the designs and deformations of rigidly and at-foldable ...the state of the art in origami design...
TRANSCRIPT
-
The designs and deformations of rigidly and flat-foldablequadrilateral mesh origami
Fan Feng1, Xiangxin Dang2, Richard D. James3, and Paul Plucinsky4
1Cavendish Laboratory, University of Cambridge, UK2State Key Laboratory for Turbulence and Complex System, Department of Mechanics and Engineering
Science, College of Engineering, Peking University, Beijing 100871, China3Aerospace Engineering and Mechanics, University of Minnesota, USA
4Aerospace and Mechanical Engineering, University of Southern California, USA
April 22, 2020
Abstract
Rigidly and flat-foldable quadrilateral mesh origami is the class of quadrilateral mesh creasepatterns with one fundamental property: the patterns can be folded from flat to fully-foldedflat by a continuous one-parameter family of piecewise affine deformations that do not stretchor bend the mesh-panels. In this work, we explicitly characterize the designs and deformationsof all possible rigidly and flat-foldable quadrilateral mesh origami. Our key idea is a rigiditytheorem (Theorem 3.1) that characterizes compatible crease patterns surrounding a singlepanel and enables us to march from panel to panel to compute the pattern and its correspondingdeformations explicitly. The marching procedure is computationally efficient. So we use it toformulate the inverse problem: to design a crease pattern to achieve a targeted shape alongthe path of its rigidly and flat-foldable motion. The initial results on the inverse problem arepromising and suggest a broadly useful engineering design strategy for shape-morphing withorigami.
1 Introduction
Origami is a popular Japanese artform that aims to achieve complex shape by the intricate foldingof an initially flat piece of paper. The state of the art in origami design incorporates mathematicalprinciples for packing problems and well-known fold operations into algorithms capable of producingcrease patterns that can be folded into a wide range of targeted shapes [13, 14, 34, 51]. While strikingin their achievements, the crease patterns emerging from these algorithms need to be folded by adelicate sequence of (often numerous) steps to reach the final shape. As an artform, this is hardlyan issue to the skilled Origamist. However, the use of origami in engineering involves modalitiesof folding such as simple external loading, motors, active materials, hinge offsets for thick origamipanels [8], and hinge mechanisms [45, 20]; none of which demonstrate capabilities anywhere close tothat of an Origamist. As a result, the utility of these algorithms for practical engineering is limitedto the simplest of designs and shapes [18].
1
arX
iv:2
003.
1289
6v2
[co
nd-m
at.s
oft]
21
Apr
202
0
-
Figure 1: The Miura-Ori crease pattern and its rigidly foldable motion.
In contrast, the famed Miura-Ori pattern shown in Fig. 1 was adapted as a concept for packagingand deploying large space membranes [38], and has since been the topic of numerous investigationsin the engineering literature. This pattern is made of a repeating corrugated unit cell (in bold)whose crease āmountain-valley assignmentsā are biased in the flat state (red/blue). Because ofthe geometry of the unit cell and this bias, the Miura-Ori has a unique (up to rigid motions),continuous one-parameter family of deformations that do not stretch or bend any of the panels;that of a uniform compressive motion from flat to folded flat. The Miura-Ori crease pattern is alsoinherently confined to a thin sheet. Thus, geometry and elasticity [10, 21] dictate that a carefullydesigned Miura-Ori, involving stiff panels and flexible creases, can achieve this large coordinatedmotion by a variety of stimuli in a fashion robust to non-idealities. In fact, every modality discussedabove has proven effective at folding a Miura-Oriāmost at a variety of length and time-scalesāandthis made the Miura-Ori a popular template for exploring functionality enabled by shape-morphing:for tunable and novel properties in compression [47], bending [56], and as a meta-material [46]; forself-folding at the microscale [40] and mesoscale [53]; for physically driven self-origanization [37];for a fast and controlled response in soft-robotics [30]; and for ease of deployment in space structureapplications [38, 42].
All this notwithstanding, only so much functionality can be squeezed out of the shape-morphingcapability of a Miura-Ori. Yet, existing predictive algorithms for producing complex shapes bysequential folding, as discussed above, are quite limited. We aim to fill the gap. To this point, we
2
-
note that the Miura-Ori is but the most basic example in the class of origami termed rigidly andflat-foldable quadrilateral mesh origami (RFFQM). This class is composed of quadrilateral meshcrease patterns with the defining property: when the crease pattern is suitably biased in the flatstate, by imposing the correct mountain-valley assignments, the pattern can fold from flat to fully-folded flat by a unique (up to rigid motion) and continuous one-parameter family of piecewise-affinedeformations that do not stretch or bend the mesh-panels. As this shape-morphing capability isdesirable in many of the aforementioned applications, we find it natural to examine the configurationspace of RFFQM in detail.
The understanding of RFFQM has evolved significantly within the past several years. Kawasaki[29] asserted a necessary condition for flat-foldability at a single degree-4 vertex, proved by Hull [25].Huffman parameterized the kinematics at flat-foldable degree-4 vertices [24]. Belacastro and Hull[26] formalized conditions of compatibility for origami using piecewise affine transformations. Tachi[50] revitalized the subject by isolating the fundamental questions of compatibility and solving themin special cases by perturbative methods, yielding surprising examples. Finally, Lang and Howell[36] made clever use of observations due to Tachi [52] to derive an algorithm that parameterizes theconfiguration space of RFFQM.
In this paper, we develop some new lines of thinking, while expanding on many ideas introducedpreviously. On the latter, it turns out that every fundamental question in origami design1 can bereduced to the question of āhow to link up neighboring folding angles by parameterizations thatdepend, in a delicate way, on the geometry of the crease patternā. To the adept Origamist, like thosereferenced previously, this is apparently a natural matter of fact. As such, it serves as a starting pointto any development in the literature. From the point of view of continuum mechanics, however, themore natural observation is: any origamiāin the purest sense of the definitionāis the result of anexplicit deformation of an initially flat crease pattern. Thus, questions and answers about designingorigami crease patterns and parameterizing their kinematics can be addressed by working directlywith deformations; specifically, by deriving and characterizing the local compatibility conditions ondeformation gradients that are necessary and sufficient for continuity of the overall deformation.
The use of the deformation as the basic kinematic object is an underdeveloped aspect of origamidesign. We develop it in Section 2 for the following purposes:
ā¢ to derive and characterize vertex compatibility, i.e., the conditions under which the creasepattern around a single vertex can be folded as origami;
ā¢ to derive panel compatibility, i.e., the conditions under which a crease pattern surrounding asingle panel can folded as RFFQM;
ā¢ to explicitly characterize the full kinematics of RFFQM when such compatibility is achieved.
The last point is a uniquely beneficial aspect of working directly with deformations. By this char-acterization, we fully determine the deformations surrounding a single vertex and surrounding asingle panel up to an overall rigid body motion of the entire crease pattern. The global deformationof the overall crease pattern must then be consistent with this local deformation.
Formalizing these necessary and sufficient conditions for compatibility brings several observa-tions and results new to the origami literature. We first show that parameterizations of the folding
1Except for open issues of global invertibility, which we do not discuss.
3
-
angles that solve vertex compatibility belong to an underlying Abelian group. We then use proper-ties of this group to characterize panel compatibility in full generality. The key ideas are: 1) Panelcompatibility is a condition that a certain composition of maps in this group is the identity map,and 2) a fundamental property of the group is that composition of maps is, in a precise sense, thesame as multiplication of parameters. Thus, an apparently hard problem of characterizing a com-position of maps is reduced to an easier problem of characterizing a multiplication of parameters.From this reduction, we derive a design theorem for RFFQM which, stated informally, addresses thefollowing question: If we know the sector angles and mountain-valley assignments at three of fourvertices surrounding the panel, can we design the fourth vertex so that the panelās crease patterncan deform as RFFQM? The answer is generically yes. In fact, there is a unique characterizationof the fourth vertex, which we provide via explicit design formulas in Theorem 3.1.
Because this theorem can be iterated, it enables a broad engineering design principle for RF-FQM (Section 4). First, we explicitly parameterize the deformations of all possible RFFQM bymarching from panel to panel and repeatedly applying the design theorem. This procedure resultsin a computationally efficient marching algorithm for the configuration space formally similar toLang and Howellās [36], but with distinctly different parameterization. Our parameterization in-volves purely referential quantities of the flat crease pattern (sector angles and mountain-valleyassignments) whereas theirs involves kinematic variables (folding angles). We find that the refer-ential quantities are natural variables to use in formulating the inverse problem. Specifically, weincorporate this algorithm into an optimization scheme for producing a crease pattern to achievea targeted shape along the path of its rigidly and flat-foldable motion. Finally, we illustrate thepower of this approach by studying slight perturbations of the Miura-Ori crease pattern in thiscontext. Note, Dudte et al. [16] achieved a variety of curvatures by perturbed Miura-Ori, but theirpatterns are not RFFQM and can only be deployed by slightly stressing the panels. In contrast, ourstrategy works with perturbations consistent with RFFQM only, yielding patterns that are (ideally)stress-free in deployment. Importantly, even under this deployability consideration, we are still ableto achieve striking examples of origami that approximate surfaces of positive Gaussian curvature,negative Gaussian curvature and changes in Gaussian curvature.
With this work, we seek to introduce a continuum mechanics approach to origami and bringgeneral and easily implementable methodologies for engineering design to the forefront. Our over-arching motivation is that applications ranging from space structure design [2, 3], to biomedicaldevices [33] and soft robotics [30] often seek a targeted final shape and the ability to achieve it bycontrolled deployment from an easily manufactured or highly compact state. In RFFQM, we havean easily manufactured state (the flat state), a highly compact state (the folded flat state), andcontrolled deployment by a single degree of freedom motion; in other words, we have a templatefor shape-morphing in engineering design. While we are in the early stages of examining the versa-tility and utility of these methods, the results hereāand those in forthcoming work on the inverseproblem [12]āpoint towards a broad engineering design tool for shape-morphing with RFFQM.
Notation. Throughout, we deal with real valued scalars Ī³ ā R, vectors v ā R3 and matricesF ā R3Ć3. We let S2 denote the set of all unit vectors on R3 and SO(3) denote the set of all rotationmatrices in R3Ć3, called simply rotations below. The vectors {e1, e2, e3} will always refer to thestandard orthonormal basis on R3. We study exclusively deformations which do not bend or stretchthe mesh-panels within a crease pattern, i.e., we study origami deformations. Denote an overallcrease pattern by ā¦ ā R2 and let y : ā¦ ā R3 be an origami deformation of this crease pattern.Then, y is defined by the following two properties:
4
-
ā¢ If we isolate any single mesh-panel P ā ā¦, then the deformation on this panel has the form
y(x1, x2) = Rx + b, x = x1e1 + x2e2 ā P (1)
for some rotation R ā SO(3) and translation b ā R3.
ā¢ The rotation and translation can change from panel to panel but only in such a way that thedeformation is continuous on the entire crease pattern ā¦.
We often refer to the rotations in the formula (1) as deformation gradients. This is a slight abuse ofnotation. Since y maps a subset of R2 to R3, the deformation gradients of origami are actually lineartransformations of the form RĢ ā O(2, 3) := {FĢ ā R3Ć2 : FĢT FĢ = I}. For our purposes, however,this distinction is unimportant, since a 3Ć2 component of the rotation matrix determines uniquelyits third column as the cross product of the first two. So we prefer to work consistently here withvectors on R3 and matrices on R3Ć3. We refer the interested reader to Conti and Maggi [10] for aprecise description of the definitions and function spaces for origami deformations.
2 Local kinematics
2.1 Compatibility at a single vertex.
We begin with a local description of the kinematics of degree-4 vertices. To motivate this study, notethat, by isolating a small ball around a single degree-4 vertex within an overall quad-mesh creasepattern (Fig. 2), we can deduce local properties on the kinematics of RFFQM. It is the consistencyof these local conditions that allows us to systematically investigate the global kinematics of thecrease pattern.
As is well-known in the origami literature [25, 35], the necessary and sufficient condition forRFFQM of an initially flat crease pattern composed of a isolated degree-4 vertex (Fig. 2(b)) isKawasakiās condition [29], i.e., the condition that opposite sector angles sum to Ļ. Since we areexclusively interested in RFFQM in this work, we therefore make repeated use of the following localdescription of the crease pattern: We label the four crease tangents of this vertex as ti ā S2 withti Ā· e3 = 0 in a counter-clockwise fashion as depicted2, we define the sector angles as
Ī± := arccos(t1 Ā· t2), Ī² := arccos(t2 Ā· t3), (2)
and we enforce Kawasakiās condition
Ī±, Ī² ā (0, Ļ), arccos(t3 Ā· t4) = Ļ ā Ī±, arccos(t4 Ā· t1) = Ļ ā Ī². (3)
Note that this is simply a necessary condition on the sector angles for the overall crease pattern toadmit RFFQM.
The rigid-folding kinematics of an isolated degree-4 vertex under Kawasakiās condition are alsowell-known in the literature [35, 50]. The earliest derivation, due to Huffman [24], employs sphericaltrigonometry. Following the continuum mechanics approach adopted here, we write explicitly theunderlying deformation (from flat). That is, we impose that the deformation gradient satisfies a
2We will also use a clock-wise labeling when convenient.
5
-
t1
t2
t3
āµļæ½
t4
I
R2(ļæ½2)R3(ļæ½3)
R2(ļæ½2)
R1(ļæ½ļæ½1)
(a) (b)
(c)
R2(ļæ½2)R3(ļæ½3)R3(ļæ½3)
R2(ļæ½2)I
II
I
II
I
I I
IR2(ļæ½2)
R2(ļæ½2)R3(ļæ½3)
R1(ļæ½ļæ½1)
Figure 2: A description of vertex compatibility. (a) Isolate a single vertex within an overall creasepattern. (b) Its kinematics is, up to an overall rigid motion, necessarily described by rotations aboutcrease tangent as depicted, and these rotations are subject to a vertex compatibility condition. (c)This can be understood by the following thought experiment: 1) cut the pattern at a crease, 2)apply sequences of rotations along creases to this cut pattern as shown, and 3) parameterize thecorresponding rotation angles so that the pattern can be glued back together again.
6
-
jump compatibility condition across each crease. Though less familiar to the origami community3,this approach has found longstanding application in the continuum mechanics modeling of mi-crostructure for materials undergoing the martensitic phase transformation [4, 6, 11, 48]. It is alsoa highly generalizable; the basic structure of the derivation is unchanged if, say, we relax Kawasakiāscondition, introduce additional creases, and/or consider non-Euclidean vertices4 [5, 43, 44, 55]. Forthese reasons, we provide a brief exposition on the topic.
The continuum mechanics approach to the kinematics can be understood heuristically as fol-lows: Rigidly foldable origami describes deformations of the crease pattern that do not involve anybending or stretching of the panels, i.e, the kinematics are localized at the creases via perfect hingemechanisms. We can therefore keep one of the panels fixed in its initially flat state by applying Ias the deformation gradient of this region5. This has the effect of 1) eliminating the degeneracy ofrigid body motion and 2) providing a convenient reference on which to build the kinematics throughdeformation gradients that are products of rotations about crease tangents. Specifically, we will bedealing with rotations Ri(Ī³i) ā SO(3) that satisfy
Ri(Ī³i) := ti ā ti + cos Ī³iPi + sin Ī³iWi for{Pi := Iā ti ā ti,Wi := e3 ā tā„i ā tā„i ā e3,
(4)
and for tā„i ā S2 defined by tā„i := ā(ti Ā· e2)e1 + (ti Ā· e1)e1. These definitions yield four right-handorthonormal frames {ti, tā„i , e3} and four rotations that satisfy Ri(Ī³i)ti = ti. Accordingly, eachrotation describes a right-hand rotation about an axis ti by an angle Ī³i ā [āĻ, Ļ].
To build the kinematics from these rotations, we follow the sketch in Fig. 2(c). From the flatstate (far-left), we apply a rotation R3(Ī³3) to the (ĻāĪ±)-sector to obtain the configuration (middle-left) which cuts the pattern along the tangent t4 but otherwise keeps the panels rigid and continuousalong every other crease. We then apply a rotation R2(Ī³2) to the Ī²-sector and deformed (Ļ ā Ī±)-sector. This yields a configuration (middle-right) which again keeps the panels rigid and continuousacross each of the t1,2,3 creases but still leaves an opening at the crease with tangent t4. We thereforeapply a final rotation R1(āĪ³1) to the (Ļā Ī²)-sector in an attempt to glue the overall rigidly-foldedcrease pattern back together again. This question of gluing naturally leads to jump compatibilityconditions on deformation gradients. Indeed, as the deformation gradients to the left and right of t4are R2(Ī³2)R3(Ī³2) and R1(āĪ³1), respectively, the necessary and sufficient conditions for gluing arethat the crease tangent t4 deforms in exactly the same way under the actions of these deformationgradients, i.e., that R2(Ī³2)R3(Ī³3)t4 = R1(āĪ³1)t4. After some algebra, this equation can be put inthe more revealing (completely equivalent) form; that of a vertex compatibility condition
Ī³1, Ī³2, Ī³3, Ī³4 ā [āĻ, Ļ] such that R1(Ī³1)R2(Ī³2)R3(Ī³3)R4(Ī³4) = I, (5)
that concretely links the four folding angles and four compatible deformation gradients at thisvertex.
The solutions R1(Ī³1),R2(Ī³2),R3(Ī³3),R4(Ī³4) to vertex compatibility (5) furnish origami defor-mations of the four panels that merge to make the degree-4 vertex. These deformations have the
3See, however, [26].4These are vertices whose sector angles do not sum to 2Ļ.5As we have done, for instance, with the Ī±-sector in Fig. 2(b).
7
-
form
y(x1, x2) =
x if x ā PĪ±,R2(Ī³2)x if x ā PĪ²,R2(Ī³2)R3(Ī³3)x if x ā PĻāĪ±,R1(āĪ³1)x if x ā PĻāĪ²,
(6)
up to an overall rigid rotation and translation. Here, x := x1e1 + x2e2, and PĪ± denotes the panelwith sector angle Ī± at the vertex (likewise PĪ², . . . respectively Ī², . . . etc.), as sketched inFig. 2(b). Furthermore, deformations of the form (6) are continuous if and only if the folding anglessatisfy (5). To this point, every vertex of a RFFQM obeys the compatibility condition (6) for itscorresponding parameters6. Thus, a complete characterization of the folding angles that satisfy (5)is needed for the designs and deformations of RFFQM.
Before proceeding to the characterization, let us comment on mountain-valley assignments.Physically, when a deformed crease pattern is viewed from an orientation consistent with the un-folded state, a fold is a valley-fold if āit looks like a valleyā and a mountain-fold if āit looks likea mountainā. This has a straightforward mathematical interpretation in our formalism7. Themountain-valley assignments simply correspond to the domains of the folding angles as
(valley-fold:) Ī³i ā (0, Ļ), (mountain-fold:) Ī³i ā (āĻ, 0). (7)
Some crease pattern have degenerate folding kinematics where the pattern can be folded-in-half andfolded-in-half again8. For RFFQM, we exclude these cases by strictly focusing on the kinematics forwhich every folding angle is participating in the deformation by being either a mountain or valleyfold. This is equivalent to enforcing the non-degeneracy condition
sin Ī³1 sin Ī³2 sin Ī³3 sin Ī³4 6= 0. (8)
In this setting, we obtain a precise characterization theorem of the rigid folding kinematics at asingle vertex.
Theorem 2.1. Assume the sector angles satisfy Kawasakiās condition in (3) and that (Ī±, Ī²) 6=(Ļ/2, Ļ/2). Then the folding angles Ī³1, Ī³2, Ī³3, Ī³4 solve vertex compatibility in (5) subject to thenon-degeneracy condition (8) if and only if
Ī³1 = Ļ, Ī³2 = Ī³ĢĻ2 (Ļ;Ī±, Ī²), Ī³3 = āĻĻ, Ī³4 = ĻĪ³ĢĻ2 (Ļ;Ī±, Ī²)
for |Ļ| ā (0, Ļ) and Ļ āMV(Ī±, Ī²) :=
ā if Ī± = Ī² 6= Ļ/2+ if Ī± = Ļ ā Ī² 6= Ļ/2Ā± if Ī± 6= Ī² 6= Ļ ā Ī²
. (9)In addition, the folding angle function Ī³ĢĻ2 : [āĻ, Ļ] ā [āĻ, Ļ], Ļ ā MV(Ī±, Ī²), satisfies the explicitrelationship
Ī³ĢĻ2 (Ļ;Ī±, Ī²) := sign(
(Ļ cos Ī² ā cosĪ±)Ļ)
arccos((āĻ1 + cosĪ± cos Ī²) cosĻ + sinĪ± sin Ī²āĻ1 + cosĪ± cos Ī² + sinĪ± sin Ī² cosĻ
), (10)
6Note, the four folding angles Ī³1, Ī³2, Ī³3, Ī³4 and two sector Ī±, Ī² angles can change from vertex to vertex.7A counter-clockwise labeling of crease tangents for which the rotations and deformations are defined by these
tangents as above.8For instance, a simple checkerboard crease pattern can be folded from flat to folded flat in numerous ways.
8
-
ļæ½ = +
ļæ½ = ļæ½
! = 0 ! = ā”! = ā”/3 ! = 2ā”/3
āµļæ½
Figure 3: The rigid folding kinematic at a degree-4 vertex satisfying Kawasakiās condition. Theorigami can fold rigidly along (generically) two mountain valley assignments (red/blue as indicated).It can also be folded from flat (Ļ = 0) to completely folded flat (Ļ = Ļ) monotonically along theseassignments.
and its inverse Ī³ĢāĻ1 : [āĻ, Ļ]ā [āĻ, Ļ] satisfies the explicit relationship
Ī³ĢāĻ1 (Ļ;Ī±, Ī²) := Ī³ĢāĻ2 (Ļ;Ī±, Ļ ā Ī²), (11)
i.e., Ī³ĢāĻ1 (Ī³ĢĻ2 (Ļ;Ī±, Ī²);Ī±, Ī²) = Ļ. Alternatively, if the sector angles satisfy Kawasakiās condition and
(Ī±, Ī²) = (Ļ/2, Ļ/2), then there is no solution to (5) under the constraint (8).
We briefly comment on a few properties related to this characterization. By fixing a Ļ āMV(Ī±, Ī²),the parameterization of the folding angles in (9) leads to a continuous one parameter family oforigami deformations (6) with Ļ as the folding parameter. As indicated by the generic examplein Fig. 3, this parameterization has several universal properties: 1) Ļ = 0 corresponds to the flatstate, 2) Ļ = Ļ corresponds to the folded flat state, and 3) Ļ ā (0, Ļ) evolves the folding anglescontinuously and monotonically from flat to folded flat along a fixed mountain-valley assignment.Additionally, Ļ 7ā āĻ simply reverses the signs of all the folding angles. Finally, the solutionsĻ ā MV(Ī±, Ī²) encapsulate all the valid mountain-valley assignments at a single degree-4 vertex.These always correspond to Maekawaās well known restriction [28] on the crease assignment thatthe number of mountain folds differs by two from the number of valley folds, i.e.,
sin Ī³1 sin Ī³2 sin Ī³3 sin Ī³4 < 0. (12)
More specifically, a crease (as highlighted in Fig. 3(far-left)) can be an assignment (blue) oppositeto the other three creases (red) if and only if its adjacent sector angles sum to a value < Ļ. Thereare generically two such assignments, each indicated by a sign Ļ = Ā± associated to the folding
9
-
angles in the parameterization. Note though, there are also special cases of vertices (Ī± = Ī² 6= Ļ/2or Ī± = Ļ ā Ī² 6= Ļ/2) that have only one valid assignment and completely degenerate vertices(Ī± = Ī² = Ļ/2) with zero valid assignments. This is what gives rise to the sector angle dependenceof the set MV(Ī±, Ī²) in (9).
An early proof of this result is due to Huffman [24] using the methods of spherical trigonometry.Espousing the continuum mechanics approach, we provide an alternative proof by working directlyand algebraically with the rotation matrices in (5).
Proof. Since Ri(Ī³i)ti = ti for each i = 1, . . . , 4, a necessary condition for (5) is that
N1(Ī³1, Ī³3) := t2 Ā·(R3(Ī³3)āR1(āĪ³1)
)t4 = (cos Ī³3 ā cos Ī³1) sinĪ± sin Ī² (13)
should vanish. Under the non-degeneracy condition (8), N1(Ī³1, Ī³3) = 0 if and only if (Ī³1, Ī³3) =(Ļ,āĻĻ) for |Ļ| ā (0, Ļ) and Ļ ā {Ā±}. We assume this parameterization in the remainder as itis necessary for the characterization in the theorem. Let P2 := I ā t2 ā t2. Another necessarycondition for (5) under (8) is that
N2(Ī³2, Ļ, Ļ) := P2
(R2(Ī³2)R3(āĻĻ)āR1(āĻ)
)t4 = R2(Ī³2)P2R3(āĻĻ)t4 āP2R1(āĻ)t4 (14)
should vanish. In this formula, notice that |P2R3(āĻĻ)t4| = |P2R1(āĻ)t4| since R3(āĻĻ)t4and R1(āĻ)t4 are unit vectors and since N1(Ļ,āĻĻ) = 0. Notice also that P2R1(āĻ)t4 6= 0since |Ļ| ā (0, Ļ). Thus, P2R1(āĻ)t4 and P2R3(āĻĻ)t4 are two vectors of equal and non-zeromagnitude in the plane with normal t2. By standard trigonometric identities, we therefore obtainthat N2(Ī³2, Ļ, Ļ) = 0 if and only if the angle Ī³2 satisifies the parameterization
cos Ī³2 =P2R1(āĻ)t4 Ā·P2R3(āĻĻ)t4
|P2R1(āĻ)t4|2=
(āĻ1 + cosĪ± cos Ī²) cosĻ + sinĪ± sin Ī²āĻ1 + cosĪ± cos Ī² + sinĪ± sin Ī² cosĻ ,
sin Ī³2 =t2 Ā· (P2R3(āĻĻ)t4 ĆP2R1(āĻ)t4)
|P2R1(āĻ)t4|2=
(Ļ cosĪ±ā cos Ī²) sinĻāĻ1 + cosĪ± cos Ī² + sinĪ± sin Ī² cosĻ .
(15)
Here, the latter equalities are by explicit calculation. For non-degeneracy (8), sin Ī³2 cannot bezero. From the parameterization above, this is evidently equivalent to (Ļ cosĪ± ā cos Ī²) 6= 0 sincesinĻ 6= 0. In other words, we conclude that (Ī±, Ī²) 6= (Ļ/2, Ļ/2) and Ļ āMV(Ī±, Ī²) are necessary forthe characterization, as stated in the theorem. Under these restrictions, we obtain Ī³2 = Ī³Ģ
Ļ2 (Ļ;Ī±, Ī²)
directly from (15).Now, since N1(Ļ,āĻĻ) = 0 and N2(Ī³ĢĻ2 (Ļ;Ī±, Ī²), Ļ, Ļ) = 0, we conclude that
R1(Ļ)R2(Ī³ĢĻ2 (Ļ;Ī±, Ī²))R3(āĻĻ)t4 = t4. (16)
As R1(Ļ)R2(Ī³ĢĻ2 (Ļ;Ī±, Ī²))R3(āĻĻ) is a rotation that fixes t4, there is a unique Ī³4 ā (āĻ, Ļ] such
that R1(Ļ)R2(Ī³ĢĻ2 (Ļ;Ī±, Ī²))R3(āĻĻ)R4(Ī³4) = I. To obtained the precise parameterization of Ī³4,
observe the following: We have parameterized Ī³2 and Ī³3 in terms of Ī³1 by a systematic procedure.We can alternatively repeat the steps of this procedure to parameterize Ī³4 and Ī³1 in terms of Ī³3.In doing this, we obtain
Ī³3 = ĻĢ, Ī³1 = āĻ ĻĢ, Ī³4 = Ī³ĢĻ2 (ĻĢ; Ļ ā Ī±, Ļ ā Ī²), |ĻĢ| ā (0, Ļ), Ļ āMV(Ļ ā Ī±, Ļ ā Ī²). (17)By matching these formula to our original parameterization, we deduce that ĻĢ = āĻĻ, Ļ = Ļ andĪ³4 = Ī³Ģ
Ļ2 (āĻĻ; Ļ ā Ī±, Ļ ā Ī²) = ĻĪ³ĢĻ2 (Ļ;Ī±, Ī²), as in the characterization of the theorem. Finally,
the inverse Ī³ĢāĻ1 of Ī³ĢĻ2 can be deduced by another reparameterization, or it can simply be verified
explicitly.
10
-
2.2 Compatibility at a single panel.
Building on the characterization at a single vertex in Theorem 2.1, we derive the necessary andsufficient conditions for RFFQM about a single panel (Fig. 4). This characterization of panelcompatibility is well-known [50], and it corresponds to a consistency condition on the folding anglefunctions when ātaking a loopā about the panel in question. We first introduce the result in theform of a theorem. Then, guided by the sketch in Fig. 4(b-d), we briefly provide a derivation usingthe continuum mechanics approach we have espoused thus far.
To introduce the result, we first isolate a crease pattern surrounding a panel and enforceKawasakiās condition at all the vertices. Such crease patterns are, in the most general setting,characterized by exactly seven independent sector angles (Fig 4(b)). These angles are restricted tothe domains
Ī±a, Ī±b, Ī±c, Ī²a, Ī²b, Ī²c, Ī²d ā (0, Ļ), 2Ļ ā Ī±a ā Ī±b ā Ī±c ā (0, Ļ),(Ī±a, Ī²a), (Ī±b, Ī²b), (Ī±c, Ī²c), (2Ļ ā Ī±a ā Ī±b ā Ī±c, Ī²d) 6= (Ļ/2, Ļ/2).
(18)
The desired result then arises as a condition on the folding angle functions that are relevant to thevertices of this panel. These functions have the form
Ī³ĢĻa2a (Ļ) := Ī³ĢĻa2 (Ļ;Ī±a, Ī²a), Ī³Ģ
āĻa1a (Ļ) := Ī³Ģ
āĻa1 (Ļ;Ī±a, Ī²a), Ļa āMV(Ī±a, Ī²a),
Ī³ĢĻb2b (Ļ) := Ī³ĢĻb2 (Ļ;Ī±b, Ī²b), Ī³Ģ
āĻb1b (Ļ) := Ī³Ģ
āĻb1 (Ļ;Ī±b, Ī²b), Ļb āMV(Ī±b, Ī²b),
Ī³ĢĻc2c (Ļ) := Ī³ĢĻc2 (Ļ;Ī±c, Ī²c), Ī³Ģ
āĻc1c (Ļ) := Ī³Ģ
āĻc1 (Ļ;Ī±c, Ī²c), Ļc āMV(Ī±c, Ī²c),
Ī³ĢĻd2c (Ļ) := Ī³ĢĻd2 (Ļ;Ī±d, Ī²d), Ī³Ģ
āĻd1d (Ļ) := Ī³Ģ
āĻd1 (Ļ;Ī±d, Ī²d), Ļd āMV(Ī±d, Ī²d),
Ī±d := 2Ļ ā Ī±a ā Ī±b ā Ī±c,
(19)
for Ļ ā [āĻ, Ļ] and for the folding angle functions in Theorem 2.1. Using these definitions, weobtain the following characterization.
Theorem 2.2. Consider the crease pattern of a single panel, as depicted in Fig. 4(b). Assumeevery vertex of this pattern satisfies Kawasakiās condition with sector angles as in (18). Then thecrease pattern admits non-degenerate9 RFFQM if and only if
Ī³ĢāĻc1c ā¦ Ī³ĢĻd2d ā¦ Ī³ĢāĻb1b ā¦ Ī³ĢĻa2a = id for some (Ļa, Ļb, Ļc, Ļd) āMVabcd, (20)
where MVabcd :=MV(Ī±a, Ī²a)ĆMV(Ī±b, Ī²b)ĆMV(Ī±c, Ī²c)ĆMV(2Ļ ā Ī±a ā Ī±b ā Ī±c, Ī²d).Here, we have employed the composition of maps notation, e.g., Ī³ĢāĻb1b ā¦ Ī³ĢĻa2a (Ļ) := Ī³ĢāĻb1b (Ī³ĢĻa2a (Ļ)).
Let us comment briefly on some important aspects of this result before turning to its derivation.In the panel compatibility condition (20), the left-hand-side is a composition of four folding anglefunctions that depend on quantities at an individual vertex: These functions depend on their vertexsector angles (Ī±(Ā·), Ī²(Ā·)) in a highly non-linear way, and their structure is indicated by a sign Ļ(Ā·)corresponding to the vertex mountain-valley assignment. As a result, the overall composition is ahighly non-linear function of all the sector angles which make up the panelās crease pattern, and allits possible mountain-valley assignments. So, satisfying (20) without simplifying assumptions onthe parameters is non-trivial. This will be our task in the coming section.
9During the process of folding, the folding angles at each vertex must satisfy the constraint (8) away from theflat and folded flat state. That is, we will not be considering the degenerate cases involving folding-in-half along acrease outlining one of the sides of the panel.
11
-
āµaļæ½a
āµbļæ½b
āµcļæ½c
ļæ½d
ta1
ta2
tb2
tb1
tc1
tc2 td2
td1
(b)
(c)
(d)
Ra1(ļæ½!a)I Rc1(!c)
I
Rd1(!d)Rb1(ļæ½!b)
II
Vertex Compatibility
Panel Compatibility
(a)
Rb2(ļæ½ļæ½Ģļæ½b2b (!b))
Rb2(ļæ½ļæ½Ģļæ½b2b (!b))Rb3(ļæ½b!b) Rd2(ļæ½Ģļæ½d2d (!d))Rd3(ļæ½ļæ½d!d)
Rd2(ļæ½Ģļæ½d2d (!d))
Rc2(ļæ½ļæ½Ģļæ½c2c (!c))
Rc2(ļæ½ļæ½Ģļæ½c2c (!c))Rc3(ļæ½c!c)Ra2(ļæ½Ģ
ļæ½a2a (!a))R
a3(ļæ½ļæ½a!a)
Ra2(ļæ½Ģļæ½a2a (!a))
!aRa1(ļæ½!a)
I
Rb1(ļæ½ļæ½Ģļæ½ļæ½b1b ļæ½ ļæ½Ģļæ½a2a (!a))
Ra2(ļæ½Ģļæ½a2a (!a))
Rc2(ļæ½Ģļæ½c2c (!a))
ļæ½Ģļæ½a2aļæ½Ģļæ½ļæ½a1c
ļæ½Ģļæ½d2d
ļæ½Ģļæ½ļæ½b1b
Figure 4: Step by step schematic to derive compatibility at a single panel. (a) Isolate a panel withinan overall crease pattern. (b) Introduce the depicted local notation for the sector angles and creasetangents at the panel. (c) Directly apply the vertex characterization theorem (i.e., Theorem 2.1)to each vertex so as to be consistent with this notation. (d) The panel compatibility condition isobtained by matching the deformation gradients on vertical and horizontal adjacent panels in (c).
12
-
For now, though, we highlight another observation potentially relevant to applications: While itwill turn out that most RFFQM crease patterns are foldable in exactly one way, this is by no meansdictated by the theorem at hand. For a given crease pattern, i.e., a given collection of sector an-gles Ī±a, . . . , Ī²d, it may turn out that multiple collections of signs (Ļa, Ļb, Ļc, Ļd), (ĻĢa, ĻĢb, ĻĢc, ĻĢd), . . . āMVabcd can be used to establish the result (20). In such cases, the panelās crease pattern can befolded from flat to folded flat along multiple distinct mountain-valley assignments, each correspond-ing to the signs (Ļa, Ļb, Ļc, Ļd), (ĻĢa, ĻĢb, ĻĢc, ĻĢd), . . . achieving panel compatibility. We will discuss thismulti-stability only in passing here. We refer the interested reader to [15] for a similar observationand some development on its applications.
We now derive panel compatibility in Theorem 2.2. We first isolate a single panel and itssurrounding crease pattern (Fig. 4(a-b)). On this panel, we assign crease tangents ta1, . . . , t
a4 ā S2
to the a-vertex, tb1, . . . , tb4 ā S2 to the b-vertex, tc1, . . . , tc4 ā S2 to the c-vertex, and td1, . . . , td4 ā S2
to the d-vertex. As indicated with Fig 4(b), the crease assignments for the a and d-vertex have acounter-clockwise labeling and the assignments for the b and c-vertex have a clockwise labeling.
If the panel is to be folded as non-degenerate rigid origami, then each individual vertex must befolded as non-degenerate rigid origami. We therefore assign the deformation gradients surroundingeach vertex as in Fig. 4(c). Considering the notation, the a and d-vertex assignments are obtainedby direct application of Theorem 2.1 using the vertex deformation (6). Alternatively, since thetangents of the b and c-vertex are labeled clockwise, we simply apply the theorem but replace allthe folding angles with their minus. This has the effect of keeping a consistent folding angle domain(7) associated to mountain and valley creases.
Deriving (20) is then simply a question of consistency of the deformation gradients surroundingthe panel (equivalently, consistency of the folding angles). As the rotations Ra,...,d1,...,4(Ā·) are 2Ļ periodic,we take the primary branch for the folding angle domains without loss of generality. Then, thisconsistency is equivalent to
Ļa = Ļc, Ļd = Ļb, Ī³ĢĻa2a (Ļa) = Ī³Ģ
Ļb2b (Ļb), Ī³Ģ
Ļc2c (Ļc) = Ī³Ģ
Ļd2d (Ļd). (21)
For instance, the a-vertex has a rotation Ra1(āĻa) on the same panel for which the c-vertex hasrotation Rc1(Ļc). The two rotations must therefore be the same; thus, yielding the condition onthe angles Ļa = Ļc since t
c1 = āta1. There are three analogous consistency conditions, so this
reasoning leads to the four total conditions on the folding angle surrounding the panel in (21).Panel compatibility (20) then emerges as a requirement for these consistency conditions: Recallthat the inverse of Ī³ĢĻ2 for Ļ āMV(Ī±, Ī²) is Ī³ĢāĻ1 . We can therefore substitute the first two equationsin (21) into the latter two, apply the inverse Ī³ĢāĻb1b to the third and Ī³Ģ
āĻc1c to the fourth, and substitute
the third equation into the fourth to arrive at the requirement
Ī³ĢāĻc1c ā¦ Ī³ĢĻd2d ā¦ Ī³ĢāĻb1b ā¦ Ī³ĢĻa2a (Ļa) = Ļa. (22)
Now let us assume this requirement holds for some collection of sector angels Ī±a, . . . , Ī²d in (18),some choice of mountain-valley assignment (Ļa, . . . , Ļd) āMVabcd, and some folding parameter Ļa.It is then possible to solve (21). Further, by substituting this solution into the deformation gradientsin Fig. 4(c), we obtain a rigid folding of the panelās crease pattern with a folding angle Ļa at theta1 crease, as indicated schematically with Fig. 4(d). Note though, we are interested in RFFQM.So not only should (22) hold for some Ļa, it should hold for all Ļa ā [āĻ, Ļ]. In other words,this composition of maps should be the identity, i.e., yield Ļa when evaluated at Ļa regardless of
13
-
the choice of this angle. Consequently, we obtain panel compatibility (20) as the necessary andsufficient condition for non-degenerate RFFQM about this panel.
3 Rigidity of the global kinematics
A classification of all designs and deformations of RFFQM emerges from a rigidity result on the so-lutions to panel compatibility. The key idea is to recognize that the folding angle functions and theircompositionsāhowever complicated is their sector angle and mountain-valley dependenceāsharean underlying group structure. This group structure then reduces the difficult problem of character-izing these compositions to a much simpler problem of characterizing a corresponding multiplicationof parameters. From this reduction, we are able to obtain all solutions to panel compatibility. Thestructure of these solutions, in turn, yields a concrete algorithm for the designs and deformationsof all RFFQM.
3.1 An Abelian group structure to folding angle functions
In an effort to have a general classification of functions which ālook likeā the folding angle functionsdefined in Theorem 2.1, we study the functions
g(Ļ;Āµ) := ā sign(ĀµĻ)
arccos((Āµ2 + 1) cosĻ + (Āµ2 ā 1)
(Āµ2 + 1) + (Āµ2 ā 1) cosĻ), Āµ ā R \ {0}. (23)
We first observe that any folding angle function can be written in this form for the suitable choiceof Āµ ā R \ {0}.Proposition 3.1. Let Ī±, Ī² ā (0, Ļ), (Ī±, Ī²) 6= (Ļ/2, Ļ/2), and Ļ āMV(Ī±, Ī²). We have
Ī³ĢĻ2 (Ļ;Ī±, Ī²) = g(Ļ; āĻ1+cosĪ± cosĪ²+sinĪ± sinĪ²
cosĪ²āĻ cosĪ±
), Ī³ĢāĻ1 (Ļ;Ī±, Ī²) = g
(Ļ; Ļ1ācosĪ± cosĪ²+sinĪ± sinĪ²ā cosĪ²+Ļ cosĪ±
). (24)
Proof. We only need to prove the result for Ī³ĢĻ2 since Ī³ĢāĻ1 is obtained from the identity in (11). In
this direction, we notice that cosĪ± ā Ļ cos Ī² 6= 0 since Ļ ā MV(Ī±, Ī²). Therefore, we may definethe parameters Ī· := āĻ1+cosĪ± cosĪ²
cosĪ²āĻ cosĪ± and Ī“ :=sinĪ± sinĪ²
cosĪ²āĻ cosĪ± , and introduce the parameterization
h(Ļ; Ī·, Ī“) := ā sign((Ī· + Ī“)Ļ
)arccos
(Ī· cosĻ + Ī“Ī· + Ī“ cosĻ
). (25)
We claim that Ī³ĢĻ2 (Ļ;Ī±, Ī²) = h(Ļ; Ī·, Ī“). Indeed, it is clear that the arccos(Ā·) part of this function isthe same as that in Ī³ĢĻ2 ; so we just need to show that sign(Ī·+ Ī“) = sign(cosĪ±āĻ cos Ī²). We observethat
sign(Ī· + Ī“) = sign(āĻ1 + cosĪ± cos Ī² + sinĪ± sin Ī²
cos Ī² ā Ļ cosĪ±)
= sign( āĻ1
cos Ī² ā Ļ cosĪ±), (26)
which yields the desired result after rearranging the latter. Thus, we have Ī³ĢĻ2 (Ļ;Ī±, Ī²) = h(Ļ; Ī·, Ī“).Next, we observe that Ī·2 ā Ī“2 = 1 by a direct calculation. We therefore define Āµ := Ī· + Ī“, and
it follows that Ī· = Āµ2+12Āµ
and Ī“ = Āµ2ā12Āµ
using the aforementioned identity. By expanding terms, wehave
Āµ =āĻ1 + cosĪ± cos Ī² + sinĪ± sin Ī²
cos Ī² ā Ļ cosĪ± . (27)
14
-
Finally, in combining all of these result, we obtain the chain of equalities
Ī³ĢĻ2 (Ļ;Ī±, Ī²) = h(Ļ;Āµ2 + 1
2Āµ,Āµ2 ā 1
2Āµ
)= g(Ļ;Āµ). (28)
This completes the proof.
Next, for definiteness, we have:
Proposition 3.2. The function g(Ļ;Āµ) is a well-defined mapping from [āĻ, Ļ] to [āĻ, Ļ] for anyĀµ ā R \ {0}.
Proof. We simply have to verify that the argument in the arccos(Ā·) is well-defined (i.e., not dividingby zero) and has magnitude ā¤ 1. In this direction, we observe that(
(Āµ2 + 1) + (Āµ2 ā 1) cosĻ)2 ā ((Āµ2 + 1) cosĻ + (Āµ2 ā 1))2
= ((Āµ2 + 1)2 ā (Āµ2 ā 1)2)(1ā cos2 Ļ) = 4Āµ2(1ā cos2 Ļ) ā„ 0.(29)
This establishes that the argument has magnitude ā¤ 1, or the argument is 0/0. The latter is,however, not a possibility. Notice that the inequality above is strict unless Ļ = 0 since Āµ 6= 0.However, also notice that (Āµ2 + 1) + (Āµ2 ā 1) cosĻ = 2Āµ2 6= 0 when Ļ = 0. Thus, we never divideby zero in this parameterization.
In light of the above definiteness of the functions in (23), we can introduce without any ambiguitythe collection of all such functions, i.e.,
G :={g : [āĻ, Ļ]ā [āĻ, Ļ]
ā£ā£ g = g(Ļ;Āµ) as above, Āµ ā R \ {0}}. (30)This collection has the following remarkable property:
Lemma 3.1. G in (30) is an abelian group under the composition of maps product. Specifically, ithas the properties:
ā¢ The functions g1 := g(Ļ;Āµ1), g2 := g(Ļ;Āµ2) ā G satisfy g1 = g2 if and only if Āµ1 = Āµ2.
ā¢ The identity map is id = g(Ļ; 1);
ā¢ For any g1 := g(Ļ;Āµ1) and g2(Ļ;Āµ2) ā G, the group product satisfies
g1 ā¦ g2 = g2 ā¦ g1 = g(Ļ, Āµ1Āµ2) ā G; (31)
ā¢ The inverse of g0 := g(Ļ, Āµ0) ā G is given by gā10 = g(Ļ, Āµā10 ) ā G.
Proof. Composition of functions is always associative, so establishing the four properties aboveverifies that G is an abelian group and that its parameterization by Āµ ā R \ {0} has no repeatedelements. For the first property, we notice that g1 = g2 only if
(Āµ21 + 1) cosĻ + (Āµ21 ā 1)
(Āµ21 + 1) cosĻ + (Āµ21 ā 1)
=(Āµ22 + 1) cosĻ + (Āµ
22 ā 1)
(Āµ22 + 1) cosĻ + (Āµ22 ā 1)
ā Ļ ā [āĻ, Ļ] (32)
15
-
since arccos(Ā·) is monotonically decreasing function of [ā1, 1]. By rearranging terms, this equationbecomes the condition (
Āµ22 ā Āµ21)
cosĻ + Āµ21 ā Āµ22 = 0, ā Ļ ā [āĻ, Ļ]. (33)
Clearly then, we require Āµ1 = Ā±Āµ2 since cosĻ is a smooth non-constant function on this interval.Taking into account the sign(Ā·) part of g(Ļ;Āµ), we obtain Āµ1 = Āµ2 as desired. For the secondproperty, we notice g(Ļ; 1) = sign(Ļ) arccos(cos(Ļ)) = sign(Ļ)|Ļ| = Ļ. For the third, we observethat (
g1 ā¦ g2)(Ļ) = g(g(Ļ;Āµ2);Āµ1)
= sign(Āµ1Āµ2Ļ) arccos
((Āµ21 + 1)( (Āµ22+1) cosĻ+(Āµ22ā1)(Āµ22+1)+(Āµ22ā1) cosĻ)+ (Āµ21 ā 1)(Āµ21 + 1) + (Āµ
21 ā 1)
((Āµ22+1) cosĻ+(Āµ
22ā1)
(Āµ22+1)+(Āµ22ā1) cosĻ
))
= sign(Āµ1Āµ2Ļ) arccos
((Āµ21+1)
((Āµ22+1) cosĻ+(Āµ
22ā1))+((Āµ22+1)+(Āµ
22ā1) cosĻ
)(Āµ21ā1)
(Āµ21+1)((Āµ22+1)+(Āµ
22ā1) cosĻ
)+(Āµ21ā1)
((Āµ22+1) cosĻ+(Āµ
22ā1)))
= sign(Āµ1Āµ2Ļ) arccos
((Āµ21Āµ
22 + 1) cos(Ļ) + (Āµ
21Āµ
22 ā 1)
(Āµ21Āµ22 + 1) + (Āµ
21Āµ
22 ā 1) cos(Ļ)
),
(34)
where the last identity is obtained simply by expanding out the terms in the second to last identity.Notice this latter identity is nothing but g(Ļ;Āµ1Āµ2). Thus, g1ā¦g2 = g(Ļ;Āµ1Āµ2) = g(Ļ;Āµ2Āµ1) = g2ā¦g1,as desired. Finally, the last property is a direct consequence of the other two.
3.2 A design theorem for rigidly and flat-foldable origami.
We now return to the study of panel compatibility in (20). Our brief digression regarding theAbelian groupāparticularly, our observation in (24) concerning the folding angle functionsāhasproven useful in the following sense: It beckons the introduction of the sector angle and mountain-valley dependent parameters
ĀµĻ2 (Ī±, Ī²) :=āĻ1 + cosĪ± cos Ī² + sinĪ± sin Ī²
cos Ī² ā Ļ cosĪ± , ĀµāĻ1 (Ī±, Ī²) := Āµ
āĻ2 (Ī±, Ļ ā Ī²), Ļ āMV(Ī±, Ī²). (35)
In keeping with the origami literature [17, 35, 50, 52], we call these parameters the fold anglemultipliers. We should point out that, although these parameters are known to this literature,here we make the direct connection to the kinematics (i.e., (24) to (35)) through an Abelian grouptransparent and rigorous.
A key step in our characterization of panel compatibility is relating the composition of foldingangle functions directly to a product of fold angle multipliers. In this direction, we introduce thevertex-dependent fold angle multipliers
ĀµĻ2a := ĀµĻ2 (Ī±a, Ī²a), Āµ
āĻ1a := Āµ
āĻ1 (Ī±a, Ī²a), Ļ āMV(Ī±a, Ī²a),
ĀµĻ2b := ĀµĻ2 (Ī±b, Ī²b), Āµ
āĻ1b := Āµ
āĻ1 (Ī±b, Ī²b), Ļ āMV(Ī±b, Ī²b),
ĀµĻ2c := ĀµĻ2 (Ī±c, Ī²c), Āµ
āĻ1c := Āµ
āĻ1 (Ī±c, Ī²c), Ļ āMV(Ī±c, Ī²c),
ĀµĻ2d := ĀµĻ2 (Ī±d, Ī²d), Āµ
āĻ1d := Āµ
āĻ1 (Ī±d, Ī²d), Ļ āMV(Ī±d, Ī²d),
Ī±d := 2Ļ ā Ī±a ā Ī±b ā Ī±c,
(36)
16
-
for admissible collections of sector angles Ī±a, . . . , Ī²d as in (18). The connection between panelcompatibility and the fold angle multipliers can now be made precise.
Lemma 3.2. For any collection of sector angle Ī±a, . . . , Ī²d in (18) and any mountain-valley as-signment (Ļa, Ļb, Ļc, Ļd) āMVabcd (recall Theorem 2.2 for the definition), the following equivalenceholds:
Ī³ĢāĻc1c ā¦ Ī³ĢĻd2d ā¦ Ī³ĢāĻb1b ā¦ Ī³ĢĻa2a = id ā ĀµāĻc1c ĀµĻd2dĀµāĻb1b ĀµĻa2a = 1. (37)Proof. Since Ī³ĢāĻc1c ā¦ Ī³ĢĻd2d ā¦ Ī³ĢāĻb1b ā¦ Ī³ĢĻa2a is the composition of functions that belong to the group G(Proposition 3.1 and Lemma 3.1), it follows that Ī³ĢāĻc1c ā¦ Ī³ĢĻd2d ā¦ Ī³ĢāĻb1b ā¦ Ī³ĢĻa2a ā G. Consequently, thereexists a Āµabcd ā R \ {0} such that
Ī³ĢāĻc1c ā¦ Ī³ĢĻd2d ā¦ Ī³ĢāĻb1b ā¦ Ī³ĢĻa2a (Ļ) = g(Ļ;Āµabcd), Ļ ā [āĻ, Ļ]. (38)
By combining the formula (24) with the product rule (31), we obtain Āµabcd = ĀµāĻc1c Āµ
Ļd2dĀµāĻb1b Āµ
Ļa2a . Since
g(Ļ;Āµ) = id if and only if Āµ = 1 (the first two properties in Lemma 3.1), we conclude (37) asdesired.
With the equivalence in (37) now established, we are prepared to develop the main character-ization theorem for RFFQM. In this characterization, we consider an arbitrary crease pattern inFig. 4(b) under Kawasakiās condition and treat quantities at three of the four vertices, i.e., sectorangles and mountain-valley assignments, as given under the hypotheses
(Design parameters:)
Ī±a, Ī±b, Ī±c ā (0, Ļ), 2Ļ ā Ī±a ā Ī±b ā Ī±c ā (0, Ļ), Ī²a, Ī²b, Ī²c ā (0, Ļ),(Ī±a, Ī²a), (Ī±b, Ī²b), (Ī±c, Ī²c) 6= (Ļ/2, Ļ/2),Ļa āMV(Ī±a, Ī²a), Ļb āMV(Ī±b, Ī²b), Ļc āMV(Ī±c, Ī²c).
(39)We then ask the fundamental design questions: Can the final sector angle Ī²d ā (0, Ļ) and mountain-valley assignment Ļd āMV(2Ļ ā Ī±a ā Ī±b ā Ī±c, Ī²d) be chosen to yield a crease pattern that admitsRFFQM. If so, how rigid or flexible is the design criterion? To this end, we obtain an explicitrigidity theorem for design.
Theorem 3.1. Consider a crease pattern as in Fig 4(b) under Kawasakiās condition, and let thesector angles Ī±a, . . . , Ī²c and mountain-valley assignments Ļa, Ļb, Ļc at the a-b-c vertices be givenunder the hypotheses in (39). The characterization of the d-vertex is based on the evaluation of
Āµabc := ĀµāĻa1a Āµ
Ļb2bĀµ
Ļc2c . (40)
ā¢ If |Āµabc| = 1, then there is no pair (Ī²d, Ļd) ā (0, Ļ) ĆMV(2Ļ ā Ī±a ā Ī±b ā Ī±c, Ī²d) yieldinga crease pattern that can be folded as non-degenerate10 RFFQM along the mountain-valleyassignments Ļa, Ļb, Ļc at the a-b-c vertices.
ā¢ Otherwise, |Āµabc| 6= 1 and there is a unique pair (Ī²d, Ļd) ā (0, Ļ)ĆMV(2ĻāĪ±aāĪ±bāĪ±c, Ī²d)yielding a crease pattern that can be folded as non-degenerate RFFQM along the mountain-valley assignments Ļa, Ļb, Ļc at the a-b-c vertices. This pair satisfies the explicit relationship
Ļd1 = ā sign(Āµ2abc ā 1
), Ī²d = arccos
(Ļd
2Āµabc ā (Āµ2abc + 1) cos(Ī±a + Ī±b + Ī±c)2Āµabc cos(Ī±a + Ī±b + Ī±c)ā (Āµ2abc + 1)
). (41)
10See the footnote on this topic in Theorem 2.2.
17
-
A couple of points with this theorem: The case |Āµabc| = 1 is highly non-generic. In fact, onecan treat Ī±a, Ī±b, Ī±c, Ī²b, Ī²c, Ļb, Ļc as given (under the hypotheses (39)) and use an argument similarto the proof of this theorem to find that there are at most two pairs (Ī²a, Ļa) ā (0, Ļ)ĆMV(Ī±a, Ī²a)which give this degeneracy. As a result, a generic set of design parameters (39) at the a-b-c verticeswill admit RFFQM by the prescription in (41). A corollary to this effect is stated in the nextsubsection. A second point is that this result, as stated, is very much mountain-valley assignmentdependent: We are testing whether a crease pattern can deform as RFFQM along specific mountain-valley assignments Ļa, Ļb, Ļc at the a-b-c vertices. Thus, when |Āµabc| = 1, one should avoid theinterpretation that the sector angles Ī±a, . . . , Ī²c cannot admit a RFFQM crease pattern; only thatthey cannot admit one whose folding corresponds to mountain-valley assignments Ļa, Ļb, Ļc at theaforementioned vertices.
There are other results in similar spirit to Theorem 3.1 in the literature. Building on ideasfor Tachi [50, 52], Lang and Howell [36] made the astute observation that one can solve panelcompatibility generically by treating the folding angles as the independent variables rather thanthe sector angles. Nevertheless, we find the sector angle to be the more natural design variables,especially for developing optimization schemes to tackle the inverse problem (see the discussion inSection 4.2). We should also point out the results of Izmestiev [27] along these lines. He classified allāKokotsakis polyhedraāāsingle panel crease patterns where no assumption are made on the sectoranglesāby polynomializing general constraints of the type in (20) and studying their zeros in thecomplex plane. While certainly a mathematical triumph, this characterization is unlike Theorem3.1 in that it does not lend itself naturally to a design principle for engineering. For instance, mostof the solutions, beyond the variety we discuss for RFFQM, involve delicate couplings between allthe vertices of the panel, making it hard to elucidate a procedure to go from a single panel to anoverall crease pattern. Some efforts to address these issues can be found in [23], but these effortsare restricted to special cases.
3.3 Derivation of the design theorem
In this section, we prove Theorem 3.1. We also state a corollary that justifies that the subcase|Āµabc| = 1 in this theorem is non-generic.Proof of Theorem 3.1. From Theorem 2.2 and Lemma 3.2, characterizing RFFQM patterns, asstated in the theorem, is equivalent to characterizing the pairs (Ī²d, Ļd) ā (0, Ļ) ĆMV(2Ļ ā Ī±a āĪ±b ā Ī±c, Ī²d) such that (ĀµāĻc1c )(ĀµĻd2d)(ĀµāĻb1b )(ĀµĻa2a) = 1. Since Ī³ĢāĻa1a is the inverse of Ī³ĢĻa2a , we observe thatĀµāĻa1a = (Āµ
Ļa2a)ā1 due to the inverse property of group elements in Lemma 3.1. Analogous observations
hold for all the other vertices. We can therefore rearrange the equation (ĀµāĻc1c )(ĀµĻd2d)(Āµ
āĻb1b )(Āµ
Ļa2a) = 1
as ĀµĻd2d = Āµabc for Āµabc given in (40). By Proposition 3.3 below, we observe that
ĀµĻd2d = Āµabc ā(ĀµĻd2d)
2 + 1
2ĀµĻd2d=Āµ2abc + 1
2Āµabc, sign
((ĀµĻd2d)2 ā 12ĀµĻd2d
)= sign
(Āµ2abc ā 12Āµabc
)(42)
since Āµabc ā R \ {0} by assumption of the given parameters.We focus on solving the latter equivalence in terms of (Ī²d, Ļd) ā (0, Ļ)ĆMV(2ĻāĪ±aāĪ±bāĪ±c, Ī²d).
By a direct calculation, we obtain
(ĀµĻd2d)2 + 1
2ĀµĻd2d=āĻd1 + cos(Ī±a + Ī±b + Ī±c) cos Ī²d
cos Ī²d ā Ļd cos(Ī±a + Ī±b + Ī±c). (43)
18
-
Consequently, we can rearrange the first equation in the latter equivalence in (42) as(2Āµabc cos(Ī±a + Ī±b + Ī±c)ā (Āµ2abc + 1)
)cos Ī²d = Ļd
(2Āµabc ā (Āµ2abc + 1) cos(Ī±a + Ī±b + Ī±c)
). (44)
Notice that(2Āµabc cos(Ī±a + Ī±b + Ī±c)ā (Āµ2abc + 1)
)2 ā (2Āµabc ā (Āµ2abc + 1) cos(Ī±a + Ī±b + Ī±c))2= (Āµ2abc ā 1)2 sin(Ī±a + Ī±b + Ī±c)2 ā„ 0 and = 0 if and only if |Āµabc| = 1,
(45)
since Ī±a + Ī±b + Ī±c ā (Ļ, 2Ļ), i.e., since sin(Ī±a + Ī±b + Ī±c) 6= 0. Thus, if |Āµabc| = 1, we divide throughby 2Āµabc cos(Ī±a + Ī±b + Ī±c) ā (Āµ2abc + 1) 6= 0 in (44) to deduce that cos Ī²d is required to equal ā1or 1. However, this is not allowed since we demand that Ī²d ā (0, Ļ). This shows that there are nosolutions if Āµabc ā {Ā±1}, as asserted in the theorem. Alternatively, if |Āµabc| 6= 1, we divide throughby 2Āµabc cos(Ī±a + Ī±b + Ī±c) ā (Āµ2abc + 1) 6= 0 in (44), and obtain that the right-hand side, after thedivision, has magnitude strictly less than 1. Therefore, Ī²d ā (0, Ļ) must have the parametrizationin (41) to solve (44) in this case.
In enforcing this parameterization, it remains only to solve the sign condition in (42) for Ļd āMV(2Ļ ā Ī±a ā Ī±b ā Ī±c, Ī²d). To this end, we first claim that any Ļd ā {Ā±} which solves the signcondition in (42) will correspond to the set Ļd āMV(2ĻāĪ±aāĪ±bāĪ±c, Ī²d). Assuming the opposite,we have cos Ī²d ā Ļd cos(Ī±a + Ī±b + Ī±c) = 0. Yet, the left-hand side is parameterized as
cos Ī²d ā Ļd cos(Ī±a + Ī±b + Ī±c) = Ļd( 2Āµabc(1ā cos2(Ī±a + Ī±b + Ī±c))
2Āµabc cos(Ī±a + Ī±b + Ī±c)ā (Āµ2abc + 1)), (46)
and neither Āµabc or 1 ā cos2(Ī±a + Ī±b + Ī±c)) are zero. This is the desired contradiction; so wesimply have to parameterize the sign condition in (42) to complete the proof. Now, by an explicitcalculation, this condition is equivalent to
sign( ā sin(Ī±a + Ī±b + Ī±c) sin Ī²d
cos Ī²d ā Ļd cos(Ī±a + Ī±b + Ī±c))
= sign(Āµ2abc ā 1
2Āµabc
)(47)
The numerator on the left side ā sin(Ī±a + Ī±b + Ī±c) sin Ī²d is strictly positive since Ī²d ā (0, Ļ) andĪ±a + Ī±b + Ī±c ā (Ļ, 2Ļ); so it can be discarded. We therefore find that
sign( ā sin(Ī±a + Ī±b + Ī±c) sin Ī²d
cos Ī²d ā Ļd cos(Ī±a + Ī±b + Ī±c))
= sign(
cos Ī²d ā Ļd cos(Ī±a + Ī±b + Ī±c))
= Ļd sign( 1ā cos(Ī±a + Ī±b + Ī±c)2
cos(Ī±a + Ī±b + Ī±c)ā (Āµ2abc+1)
2Āµabc
) (48)by routine manipulations and (46). Finally, notice that cos(Ī±a +Ī±b +Ī±c)
2 < 1 since Ī±a +Ī±b +Ī±c ā(Ļ, 2Ļ) and
ā£ā£ā£ (Āµ2abc+1)2Āµabc ā£ā£ā£ > 1 since |Āµabc| 6= 1. Thus, this result can be manipulated further; specifically,we have the identity
Ļd sign( 1ā cos(Ī±a + Ī±b + Ī±c)2
cos(Ī±a + Ī±b + Ī±c)ā (Āµ2abc+1)
2Āµabc
)= āĻd sign
(Āµ2abc + 12Āµabc
). (49)
19
-
As a result of all of the manipulations in (48) and (49), the identity in (47) demands that
Ļd = ā sign(Āµ2abc + 1
2Āµabc
)sign
(Āµ2abc ā 12Āµabc
)= ā sign(Āµ2abc ā 1). (50)
This is the desired result. We have solved the equation in (42) for |Āµabc| 6= 1 by the parameterizationsin (41), and these parameterizations are clearly necessary.
Proposition 3.3. Let ĀµĢ, ĀµĢ ā R \ {0}. Then ĀµĢ = ĀµĢ if and only if
ĀµĢ2 + 1
2ĀµĢ=ĀµĢ2 + 1
2ĀµĢ, sign
( ĀµĢ2 ā 12ĀµĢ
)= sign
( ĀµĢ2 ā 12ĀµĢ
)(51)
Proof. Clearly, ĀµĢ = ĀµĢ ā R\{0} implies (51). For the other direction, we first claim that (51) implies
ĀµĢ2 ā 12ĀµĢ
=ĀµĢ2 ā 1
2ĀµĢ. (52)
Indeed, as a direct consequence of the first equality and some algebraic manipulation, we have thechain of equalities ( ĀµĢ2 ā 1
2ĀµĢ
)2=( ĀµĢ2 + 1
2ĀµĢ
)2ā 1 =
( ĀµĢ2 + 12ĀµĢ
)2ā 1 =
( ĀµĢ2 ā 12ĀµĢ
)2. (53)
Here, the first and last equality are simply by rearrangement of terms and the second uses the firstidentity in (51). Thus, with the second identity in (51), we obtain (52). We therefore have
ĀµĢ
2=ĀµĢ2 + 1
2ĀµĢ+ĀµĢ2 ā 1
2ĀµĢ=ĀµĢ2 + 1
2ĀµĢ+ĀµĢ2 ā 1
2ĀµĢ=ĀµĢ
2(54)
using the first identity in (51) and the identity in (52). This completes the proof.
Corollary 3.1. Let Ī±a, Ī±b, Ī±c ā (0, Ļ) with 2Ļ ā Ī±a ā Ī±b ā Ī±c ā (0, Ļ); let Ī²b, Ī²c ā (0, Ļ); letĻb āMV(Ī±b, Ī²b), Ļc āMV(Ī±c, Ī²c); define Āµbc := ĀµĻb2bĀµĻc2c .ā¢ If |Āµbc| = 1, then there are no pairs (Ī²a, Ļa) ā (0, Ļ)ĆMV(Ī±a, Ī²a) such that |Āµabc| = 1.
ā¢ Otherwise, |Āµbc| 6= 1 and there are exactly two pairs (Ī²Ā±a , Ļa) ā (0, Ļ)ĆMV(Ī±a, Ī²Ā±a ) such that|Āµabc| = 1. These pairs satisfy
Ļa1 = ā sign(Āµ2bc ā 1
), Ī²Ā±a = arccos
(ĻaĀ±2Āµbc ā (Āµ2bc + 1) cosĪ±aĀ±2Āµbc cosĪ±a ā (Āµ2bc + 1)
). (55)
4 Applications
The design theorem for compatibility at a single panel (Theorem 3.1) lends itself naturally toengineering design principles for RFFQM. Particularly, we show how these results enable a marchingalgorithm for explicitly and efficiently exploring of the configuration space of all RFFQM. We thendiscuss how this algorithm can be incorporated into an optimization scheme to address the inverseproblem:
20
-
ā¢ Can we achieve a targeted three dimensional surface by the deployment of a RFFQM creasepattern?
Here we formulate the inverse problem and give some examples, mainly as a means to demonstratethe utility of our theoretical results. In a forthcoming paper [12], we address the inverse problem ina comprehensive manner, with a view towards broad engineering appeal and potential applications.
4.1 A marching algorithm
We introduce the marching algorithm by following the schematic in Fig. 5. Recall that the designtheorem (Theorem 3.1) essentially states that, if we know the sector angles and mountain-valleyassignments at three of four vertices of a panel, then for RFFQM of the panelās crease pattern:1) there is a unique solution for the fourth vertex or 2) there is no solution. With this in mind,suppose we are interested in a RFFQM crease pattern with N rows of panels and M columns ofpanels, so that the total number of panels for the quad-mesh crease pattern is N ĆM . We thenhave a design problem for a RFFQM crease pattern involving N Ć M applications of Theorem3.1, which results in marching procedure to obtain the crease pattern and kinematics as a functionof certain boundary sector angles and mountain-valley assignments. Note, Lang and Howell [36]derived a formally similar marching algorithm, except that it involves treating folding angles, notsector angles, as the independent variables. The advantages of a sector angle parameterization ofmarching will become apparent in our discussion of the inverse problem (Section 4.2).
A couple of additional points of emphasis are:
ā¢ The marching algorithm for the crease pattern and its kinematics is computationally efficient.Precisely, the algorithm computes the design and kinematics of an N ĆM RFFQM creasepattern with a total computational time of
Computational Time ā¼ O(N)ĆO(M)ĆO(# of computed deformation). (56)
Recall that each RFFQM has a one parameter family of origami deformations from flat tofolded flat. So the ā# of computed deformationsā above is simply the discretization of thisone parameter family.
ā¢ The marching algorithm for the crease pattern and its kinematics is also completely explicit.In particular, every quantity of interest is obtained by a direct formula, not an optimization.We emphasize this point below by explicitly enumerating all the formulas that are necessaryand sufficient for the marching procedures (i.e., with (57)-(66) for the crease pattern and(70)-(80) for the kinematics). No detail, large or small, is suppressed. As such, these formulacan be directly implemented into any standard computational solver to obtain the designsand deformations of RFFQM.
To explain the procedure, we will always suppose the scheme below marches from panel topanel using the ordering (i, j) = (1, 1), . . . , (1,M), (2, 1), . . . , (N,M), consistent with the sketch inFig. 5(c).
Marching to obtain the crease pattern. We populate the local sector angles, mountain-valley assignments and side lengths for this design problem. The notation here follows Fig. 5(a).
21
-
(c) (d)
(e)
āµ11aļæ½11a
āµi1aļæ½i1a
ļæ½1ja āµ1ja
ļæ½11a
ļæ½i1a
ļæ½1ja
`i1ab
`11ab
`11ac`1jac
(a) (b)
āµija
āµijb
āµijcļæ½ija
ļæ½ijc
ļæ½ijbļæ½ijd
ļæ½ija
ļæ½ijb
ļæ½ijc
ļæ½ijd
Ri(jļæ½1)!
t1 t1
t2t2
R(iļæ½1)j!
`ijac
`ijab
`ijbd
`ijcd
Ri(jļæ½1)! R2(ļæ½ļæ½ij2!) R(iļæ½1)j! R1(ļæ½ij1!)
! = 0 ! = ā”
xija
xijb
xijc
xijd
xija
xijb
xijc
xijd
Figure 5: A diagram describing the marching procedure. (a) Notation for the geometry of the creasepattern at each panel. (b) Notation for the deformation gradients at each panel. (c). The inputdata for the marching algorithm used to obtain the crease pattern (d) and its kinematics (e).
22
-
We consider the (i, j)-panel for i, j > 1, and let us assume that the āmarchingā (by our to bedetermined scheme) has succeeded at every step prior to this panel. Then, most of the sector angles,mountain-valley assignments and side lengths at the (i, j)-panel are informed by the previouslydetermined panels. Specifically, the sector angles at this panel satisfy
Ī±ija = Ī²i(jā1)c , Ī²
ija = Ī±
i(jā1)c , Ī±
ijb = Ī²
i(jā1)d , Ī²
ijb = 2Ļ ā Ī±i(jā1)a ā Ī±
i(jā1)b ā Ī±i(jā1)c ,
Ī±ijc = Ļ ā Ī²(iā1)jd , Ī²ijc = āĻ + Ī±(iā1)ja + Ī±(iā1)jb + Ī±
(iā1)jc
(57)
due to Kawasakiās conditions, the mountain-valley assignments satisfy
Ļija = Ļi(jā1)c , Ļ
ijb = Ļ
i(jā1)d , Ļ
ijc = Ļ
(iā1)jd , (58)
and the side length satisfy
`ijab = `i(jā1)cd , `
ijac = `
(iā1)jbd . (59)
In contrast, the sector angle Ī²ijd , mountain-valley assignment Ļijd , and the side lengths `
ijbd and `
ijcd
are not determined from the panelās neighbors; instead, they are constrained by foldability andconsistency considerations.
Particularly, the pair (Ī²ijd , Ļijd ) at the panel is evaluated based on a direct application of Theorem
3.1 to enforce a RFFQM crease pattern. Explicitly, we check that
(first check:) 2Ļ ā Ī±ija ā Ī±ijb ā Ī±ijc ā (0, Ļ), (60)
as this is required for the interior angles of the (i, j)-panel sum to 2Ļ. If successful, we then proceedto define the folding angle multiplier
Āµijabc := ĀµāĻija1 (Ī±
ija , Ī²
ija )Āµ
Ļijb2 (Ī±
ijb , Ī²
ijb )Āµ
Ļijc2 (Ī±
ijc , Ī²
ijc ), (61)
and check that(second check:) |Āµijabc| 6= 1. (62)
Assuming success on the latter, we then determine the pair (Ī²ijd , Ļijd ) ā (0, Ļ)ĆMV(Ī±ijd , 2ĻāĪ±ija ā
Ī±ijb ā Ī±ijc ) uniquely and explicitly by the design formulas
Ļijd 1 = ā sign((Āµijabc)
2 ā 1), Ī²ijd = arccos
(Ļijd
2Āµijabc ā ((Āµijabc)2 + 1) cos(Ī±ija + Ī±ijb + Ī±ijc )2Āµijabc cos(Ī±
ija + Ī±
ijb + Ī±
ijc )ā ((Āµijabc)2 + 1)
). (63)
Finally, we turn our attention to the side lengths `ijbd and `ijcd. Since all the interior angles of the (i, j)-
panel are known, these angles, combined with the left and bottom side lengths in (59), determine`ijbd and `
ijcd. Precisely,(
`ijcd`ijbd
)= L(Ī±ija , Ī±
ijb , Ī±
ijc )
(`ijab`ijac
), L(Ī±a, Ī±b, Ī±c) :=
(ā sinĪ±b
sin(Ī±a+Ī±b+Ī±c)sin(Ī±a+Ī±b)
sin(Ī±a+Ī±b+Ī±c)sin(Ī±a+Ī±c)
sin(Ī±a+Ī±b+Ī±c)ā sinĪ±c
sin(Ī±a+Ī±b+Ī±c)
). (64)
As these are āside lengthsā, this calculation is of course subject to the constraints
(final check:) `ijcd > 0, `ijbd > 0. (65)
23
-
In total, the collective calculations in (60)-(65) reveal an explicit procedure to obtain the uniquecrease pattern at the (i, j)-panel that admits RFFQM for the pre-determined sector angles (57),mountain-valley assignments (58) and side lengths (59).
Clearly then, with the outline above, we are building the core structure of a marching algorithmto obtain an overall N Ć M RFFQM crease pattern. There are, however, some finer points toaddress: The first point is the initial input. Notice that, for i or j = 1, some of quantities in (57),(58) and (59) cannot be prescribed by their neighboring panels (as there is no such correspondingneighboring panel). Instead, we treat these parameters, i.e.,
(input data:)
(Ī±11a , Ī²11a ), . . . , (Ī±
(N+1)1a , Ī²
(N+1)1a ) ā (0, Ļ)2 and 6= (Ļ/2, Ļ/2),
(Ī±12a , Ī²12a ), . . . , (Ī±
1(M+1)a , Ī²
1(M+1)a ) ā (0, Ļ)2 and 6= (Ļ/2, Ļ/2),
Ļ11a āMV(Ī±11a , Ī²11a ), . . . , Ļ(N+1)1a āMV(Ī±(N+1)1a , Ī²(N+1)1a ),Ļ12a āMV(Ī±12a , Ī²12a ), . . . , Ļ1(M+1)a āMV(Ī±1(M+1)a , Ī²1(M+1)a ),`11ab, . . . , `
N1ab > 0,
`11ac, . . . , `1Mac > 0,
(66)
as input data to the marching algorithm. Then, by applying all consistency conditions with theseparameters (e.g., Ī±i1b = Ī±
(i+1)1a , . . .), we can initialize the procedure in (57)-(65) at the (1, 1)-panel
and march without any inconsistency. The general physical picture is sketched in Fig. 5(c). Theinput data in (66) serves to parameterize the left and bottom boundaries of the crease pattern.By marching, we simply discover the overall RFFQM crease pattern consistent with this data(Fig. 5(d)).
To the last point, we can now properly interpret the checks (60), (62) and (65). Given the inputparameters (66) on the left and bottom boundaries and following through on the reasoning in thedesign theorem and its application to marching, there are only two possibilities:
ā¢ There is a unique N ĆM RFFQM crease pattern consistent with the prescribed geometryand mountain-valley assignments on the left and bottom boundary.
ā¢ Or there is no such crease pattern consistent ā. . . ā boundary.
The former statement is simply the result of marching, via (57)-(65), without failure for the input.Alternatively, the latter is deduced when one of the checks in (60), (62) and (65) is violated duringthe process of marching. In this case, the algorithm cannot continue, and we are left with the factthat the chosen input data is incompatible with RFFQM.
A linear program for the side lengths. The aforementioned incompatibility is somewhatunsatisfying; really, it calls into question of robustness of the algorithm when dealing with creasepatterns that have a large number of panels. There are, however, some convenient tools to addressincompatibility and robustness. Along this line, we build on an idea of Lang and Howell [36].The basic heuristic is that this incompatibility is unfortunately extremely sensitive to the sidelength conditions (65). To this point, notice that the calculation of sector angle and mountain-valley assignments (60)-(63) does not require the lengths. We may therefore compute by marchingall the sector angles Ī±Tot := (Ī±
11a , Ī±
11b , . . . , Ī±
NMb , Ī±
NMc ) and all the mountain-valley assignments
Ļ := (Ļ11a , . . . , ĻNMa ) prior to dealing with the lengths. Then, rather than treat the side lengths
24
-
`11ab, . . . , `N1ab > 0 and `
11ac, . . . , `
1Mac > 0 as input, we can attempt to optimize these subject to the
sensitive criterion in (65) that all the lengths of the overall crease pattern are positive.This optimization can conveniently be formulated as a standard linear program. In this direction,
we introduce the vector ` := (`11ab, . . . , `N1ab , `
11ac, . . . , `
1Mac ). It then follows from (64) that there exists
a linear transformation LijĪ±Tot , depending only on the sector angles Ī±Tot, such that
(`ijcd, `ijbd) = L
ijĪ±Tot
`. (67)
By also introducing the vectors 1 = (1, . . . , 1) and `Īµ with the same dimension as `, we obtain thestandard linear program11 in `Īµ,
min `Īµ1T
subject to LijĪ±Tot`Īµ ā„ Īµ(1, 1) for i ā {1, . . . , N}, j ā {1, . . . ,M},and `Īµ ā„ 0
(68)
for Īµ ā„ 0. As is well-known, standard and computationally efficient algorithms can quickly evaluatethe feasibility of the domain of the inequalities above and, if feasible, compute a solution to theoptimization. In short, this linear program provides a robust and efficient framework for ensuring avalid crease pattern or declaring that such a pattern is infeasible. Particularly, in assuming a caseĪµ > 0 is feasible and has solution `?Īµ to (68), we can set
(`11ab, . . . , `N1ab , `
11ac, . . . , `
1Mac ) = `
?Īµ + Ī“1. (69)
By continuity and using that `?Īµ obeys the constraints in (68), there is a Ī“Īµ > 0 such that for allĪ“ ā (0, Ī“Īµ), the side lengths (`ijcd, `ijbd) := LijĪ±Tot(`
?Īµ+Ī“1) are strictly positive for all i, j. So by choosing
Ī“ accordingly, we obtain a valid crease pattern by this procedure for a given set of sector anglesĪ±Tot. On the other hand, if this linear program fails the feasibility test for the choice Īµ = 0, thenwe can be assured that the sector angles Ī±Tot, obtained by marching, cannot produce a valid creasepattern.
In practice, while capable of quickly assessing the feasibility of inputed sector angles Ī±Tot, thisprogram is no substitute for sound engineering judgment or an āOrigamistā intuition. In fact, fromour exploration of the configuration space by this method, it appears that most feasible sectorangles will result in a crease pattern with panels of disparate aspect ratios. To illustrate thispoint, consider the example in Fig. 6. Here, we input boundary sector angles in (66) by a uniformdistribution whose average gives the classical Miura-Ori and whose support is Ā±0.05Ļ to the left andright of this average. We also input the mountain-valley assignments consistent with a Miura-Ori.We then march, as above, to obtain all the sector angles Ī±Tot and mountain-valley assignments Ļ,and we use the sector angles to compute the side lengths via the optimization (68). One such resultof this procedureāand a seemingly generic one at thatāis the crease pattern in Fig. 6. Notice thatsome of the side lengths emerging from the optimization are driven close to zero, while others takeon rather large values. This is hardly in keeping with manufacturability considerations.
A natural next step might be to replace the (relatively arbitrary) linear objective function in (68)with a non-linear one that attempts to keep the aspect ratios of the panels as uniform as possible,while optimizing over the same feasible set. However, this generalization would be a significant
11In keeping with the literature on linear programming, the inequalities here represent an element-wise constraint.
25
-
departure from the well-developed and efficient algorithms used for the linear program. So we leaveit a topic of future research. In any case, one can always start from a well-known crease pattern(e.g., the Miura-Ori), and build a new pattern by slight perturbations that do not necessarily requireimplementation of the linear program for robustness. In fact, we follow this approachāwith markedsuccessāwhen discussing the inverse problem in the next section.
Marching to obtain the kinematics. Let us assume we have completed the marching pro-cedure aboveāeither by incorporating the linear program or notāand obtained all sectors angels,mountain-valley assignments and side lengths at every (i, j)-panel in the N ĆM crease pattern. Itfollows that every vertex of the crease pattern is determined up to rigid motion. For an explicitcalculation, we introduce the rotation Re3(Ī±) := e3āe3 +cosĪ±(Iāe3āe3)+sinĪ±(e2āe1āe1āe2)and prescribe the (1, 1)-panel by the vertices on R3,
x11a = 0, x11c = `
11ace1, x
11b =
`11ab`11ac
Re3(Ī±11a )(x
11c ā x11a ), x11d = xb +
`11bd`11ab
Re3(Ī±11b )(x
11a ā x11b ) (70)
(see Fig. 5(a-b) for the vertex notation). This then fixes the rigid body motion and enables amarching procedure for all the other vertices. In particular, the remaining panels along the firstrow are described by the vertices
x1ja = x1(jā1)c , x
1jb = x
1(jā1)d , x
1jc = x
1ja +
`1jac`1jab
Re3(āĪ±1ja )(x1jb ā x1ja ),
x1jd = x1jb +
`1jbd`1jab
Re3(Ī±1jb )(x
1ja ā x1jb ),
(71)
and the panels for the rest of the rows are described by the vertices
xija = x(iā1)jb , x
ijc = x
(iā1)jd , x
ijb = x
ija +
`ijab`ijac
Re3(Ī±ija )(x
ijc ā xija ),
xijd = xijb +
`ijbd`ijab
Re3(Ī±ijb )(x
ija ā xijb ), i > 1.
(72)
From the vertices, we can build the full kinematics of the RFFQM by making repeated use of thelocal description of the origami in Theorem 2.1 and following the schematic in Fig. 5(b). Basicallyat every xija vertex, we can define the local tangents as
t1 :=xijc ā xija|xijb ā xija |
, t2 :=xijb ā xija|xijb ā xija |
, t3 := Re3(Ī²ija )t2, t4 := Re3(Ī²
ija ā Ļ)t1. (73)
(Here, we do not label them with an ij since there is never a need to store these quantities.) Wecan then consider the four folding angles along the creases associated to these tangent, i.e., Ī³ij1 , Ī³
ij2 ,
Ī³ij3 and Ī³ij4 , and observe that these angles have to be consistent with the local description in the
theorem and with each other when marching from neighbor to neighbor. On the last point, noticethat Ī³ij3 must coincide with Ī³
i(jā1)1 since they both describe the same crease. Likewise Ī³
ij4 = Ī³
(iā1)j2 .
This leads to an explicit procedure. We define
Ī³111Ļ = Ļ, Ī³112Ļ = Ī³Ģ
Ļ11a2 (Ī³
111Ļ;Ī±
11a , Ī²
11a ), Ī³
113Ļ = āĻ11a Ī³111Ļ, Ī³114Ļ = Ļ11a Ī³112Ļ (74)
at the x11a vertex to initialize a procedure for these kinematics for some Ļ ā [āĻ, Ļ]. The remainingfolding angles at the bottom boundary are then determined by
Ī³1j3Ļ = Ī³1(jā1)1Ļ , Ī³
1j1Ļ = āĻ1ja Ī³1j3Ļ, Ī³1j2Ļ = Ī³ĢĻ
1ja
2 (Ī³1j1Ļ;Ī±
1ja , Ī²
1ja ), Ī³
1j4Ļ = Ļ
1ja Ī³
1j2Ļ, (75)
26
-
! = 0 ! = ā”
Figure 6: A āgenericā RFFQM crease pattern obtained the marching algorithm. The input sectorangles are chosen from a uniform distribution that perturbs around a classical Miura-Ori, theside lengths are determined by optimization using the linear program, and the mountain-valleyassignment is consistent with the Miura-Ori.
27
-
and the rest of the folding angles needed to build the kinematics are completely determined by
Ī³ij4Ļ = Ī³(iā1)j2Ļ , Ī³
ij2Ļ = Ļ
ija Ī³
ij4Ļ, Ī³
ij1Ļ = Ī³Ģ
āĻija1 (Ī³
ij2Ļ;Ī±
ija , Ī²
ija ), Ī³
ij3Ļ = āĻija Ī³ij1Ļ, i > 1. (76)
From the explicit description of the folding angles, we can compute the deformation gradientsby a marching procedure. To this end, let us focus on the sketch in Fig. 5(b) for an (i, j)-panel
that is not the (1, 1)-panel. We have access to either the deformation gradient Ri(jā1)Ļ of the left
neighbor or the deformation gradient R(iā1)jĻ of the bottom neighbor (or both), as depicted. We
also know that the deformation gradients must, up to an overall rigid rotation, be as indicated inFig. 2(b) for this vertices folding angles. Given the local tangents in (73), the rotations in (4), andfolding angles (74)-(76), we therefore work out that
RijĻ =
{Ri(jā1)Ļ R2(āĪ³ij2Ļ) if Ri(jā1)Ļ is known,
R(iā1)jĻ R1(Ī³
ij1Ļ) if R
(iā1)jĻ is known.
(77)
Note, when both deformation gradients are known, i.e., i, j > 1, the two prescriptions for RijĻ abovewill coincide since the sector angles and mountain valley assignments have been chosen to satisfypanel compatibility. We therefore set R11Ļ = I to eliminate the degeneracy of a rigid rotation anduse this to initialize the formulas in (77) to compute the deformation gradients at all panels.
With the deformation gradients, we are finally able to compute the deformations. Explicitly, wetake the first panel to be undeformed, i.e.,
y11aĻ = x11a , y
11bĻ = x
11b , y
11cĻ = x
11c , y
11dĻ = x
11d , (78)
as R11Ļ = I. Then, every other panel along the first row has deformed vertices
y1jaĻ = y1(jā1)cĻ , y
1jbĻ = y
1jaĻ + R
1jĻ (x
1jb ā x1ja ), y1jcĻ = y1jaĻ + R1jĻ (x1jc ā x1ja ),
y1jdĻ = y1jaĻ + R
1jĻ (x
1jd ā x1ja ).
(79)
Finally, every remaining panel has deformed vertices
yijaĻ = y(iā1)jbĻ , y
ijbĻ = y
ijaĻ + R
ijĻ (x
ijb ā xija ), yijcĻ = yijaĻ + RijĻ (xijc ā xija ),
yijdĻ = yijaĻ + R
ijĻ (x
ijd ā xija ).
(80)
This completes the marching procedure to obtain the kinematics for a given Ļ ā [āĻ, Ļ]. Byrepeating the procedure monotonically from Ļ = 0 to Ļ = Ļ, we obtain the complete kinematicaldescription of the rigidly and flat-foldable motion of the crease pattern (e.g., Fig. 5(e)).
4.2 Discussion on the inverse problem
Beginning with the pioneering work in [31] on non-Euclidean plates, the design of shape-morphingstructures and materials has blossomed into an active topic of research. In this context, the forwardproblem refers to computing the shapes achievable given an explicit design of the material orstructure. This is essentially what we have outlined thus far in the context of RFFQM. Alternatively,the inverse problem refers to designing a material or structure to achieve a targeted shape. This is
28
-
v
u
Oy
z
Ox
u=i, v=j v=ju=i r (i, j)
y
z
Ox
(a) (b) (c)
Figure 7: Sketch of the parameterizations in the inverse design problem. (a) The targeted surface isdiscretized by a parameterization on rectangular subset of the Z2 lattice. (b) Likewise the origamiis discretized on this lattice, but with the corner vertices of 2Ć 2 sets of panels forming the imagepoints of the lattice.
a far more difficult task. Nevertheless, some results are beginning to emerge in this direction forstimuli-responsive materials [1, 22, 44, 54], kirigami [9], and origami [14, 16, 34].
We focus here on a brief introduction of the inverse problem for RFFQM, with more compre-hensive exposition to follow [12]. In contrast to these previous works, the shape-morphing achievedby RFFQM has two novel properties:
ā¢ For a computed design, the full (ideal) kinematics along a prescribed mountain-valley assign-ments are known; apply the procedure outlined in the previous section.
ā¢ The structure is deployable; that is, it can deform from flat to folded flat as rigidly foldableorigami.
These aspects are expected to be important in many aforementioned applications [2, 3, 30, 33, 38,42], as they enable predictable deployment from both an the easy-to-manufacture flat state and afully-folded compact state.
We start simple to build some understanding of feasibility in the inverse problem. Recall thatthe configuration space for RFFQM depends on the sector angles, side lengths and mountain-valley assignments at the left and bottom boundary (equation (66) and Fig. 5(c)). Recall also thatthe algorithm to explore this configuration space may not admit a valid crease pattern, or, evenwhen it does admit one, there is an entire one-parameter family of deformations associated to thiscrease pattern. In other words, the configuration space is simultaneously massive and fraught withrobustness issues. To alleviate these concerns, we restrict our focus here to patterns consisting oftwo properties:
29
-
(i) We study slight perturbations of the boundary sector angles and lengths of a basic Miura-Ori.
(ii) We fix the boundary mountain-valley assignments to be consistent with a basic Miura-Ori.
These assumptions are built on the heuristic that, while a crease pattern can look qualitativelysimilar to a basic Miura-Ori in the flat state, it can be dramatically different by comparison in itspartially folded states due to the highly non-linear nature of origami kinematics. We show that thisādramatic differenceā can be harnessed to achieve a targeted shape.
Our basic formulation for the inverse design problem follows the sketch in Fig. 7. Consider atargeted surface as the image of a sufficiently smooth map rĢ : (0, N)Ć (0,M)ā R3, as depicted ingrey. At every interior pair of integers, we compute, by a finite difference scheme, discrete versionsof quantities describing the shape of target surface (e.g., the first and second fundamental forms),which we collect in a vector denoted SĢi,j. We also study 2N Ć 2M origami tessellations obtainedby the marching procedure above, and we index the vertices of these partially folded origami asyijĪ±,Ī²,`,Ļ for Ī±,Ī², ` parameterizing the relevant boundary data in
12 (66) and Ļ parameterizing thefolding parameter. We then introduce the reparameterization of the origami, rĪ±,Ī²,`,Ļ : {0, 1, . . . , N}Ć{0, 1, . . . ,M} ā R3, such that
rĪ±,Ī²,`,Ļ(i, j) := y(2i)(2j)Ī±,Ī²,`,Ļ. (81)
Importantly, this new parameterization serves to smooth out some of the oscillatory behavior of theorigami by taking 2Ć2 sets of panels as the fundamental meshing. As a result, we have a reasonablecomparison between the vector SĢi,j of our targeted surface and the analogous vector Si,j(Ī±,Ī², `, Ļ)obtained by computing the same finite difference quantities for rĪ±,Ī²,`,Ļ. In this comparison, we treatour objective function as
Loss(Ī±,Ī², `, Ļ) ā¼ā
i=0,...,M
āj=0,...,N
|Si,j(Ī±,Ī², `, Ļ)ā SĢi,j|2, (82)
up to some numerical considerations not developed here (see [12]). Importantly, Loss(Ī±,Ī², `, Ļ) = 0provides two discretizations rĢ and rĪ±,Ī²,`,Ļ that are, essentially, the same up to rigid body motion.As a consequence, the global minimizers to (82) represent a reasonable notion of solutions to theinverse problem. While global minimality is desirable, we aim for the more modest goal of computinglocal minimizers to (82); particularly, for the small subset RFFQM crease patterns that are slightperturbations of the basic Miura-Ori as stated in (i-ii). This setting allows us to investigate theinverse problem by means of standard optimization tools.
It is well-known that smooth isometric embeddings are extremely rigid and incapable of adeptlyapproximating surfaces with non-zero Gaussian curvature (Gaussās Theorema Egregium). However,this rigidity breaks down when the isometric condition or smoothness is relaxed, as Nashās famedembedding theorem [41] shows. To achieve non-trivial Gaussian curvature, one can exploit non-isometric deformation induced by active materials [39, 32], or non-smooth isometric embeddingsinduced by origami/kirigami constructions [7]. RFFQM is a version of the latter approach, and sothere is no apparent mathematical obstacle to their utility as approximations of arbitrary surfaces.Yet, approximation theorems involving origami mapsāof which the most general can be found in[10]āoften involve significant refinement, making their application ill-suited to practical engineeringdesign. We are therefore motivated to study whether structured flat-foldable origami, as exhibited
12replacing N by 2N and M by 2M for consistency, of course
30
-
(a)
(b)
(c)
Figure 8: Examples of RFFQM to achieve a Spherical Cap (a), a Hyperboloid (b) and a QuarterVase (c). (Left:) The smooth targeted surface and its discrete mesh. (Middle:) A RFFQM in itoptimized partially folded state. (Right:) The flat crease pattern that is slightly perturbed from aMiura-Ori pattern and rigidly folded to achieve the middle state.
by the assumption (i-ii), can serve as good approximations to surfaces of varied and non-zeroGaussian curvature.
In this direction, we test the design methodology outlined above on a series of simple examplesthat illustrate the range of Gaussian curvatures achievable: a Spherical Cap to achieve positiveGaussian curvature, a Hyperboloid to achieve negative Gaussian curvature, and finally a QuarterVase to exhibit changes in Gaussian curvature. The results of our optimization are provided inFig. 8. We find striking qualitative agreement in these examples, especially given the fact that,
31
-
with properties (i-ii), we have confined ourselves to only a small fraction of the configuration spaceof RFFQM.
To this point, natural extensions to the underlying assumptions (i-ii) can be made by incorpo-rating ideas of objective origami, which we first explored in [19] and are developing in more detailin [49]. By incorporating these extensions and tools from data-driven engineering, we are in theprocess of gaining a more complete picture of the configuration space. The results given here suggestthis is a promising framework for shape-morphing based on inverse design principles.
Acknowledgement
F.F., R.D.J., and P.P. gratefully acknowledge the support of the Air Force Office of ScientificResearch through the MURI grant no. FA9550-16-1-0566. The authors would like to thank RobertJ. Lang for many helpful discussions and useful comments on this manuscript.
References
[1] H. Aharoni, Y. Xia, X. Zhang, R. D. Kamien, and S. Yang. Universal inverse design ofsurfaces with thin nematic elastomer sheets. Proceedings of the National Academy of Sciences,115(28):7206ā7211, 2018.
[2] M. Arya. Packaging and deployment of large planar spacecraft structures. PhD thesis, CaliforniaInstitute of Technology, 2016.
[3] M. Arya, N. Lee, and S. Pellegrino. Ultralight structures for space solar power satellites. In3rd AIAA Spacecraft Structures Conference, page 1950, 2016.
[4] J. M. Ball and R. D. James. Fine phase mixtures as minimizers of energy. In Analysis andContinuum Mechanics, pages 647ā686. Springer, 1989.
[5] M. Berry, M. Lee-Trimble, and C. Santangelo. Topological transitions in the configurationspace of non-euclidean origami. arXiv preprint arXiv:1910.01008, 2019.
[6] K. Bhattacharya et al. Microstructure of martensite: why it forms and how it gives rise to theshape-memory effect, volume 2. Oxford University Press, 2003.
[7] S. J. Callens and A. A. Zadpoor. From flat sheets to curved geometries: Origami and kirigamiapproaches. Materials Today, 21(3):241 ā 264, 2018.
[8] Y. Chen, R. Peng, and Z. You. Origami of thick panels. Science, 349(6246):396ā400, 2015.
[9] G. P. Choi, L. H. Dudte, and L. Mahadevan. Programming shape using kirigami tessellations.Nature materials, 18(9):999ā1004, 2019.
[10] S. Conti and F. Maggi. Confining thin elastic sheets and folding paper. Archive f