IMA Journal of Applied Mathematics (2011) 76, 57−66

doi:10.1093/imamat/hxq065
Advance Access publication on December 31, 2010

The stiffness of tensegrity structures

S. D. G UEST
Department of Engineering, University of Cambridge, Trumpington Street,
Cambridge CB2 1PZ, UK

Downloaded from http://imamat.oxfordjournals.org/ at University of Prince Edward Island on November 8, 2014

sdg@eng.cam.ac.uk
[Received on 12 March 2010; revised on 8 July 2010; accepted on 16 November 2010]

The stiffness of tensegrity structures comes from two sources: the change of force carried by members as
their length is changed, and the reorientation of forces as already stressed members are rotated. For any
particular tensegrity, both sources of stiffness may have a critical role to play. This paper explores how
the stiffness of two example tensegrity structures changes as the level of prestress in a member varies. It is
shown that, for high levels of prestress, an originally stable tensegrity can be made to have zero stiffness
or indeed be made unstable.

Keywords: tensegrity; prestress; tangent stiffness.

1. Introduction
Tensegrities form remarkable structures. They are frequently visually arresting (Heartney, 2009); and
they can be designed to give ‘optimal’ structures (Masic et al., 2006). The present paper will discuss the
stiffness of tensegrity structures—the first-order change of force carried as a structure is deformed. There
are two competing sources for a tensegrity’s stiffness, and the balance between these sources changes
as the prestress varies. Thus, for instance, the paper will show that, for a particular ‘stable’ tensegrity,
increasing a low level of prestress will increase the stiffness; while for a high level of prestress, a further
increase in prestress may reduce the stiffness and even lead to a structure with zero or negative stiffness.
The definition of ‘tensegrity’ is a subject of debate (see, e.g. Motro, 2003). At one extreme is the
mathematical definition (Roth & Whiteley, 1981; Connelly & Whiteley, 1996) that a tensegrity is a
structure consisting of ‘cables’ (members only able to resist tension), ‘struts’ (members only able to
resist compression, e.g. a contact force) and ‘bars’ (members able to resist tension and compression.
Others might insist that a tensegrity must have compression members that do not touch or must have
an infinitesimal mechanism. The present paper will not enter the debate on definition, except to note
that the basic formulation used here is valid for any prestressed structure, and that at least one of
the tensegrities used as examples satisfies even the most stringent definition of tensegrity.
By way of example, the paper will show results for the two tensegrity structures shown in Fig. 1.
Tensegrity A is the classic example described by Pugh (1976) as the ‘expanded octahedron’ tensegrity.
It consists of j = 12 nodes and b = 30 members, made up of 24 cables and 6 struts. Using an extended
Maxwell rule (Calladine, 1978) relating the number of infinitesimal mechanisms m and states of self-
stress s gives
m − s = 3 j − b − 6 = 0. (1)
Tensegrity A has Th symmetry (in the Schoenflies notation, see e.g. Altmann & Herzig, 1994), with
symmetry elements that consist of four three-fold axes, three two-fold axes and three planes of reflec-
tion. Symmetry defines the position of all nodes in terms of one reference node, shown in Fig. 2: for

c The Author 2010. Published by Oxford University Press on behalf of the Institute of Mathematics and its Applications. All rights reserved.
58 S. D. GUEST

Downloaded from http://imamat.oxfordjournals.org/ at University of Prince Edward Island on November 8, 2014

FIG. 1. Two tensegrities used as examples. Tensegrity A is the classic ‘expanded octahedron’ tensegrity, which has one state of
self-stress and one infinitesimal mechanism. Tensegrity B has the same arrangement of struts and cables as Tensegrity A, except
that it also has additional cables pulling pairs of nodes closer together; and Tensegrity B does not have an infinitesimal mechanism.

FIG. 2. The coordinates of a single node for both Tensegrity

√ A when p = 1/2 and Tensegrity B when p = 1/3. For comparison,
the vertices of a regular icosahedron have p = 2/(1 + 5) = 0.618.

Tensegrity A to be prestressed (s 6= 0), the parameter p must take the value 0.5, which can be con-
firmed by simple statics, or the use of matrix methods, as described by Pellegrino & Calladine (1986)
and Pellegrino (1993). The presence of a state of self-stress guarantees, from (1), the existence of an in-
finitesimal mechanism (for this mode, to first order, nodal movement results in zero extension of every
member).
 THE STIFFNESS OF TENSEGRITY STRUCTURES 59

Tensegrity B is a variation on Tensegrity A, designed to not have an infinitesimal mechanism. Six

cables have been added between nodes so that the cable net forms the edges of an (irregular) icosahedron.
Thus, the structure consists of j = 12 nodes and b = 36 members, made up of 30 cables and 6 struts,
and the extended Maxwell rule gives

m − s = 3 j − b − 6 = −6. (2)

In fact, m = 0 is guaranteed in this case as the cable net alone forms the edges of a convex triangulated
polyhedra (Cauchy, 1813; Dehn, 1916), and the addition of six internal struts can only add to the states

Downloaded from http://imamat.oxfordjournals.org/ at University of Prince Edward Island on November 8, 2014

of self-stress to give s = 6. For Tensegrity B, the value p = 1/3 is chosen, as in this case, the totally
symmetric state of self-stress (which could be found by, e.g. the methods described by Kangwai &
Guest, 2000) has equal tension coefficients (tension/length) in all cables.
Both Tensegrities A and B are ‘super-stable’, in the sense defined by Connelly (1999). This means
that prestress properties could not be more benign—but despite this, it will be shown in Section 6 that
both structures can be made unstable for sufficiently high levels of prestress.
In this paper, for both Tensegrities A and B, it will be assumed that the struts are axially rigid but
the cables are axially flexible. For each tensegrity, two contrasting material properties for the cables
will be considered. Firstly, a set of stiff cables will be considered for which a typical graph of tension
against length is shown in Fig. 3(a). A key dimensionless parameter in the stiffness formulation used in
this paper, as described in Section 2, is the ratio of the tension coefficient, tˆ = t/l to the axial stiffness
g = dt/dl,
t/l
ε= . (3)
dt/dl
Locally to the working point of the cable, we can define a Young’s modulus E for the material, and
a cross-sectional area of the cable A, so that the axial rigidity is g = AE/l. Thus, we can define the
parameter ε as a nominal strain
t
ε= . (4)
AE
For metallic cables, the slope g will be essentially linear before yield, and hence for such stiff cables, ε
must be less than the yield strain and consequently ε  1. In Section 3, we will assume ε = 0.01.
A contrasting set of material properties will also be considered, when the cables are compliant, being
made of e.g. rubber or some other elastomer. For these materials, a typical graph of tension against length
is shown in Fig. 3(b). Now, the dimensionless parameter ε is not limited to being much less than one. In
Section 4, a value of 0.6 will be assumed.

2. Stiffness formulation
The basic stiffness formulation that will be used is described in Guest (2006); and an identical formu-
lation with an alternative notation is described in Skelton & de Oliveira (2009). The tangent stiffness
matrix K relates, to first order, the displacements at each of the twelve nodes in the x-, y- and
z-directions, written as a vector d, to the applied load at each of the nodes, written as a vector p,

Kd = p. (5)

The matrix K depends on the configuration of the structure, the axial stiffness of the members (the slope
g = dt/dl shown in Fig. 3) and the tension coefficient carried by the members (the slope tˆ = t/l shown
60 S. D. GUEST

Downloaded from http://imamat.oxfordjournals.org/ at University of Prince Edward Island on November 8, 2014

FIG. 3. Material properties for two contrasting sets of cables. In (a), the cables are stiff, made of, for instance, steel, and t/l 
dt/dl. In (b), the cables are assumed to be compliant, made of, in this instance, an elastomer where t/l 6 dt/dl. Note that, for
a cable with the same cross-sectional area, the values of force in (a) are likely to be very much higher than those in (b), as the
Young’s modulus of a steel will be the order of a hundred times stiffer than the Young’s modulus of an elastomer.

in Fig. 3). It can be written as

K = AĜA> + S. (6)

In (6), A is the ‘equilibrium matrix’ for the structure—a matrix of direction cosines describing the equi-
librium relationship between internal forces in the members t and applied loads at nodes p, At = p
(Pellegrino, 1993); A> equivalently describes the first-order kinematic relationship between the dis-
placement of nodes d and the extensions of members e. Ĝ is a diagonal matrix of modified axial stiff-
nesses, with an entry for each member i (1 6 i 6 b),

ĝi = gi − tˆi , (7)

which can be written in terms of the nominal strain for the member, εi , as

ĝi = gi (1 − εi ). (8)

S is the (large) ‘stress matrix’ for the structure. S can be written as the Kronecker product of a ‘small’
or ‘reduced’ stress matrix Ω and a 3D identity matrix I

S=Ω ⊗I (9)

and the coefficients of the small stress matrix are given by


P −tˆi, j = −tˆj,i if i 6= j, and {i,j} a member,
Ωi j = tˆik if i = j, (10)
 k6=i
0 if there is no connection between i and j.

In this formulation, tˆi, j is the tension coefficient (t/l) in the member that runs between nodes i and j
(there will be a unique mapping between the pair (i, j) and the bar numbering described above for Ĝ).
It was shown in Schenk et al. (2007) that, for a self-stressed structure, the stress matrix S must have a
 THE STIFFNESS OF TENSEGRITY STRUCTURES 61

Downloaded from http://imamat.oxfordjournals.org/ at University of Prince Edward Island on November 8, 2014

FIG. 4. The two modes considered in detail in Sections 3, 4 and 5. The modes are shown for Tensegrity A, but almost identical
modes can be defined for Tensegrity B. (a) Mode 1 is the infinitesimal mechanism for Tensegrity A. (b) Mode 2 is a shear mode,
with the structure shearing in the x–z plane; two other identical modes in the x–y and x–z planes exist.

nullity of at least 12 (and, it turns out, exactly 12 for both Tensegrities A and B) and can provide no
stiffness to any of the 6 rigid body modes or to any of the 6 ‘affine deformation’ modes, i.e. modes in
which the body is deformed uniformly by stretching or shear.
For both Tensegrities A and B, the stiffness matrix K is a symmetric matrix of dimension 36 × 36. In
fact, for the results reported here, a change of coordinates was used to condense out two sets of freedoms
from the original 36: the 6 rigid body modes, and the 6 modes that correspond to extension of the struts.
The effect of this is to leave a 24 × 24 symmetric matrix, for which the eigenvalues are then found.
These 24 eigenvalues have at most 10 distinct values (as can be predicted by a symmetry analysis of the
original system, as described by Kangwai et al., 1999). Condensing out the six modes corresponding to
extensions of struts essentially makes the assumption that the struts are rigid. An alternative procedure
would have been to have given the struts a stiffness of, say, 1000 times the value of the stiffness of the
cables, and worked with the original 36 × 36 matrix. This would have given essentially the same set
of 24 eigenvalues, plus an additional 6 eigenvalues which are approximately 1000 times as large, and 6
zero eigenvalues corresponding to rigid body modes.
The complete set of eigenvalues for varying levels of prestress will be reported in Section 6, but first,
the paper will concentrate on just two modes, shown in Fig. 4. Mode 1 corresponds to the infinitesimal
mechanism for Tensegrity A and is an eigenmode of the stiffness matrix for all levels of prestress. Mode
2 is a shear mode and is actually an eigenmode only for a prestress corresponding to a nominal strain ε = 1;
however, for other levels of prestress, results are reported for the eigenmode which is closest to this.

3. Stiff cables/low relative prestress

This section will consider the results that are appropriate for cases where the stiffness of the cables
g is much greater than the current tension coefficient, t/l, for instance, the case shown in Fig. 3(a).
This would be typical of tensegrities constructed with, e.g. steel cables. We will assume a nominal
62 S. D. GUEST

TABLE 1 Eigenvalues of the stiffness matrix for two levels

of prestress for the two tensegrities shown in Fig. 1. Results
are presented for the two modes shown in Fig. 4

All ×AE Tensegrity A Tensegrity B

(a) Low prestress, ε = 0.01
Mode 1 0.03 3.15
Mode 2 0.48 0.48

Downloaded from http://imamat.oxfordjournals.org/ at University of Prince Edward Island on November 8, 2014

(b) High prestress, ε = 0.6
Mode 1 1.96 4.97
Mode 2 0.34 0.40

strain ε = t/AE = 0.01 in the cables of Tensegrity A, and the 24 equivalent cables of Tensegrity
B (the additional cables in Tensegrity B have a lower level of tension for equilibrium, and hence a
lower nominal strain). In fact, a value of ε = 0.01 may be very large in these circumstances and would
correspond to a high tensile steel cable being stressed close to yield.
The results for the eigenvalues associated with Modes 1 and 2 are presented in Table 1(a).
For Tensegrity A, Mode 1 is the most flexible mode (has the smallest eigenvalue), which reflects the
fact that this is an infinitesimal mechanism, and to the first-order approximation of the stiffness matrix,
there is no change in the length of any member for this mode. Thus, the ‘material’ stiffness can contribute
nothing to the stiffness, and the stiffness is entirely generated from the reorientation of already stressed
members. The stiffness of Mode 1 is proportional to the level of prestress in the structure, which will be
clearly shown later in Fig. 6(a.i).
For Tensegrity B, Mode 1 is not the most flexible mode. The additional cables added, when compared
with Tensegrity A, have ensured that Mode 1 is no longer an infinitesimal mechanism, and the stiffness
of this mode is now far higher than the stiffness of the shear mode, Mode 2.

4. Compliant cables/high relative prestress

This section will consider the results that are appropriate for cases where the stiffness of the cables g
is of a similar order to the value of the current tension coefficient t/l, for instance, the case shown in
Fig. 3(b). This would be typical of tensegrities constructed with cables made of rubbers or other elas-
tomers or perhaps helically wound springs. Structures constructed in this way are unlikely to be used
for civil engineering structures but might be appropriate for highly compliant structures, e.g. tensegrity
robots (Aldrich et al., 2003; Mirats Tur & Hernàndez Juan, 2009) or tensegrity springs (Azadi et al.,
2009). Furthermore, demonstration models are often constructed with elastomeric cables (Pugh, 1976;
Connelly & Back, 1998). We will assume a nominal strain ε = t/AE = 0.6 in the cables of Tenseg-
rity A, and the 24 equivalent cables of Tensegrity B.
The results for the eigenvalues associated with Modes 1 and 2 are presented in Table 1. For both
Tensegrities A and B, the shear mode, Mode 2, is now the most flexible mode. Now, the infinitesimal
mechanism no longer dominates the behaviour of Tensegrity A, and the stiffness of this mode, Mode 1,
is not as markedly different between Tensegrities A and B.

5. Zero-free-length cables
An extreme value of prestress is considered in this Section. When springs are wound helically, it is pos-
sible for them to be wound with pretension, where the coils of the spring are pressed against themselves
 THE STIFFNESS OF TENSEGRITY STRUCTURES 63

Downloaded from http://imamat.oxfordjournals.org/ at University of Prince Edward Island on November 8, 2014

FIG. 5. Material properties for a ‘zero-free-length’ spring. The spring is prestressed when coiled at length l0 . Initially, as tension
is applied, this prestress is removed at approximately constant length. Then, within the working range, the spring has a tension
proportional to its length, and hence tˆ = t/l = dt/dl = g.

when the spring is not loaded. For the correct level of pretension, the spring can be wound so that it has
the tension/length properties shown in Fig. 5: such springs are commonly used for static balancing, see,
e.g. French & Widden (2000) and Herder (2001).
If zero-free-length springs were used for either Tensegrity A or Tensegrity B, then the resultant
structure has a zero-stiffness mode. For this case, ε = 1 for all cables, and (neglecting the rigid struts)
Ĝ = 0. Thus, in the formulation given in (6), the term AĜA> is zero, and only the stiffness resulting
from the stress matrix S remains. However, this can provide no stiffness for shear modes, and thus,
Mode 2 has zero stiffness. Furthermore, the results of Schenk et al. (2007) show that this is not just a
local phenomenon, and the structure could be deformed without limit, without any load being applied.
In practice, of course, friction, and the limitations of the working length of the springs, will become
important (Schenk et al., 2006).
Note that the existence of a zero-stiffness shear mode for ε = 1 cannot be generalized to all tenseg-
rities, as it depends critically on the orientation of the rigid struts. This is further discussed in Schenk
et al. (2007).

6. Discussion
Sections 3, 4 and 5 have shown how the dominant (softest) modes of Tensegrities A and B change as
the relative level of prestress changes, described by the nominal strain ε. A complete overview of the
stiffness changes is provided by the plots of the eigenvalues of the stiffness matrix for 0 6 ε 6 1 given
in Fig. 6. In these plots, the eigenvalues have been normalized in two ways. In (a), the stiffness of the
material remains constant, and ε is changed by varying the tension, t; however, to compare different
materials, it may be more realistic to consider (b), where the tension t carried by the cables is held
constant, and ε is changed by varying cable stiffness AE.
Figure 6 clearly shows that, for Tensegrity A, the most flexible mode for low values of ε is the
infinitesimal mechanism shown as Mode 1, while for high values of ε, the most flexible mode is the
shear mode shown as Mode 2. In Tensegrity A, all cables are symmetrically equivalent, and so this
example shows in a particularly clean way what we would expect to see for every tensegrity. For small
64 S. D. GUEST

Downloaded from http://imamat.oxfordjournals.org/ at University of Prince Edward Island on November 8, 2014

FIG. 6. The complete set of eigenvalues λ of the stiffness matrix K for Tensegrities A (a.i and a.ii) and B (b.i and b.ii) for
varying values of ε = t/ AE. The eigenvalues for Modes 1 and 2 are shown bold. For (a.i) and (b.i), the results are presented as
λ × L/E A, i.e. the stiffness of the cables is preserved as level of prestress varies. For (a.ii) and (b.ii), the results are presented as
λ × L/E A × 1/ε = λ × L/t, i.e. the tension in the cables is preserved as the cable stiffness varies. Assuming that the struts are
rigid, and neglecting rigid-body modes, there are 24 eigenvalues in each plot, although symmetry ensures that there are at most
only ten distinct values.

values of ε, the key understanding of the structural behaviour comes about from understanding the
equilibrium of the structure and the material properties, and hence the ‘material’ stiffness AGA> , where
G = Ĝ for t = 0. By contrast, for large values of ε, the understanding of the stiffness that comes from
the stress matrix S is key, i.e. it is dominated by the stiffness that results from the reorientation of already
stressed members.
If we were to extend the graphs in Fig. 6 for ε > 1, we can see that both Tensegrities A and B would
have a stiffness matrix with a negative eigenvalue, and hence even these super-stable tensegrities can be
made unstable.
It should be noted that the results in Fig. 6 are actually not valid for ε = 0, except for the zero value
of the eigenvalue corresponding to Mode 1 for Tensegrity A. For any other mode of deformation, the
 THE STIFFNESS OF TENSEGRITY STRUCTURES 65

calculation assumes some cables will go into compression, when in reality they would become slack.
Different cables will go slack for different modes, and there is no longer a consistent tangent stiffness
at this point. However, Roth & Whiteley (1981) show that all of these deformations will in fact have a
positive stiffness.

7. Conclusion
The present paper shows that the stiffness of structures, and in particular tensegrity structures, depends

Downloaded from http://imamat.oxfordjournals.org/ at University of Prince Edward Island on November 8, 2014

not only on the connectivity, geometry and material properties of the structure but also upon the level
of prestress that the structure carries, with the dimensionless parameter ε playing a key descriptive role.
The key novelty of the present paper is to introduce the parameter ε and show how the stiffness of
two carefully chosen tensegrity structures vary as ε varies from 0 to 1. Traditional structural materials
require ε  1, but with the development of soft and compliant structures, choice of ε might be a critical
factor in design.

Acknowledgement
I would like to thank R. Pandia Raj for his help with plotting the pictures of tensegrities.

R EFERENCES
A LDRICH , J. B., S KELTON , R. E. & K REUTZ -D ELGADO , K. (2003) Control synthesis for a class of light and
agile robotic tensegrity structures. Proceeding IEEE American Control Conference, Denver, CO, vol. 6, pp.
5245–5251.
A LTMANN , S. L. & H ERZIG , P. (1994) Point-Group Theory Tables. Oxford: Clarendon Press.
A ZADI , M., B EHZADIPOUR , S. & FAULKNER , G. (2009) Antagonistic variable stiffness elements. Mech. Mach.
Theory, 44, 1746–1758.
C ALLADINE , C. R. (1978) Buckminster Fuller’s “Tensegrity” structures and Clerk Maxwell’s rules for the
construction of stiff frames. Int. J. Solids Struct., 14, 161–172.
C AUCHY, A. L. (1813) Recherche sur les polyèdres—premier mémoire. J. l’École Polytechnique, 9, 66–86.
C ONNELLY, R. (1999) Tensegrity structures: why are they stable? Rigidity Theory and Applications
(M. F. Thorpe & P. M. Duxbury eds). New York: Kluwer Academic, pp. 47–54.
C ONNELLY, R. & BACK , W. (1998) Mathematics and tensegrity. Am. Sci., 86, 142–151.
C ONNELLY, R. & W HITELEY, W. J. (1996) Second-order rigidity and prestress stability for tensegrity frameworks.
SIAM J. Discrete Math., 7, 453–491.
D EHN , M. (1916) Über die starreit konvexer polyeder. Math. Ann., 77, 466–473.
F RENCH , M. J. & W IDDEN , M. B. (2000) The spring-and-lever balancing mechanism, George Carwardine and
the Anglepoise lamp. Proc. Inst. Mech. Eng. C J. Mech. Eng. Sci., 214, 501–508.
G UEST, S. D. (2006) The stiffness of prestressed frameworks: a unifying approach. Int. J. Solids Struct., 43,
842–854.
H EARTNEY, E. (2009) Kenneth Snelson: Forces Made Visible. Manchester, VT: Hudson Hills Press LLC.
H ERDER , J. L. (2001) Energy-free systems. Theory, conception and design of statically balanced spring mecha-
nisms. Ph.D. Thesis, Delft University of Technology.
K ANGWAI , R. D. & G UEST, S. D. (2000) Symmetry-adapted equilibrium matrices. Int. J. Solids Struct., 37,
1525–1548.
K ANGWAI , R. D., G UEST, S. D. & P ELLEGRINO , S. (1999) An introduction to the analysis of symmetric struc-
tures. Comput. Struct., 71, 671–688.
M ASIC , M., S KELTON , R. E. & G ILL , P. E. (2006) Optimization of tensegrity structures. Int. J. Solids Struct., 43,
4687–4703.
66 S. D. GUEST

M IRATS T UR , J. M. & H ERN ÀNDEZ J UAN , S. (2009) Tensegrity frameworks: dynamic analysis review and open
problems. Mech. Mach. Theory, 44, 1–18.
M OTRO , R. (2003) Tensegrity Structural Systems for the Future. London: Kogan Page Science.
P ELLEGRINO , S. (1993) Structural computations with the singular value decomposition of the equilibrium matrix.
Int. J. Solids Struct., 30, 3025–3035.
P ELLEGRINO , S. & C ALLADINE , C. R. (1986) Matrix analysis of statically and kinematically indeterminate
frameworks. Int. J. Solids Struct., 22, 409–428.
P UGH , A. (1976) Introduction to Tensegrity. Berkeley, CA: University of California Press.

Downloaded from http://imamat.oxfordjournals.org/ at University of Prince Edward Island on November 8, 2014

ROTH , B. & W HITELEY, W. (1981) Tensegrity frameworks. Am. Math. Soc., 265, 419–446.
S CHENK , M., G UEST, S. D. & H ERDER , J. L. (2007) Zero stiffness tensegrity structures. Int. J. Solids Struct., 44,
6569–6583.
S CHENK , M., H ERDER , J. L. & G UEST, S. D. (2006) Design of a statically balanced tensegrity mechanism.
Proceedings of DETC/CIE 2006, ASME 2006 International Design Engineering Technical Conferences &
Computers and Information in Engineering Conference, 10–13 September 2006, Philadelphia, PA, vol 2,
pp. 501–511.
S KELTON , R. & DE O LIVEIRA , M. (2009) Tensegrity Systems. Dordrecht, The Netherlands: Springer.