoided a complicated tensor-field value of δc relative to `. description of our structured elastic metamaterial, in We now apply our previous assumption that we will favor of a hinge-based energetic formalism based on a avoid extreme, high energy configurations of the hinge greatly simplified set of degrees of freedom: the positions where material fracture becomes likely. Quantitatively, and orientations of the square elements. In this formal- this is the condition δc 1. We may use this condi- ` ism, energetics and scaling permit analytic descriptions tion to expand the energy in Eq. (A19) to second order, which prove helpful for coarse graining. yielding

2 EH =` gT (T ) 4. Coarse-Graining Procedure for Nonlinear (δc · cˆ)2 (δc · cˆ )2 + k (T ) + ⊥ k (T ) Nonuniform Mechanics of the Checkerboard Lattice 2 s 2 µ (A20) δc3 + O , We now derive the continuum elasticity theory for the ` rotating-square (RS) lattice, as presented in the main text. Following Section.A3, we approximate the elastic The twist-dependent coefficients may be written in terms RS metamaterial as a collection of rigid square elements of g1 via connected by comparatively flexible hinges. The hinges g (T ) = g (T, 0,T ) and square elements may be made out of the same ma- T 1 0 terial, while the comparative flexibility arises due to the 2 ∂ g1 small size of the hinge. ks(T ) = ∂(δc · cˆ)2 T,δc=0,T0 (A21) We establish the local continuum energy penalty cre- 2 ated by a mapping x → X(x). This mapping will control ∂ g1 kµ(T ) = 2 . the location of the square centers, while the orentations ∂(δc · cˆ⊥) T,δc=0,T0 are left free to realize torque balanced configurations. Using Eqs. A17& A21, we can convert this energy to the Following the main text result that soft deformations of angular parametrization. This becomes the RS metamaterial are near-conformal, we restrict our focus to deformations composed of a large nonlinear di- 2 lation, a finite local rotation, and a small generic shear. EH =` g(˜c) This prescription is captured by the deformation tensor a2k (˜c) + 1 (δθ2 + δθ2 ) + a2k (˜c)δθ δθ A B 2 A B (A22) 2 s1(x) (3) s2(x) (1) 3 F (x) = R(φ(x)) α(x) δ + σ + σ δc ij ik kj 2 kj 2 kj + O , ` (A24) 14 where cell to get a continuum energy density. To identify the torque-balanced square orientations, we note that there 0 1 1 0 σ(1) = & σ(3) = (A25) are only two squares per unit cell. We therefore assume 1 0 0 −1 that square orientations are controlled by two angular fields which vary slowly through space. A unit cell con- are Pauli matrices and R(φ) a standard rotation by φ. tains squares A and B which after deformation are lo- To understand this tensor, we break it down piece by cated at rA and rB and have orientations θA and θB. piece. First, recall that by definition this tensor de- There is already some intuition for the expected orienta- scribes the Jacobian of the deformation map x → X(x) tions in terms of the counter-rotation mechanism of the ∂Xi via Fij = . The infinitesimal material vectors are ∂xj squares arising from the local lattice dilation factor α(x). therefore understood to transform under deformation via This is expressed by writing the orientations as dx → dX = F · dx. Consider, then, the action of each el- √ ement of F going from right to left, in the order that they θA = T ( 2aα(rA)) + φ(rA) + ∆T (rA) + ∆φ(rA) act on dx. The first (rightmost, in square brackets) term √ θ = −T ( 2aα(r )) + φ(r ) − ∆T (r ) + ∆φ(r ) , applies a linear strain composed solely of shears s1(x) B B B B B and s2(x) defined as (A28) where T () is the same mechanism function from App.A3, s1 = ∂xux − ∂yuy while ∆T, ∆φ are the two smooth fields required to cap- (A26) ture the two orientation degrees of freedom in each unit s2 = ∂xuy + ∂yux cell. Note that while Eq. (A28) relies on a well-motivated where ui is the i-th component of displacement, defined guess, there is no loss of generality due to the presence at this intermediate deformation for the purposes of un- of the correction fields ∆T, ∆φ, and conveniently that if derstanding. Defined in this way, s1 and s2 correspond to our guess is correct, then enforcing torque balance will the traditional notions of pure shear and simple shear, re- simply set the value of these fields locally to zero. In the spectively. Following the application of the linear shear, limit of vanishing shears and vanishing gradients of α and the material vectors undergo an isotropic dilation α(x) φ, we find that ∆T and ∆φ also vanish and therefore the and are then rotated by φ(x), bringing us to the final magnitude of these angular correction fields must be on configuration. While this is a general form for such a the order of the shears or ∇φ or smaller. (large dilation plus small shears) deformation, we note We are now equipped with a sufficient description to that these new variables φ, α, s1, s2 cannot vary arbitrar- specify the energy (Eq. A22) of an arbitrary hinge con- ily through space, as they must still obey a closure connecting the square initially centered at x to the square 0 dition to guarantee geometric (mechanical) compatibil- initially centered at x + cν in terms of our continuum ity. This is because a generic, spatially varying candidate quantities (φ, α, s1, s2, ∆T, ∆φ). Constructing the re- strain field may not correspond to a valid displacement quired geometric quantities to evaluate the hinge energy, field, just as a generic candidate electrical field does not we first find via application of the deformation tensor always correspond to an electrostatic potential. In the that the length of cν changes as vanishing shear limit the closure condition takes the form 1 0 |cν | → |cν |α(x + cν )× (A29) 0 = ∂iα − αij∂jφ , (A27) 2 " 0 2 0 2 s1(x) (cν · eˆ1) − (cν · eˆ2) where ij is the 2-by-2 antisymmetric unit tensor. Reas- 1 + 0 2 suringly, Eq. (A27) captures the Cauchy-Riemann condi- 2|cν | tion for a deformation to be conformal. To see this, note 0 0 # s2(x)(cν · eˆ1)(cν · eˆ2) that the derivative of a conformal function may be writ- + 0 2 , 0 |cν | ten in modulus-argument form as ∂zf = f = α exp[iφ]. Then, as conformal functions must be functions of z 0 only, we must have ∂z¯f = 0. Writing out the real and wheree ˆ1, eˆ2 are the orthogonal basis vectors (x- and y- imaginary parts of this condition leads to the x- and y- directions, respectively) in the reference space. The de- components of Eq. (A27). formation also produces a reorientation of this local in- We now choose the vectors cν connecting neighbor- finitesimal via ing square centers (and where ν = {1, 2, 3, 4} indexes 1 which neighbor) to represent the infinitesimal vector el- Arg(c ) → Arg(c ) + φ x + c ν ν 2 ν ements of the effective continuum material, and thus Fij prescribes all the relative square locations after deforma- (A30) tion. If we can now find a similar recipe for writing down s (x) (c · eˆ )2 − (c · eˆ )2 + 2 ν 1 ν 2 the square orientations in terms of continuum quantities, 2|c |2 then our description of the metamaterial configuration ν is effectively complete. We will then be able to write s1(x)(cν · eˆ1)(cν · eˆ2) − 2 . down the hinge energies and sum these over the unit |cν | 15

This information, combined with Eq. (A28), allows us to formulation rather than the vector formulation where write the angles of deviation z ≡ x1 + ix2, and similar for vector quantities such as cν . Our continuum energy density may now be constructed (ν) ∂T by summing the hinge energy Eq. (A16) over the four δθ =∆T + ∆φ − Re [cν ∂zα] − Re [cν ∂zφ] A ∂α hinges in the unit cell via 2 2 Re[cν ] ∂T Im[cν ] ∂T − 2 αs1 + s2 + 2 s1 − s2 α 2|cν | ∂α 2|cν | ∂α 4 1 X (ν) (ν) ∂T Φ(z) = EH (A32) δθ =∆T − ∆φ + Re [c ∂ α] − Re [c ∂ φ] Acell B ∂α ν z ν z ν=1 2 2 Re[cν ] ∂T Im[cν ] ∂T + s − αs − s + s α 2 2 2 1 2 1 2 where A = 4a cos(T ) is the area of the unit cell in 2|cν | ∂α 2|cν | ∂α cell 0 (ν) (A31) the reference space and EH is the energy from Eq. (A22) which will be input to determine the energy of hinge for hinge ν. Inserting Eqs. A31 and performing the sum ν. For convenience, we have moved to the complex yields

`2 g(α) (k (α) + k (α))T 0(α)2α2 (k (α) − k (α)) Φ(z) = + 1 2 s2 + 1 2 s2 a2 cos(T )2 4cos(T )2 1 4cos(T )2 2 0 0 0 (A33) 2 2 a k1(α) 1 0 2 a k2(α) 1 0 2 2 1 2 2 + ( 2 + T (α) )) + ( 2 − T (α) )) |∇α| + (k1(α) + k2(α)) ∆T + ∆φ . 4 α 4 α cos(T0)

The energy in Eq. (A33) includes all terms which are the small hinge limit, the strain gradient modulus may not higher order in either `/a or the shears or |∇α|. To be inferred from the shear moduli via the simple relation reduce to an elasticity theory in terms of spatial defor- 2 2 4 mation variables, we minimize Eq. (A33) with respect ˜ 1 − α cos (T0) cos (T0) B(α) = 4 G1(α) + 2 2 G2(α), to the internal degrees of freedom ∆T and ∆φ. This α 1 − α cos (T0) leads to ∆T = 0 and ∆φ = 0, which indicates that (A39) our mechanism-based guess for the square orientations where we have used the analogue for the mechanism in in Eq. (A28) was correct. The continuum elastic theory Eq. (A13) to replace for the RS lattice is then Z 2 0 −cos(T0) 2 1 ` 2 2 T (α) = (A40) E = d x M(α) + a M˜ (α)|∇α| + p 2 2 2 a2 1 − α cos (T0) 2 2 excluding higher order terms proportional to `/a. G1(α)s1 + G2(α)s2 . (A34) Where the moduli 5. Symmetry-based construction of the energy 2g(α) functional B(α) = 2 , cos(T0) (A35) Here, we present an alternate derivation of an energy functional of the form Eq. (A34) without recourse to the 0 2 2 (k1(α) + k2(α))T (α) α exact RS microstructure, based instead upon its funda- G1(α) = 2 , (A36) 2cos(T0) mental symmetries. (k1(α) − k2(α)) We consider, then, a general elastic system that: G2(α) = 2 , 2cos(T0) (A37) • has a dilational mechanism k1(α) 1 • has standard elastic translational and rotational B˜(α) = + T 0(α)2) 2 α2 symmetries (A38) • has a four-fold rotational symmetry k (α) 1 + 2 − T 0(α)2) , 2 α2 • has an x → −x mirror symmetry capture the energy cost of dilation, pure shear, simple • has a y → −y mirror symmetry (as implied by the shear, and dilation gradients, respectively. Note that in previous two symmetries) 16

s1 s2 ∂xα ∂yα nature of the RS mechanism or material properties to C4 −s1 −s2 −∂yα ∂xα derive this elastic energy, is it specific to the point sym- x-mirror s1 −s2 −∂xα ∂yα metry group of the lattice and a different energetic form y-mirror s1 −s2 ∂xα −∂yα is expected for other dilational metamaterials such as the TABLE I. Action of symmetry operations on deformation kagome lattice. fields

6. Analytic prediction of deformation These last three properties arise from the p4g wallpaper symmetry group, which applies to all states of the lattice along the uniform dilational mechanism. Having identified the effective continuum energy which A uniform application of the mechanism will gener- will govern the deformation of the checkerboard meta- ate an energy density that we again label (`/a)2M(α), material, we would like to use it to predict material re- having identified the scaling with hinge thickness in Sec- sponse to applied loads. Taking the limits of small hinge tionA3. Additionally, energy terms may arise from other and small lattice spacing in Eq. (A34) leads to an en- components of the deformation tensor: pure shear (s ), ergy which is comparatively very stiff against deforma- 1 tions that include shears, and the shear terms dominate simple shear (s2) and rotation (φ). However, rotation in particular can be easily eliminated: the system is invari- the energy functional. In this limit, the material will ant under a uniform rotation and Eq. (A27) (Cauchy- choose to expel shear whenever possible, leading directly Riemann) shows that for dilation-dominated deforma- to the conformal maps (which definitionally exclude lo- tions, rotation gradients can be expressed in terms of cal shear) as the lowest energy deformations. With only dilation gradients instead (up to a small correction from shear terms in the energy, the conformal deformations the shears). Overall, our energy density can therefore be form a highly degenerate space of ground states. The point displacements applied in the Finite Element simu- expressed in terms of three fields: (s1, s2, α) and gradients thereof. In general, this allows for terms propor- lations from SectionA1 are not sufficient to fully break 2 the degeneracy — an infinite space of deformations sat- tional to s1, s1s2, ∂xα∂yα etc. Terms such as ∂xs1∂ys1 2 isfy the (point) boundary conditions without any shears, and s2∂xα are also permitted, but these will be higher order in gradients and shears which are small due to the leaving no single prediction for the deformation. comparative size of the unit cell and design based on a To break this degeneracy, we turn to perturbation the- mechanism, respectively. ory, adding back in the dilation M(α) and dilation gra- However, such terms are only permitted if they are in- dient M˜ (α) terms. Predicting the deformation then be- variant under the discrete symmetries of the undeformed comes a search for the conformal map which minimizes or uniformly dilated lattice: a four-fold rotation about the perturbative energy the center of a square and two mirror symmetries about Z 1 ` the center of an open pore. E ∼ ∆E ≡ dx M(α) + a2M˜ (α)|∇α|2 . (A42) The action of the symmetry operations on the rel- 2 a evant fields are listed in Table.I. Enforcing that the energy remain the same before and after applying the Shown in Fig.9-a, this reduced form is a reasonable ap- x-mirror symmetry eliminates (s2, ∂xα, s1s2, ∂xα∂yα, proximation to the true energy, with the shear energy s2∂yα, s1∂xα ) from the energy density. Similarly, the roughly an order of magnitude smaller at small hinge y-mirror symmetry eliminates the terms (∂yα, s2∂xα, size. This energy gap will be even more significant, and s1∂yα). Finally, the C4 symmetry eliminates S1, and fur- the perturbative approximation better, for larger mate- 2 ther enforces that the energetic coefficient of the (∂xα) rial samples with more unit cells which are closer to the 2 term match the coefficient of (∂yα) . We do not con- continuum limit. As described in the main text, this sider terms involving second gradients of the dilation process of obtaining and using this effective theory con- field, which are ruled out in the coarse-grained micro- stitutes our conformal elasticity. scopic theory. With this, the energy functional to lowest order in the small shears and dilation gradients must take the form a. Analytic prediction of linear deformation using effective conformal theory Z 2 2 1 ` 2 ˜ 2 E = d x 2 M(α) + a M(α)|∇α| + 2 a The energy in Eq. (A42) is highly nonlinear and min- 2 2 imization requires prior knowledge of the nonlinear stiff- G1(α)s1 + G2(α)s2 . nesses k1(α), k2(α), g(α), which are not easily obtained. (A41) Even with knowledge of these moduli, analytic solu- where the coefficients have been named to connect to the tions to such nonlinear mechanics problems are hard to coarse grained theory in Eq. (A34). come by. However, taking the theory to lowest order in Importantly, while we have not invoked the specific strains and rewriting in terms of a complex displacement 17

a. b.

- -

- - - - -

- -

- - - -

FIG. 9. Conformal elasticity (a) We show that going from Eq. (A34) to Eq. (A42) is a good approximation by showing that the contribution to the energy from the shear terms is an order of magnitude smaller (or more) in the limit of small hinges. Energies are normalized by E0, which is the amount of energy required to create the same coarse strain magnitude in the absence of the pore structure, and which showcases also the comparative softness of the porous metamaterial. The energy in Eq. (A42) may also be used to generate predictions of small deformations which have very small fractional mean squared error shown in (b).

u(z) = f(z) − z = ux(z) + iuy(z) yields offers a relation between these moduli which is exact in the limit of small hinges Z 2 1 ˜ 2 2 E = dzd¯z KRe(∂zu) + K|∂z u| + Econstr 2 2 ˜ 2 2 cos(T0) K = a sin(T0) G1(α = 1) + 2 G2(α = 1) . (A43) tan(T0) where ∂z = 0.5(∂x − i∂y) is a standard complex deriva- (A46) tive [32] and With this, the strain level moduli (K,G1,G2) obtained in the methods section of the main text constitute all neces- `2g00(1) K = sary information to make linear deformation predictions. 4a2cos(T )2 0 (A44) For simulations with hinge thickness `0 = 0.1mm these 3 K˜ = a2M˜ (α = 1) moduli are (K, K˜ ) = (35Pa*m, 125Pa*m ). To find predictions at different hinge thicknesses `, we rely on the are the bulk and bulk gradient moduli, respectively. Note coarse graining results, which indicate that K ∼ `2 while the unknown function g, defined in SectionA3, which all other moduli G1,G2, K˜ are independent of `. There- captures the mechanism energy. In order to incorporate `2 fore, we write K(`) = `2 K(`0) in terms of the known the applied point motions (constraints), Lagrange mul- 0 tiplier terms have been added to the energy functional K(`0) and use these values to solve the linear algebra as problem for the coefficients {Cn} and therefore our prediction for u(z). These predictions capture all but a small Z ( ) 2 X portion ∆pred of the error, as displayed in Fig.9-b. This Econstr = dzd¯z λk(uk − u(zk)) , (A45) completes k where {uk} are the displacements prescribed at locations 7. Boundary method for analytic prediction of {zk}. The energy in Eq. (A43) will be used to select between conformal deformation conformal displacements u which may be written as an P n expansion u(z) = n Cnz . As higher coefficients Cn As described in the main text, materials in the con- generate sharper and sharper features, we can cut off formal limit admit a novel boundary method for analytic the expansion and then solving the equations of energy control of deformation. Here we give more explicit ana- minimization is reduced to a linear algebra problem. lytic recipes both for continuous analytic description of The above is not yet sufficient to generate predictions, the metamaterial and for a discrete approximation. as the moduli K and K˜ remain undetermined. Conve- As noted in the main text, a planar conformal defor- niently, the coarse-graining procedure from Section.A4 mation may be described by a complex analytic function 18 z → f(z). The function f(z) obeys ∂z¯f = 0 (¯z is the • solve for α(x, y) using the boundary conditions and complex conjugate of z), and as a result admits well de- the equation ∇2 ln(α) = 0 fined z-derivatives everywhere in its domain and can be • insert to determine the function f 0(z) = expanded in a series as in Eq. (A8). These analytic prop- z z 0 exp 2 ln(α( , )) + φ0i where φ0 is the undeter- erties are preserved when the derivative is taken f ≡ ∂zf 2 2i and when we take the logarithm, defining g ≡ ln(f 0). mined global rotation. Note that the logarithm has the potential to introduce • integrate to find f(z) nonanalyticity, yet avoiding nonphysical material configurations such as inversion and collapse into a point avoids We remark that the analytic viability of this recipe de- any such problems. The function f 0 may be written in pends on the ability to generate solutions to the Laplace modulus-argument form as equation on the chosen domain, and to integrate them. The examples of square and circular domains provide f 0(z) = α(z, z¯)exp(iφ(z, z¯)) . (A47) convenient solvable examples. However, the true strength of our method lies in the reverse application. Given a de- In finite strain theory, f 0 captures the deformation tensor, sired conformal deformation f(z), this may be prescribed the function α will capture the local isotropic rescaling of by simply taking the derivative and evaluating the mag- area dA → α2dA due to deformation, and φ captures the nitude at the boundary, yielding the appropriate bound- coarse material rotation (e.g. reorientation of the entire ary dilation pattern to actuate. It is the uniqueness of unit cell). Taking the logarithm, our function g may be the inverse recipe shown here (up to overall translations written and rotations) which guarantees only the target map can g = ln(α) + iφ . (A48) satisfy these boundary conditions. Finally, the particular choice of mechanism sets limita- The real part of this function now depends only on the tions to the amount the metamaterial can be dilated and scaling factor α while the imaginary part depends only contracted to avoid self intersection and material over- on the rotation φ. It is well known that the real and extension. For instance, the RS mechanism√ requires α imaginary parts of a complex analytic function each sep- to be everywhere confined to the range 1/ 2 < α < 1, arately satisfy the Laplace equation, which appears as when starting from the fully dilated configuration. How- 4∂z∂z¯ln(α) = 0 in complex form for the real part of ever, very conveniently, it turns out that any sufficiently g. The Laplace equation admits a unique solution given smooth choice of boundary dilations which do not over- Dirichlet data (the local value of ln(α)) all along a closed or under-dilate the RS lattice on the boundary, will also boundary. Solving this for the real part of g, the bound- generate a conformal deformation in the interior which ary data for ln(α) comes from local lattice dilations along does not go beyond the valid dilation range of the specific the boundary. Therefore, given prescribed dilation data mechanism. This is due to the maximum modulus prin- all along the boundary of a closed material domain, the ciple which states that both the maximum and minimum dilations throughout the bulk are determined by the dilations of a conformal map will occur at the boundary. Laplace equation. Next we note that full knowledge of Re[g] across the domain may be used to infer the full analytic function g up a. Discrete Inference Method to a purely imaginary constant. Recipes to infer g such as the method of Milne-Thompson [33] or that applied by For most material domain shapes and boundary dila- John d’Angelo in Ref. [34] rely on the Cauchy-Riemann tion patterns, a closed-form solution for f(z) is not guar- equations. Employing the method of John d’Angelo we anteed. We therefore illustrate an alternative scheme may write the full function that allows for quick and accurate inference of a map z z from discrete boundary data, and which generates the g(z) = 2 ln α x → , y → + iφ (A49) 2 2i 0 predictions in main text Fig. 4. in terms of the internal dilation field α(x, y) up to an In this method, we first identify the boundary via a set of discrete points {zk} and the corresponding local dila- undetermined material rotation φ0. Finally, taking the exponential f 0 = exp(g) yields a complete prediction of tions {αk}. Points should be densely spaced compared to the deformation tensor throughout the bulk. Integrating the lengthscale of variation of αk. And, naturally, they 0 should approximately trace a single continuous path en- f along a path from zi yields a prediction for the relative final position of the material point initially located at z closing a simply-connected region of the plane. Following f this, we choose a cutoff N in the number of coefficients Z zf of f(zf ) − f(zi) = exp [g] dz . (A50) zi N−1 X n Importantly, the conformal property guarantees that this g(z) = Cnz . (A51) integral is not path dependent. n=0 In summary, the full recipe for inferring the spatial N should be significantly less than half the number of mapping f(z) from the boundary is as follows: boundary points M in order to avoid overfitting. 19

Next we demand that our function g(z) satisfy the where boundary conditions by minimizing the error M X n+l Qln = rk cos(nθk)cos(lθk) k M X n+l Rln = − rk sin(nθk)cos(lθk) k M X n+l Sln = − rk cos(nθk)sin(lθk) k (A54) M M X 2 err = (Re[g(zk)] − ln(αk)) . (A52) X n+l Tln = rk sin(nθk)sin(lθk) k k M (A) X l fl = − ln(αk)rkcos(lθk) k M (B) X l fl = ln(αk)rksin(lθk) , k

iθ and we have expressed zk = rke k in a complex polar Inserting Eq. (A51), we minimize this error with respect form. Eq. (A53) reduces the inference of g to a linear algebra problem which may be readily solved using built-in to the coeﬃcients Cn = An + iBn yielding the equations tools in, e.g., Mathematica. Note importantly that the row and column corresponding to B0 = φ0 (the undetermined global rotation) are zero throughout. This row- column pair is, in general, the only one that needs to be removed before numerically solving. The above yields the coeﬃcients Cn of g(z), which is inserted into the exponential function to yield a prediction for ∂zf and therefore constitutes a full prediction of the deformation tensor everywhere inside the domain. N N ∂[err] (A) X X In order to generate the displacement predictions shown 0 = = fl + AnQln + BnRln in main text Fig. 3, we employ the built-in numerical ∂Al n n (A53) integrators in Mathematica. N N ∂[err] (B) X X 0 = = f + A S + B T , ∂B l n ln n ln l n n

[1] MIURA and Koryo, The Institute of Space and Astro- 117, 483 (1995). nautical Science report 618, 1 (1985). [10] E. T. Filipov, T. Tachi, G. H. Paulino, and D. A. Weitz, [2] K. Bertoldi, P. M. Reis, S. Willshaw, and T. Mullin, Proceedings of the National Academy of Sciences of the Advanced Materials 22, 361 (2010). United States of America 112, 12321 (2015). [3] J. Shim, C. Perdigou, E. R. Chen, K. Bertoldi, and P. M. [11] B. Deng, V. Tournat, P. Wang, and K. Bertoldi, Physical Reis, Proceedings of the National Academy of Sciences of Review Letters 122, 044101 (2019). the United States of America 109, 5978 (2012). [12] B. Deng, J. Li, V. Tournat, P. K. Purohit, and [4] B. Florijn, C. Coulais, and M. van Hecke, Physical Re- K. Bertoldi, Journal of the Mechanics and Physics of view Letters 113, 175503 (2014). Solids , 104233 (2020). [5] J. L. Silverberg, A. A. Evans, L. McLeod, R. C. Hayward, [13] L. Jin, R. Khajehtourian, J. Mueller, A. Rafsanjani, T. Hull, C. D. Santangelo, and I. Cohen, Science 345, V. Tournat, K. Bertoldi, and D. M. Kochmann, Proceed- 647 (2014). ings of the National Academy of Sciences of the United [6] C. L. Kane and T. C. Lubensky, Nature Physics 10, 39 States of America 117, 2319 (2020). (2014). [14] K. Bertoldi, V. Vitelli, J. Christensen, and M. van Hecke, [7] J. Paulose, B. G. G. Chen, and V. Vitelli, Nature Physics Nature Reviews Materials 2, 17066 (2017). 11, 153 (2015), arXiv:1406.3323. [15] J. N. Grima and K. E. Evans, Journal of Materials Sci- [8] A. Saremi and Z. Rocklin, Physical Review X 10, 011052 ence Letters 19, 1563 (2000). (2020), arXiv:1908.07499. [16] T. Mullin, S. Deschanel, K. Bertoldi, and M. C. Boyce, [9] G. W. Milton and A. V. Cherkaev, Journal of Engineer- Physical Review Letters 99, 084301 (2007). ing Materials and Technology, Transactions of the ASME 20

[17] C. Coulais, C. Kettenis, and M. van Hecke, Nature [26] Y. Suzuki, G. Cardone, D. Restrepo, P. Zavattieri, Physics 14, 40 (2018). T. Baker, and F. Tezcan, Nature 2016 533:7603 533, [18] K. Sun, A. Souslov, X. Mao, and T. Lubensky, Pro- 369 (2016). ceedings of the National Academy of Sciences 109, 12369 [27] M. Konakovic, K. Crane, B. Deng, S. Bouaziz, (2012). D. Piker, and M. Pauly, ACM Trans. Graph. 35 (2016), [19] V. Riva and J. Cardy, Physics Letters B 622, 339 (2005). 10.1145/2897824.2925944. [20] C. D. Santangelo, Soft Matter 9, 8246 (2013), [28] G. P. T. Choi, L. H. Dudte, and L. Mahadevan, Nature arXiv:1210.1488. Materials 18, 999 (2019). [21] D. Prall and R. S. Lakes, International Journal of Me- [29] M. Pretko and L. Radzihovsky, Physical Review Letters chanical Sciences 39, 305 (1997). 120, 195301 (2018), arXiv:1711.11044. [22] G. W. Milton, Journal of the Mechanics and Physics of [30] C. Coulais, International Journal of Solids and Structures Solids 61, 1543 (2013), arXiv:1205.6213. 97-98, 226 (2016). [23] Y. Cho, J. H. Shin, A. Costa, T. A. Kim, V. Kunin, [31] D. M. J. Dykstra, J. Busink, B. Ennis, and J. Li, S. Y. Lee, S. Yang, H. N. Han, I. S. Choi, and C. Coulais, Journal of Applied Mechanics 86 (2019), D. J. Srolovitz, Proceedings of the National Academy 10.1115/1.4044036. of Sciences of the United States of America 111, 17390 [32] A. England, Complex Variable Methods in Elasticity, (2014). Dover Books on Mathematics Series (Dover Publications, [24] R. Gatt, L. Mizzi, J. I. Azzopardi, K. M. Azzopardi, 2003). D. Attard, A. Casha, J. Briﬀa, and J. N. Grima, Sci- [33] L. M. Milne-Thomson, The Mathematical Gazette 21, entiﬁc Reports 5, 1 (2015). 228 (1937). [25] C. P. Goodrich, A. J. Liu, and S. R. Nagel, Physical [34] M. Stone and P. Goldbart, Mathematics for Physics: A Review Letters 114, 225501 (2015). Guided Tour for Graduate Students (Cambridge Univer- sity Press, 2009).