Importance and Effectiveness of Representing the Shapes of Cosserat Rods and Framed Curves as Paths in the Special Euclidean Algebra
 829 Downloads
 1 Citations
Abstract
We discuss how the shape of a special Cosserat rod can be represented as a path in the special Euclidean algebra. By shape we mean all those geometric features that are invariant under isometries of the threedimensional ambient space. The representation of the shape as a path in the special Euclidean algebra is intrinsic to the description of the mechanical properties of a rod, since it is given directly in terms of the strain fields that stimulate the elastic response of special Cosserat rods. Moreover, such a representation leads naturally to discretization schemes that avoid the need for the expensive reconstruction of the strains from the discretized placement and for interpolation procedures which introduce some arbitrariness in popular numerical schemes. Given the shape of a rod and the positioning of one of its cross sections, the full placement in the ambient space can be uniquely reconstructed and described by means of a base curve endowed with a material frame. By viewing a geometric curve as a rod with degenerate pointlike cross sections, we highlight the essential difference between rods and framed curves, and clarify why the family of relatively parallel adapted frames is not suitable for describing the mechanics of rods but is the appropriate tool for dealing with the geometry of curves.
Keywords
Cosserat rod Framed curve Euclidean algebra Shape discretizationMathematics Subject Classification (2010)
74K10 53A041 Motivation and Main Results
Over the past century, rod theory has undergone a systematic development and has provided a platform for endless applications. We regard Antman’s [1] presentation of the subject as the definitive reference for both the physical and mathematical foundations of the theory. As for applications, a wealth of specialized references can be found. Here, we only mention models of elastic beams in structural engineering, studies of the shapes and instabilities of cables and cords, simulations of hair strands in computer graphics, and investigations of DNA supercoiling as evidence of the widespread usage of rod theory. In each of these applications what is used is the special Cosserat theory of rods, as introduced by the brothers Cosserat [2, 3] in 1907. The first aim of the present paper is to put in evidence some features of the Lie algebraic structure that is implicit in the treatment of rod theory given by Antman [1]. That structure, while being only accessory to most analytical developments, is extremely relevant to the construction of discretization schemes able to capture some important traits of the theoretical framework.
We show that the shape of a rod, namely those features that are invariant under direct isometries of the threedimensional ambient space, can be identified with a squareintegrable path in the special Euclidean algebra. As explained in Sect. 2, this emerges because the cross sections of a rod are assumed to be rigid and the special Euclidean group is the Lie group that describes the possible placements of a rigid body in threedimensional space. By virtue of the tacit continuity assumptions of rod theory, a purely Lie algebraic description of the rod shape is available. The main feature of this approach is that information about the shape of a rod is not encoded in a description of what is seen in the ambient space but instead stems from a description of the procedure that must be followed to redraw what is seen.
Such a representation of the rod shape, though not intuitive, appears to be extremely natural once it is recognized that it is defined in terms of the same strain fields that are most commonly used to describe the material response of the rod. We show that the six strain fields are the only degrees of freedom necessary to determine the shape of a rod (accompanied, of course, by a description of the cross sections as twodimensional sets).
Significantly, the general variational approach devised by Schuricht [4] to study the equilibria of nonlinearly elastic rods with topological constraints (and recently adopted by Giusteri, Lussardi and Fried [5] to study the Kirchhoff–Plateau problem) is tacitly based on the same Lie algebraic representation of the rod shape. Moreover, the role of the special Euclidean algebra is also essential in connection with the geometric mechanical concepts described, for instance, in the works by Simo, Marsden and Krishnaprasad [6], Simo, Posbergh and Marsden [7], Holm, Noakes and Vankerschaver [8], and Eldering and Vankerschaver [9] and with the \(G\)strand equations discussed by Holm and Ivanov [10]. It should be noted, however, that these authors apply geometric concepts to study the dynamics of rods, whereas we focus on the description of shapes.
Due to the basic role played by the strain fields, simulation strategies based on this representation offer an easier management of the relevant physical information. We present, in Sect. 3, a very intuitive and yet powerful discretization scheme, that generalizes to special Cosserat rods the approach devised by Bertails, Audoly, Cani, Querleux, Leroy and Lévêque [11] for Kirchhoff rods. The major advantage of this approach is that, operating directly at the level of the Lie algebra, it is never necessary to interpolate between different elements of the special Euclidean group. Interpolation or discrete differentiation are usually necessary to retrieve differential information about the shape of a rod—information that is essential to compute the material response—from the placements in the ambient space of a finite number of cross sections of the rod. Unfortunately, there is no unique way to reconstruct that information. By contrast, we introduce a finiteelement discretization of the rod shape in which the essential information is always available and from which the placements in the ambient space of the cross sections are uniquely determined. In Sect. 4, we illustrate the effectiveness of that discretization by solving boundaryvalue problems to find the equilibrium shapes of special Cosserat rods.
The same Lie algebraic construction applies to the theory of framed curves. The points of any such curve are endowed with a triad of orthonormal vectors that constitute a frame field varying along the curve. Framed curves have been used to study topological and geometric invariants and as basic models for describing the kinematics of slender bodies. In this context, they are sometimes considered equivalent to special Cosserat rods, but this commingling should be avoided. Indeed, as the name suggests, the notion of framed curve rests upon the geometry of a curve as the basic constituent, according to which the frame field should be constructed. In contrast, the basic objects in rod theory are the material cross sections.
The second objective of this paper is thus to clarify the distinction between special Cosserat rods and framed curves. By deriving, in Sect. 5, the theory of framed curves as a limiting case of the Cosserat theory, we show that the former theory is not adequate to describe the mechanics of rods, since it is incapable of tracking twisting and shearing deformations and completely neglects any effect due to the actual shapes of the cross sections. Even in those cases in which the frame along the curve is chosen to represent the material frame (and not merely determined by the curve geometry), the essential role accorded to the base curve makes it difficult to factor out global isometries. It also imposes viewing the strain fields as derived degrees of freedom, at odds with their primary role in the mechanical theory of rods.
Our derivation of the theory of framed curves highlights the relevance of the results presented by Bishop [12] in 1975, results which are still surprisingly ignored in some recent publications. We generalize his construction of relatively parallel adapted frames to the case of continuously differentiable regular curves. We show that the corresponding family of frame fields is uniquely determined by the geometric invariants of a generic curve. We also identify such geometric invariants with a squareintegrable curvature field and a measurevalued torsion field, the regularity of which cannot be improved without imposing additional assumptions. We conclude by remarking that, when treating purely geometric questions surrounding space curves, relatively parallel adapted frames are the appropriate tool, and any use of the Frenet frame should be abandoned.
2 Describing a Thin Rod
When modeling a filament or rod as a continuous body, we can mathematically express its slenderness by saying that, at any of its points, we can identify a direction in which the boundary of the body appears to be much farther away than it does in the two remaining orthogonal directions. If this is the case, we can represent the body as the collection of planar twodimensional rigid bodies, named cross sections. The special Cosserat theory of rods (as presented, for instance, by Antman [1]) is predicated on the assumption that these cross sections are rigid and can only rotate or translate in space when the rod deforms. It is then clear that the configuration of a special Cosserat rod (henceforth referred to simply as a rod) is fully described by assigning a family of twodimensional sets, describing the material cross sections, and specifying how those sets are placed in threedimensional ambient space. On the other hand, the shape of a rod is invariant under isometries of the ambient space and it is encoded in the relative placement of infinitesimally close cross sections.
For definiteness, we describe the family of cross sections, parametrized by \(s\) in the interval \([0,L]\), as given by compact simply connected subsets \(\mathscr {A}(s)\) of \(\mathbb{R}^{2}\). It is important to clearly state a continuity assumption to make sure that any positioning of the collection of cross sections in space forms a continuous body. A first step toward guaranteeing continuity is to assume that the origin \(\boldsymbol{0}_{2}\) of \(\mathbb{R}^{2}\) belongs to the interior of \(\mathscr {A}(s)\) for every \(s\). Although the choice of \(\boldsymbol{0}_{2}\) is convenient, we emphasize that it is completely arbitrary. Using any other point of \(\mathbb{R}^{2}\) is allowed and it is also possible to devise different conditions.
The strain fields \(u_{i}\) and \(v_{i}\), for \(i=1,2,3\), have the following geometric interpretations. Indicating by \(ds\) an infinitesimal increment of arclength, \(u_{i}(s)\) represents the differential rotation about \(\boldsymbol{d}_{i}(s)\) needed to bring the material frame at \(s\) onto the material frame at \(s+ds\); \(u_{1}(s)\) and \(u_{2}(s)\) thus concern flexural deformations of the collection of cross sections, while \(u_{3}(s)\) is associated with twisting deformations. Meanwhile, \(v_{i}(s)\) represents the differential translation in the direction of \(\boldsymbol{d}_{i}(s)\) needed to bring the image of the origin at \(s\) onto the image at \(s+ds\); \(v_{1}(s)\) and \(v_{2}(s)\) thus concern shearing between adjacent cross sections, while \(v_{3}(s)\) governs the differential distance between them, since \(\boldsymbol{d}_{3}(s)\) is normal to the cross section at \(s\).
In view of the differentiability assumption, the mapping \(\boldsymbol{x}\) describes a differentiable curve in \(\mathbb{R}^{3}\), parametrized by \(s\) in the interval \([0,L]\). Such a base curve (called also midline, centerline, etc.) gives a first approximation of the rod configuration and it is most often taken as a starting point in the description of a rod. Nevertheless, we think that this point of view (albeit followed in our previous related publications) is misleading, since that curve is only expedient in describing the placement of the cross sections in space. We will return to analogies and differences between a rod and a framed curve in Sect. 5, but, to appreciate the immaterial nature of the base curve, it is enough to observe that it is possible to choose the sets describing the cross sections in such a way that the origin \(\boldsymbol{0}_{2}\) of \(\mathbb{R}^{2}\) never belongs to \(\mathscr {A}(s)\), clearly showing that the points of the image of the base curve do not belong to the material points that constitute the rod as a continuous body. A thorough analysis of the role of the base curve in rod theory is given by Antman and Schuricht [13], and analogous considerations for the case of shells were earlier provided by Naghdi [14].
From the construction above, we conclude that the shape of a rod, namely those features that are invariant under direct isometries of threedimensional space, is fully encoded in the strain fields \(u_{i}\) and \(v_{i}\), \(i=1,2,3\), that determine the operator \({\mathsf{L}}\). At the same time, we see that the way in which a rod is rigidly translated and rotated in space depends solely on the initial conditions given by \({\mathscr {R}}_{0}\).
2.1 The Lie Algebra and the Lie Group Associated with the Rod Description
Since a rod is defined by a collection of planar rigid cross sections continuously positioned in space, it is not surprising that the Lie algebra used to describe this system corresponds to the one needed to describe the positioning of rigid bodies in three spatial dimensions: it is the special Euclidean algebra \({\mathfrak{se}(3)}\), which is associated with the special Euclidean group \({\mathsf{SE}(3)}\) generated by rotations and translations of threedimensional space.
Also the corresponding Lie group \({\mathsf{SE}(3)}\) can be represented as a subgroup of \(\mathsf{GL}_{4}(\mathbb{R})\). Its elements can be obtained by applying the exponential map to the linear combinations of the generators of \({\mathfrak{se}(3)}\). Other representations of \({\mathsf{SE}(3)}\) have been devised with the objective of reducing memory usage in computational settings (see, for instance, the presentation by Murray, Li and Sastry [16]), but they are not needed in the present treatment.
The importance of the special Euclidean group \({\mathsf{SE}(3)}\), as a subgroup of the affine group, for the discussion of motion and shape representations in computer graphics and geometric modeling is presented, for example, by Agoston [17]. The relevance of \({\mathsf{SE}(3)}\) and the associated algebra \({\mathfrak{se}(3)}\) to rod theory is acknowledged by Sander [18], discussed in a review by Chirikjian [19], and exploited in beam modeling by Sonneville, Cardona and Brüls [20, 21]. These authors base their approaches on representing a rod through elements of the group \({\mathsf{SE}(3)}\), but we propose that the algebra \({\mathfrak{se}(3)}\) provides a representation that is more naturally and directly related to the shape of a rod.
We have already observed that a full description of the placement of a rod in space corresponds to a continuous path \(\{{\mathscr {R}}(s):s \in [0,L]\}\) in \(\mathbb{R}^{12}\), accompanied by a description of the material cross sections, since these fully determine the placement map \(\boldsymbol{p}\). Based on the decomposition (5) of \({\mathscr {R}}(s)\) as the action of the propagator \({\mathsf{U}}(s;0)\) on the initial point \({\mathscr {R}}_{0}\), it is possible to factor out global rigidbody motions, encoded in \({\mathscr {R}}_{0}\), and identify the corresponding equivalence class of placements with the path \(\{{\mathsf{U}}(s;0):s\in [0,L]\}\) in \(\mathsf{GL}_{12}(\mathbb{R})\). Specifically, since the operator \({\mathsf{U}}(s;0)\) belongs, for any \(s\), to a representation of \({\mathsf{SE}(3)}\) within \(\mathsf{GL} _{12}(\mathbb{R})\), we can identify the placement of the rod modulo rigid transformations with the continuous path \(\{{\mathsf{U}}(s;0):s \in [0,L]\}\) in \({\mathsf{SE}(3)}\). It is also immediately evident that, having obtained \({\mathsf{U}}(s;0)\) by solving (4), the essential information encoding the shape of the rod can be identified with the possibly discontinuous path \(\{{\mathsf{L}}(s):s\in [0,L]\}\) in \({\mathfrak{se}(3)}\).
A rod is sometimes described as the juxtaposition of a path in \(\mathbb{R}^{3}\) (the base curve) and a path in \(\mathsf{SO}(3)\) representing the collection of material frames. This point of view, popularized by the works of Simo, Marsden and Krishnaprasad [6] and Simo, Posbergh and Marsden [7], does not seem to advance the objective of distinguishing between the shape of the rod and its placement in space; indeed it is akin to choosing the first three components of \({\mathscr {R}}(s)\) and the rotational part of \({\mathsf{U}}(s;0)\) to describe the rod and thereby introducing an unnecessary asymmetry.
All of the mentioned identifications—which exploit either ℛ, \({\mathsf{U}}(\cdot ;0)\), or \({\mathsf{L}}\)—are relevant to the construction of computational schemes for the simulation of rods and different discretized representations of a rod can be interpreted as different ways to discretize those paths. Effective discretizations of rods to model slender bodies have been developed, among others, by Cao, Liu and Wang [22], Spillmann and Teschner [23], Bergou, Wardetzky, Robinson, Audoly and Grinspun [24], Bergou, Audoly, Vouga, Wardetzky and Grinspun [25], Audoly, Clauvelin, Brun, Bergou, Grinspun and Wardetzky [26], Jung, Leyendecker, Linn and Ortiz [27], Lang, Linn and Arnold [28], and Linn [29]. A vast literature also exists in which rod theory is applied to the computational mechanics of beams. These approaches are characterized by the fact that translational and rotational degrees of freedom are often considered separately and the beam shape is reconstructed by means of interpolation procedures. A selection of methods can be found in the works by Simo and VuQuoc [30], Borri and Bottasso [31], Ibrahimbegović [32], Betsch and Steinmann [33], Meier, Popp and Wall [34, 35], Gaćeša and Jelenić [36], Bauer, Breitenberger, Philipp, Wüchner and Bletzinger [37], Yilmaz and Omurtag [38], and Zupan and Zupan [39].
In all the foregoing examples, the discretization is performed at the level of either ℛ or \({\mathsf{U}}(\cdot ;0)\), that is, by considering the placement of the rod in space. An important exception to this general trend can be found in the works by Zupan and Saje [40, 41], Češarek, Saje and Zupan [42] (mainly addressing linearized beam equations), Su and Cesnik [43], and Schröppel and Wackerfuß [44]. There, discretization is performed at the level of Lie algebraic fields, called strains, but nodal values are of the essence and interpolation schemes are again needed to reconstruct the shape of a rod.
In Sect. 3, we introduce a discretization of the shape of a rod viewed as a path in the algebra \({\mathfrak{se}(3)}\), as defined by \({\mathsf{L}}\). This generalizes to special Cosserat rods the approach used by Bertails, Audoly, Cani, Querleux, Leroy and Lévêque [11] for Kirchhoff rods and can be viewed as a bridge between the methods of Sander [18], Chirikjian [19], and Sonneville, Cardona and Brüls [20, 21], based on the special Euclidean group, and the aforementioned ones, based on Lie algebraic quantities. A distinguishing feature of the present approach is that it obviates the need for any interpolation associated with the reconstruction of the shape of a rod from a finite sampling of its placement in space.
2.2 Constraints on the Placement and on the Shape of a Rod
We consider two classes of constraints: constraints on how a rod is positioned in space and constraints on the shape of a rod, usually termed internal constraints. Internal constraints are more easily represented as conditions on the path traced by the operator \({\mathsf{L}}\) in \({\mathfrak{se}(3)}\), whereas constraints on the placement of the rod are nicely enforced on the path given by ℛ in \(\mathbb{R}^{12}\).
2.2.1 Placement Constraints
It is possible to express the clamping conditions in terms of \({\mathscr {R}}_{0}\) and of the path traced by \({\mathsf{L}}\) in \({\mathfrak{se}(3)}\) only by means of the nonlinear and nonlocal expression of \({\mathsf{U}}\) in terms of \({\mathsf{L}}\). This shows that enforcing the clamping conditions can be a delicate issue when this representation of a rod is used. Moreover, the relation between \({\mathsf{U}}\) and \({\mathsf{L}}\) can be made explicit only in particular cases. It is fortunate that those cases can be exploited to set up computational schemes, as we will show in Sect. 3.
Notably, the foregoing clamping conditions can also be used to describe closed rods and they can be adapted, as discussed at the end of Sect. 5, to express the closure constraint when dealing with framed curves.
2.2.2 Internal Constraints
Evidently, there is a considerable simplification in the model, since only three scalar fields determine the shape of a rod constrained in accord with (8) and (9). Perhaps surprisingly, however, there is absolutely no simplification in the Lie algebra and group necessary to describe the system. Indeed, due to the commutation relation \([V_{i},U_{j}]=\varepsilon_{ijk}V_{k}\), the unavoidable presence of the generator \(V_{3}\) in the algebra associated with (10) requires that \(V_{1}\) and \(V_{2}\) both remain in the picture. Hence, \({\mathfrak{se}(3)}\) and \({\mathsf{SE}(3)}\) are again the relevant mathematical structures to be considered.
3 Discretizing the Rod Shape in \({\mathfrak{se}(3)}\)
In this section, we introduce a discretization of the shape of a rod based on its representation as a path in the special Euclidean algebra \({\mathfrak{se}(3)}\). We also discuss the advantages and limitations of this approach, with particular reference to variational descriptions of the rod elasticity. For the special case of a Kirchhoff rod, this discretization scheme reduces to the one used by Bertails, Audoly, Cani, Querleux, Leroy and Lévêque [11]. We moreover discuss the connection between our approach and the interpolation of affine transformations introduced very recently by Kaji and Ochiai [47] in the context of computer graphics applications.
The most important feature of our perspective is that it does not rest on discretizing the placement of a rod in space. We instead discretize the shape of the rod. The placement in space is uniquely determined by the shape of a rod and the placement of one of its cross sections and can be easily reconstructed. The converse is not true, and this shows the major advantage of the present method. Indeed, there is no unique way to reconstruct the shape of a rod from a discretization of its placement in space, as testified by the large number of interpolation techniques proposed in the literature (reviewed, for instance, by Romero [48] and Bauchau and Han [49]).
The starting point for the scheme is the observation that the solution of a firstorder linear ordinary differential equation with constant coefficients can be represented explicitly using the matrix exponential map. As we already mentioned, the representation is not explicit in the general case of nonconstant coefficients, but it remains explicit for piecewise constant coefficients.
Relation (11) uniquely defines discrete propagators \({\mathsf{U}}_{k}\), \(k=1,\ldots ,N\), that can be used to reconstruct the (discretized) placement of the cross sections of the rod in space, by giving an initial cross section as a vector \({\mathscr {R}}_{0}\) in \(\mathbb{R}^{12}\) and successively applying equation (5). We then see that a piecewise constant finiteelement approximation of the operator field \({\mathsf{L}}\)—which is a path in \({\mathfrak{se}(3)}\)—provides a uniquely defined approximation of the placement of a rod through a discretization of its shape.
3.1 Advantages and Disadvantages of the Discretization
It is now worth commenting on the connection between our approach and the work of Kaji and Ochiai [47]. As a particular case, their results provide a parametrization of the group \({\mathsf{SE}(3)}\) in terms of the algebra \({\mathfrak{se}(3)}\). That parametrization can be used to describe the cross sections that represent the nodes of a discretization of the placement of a rod. This is only part of the information contained in the shape of a rod, since no strategy for going from one cross section to another is specified. The function “Blend” is used by Kaji and Ochiai [47, Sect. 5.2] to interpolate between two cross sections in a way which is consistent with additional information about the “shape” of the interpolation. For instance, they are free to prescribe the total twist accumulated between two cross sections. Although their tool is clearly very flexible and useful for graphics manipulations, their approach cannot be used to render a rod without providing additional information about its shape. Our perspective differs because we take the discretized shape of a rod as primitive information and then uniquely reconstruct the rod placement in space. Whereas Kaji and Ochiai [47] use the elements of \({\mathfrak{se}(3)}\) to parametrize \({\mathsf{SE}(3)}\), we use piecewise constant paths in \({\mathfrak{se}(3)}\) to encode the shape of a rod and reconstruct elements of \({\mathsf{SE}(3)}\), such as \({\mathsf{U}}_{k}\), only when necessary.
It should now be clear that the present discretization scheme is particularly useful when the information about the shape of a rod (and not its placement in space) is of central importance. This is particularly true whenever elastic beams or filaments are modeled by variational methods. Such methods always require the definition of an energy functional, the form of which characterizes the elastic response of the rod, and the main contribution to the stored elastic energy of a rod is always related to its shape. In this context, the need to reconstruct the information about the shape from the discretized placement is a potential source of difficulty. On the contrary, our construction is directly expressed in terms of shape parameters, the strain fields \(u_{i}\) and \(v_{i}\), \(i=1,2,3\), that uniquely determine the placement. In simple words, it is always clear how we go from a cross section to the adjacent one and this determines the elastic energy.
Another advantage of the present scheme is that, in each of the discretization intervals, the portions of a rod are generic helical segments. Hence, we can describe without any approximation certain curved configurations, as long as their shapes correspond to piecewise constant paths in \({\mathfrak{se}(3)}\). Moreover, the internal constraints of unshearability (8) and inextensibility (9) discussed in Sect. 2.2 can be imposed exactly because they are compatible with piecewise constant values of the fields \(v_{1}\), \(v_{2}\), and \(v_{3}\).
3.2 Examples
Here, we present a few examples of discretized rod shapes and renderings of the corresponding placements in threedimensional space. To avoid situations in which the cross sections are trivially superimposed (and which thus are of no physical interest) we stipulate that \(v_{3}(s)=1\) for each value of \(s\). Although this is not enough to guarantee noninterpenetration of matter, it rules out some trivial cases where this occurs. For the clamping condition at \(s=0\), we always assume that the base curve starts at the origin and that the material frame is aligned with a fixed orthonormal reference frame. We will first present “exact approximations”, namely cases in which the strain fields are uniform on the entire interval \([0,L]\).

Taking \(u_{1}(s)=c_{1}\neq 0\), \(u_{2}(s)=c_{2} \neq 0\), and \(u_{3}(s)=0\) for each \(s\) in \([0,L]\), we arrive at a description of a twistfree circular arc with scalar curvature \(\kappa \) of the base curve given by \(\kappa =(c_{1}^{2}+c_{2}^{2})^{1/2}\) (Fig. 1c).

Taking \(u_{1}(s)=u_{2}(s)=0\) and \(u_{3}(s)=c_{3} \neq 0\) for each \(s\) in \([0,L]\), we arrive at a description of a straight rod with total twist \(T\) given by \(T=c_{3}L\) (Fig. 1b).
 Taking \(u_{1}(s)=c_{1}\neq 0\), \(u_{2}(s)=c_{2} \neq 0\), and \(u_{3}(s)=c_{3}\neq 0\) for each \(s\) in \([0,L]\), we arrive at a description of a helical segment (Fig. 2a).

Taking constant values of \(v_{1}\) and \(v_{2}\) with \(v_{3}=1\) and requiring that \(u_{1}\), \(u_{2}\), and \(u_{3}\) vanish, we arrive at a description of a straight sheared rod (Fig. 1d).

Taking nonvanishing constant values for \(v_{1}\) and \(u_{3}\) with \(v_{3}=1\), we arrive at a description of a helix without flexural deformations (Fig. 2b).

Taking nonvanishing constant values for \(v_{1}\) and \(u_{2}\) with \(v_{3}=1\), we arrive at a description of a sheared circular arc (Fig. 2c).

Taking nonvanishing constant values for \(v_{1}\) and \(u_{1}\) with \(v_{3}=1\), we arrive at a description of another helical shape (Fig. 2d).
Any rod represented with our discretization is an assembly of segments with shapes of the kind described. The strain fields need not be continuous at the joints, so it is possible to exactly describe a rod formed, for instance, by two circular arcs lying in distinct planes using a partition of \([0,L]\) in just two subintervals. It should be noted, however, that whereas the tangent field to the base curve of a Kirchhoff rod is continuous by construction—the base curve is a regular curve—that field may be discontinuous for a shearable Cosserat rod.
4 Application to Shape Relaxation
The representation and discretization of the rod shape presented in the previous sections is particularly effective in dealing with shape relaxation problems. Here we provide a few examples. Our objective is to illustrate the main advantage of the proposed approach, namely that operating directly at the level of the strain fields makes it possible to easily treat problems in which all the deformation modes of a rod are combined and also to represent with a small number of elements nontrivial curvilinear configurations. Since our emphasis is on the representation of shapes and not on the solution procedure, we employ a simple and reliable gradient flow algorithm to find equilibrium configurations, but we do not aim to optimize the efficiency of the implementation.
The intrinsic shape defined by the fields \(\bar{u}_{i}\) and \(\bar{v}_{i}\), \(i=1,2,3\), obviously represents the unique minimizer of the functional ℰ in the absence of additional constraints on the strain fields (up to a correction to \(\bar{v}_{3}\) of order \(\epsilon \)). However, if clamping conditions are imposed at both ends of the rod, those conditions constitute a nonlinear constraint that defines the submanifold of \({\mathfrak{se}(3)}\) of admissible strain fields. If the intrinsic shape is not compatible with the clamping conditions, then the minimizer of ℰ is no longer obvious. We can also encounter situations in which multiple local minima are present, since the constraint is not convex.
To approximate energy minima numerically, we apply a gradient flow scheme to the functional ℰ while exploiting the discretization of strain fields discussed in the previous section. Given that the constraint manifold is, in general, nonlinear, a strategy for enforcing the constraint at each iteration is needed. Since the exponential map from \({\mathfrak{se}(3)}\) to \({\mathsf{SE}(3)}\) involved in the definition of the clamping constraint admits a closed form expression (see, for instance, the work of Kaji and Ochai [47]), it is possible to explicitly compute its gradient and apply a manifold projection method, as discussed by Hairer [50].
5 Framed Curves
In this section, we derive a description of framed curves as degenerate rods. This provides an appealing way to discern the nature of the geometric invariants associated with a regular curve, namely a curve for which the tangent field is welldefined and everywhere continuous. Although Bishop [12] carried out a careful analysis of this subject more than forty years ago, we show that his construction is valid under weaker assumptions, made clear by our alternate derivation. Moreover, we believe that our analysis provides additional compelling motivations for the exclusive use of parallel adapted frames for describing the kinematics of curves.
We view framed curves as rods with cross sections that shrink to single points. Those crosssections are thus clearly invariant under rotations. This degeneracy is reflected by the loss of meaning of the shear and twisting parameters. Indeed, to bring a cross section at \(s\) (a point in space) onto an adjacent cross section at \(s+ds\) it is necessary only to adjust the direction of the movement, as specified by assigning \(u_{1}(s)\) and \(u_{2}(s)\), and the intensity of the movement, as specified by \(v_{3}(s)\). This suffices to rigidly move from one point to another, thus making unnecessary the use of \(v_{1}(s)\), \(v_{2}(s)\), or \(u_{3}(s)\).
It is now worth commenting on the regularity needed for the forgoing construction. In particular, the fields \(u_{1}\) and \(u_{2}\) need not be continuous for (15) to have a unique solution. Correspondingly, granted that \(\boldsymbol{t}'\) is a squareintegrable field, the solutions \(u_{2}\) and \(u_{1}\) of (18) are also squareintegrable fields, as discussed by Tricomi [52]. A choice that is convenient for most practical purposes is to view \(u_{1}\), \(u_{2}\), and \(\boldsymbol{t}'\) as piecewisecontinuous fields, but weaker regularity is also allowed. We have thus shown that the prescription of a continuouslydifferentiable curve \(\boldsymbol{x}:[0,L] \to \mathbb{R}^{3}\), such that \(\boldsymbol{x}'(s)>0\) for any \(s\) in \([0,L]\), together with a choice for the value \(\boldsymbol{d}_{1}(0)\) of the material director \(\boldsymbol{d}_{1}\) at one end of the curve (since \(\boldsymbol{d}_{2}(0)\) is simply given by \(\boldsymbol{t}(0)\times \boldsymbol{d}_{1}(0)\)) uniquely determines the scalar fields \(v_{3}\), \(u_{1}\), and \(u_{2}\) that, in turn, suffice to build a relatively parallel adapted frame by solving (15).
It is possible to state in terms of the strain fields \(u_{1}\) and \(u_{2}\) (assuming \(v_{3}\) is equal to unity) also the famous problem raised independently by Efimov [54] and Fenchel [55] of identifying necessary and sufficient conditions on curvature and torsion for the curve parametrized by \(\boldsymbol{x}\) to be closed. This amounts to requiring that a suitable restriction of the operator \({\mathsf{U}}(L;0)\) defined in (5) be equal to the identity map. If we simply want the curve to be closed, we consider only the action of \({\mathsf{U}}(L;0)\) on the components of the field \(\boldsymbol{x}\). If we require a smooth closure, we then consider the restriction of \({\mathsf{U}}(L;0)\) to the components of \(\boldsymbol{x}\) and \(\boldsymbol{t}=\boldsymbol{d}_{3}\). This general solution of the closed curve problem is identical in spirit to that provided by Schmeidler [56] (and later by Hwang [57]) in terms of continuous curvature and torsion fields, but it readily shows that the results extend to the case of squareintegrable curvature and measurevalued torsion. As we previously observed, an explicit expression of \({\mathsf{U}}(L;0)\) in terms of the strain fields is available only in very special cases, limiting the scope of applicability of the general closure conditions.
Consistent with our construction of framed curves and with the emphasis on the curve geometry appropriate to these objects, we emphasize the importance of maintaining a clear distinction between framed curves and special Cosserat rods. Even though the two concepts have been successfully combined in the context of Kirchhoff rods (see, for instance, the papers of Langer and Singer [58] and Goriely and Tabor [59], the models for DNA reviewed by Swigon [60], and the recent contributions by Kawakubo [61]), this was possible because relatively parallel adapted frames were correctly used to describe the geometry of the base curve, while an additional material frame was employed to keep track of the mechanical twist. Nevertheless, we have shown that the geometry of the base curve is not a necessary starting point to define the shape of a rod, since the strain fields \(u_{i}\) and \(v_{i}\), \(i=1,2,3\), are the only degrees of freedom needed to characterize that shape. In summary, it seems more appropriate to explicitly use rod theory when dealing with mechanical models (as exemplified by the works of Domokos [62] and Domokos and Healey [63], and many others cited above) and framed curves when focusing on geometry (as exemplified by the works of Starostin and van der Heijden [64], Bohr and Markvorsen [65], da Silva [66], and Honda and Takahashi [67]).
We finally stress that, even in treatments of purely geometrical questions connected to space curves, the classical Frenet frame is not a suitable tool for two important reasons. First, whenever a curve has a straight portion, the curvature \(\kappa\) vanishes and the Frenet normal is not defined even if the geometric invariants \(\kappa\) and \(\tau\) are welldefined everywhere. Second, even when \(\kappa>0\) at all points of a curve, it may be no more than squareintegrable (with a measurevalued \(\tau\)), so that the corresponding Frenet frame could possibly be discontinuous. (Consider, for example, the properties of the base curve of a Möbius band, as described by Randrup and Røgen [68].) We thus see that the family of relatively parallel adapted frames, being uniquely determined by the geometric invariants of a regular curve and containing only globallydefined and continuous frames, should always be preferred over the Frenet frame. This, however, should not obscure the fact that \(\kappa\) and \(\tau\) are the true geometric invariants of the curve, while \(u_{1}\), \(u_{2}\), and \(v_{3}\) provide a convenient parametrization of the shape of the curve, having selected \(\boldsymbol {d}_{1}(0)\).
6 Conclusions
We have described how the essential degrees of freedom that encode the shape of a special Cosserat rod, namely those geometric features that are invariant under isometries of the threedimensional ambient space, correspond to a path traced in the special Euclidean algebra. The typical regularity of such path, relevant for physical applications, is that of a squareintegrable map. Being given directly in terms of the strain fields that underpin the elastic response of special Cosserat rods, this representation of shape is intrinsic to the description of the mechanical properties of such rods.
The Lie algebraic description of the rod shapes leads to an appealing discretization scheme that can be successfully applied to the analysis of shape relaxation problems under strongly nonlinear and nonconvex geometric constraints, such as the clampedends and closure requirements.
We have recovered the notion of a framed curve as a Cosserat rod with pointlike cross sections. That degeneracy is reflected on the actual degrees of freedom of the system. From this standpoint, we have highlighted the essential difference between rods and framed curves, and we have clarified why the family of relatively parallel adapted frames is not suitable for describing the mechanics of rods but it is the appropriate tool for dealing with the geometry of curves.
Notes
Funding
The authors acknowledge support from the Okinawa Institute of Science and Technology Graduate University with subsidy funding from the Cabinet Office, Government of Japan.
References
 1.Antman, S.S.: Nonlinear Problems of Elasticity, 2nd edn. Applied Mathematical Sciences, vol. 107. Springer, New York (2005) zbMATHGoogle Scholar
 2.Cosserat, E., Cosserat, F.: Sur la statique de la ligne déformable. C. R. Acad. Sci. Paris 145, 1409–1412 (1907) zbMATHGoogle Scholar
 3.Cosserat, E., Cosserat, F.: Théorie des Corps Déformable. Hermann, Paris (1909) zbMATHGoogle Scholar
 4.Schuricht, F.: Global injectivity and topological constraints for spatial nonlinearly elastic rods. J. Nonlinear Sci. 12(5), 423–444 (2002) ADSMathSciNetCrossRefzbMATHGoogle Scholar
 5.Giusteri, G.G., Lussardi, L., Fried, E.: Solution of the Kirchhoff–Plateau problem. J. Nonlinear Sci. 27(3), 1043–1063 (2017) ADSMathSciNetCrossRefzbMATHGoogle Scholar
 6.Simo, J.C., Marsden, J.E., Krishnaprasad, P.S.: The Hamiltonian structure of nonlinear elasticity: the material and convective representations of solids, rods, and plates. Arch. Ration. Mech. Anal. 104(2), 125–183 (1988) MathSciNetCrossRefzbMATHGoogle Scholar
 7.Simo, J.C., Posbergh, T.A., Marsden, J.E.: Stability of coupled rigid body and geometrically exact rods: block diagonalization and the energymomentum method. Phys. Rep. 193(6), 279–360 (1990) ADSMathSciNetCrossRefGoogle Scholar
 8.Holm, D.D., Noakes, L., Vankerschaver, J.: Relative geodesics in the special Euclidean group. Proc. R. Soc. Lond., Ser. A, Math. Phys. Eng. Sci. 469(2158), 20130297 (2013), 21 ADSMathSciNetCrossRefzbMATHGoogle Scholar
 9.Eldering, J., Vankerschaver, J.: A distance on curves modulo rigid transformations. Differ. Geom. Appl. 36, 149–164 (2014) MathSciNetCrossRefzbMATHGoogle Scholar
 10.Holm, D.D., Ivanov, R.I.: Matrix Gstrands. Nonlinearity 27(6), 1445–1469 (2014) ADSMathSciNetCrossRefzbMATHGoogle Scholar
 11.Bertails, F., Audoly, B., Cani, M.P., Querleux, B., Leroy, F., Lévêque, J.L.: Superhelices for predicting the dynamics of natural hair. ACM Trans. Graph. 25, 1180–1187 (2006) CrossRefGoogle Scholar
 12.Bishop, R.L.: There is more than one way to frame a curve. Am. Math. Mon. 82(3), 246–251 (1975) MathSciNetCrossRefzbMATHGoogle Scholar
 13.Antman, S.S., Schuricht, F.: The critical role of the base curve for the qualitative behavior of shearable rods. Math. Mech. Solids 8(1), 75–102 (2003) MathSciNetCrossRefzbMATHGoogle Scholar
 14.Naghdi, P.M.: On the formulation of contact problems of shells and plates. J. Elast. 5(3–4), 379–398 (1975) MathSciNetCrossRefzbMATHGoogle Scholar
 15.Hartman, P.: Ordinary Differential Equations, 2nd edn. Birkhäuser, Boston (1982) zbMATHGoogle Scholar
 16.Murray, R.M., Li, Z., Sastry, S.S.: A Mathematical Introduction to Robotic Manipulation. CRC Press, Boca Raton (1994) zbMATHGoogle Scholar
 17.Agoston, M.K.: Computer Graphics and Geometric Modelling: Mathematics. Springer, New York (2005) zbMATHGoogle Scholar
 18.Sander, O.: Geodesic finite elements for Cosserat rods. Int. J. Numer. Methods Eng. 82(13), 1645–1670 (2010) MathSciNetzbMATHGoogle Scholar
 19.Chirikjian, G.S.: Group theory and biomolecular conformation: I. Mathematical and computational models. J. Phys. Condens. Matter 22(32), 323103 (2010) CrossRefGoogle Scholar
 20.Sonneville, V., Cardona, A., Brüls, O.: Geometric interpretation of a nonlinear beam finite element on the Lie group \(\mathit{SE}(3)\). Arch. Mech. Eng. 61(2), 305–329 (2014) CrossRefzbMATHGoogle Scholar
 21.Sonneville, V., Cardona, A., Brüls, O.: Geometrically exact beam finite element formulated on the special Euclidean group \(\mathit{SE}(3)\). Comput. Methods Appl. Mech. Eng. 268, 451–474 (2014) ADSMathSciNetCrossRefzbMATHGoogle Scholar
 22.Cao, D.Q., Liu, D., Wang, C.H.T.: Threedimensional nonlinear dynamics of slender structures: Cosserat rod element approach. Int. J. Solids Struct. 43(3–4), 760–783 (2006) CrossRefzbMATHGoogle Scholar
 23.Spillmann, J., Teschner, M.: CoRdE: Cosserat Rod Elements for the dynamic simulation of onedimensional elastic objects. In: Proceedings of the 2007 ACM SIGGRAPH/Eurographics Symposium on Computer Animation, SCA ’07, AirelaVille, Switzerland pp. 63–72 (2007). Eurographics Association Google Scholar
 24.Bergou, M., Wardetzky, M., Robinson, S., Audoly, B., Grinspun, E.: Discrete elastic rods. ACM Trans. Graph. 27(3), 63 (2008) CrossRefGoogle Scholar
 25.Bergou, M., Audoly, B., Vouga, E., Wardetzky, M., Grinspun, E.: Discrete viscous threads. ACM Trans. Graph. 29(4), 116 (2010) CrossRefGoogle Scholar
 26.Audoly, B., Clauvelin, N., Brun, P.T., Bergou, M., Grinspun, E., Wardetzky, M.: A discrete geometric approach for simulating the dynamics of thin viscous threads. J. Comput. Phys. 253, 18–49 (2013) ADSMathSciNetCrossRefzbMATHGoogle Scholar
 27.Jung, P., Leyendecker, S., Linn, J., Ortiz, M.: A discrete mechanics approach to the Cosserat rod theory—Part 1: static equilibria. Int. J. Numer. Methods Eng. 85(1), 31–60 (2011) MathSciNetCrossRefzbMATHGoogle Scholar
 28.Lang, H., Linn, J., Arnold, M.: Multibody dynamics simulation of geometrically exact Cosserat rods. Multibody Syst. Dyn. 25(3), 285–312 (2011) MathSciNetCrossRefzbMATHGoogle Scholar
 29.Linn, J.: Discrete kinematics of Cosserat rods based on the difference geometry of framed curves. In: Proceedings of the 4th Joint International Conference on Multibody System Dynamics, Montréal, Canada (2016). URL: https://www.researchgate.net/publication/303941908 Google Scholar
 30.Simo, J.C., VuQuoc, L.: A threedimensional finitestrain rod model. Part II: computational aspects. Comput. Methods Appl. Mech. Eng. 58(1), 79–116 (1986) ADSCrossRefzbMATHGoogle Scholar
 31.Borri, M., Bottasso, C.: An intrinsic beam model based on a helicoidal approximation—Part I: formulation. Int. J. Numer. Methods Eng. 37(13), 2267–2289 (1994) CrossRefzbMATHGoogle Scholar
 32.Ibrahimbegović, A.: On finite element implementation of geometrically nonlinear Reissner’s beam theory: threedimensional curved beam elements. Comput. Methods Appl. Mech. Eng. 122(1), 11–26 (1995) ADSCrossRefzbMATHGoogle Scholar
 33.Betsch, P., Steinmann, P.: Frameindifferent beam finite elements based upon the geometrically exact beam theory. Int. J. Numer. Methods Eng. 54(12), 1775–1788 (2002), 8 MathSciNetCrossRefzbMATHGoogle Scholar
 34.Meier, C., Popp, A., Wall, W.A.: An objective 3D large deformation finite element formulation for geometrically exact curved Kirchhoff rods. Comput. Methods Appl. Mech. Eng. 278, 445–478 (2014) ADSMathSciNetCrossRefGoogle Scholar
 35.Meier, C., Popp, A., Wall, W.A.: A lockingfree finite element formulation and reduced models for geometrically exact Kirchhoff rods. Comput. Methods Appl. Mech. Eng. 290, 314–341 (2015) ADSMathSciNetCrossRefGoogle Scholar
 36.Gaćeša, M., Jelenić, G.: Modified fixedpole approach in geometrically exact spatial beam finite elements. Finite Elem. Anal. Des. 99, 39–48 (2015) MathSciNetCrossRefGoogle Scholar
 37.Bauer, A.M., Breitenberger, M., Philipp, B., Wüchner, R., Bletzinger, K.U.: Nonlinear isogeometric spatial Bernoulli beam. Comput. Methods Appl. Mech. Eng. 303, 101–127 (2016) ADSMathSciNetCrossRefGoogle Scholar
 38.Yilmaz, M., Omurtag, M.H.: Large deflection of 3D curved rods: an objective formulation with principal axes transformations. Comput. Struct. 163, 71–82 (2016) CrossRefGoogle Scholar
 39.Zupan, E., Zupan, D.: Velocitybased approach in nonlinear dynamics of threedimensional beams with enforced kinematic compatibility. Comput. Methods Appl. Mech. Eng. 310, 406–428 (2016) ADSMathSciNetCrossRefGoogle Scholar
 40.Zupan, D., Saje, M.: Finiteelement formulation of geometrically exact threedimensional beam theories based on interpolation of strain measures. Comput. Methods Appl. Mech. Eng. 192(49–50), 5209–5248 (2003) ADSMathSciNetCrossRefzbMATHGoogle Scholar
 41.Zupan, D., Saje, M.: The linearized threedimensional beam theory of naturally curved and twisted beams: the strain vectors formulation. Comput. Methods Appl. Mech. Eng. 195(33–36), 4557–4578 (2006) ADSCrossRefzbMATHGoogle Scholar
 42.Češarek, P., Saje, M., Zupan, D.: Dynamics of flexible beams: finiteelement formulation based on interpolation of strain measures. Finite Elem. Anal. Des. 72, 47–63 (2013) MathSciNetCrossRefzbMATHGoogle Scholar
 43.Su, W., Cesnik, C.E.S.: Strainbased geometrically nonlinear beam formulation for modeling very flexible aircraft. Int. J. Solids Struct. 48(16–17), 2349–2360 (2011) CrossRefGoogle Scholar
 44.Schröppel, C., Wackerfuß, J.: Introducing the Logarithmic finite element method: a geometrically exact planar Bernoulli beam element. Adv. Model. Simul. Eng. Sci. 3(1), 1–42 (2016) CrossRefGoogle Scholar
 45.Kirchhoff, G.: Ueber das Gleichgewicht und die Bewegung eines unendlich dünnen elastischen Stabes. J. Reine Angew. Math. 56, 285–313 (1859) MathSciNetCrossRefGoogle Scholar
 46.Dill, E.H.: Kirchhoff’s theory of rods. Arch. Hist. Exact Sci. 44(1), 1–23 (1992) MathSciNetCrossRefzbMATHGoogle Scholar
 47.Kaji, S., Ochiai, H.: A concise parametrization of affine transformation. SIAM J. Imaging Sci. 9(3), 1355–1373 (2016) MathSciNetCrossRefzbMATHGoogle Scholar
 48.Romero, I.: A comparison of finite elements for nonlinear beams: the absolute nodal coordinate and geometrically exact formulations. Multibody Syst. Dyn. 20(1), 51–68 (2008) MathSciNetCrossRefzbMATHGoogle Scholar
 49.Bauchau, O.A., Han, S.: Interpolation of rotation and motion. Multibody Syst. Dyn. 31(3), 339–370 (2014) MathSciNetCrossRefzbMATHGoogle Scholar
 50.Hairer, E., Lubich, C., Wanner, G.: Geometric Numerical Integration, 2nd edn. Springer Series in Computational Mathematics, vol. 31. Springer, Berlin (2006) zbMATHGoogle Scholar
 51.Kress, R.: Linear Integral Equations, 3rd edn. Applied Mathematical Sciences, vol. 82. Springer, New York (2014) CrossRefzbMATHGoogle Scholar
 52.Tricomi, F.G.: Integral Equations. Dover, New York (1985) Google Scholar
 53.Hasimoto, H.: A soliton on a vortex filament. J. Fluid Mech. 51(3), 477–485 (1972) ADSMathSciNetCrossRefzbMATHGoogle Scholar
 54.Efimov, N.V.: Some problems in the theory of space curves. Usp. Mat. Nauk 2, 193–194 (1947) Google Scholar
 55.Fenchel, W.: On the differential geometry of closed space curves. Bull. Am. Math. Soc. 57, 44–54 (1951) MathSciNetCrossRefzbMATHGoogle Scholar
 56.Schmeidler, W.: Notwendige und hinreichende Bedingungen dafür, dass eine Raumkurve geschlossen ist. Arch. Math. (Basel) 7, 384–385 (1956) MathSciNetCrossRefzbMATHGoogle Scholar
 57.Hwang, C.C.: A differentialgeometric criterion for a space curve to be closed. Proc. Am. Math. Soc. 83(2), 357–361 (1981) MathSciNetCrossRefzbMATHGoogle Scholar
 58.Langer, J., Singer, D.A.: Lagrangian aspects of the Kirchhoff elastic rod. SIAM Rev. 38(4), 605–618 (1996) MathSciNetCrossRefzbMATHGoogle Scholar
 59.Goriely, A., Tabor, M.: The nonlinear dynamics of filaments. Nonlinear Dyn. 21(1), 101–133 (2000) MathSciNetCrossRefzbMATHGoogle Scholar
 60.Swigon, D.: The mathematics of DNA structure, mechanics, and dynamics. In: Benham, C.J., Harvey, S., Olson, W.K., Sumners, D., Swigon, D. (eds.) Mathematics of DNA Structure, Function and Interactions. IMA Vol. Math. Appl., vol. 150, pp. 293–320. Springer, New York (2009) CrossRefGoogle Scholar
 61.Kawakubo, S.: Kirchhoff elastic rods in threedimensional space forms. J. Math. Soc. Jpn. 60(2), 551–582 (2008) MathSciNetCrossRefzbMATHGoogle Scholar
 62.Domokos, G.: A grouptheoretic approach to the geometry of elastic rings. J. Nonlinear Sci. 5(6), 453–478 (1995) ADSMathSciNetCrossRefzbMATHGoogle Scholar
 63.Domokos, G., Healey, T.: Hidden symmetry of global solutions in twisted elastic rings. J. Nonlinear Sci. 11(1), 47–67 (2001) ADSMathSciNetCrossRefzbMATHGoogle Scholar
 64.Starostin, E.L., van der Heijden, G.H.M.: Characterisation of cylindrical curves. Monatshefte Math. 176(3), 481–491 (2015) MathSciNetCrossRefzbMATHGoogle Scholar
 65.Bohr, J., Markvorsen, S.: Autorotation. Phys. Scr. 91(2), 023005 (2016) ADSCrossRefGoogle Scholar
 66.da Silva, L.C.B.: Moving frames and the characterization of curves that lie on a surface. J. Geom. (2017). doi: 10.1007/s0002201703987 MathSciNetzbMATHGoogle Scholar
 67.Honda, S., Takahashi, M.: Framed curves in the Euclidean space. Adv. Geom. 16(3), 265–276 (2016) MathSciNetCrossRefzbMATHGoogle Scholar
 68.Randrup, T., Røgen, P.: Sides of the Möbius strip. Arch. Math. (Basel) 66(6), 511–521 (1996) MathSciNetCrossRefzbMATHGoogle Scholar
Copyright information
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.