Formfinding by the

“Direct Approach”
and Pertinent Strategies
for the Conceptual Design of
P re s tressed and Hanging
Structure s
K. Linkwitz
University of Stuttgart, Obere Tannenbergstr 24, D-71229 Leonberg, Germany
(Received 12th November 1998, revised version received 30th April 1999)

Contents
1. Some remarks regarding the process of formfinding for
prestressed and hanging structures.
1.1 Figures of equilibrium, models, and measurement of
models
1.2 Different (analytical) approaches to the problem
2. The Essence of the ‘Direct Formulation’ of the “Direct
Approach”
3. ‘Indirect Formulation’ of the “Direct Approach” using Concepts
of Graph Theory
3.1 Tools of Graph Theory
3.1.1 Topological Description of a Netlike Structure
3.1.2 Elementary Relationships
3.1.3 Towards a general solution of network analysis
3.2 Transfer of the generalised solution to prestressed netlike
structures
4. Comparison of Direct and Indirect Solution: Complete
Agreement
5. Interpretation, discussion, and consequent strategies
5.1Essential Properties of the Solution
5.2 Subsequent Strategies for Practical Applications
5.2.1 Limitations of the Unaltered Solution
5.2.2 Extension of the Basic Equations
5.2.3 Further Extension: The Approach via Least Squares
5.2.4 Least Squares and Additional Constraints
6. Appendix: Mathematical Notation
7. Literature

International Journal of Space Structures Vol. 14 No. 2 1999 73

 Formfinding by the "Direct Approach"

1. SOME REMARKS REGARDING THE Different approaches were chosen by different

PROCESS OF FORMFINDING FOR scientists/engineers for the same problem of
PRESTRESSED AND HANGING generating shapes of equilibrium, analysing them,
STRUCTURES. and determining the cutting patterns. In Great Britain
The geometrical shapes of prestressed and hanging M. Barnes used the method of “dynamic relaxation”,
structures can be interpreted and consequently be widely applied still to-day by British and US
described as ‘figures of equilibrium’ of the tension structural engineers. In Stuttgart almost
forces acting in them and the forces of the dead loads. simultaneously and entirely independently three new
Depending on the position of the exterior anchoring methods were discovered [2], ([4], [3], [7]),
points – at high masts and at low fixed fundaments – , especially under the impact of projects. J. Argyris and
the type of meshes/ and their orientation, and the his colleagues [3] expanded the method of finite
amount of prestress induced, an unlimited number of elements to prestressed cable-nets. G. Brinkmann’s
different shapes can be generated. Of these only a few [7] point of departure was the continuum mechanics
will find the approval of the scrutinising architect, analysis of thin films and their translation into net like
who would like the criteria of function and aesthetics structures, permitting direct analytical solutions –
to be fulfilled. The engineer – apart from conceptual even on a small computer – for nets of particular
design – must find a compromise between admissible types.
forces, uniformity in their distribution, properties of The author, deeply involved as a consulting
chosen materials, load-deflection behaviour of the engineer in the project “Olympic Roofs” Munich 1972
structure, and design of structural details. and pressured to fulfil an engineering contract
Typical for this type of structure, shape and forces awarded to him stipulating the determination of the
are intrinsically interwoven. The conventional design cutting pattern for all the roofs in due time, pondered
procedure, based on a mental picture of the shape about a new applicable solution to the problem, and
which then has to be translated to sketches and invented the “Direct Approach” [2], [4], deviating
subsequent working drawings using paper and pencil sharply from the classical way of thinking in structural
only, no longer holds. Shapes have to be taken from engineering. He based his ideas on the method of least
physical models or/and from interactively used squares as it has been used in physics and geodesy to
analytically working computer tools. evaluate redundant observations. He started from the
These facts make the design of such structures a fact – earlier already thoroughly investigated –
unique task, differing essentially from conventional LINKWITZ [1960] -that the figure and geometry of
designs that architects and engineers are accustomed an elastomechnical net correspond exactly to the
to. The classical development of a design via drawings figure and geometry of a geodetic network in space
– even when supported by CAD – is not valid. containing redundant observations. This approach was
Until about 1965, – the advent of analytical, also the start of departure for the discovery and
computer based methods – shapes of equilibrium formulation of the force density method [LINKWITZ,
could only be transformed into executable structures SCHEK 1971].
after approximating them through models. Gaudi The method passed its crucial test by providing the
(1852 – 1926) of Barcelona was one of the first to structural analysis and cutting pattern for the
apply this technique. In concrete shell construction prestressed nets of the stadium and the delicate
Heinz Isler of Switzerland achieved a decisive break intermediate parts of the Olympic buildings and the
through by using forms derived not only from models cutting patterns also for remaining buildings.
but especially from a variety of physical experiments It then found its scientific home in the University of
involving forms generated naturally in hanging wet Stuttgart laboratory of the author in the context of the
and then frozen cloth. two Special Research Fields 64, 230 from 1972 –
Up to the middle of the sixties it was considered 1995. The progress in these further investigations is
impossible to directly generate figures and shapes of reflected in two Habilitation Thesis [11], [21], 9
equlibrium with a computer, notwithstanding the fact, Doctor Thesis [9], [14], [15], [17], [25], [26], [30],
that already mainframes with high performance were [33], [34], a Diploma Thesis [20] showing already the
available. Adequate computer methods only extension of the method of force densities to
developed in the late sixties, when engineers were membranes, and many advanced students’ works. It
challenged by new large projects. was also applied in a number of projects.

74 International Journal of Space Structures Vol. 14 No. 2 1999

K. Linkwitz

2. THE ESSENCE OF THE ’DIR E C T In both forms written the elastic elongation of a bar
F OR MUL AT I O N’ OF THE “DIRECT ei is a function of its slack length li, the area of the cross
AP P ROACH” section Fi, the stiffness of the material Ei, and the
The basic equations of the “Direct Approach” are tension force Si. Reversibly the tension force Si may be
characterised by the fact, that they describe directly what formulated as a function of slack length, area of cross
is visible when looking contemplatively at a whole or at sections, stiffness of material, and elastic elongation.
a part of a prestressed net, remembering simultaneously
elementary rules of applied mechanics. 2 Compatibility between geometry of the whole
To begin with, we formulate our prerequisites (and net and the geometry of each individual bar: In
limitations!): the net under prestress each individual
elastically extended bar has exactly the length
1. A prestressed net is assumed to be discretised as
as is the distance between those adjacent nodes
a pin-jointed net of individual bars subjected to
between which it is spanned: There is
tension forces; i.e. no bending and no sagging of
compatibility throughout between the lengths
the individual elements due to dead loads exists.
of the stress-extended bars and the distances
In the overwhelming number of practical
between appertaining nodes. This we can
applications this assumption describes reality
describe by using Pythagoras’ law for each bar
sufficiently closely.
with the equation
2. The law of tension-force/bar-extension behaviour
of all bars is known, be it linear or non-linear; but
no plastic deformations occur: Without restriction (2)
to generalisation we assume here – for the
moment and due to lack of space for a more
extended development – Hooke’s linear law to be
applicable.
Regarding an individual node and its four neighbours and, regarding the whole net, with the matrix equation.
in a net under its due prestress we can formulate three
essential equations governing each node of the net.
(l+e) = w = f(x) =(2)
1. Under the impact of the tension force acting in
it, each individual bar is extended with respect to
3. Equilibrium at each node: At each node there must
its slack length (the length to which the bar was
be equilibrium between the four tension forces acting
pre-cut at assembly time before prestressing the
in the four (in the boundary zone three) attached bars
net). The difference of extended and slack length
and the exterior or dead load acting upon the node.
is determined by the arbitrary ‘Material Law’.
Mathematically formulated this means the sum of the
Assuming for the moment Hooke’s Law (vide
four tension vectors + the downwards pointing vector
above) we may write it in the two forms.
of deadloads and exterior load must be zero.
Decomposing these vectors into their x-, y-, and z-
components we have three equations of equilibrium at
each point, namely:

(3)

Again, regarding the whole net, and using matrices

(1) we can write these equations of equilibrium in the
form:
(The left part of equ. (1) is the formulation for an individual bar, whereas the (=3)
right hand part, using matrix notation, is the formulation for the whole net.)

International Journal of Space Structures Vol. 14 No. 2 1999 75

 Formfinding by the "Direct Approach"

Now we have to formulate the above three basic 3. THE ‘INDIRECT FORMULATION’ OF
facts in one equation by appropriate substitutions. THE ST UTTGART APPROACH USING
First, we substitute the forces in (3) by the GRAPH THEORY
equivalent elastic elongations following from the
material law (1)
3.1 Some excerpts from graph theory
3.1.1 Topological description of a network
s = H*L–1* e We may start by regarding again the form of equations
(2): the coordinate-components x-,y-,z have a single
Then we substitute the elastic elongations by their index, corresponding to the “point-number” of the
expression (2) in the compatibility equations. node, whereas we have written the bars and their
elongations with a double index, namely the “point-
numbers” of the ends of each bar.
e = f(x)–1 We gain considerable advantages in our further
analysis if we use a different way of indexing,
As a result we get the equivalent and unique avoiding double indexes altogether, by assigning one
equations of equilibrium for the whole net unique number to each node and to each bar. The
“topology” of a net thus indexed is exhaustively
described by the incidence matrix of graph theory, as
(4) is shown in the following small example:
The ‘topology’ of the small net in fig 1 consists of
6 nodes and 7 branches . It can be described digitally
If the net has h free nodes, then (4) is a (highly
in the branch-node list of Fig. 2
non-linear) system of equations with 3h unknown
coordinate components, describing thus the shape
of the net completely. We can also interpret (4) as
the equations of the appertaining “figure of
equilibrium”.
To arrive at “practical solutions” we can proceed
further, always using equations (4) as the point of
departure, in different ways.
With regard to “practical solutions” we have to
keep in mind two essential points of view Figure 1. Graph of 6 nodes and 7 branches

1. We have to linearise (4) in an appropriate

manner to come to numerical solutions

2. We can – and have to – interpret (4) differently

if we want to arrive at “practical solutions”
allowing certain strategies in the process of
formfinding.

We will deal with both aspects further down. Figure 2. List of branches and nodes
Especially to enrich the potential of interpreting (4)
in different ways we choose an approach equivalent
in each respect to the above equations (1) ... (4), but
using basic equations from graph theory to describe
the netstructure. This is also of great practical
impact, because this type of description leads to most
useful file-formats in the subsequent computer
solutions.

Figure 3. Branch-node matrix of graph

76 International Journal of Space Structures Vol. 14 No. 2 1999

K. Linkwitz

Naturally, the “branch-node” matrix may be Finally, as can be shown, in every physical realisation
manipulated. Thus we get the two matrices of a network, it is always possible to formulate a (linear
or non-linear) relationship between the branch variables
CTC , CCT t and the differences u of the node variables
t = f(u) (9)
which we will use later.
At the outset the netlike structure described by the Substituting (9) into (8) we can write
branch-node matrix C is an abstract entity, which may
be physically realised as an electrical network, a
geodetic network, a network of transportation.. etc. , CT*f(u) = r < = > C T*f(u) – r = 0 (10)
and a prestressed cable-net. To prepare the necessary
transit from the abstract net and a real world net we In many physically realised types of netlike
formally introduce as further variables structures we can interpret (10) as ‘equations of
“node-variables” xT = (x1, x2,..., xn), which we equilibrium’ in each node.
assign to the nodes,
“branch variables” tT = (t1, t2,..., tn), which we 3.1.3 Towards a general solution of network
assign to the branches, i.e. the bars, analysis
leaving it for the moment completely open, what We set out to formulate a general solution of (10) with
we have to understand by these variables semantically. the aim of finding the node variables x or,
In the case that the variables have more than 1 equivalently, the differences u of the node variables.
dimension the vectors of the node variables may be Introducing for them starting values x0. or . u0, thus
written
x = x0 + Dx0 and/or u = u0 +Du0

we can apply Newton’s method to solve (10) in a

numerical process of iteration. (Approximate values
for the unknowns may be taken from a model,
The same expansion applies to C and the branch estimated, or ‘precalculated’). Using the linear term of
node matrix becomes expanded a Taylor’s expansion of (10) and working with the
Jacobian matrix derived from f(u) we have the
relationship

(11)

and thus the system of equations to be solved for the Dx

3.1.2 Elementary relationships in netlike
structure s =(11)
Playing with the mathematical material thus gained,
we find easily two important equations, namely
in the first and consequent further steps of iteration. In
these, naturally, the Jacobian matrix always has to be
u:=C*x < = > C*x = u (7)
updated according to the actual unknowns in each step.
– stating, that the multiplication of the vector of node- Applying the chain rule of analysis and
variables x with the branch node matrix C yields the remembering that from equ. (7) above it follows
differences u of the node variables, and

r = CT*t (8)
we find for the Jacobian in (11)
– stating, that the multiplication of the transposed
incidence matrix C T with the branch variables t yields (12)
the sums r of the branch variables in each node

International Journal of Space Structures Vol. 14 No. 2 1999 77

 Formfinding by the "Direct Approach"

so that substituting (12) into (11) yields the equations adjacent nodes
for the first step of iteration Sums r sums of forces in each node

(13) 2nd step: setting up appropriate relationships

appertaining to the physical net investigated

Under prestress the lengths of the extended bars

Equations (13) allow the analysis of arbitrary
between adjacent nodes are
physical realisations of network like entities.
However, three remarks are necessary in this context.

1. By applying the chain rule for Jacobians (12) and the equations of equilibrium in the nodes can be
and substituting this into (11) it is possible to written as
reintroduce the branch node matrix C into the CT *t = r (2.1)
equations of solutions. Thus (13) also reflects
the discrimination between the topology of the Defining s = absolute forces in the bars, we can
net – contained in C – and its physical write their components in x-,y-,z-direction in the
realisation- contained in the Jacobian form
t: = UT * W–1*s (2.2)
2. In the generalised equations (13) for the numerical conforming to the formal expression of t = f(u) of (9)
solution the Jacobian (¶t/¶u) has to be found. above.
This constitutes the crucial point in the In (2.2) the direction-cosines, transforming the
individual realisations of nets. In the case that absolute forces s into coordinate orientated
this realisation is a prestressed cable net it will components, are given by the expression
be shown later how this determination is done.

3. The degree of ‘exactness’ and of interpretation

of the Jacobian determines characteristic
properties of certain individual cases of the
solution. Also this is demonstrated further down.
in which UT and W–1 are diagonal matrices with the
3.2 Transfer of the generalised solution to elements of the vectors u and w.
prestressed netlike structures: Networks of Using the formulation (1) above of Hooke’s law
pin jointed truss in space, prestressed cable nets, and remembering that the elastic elongations e are the
‘appropriately’ discretised membranes difference between extended lengths w and slack
Now, as we are dealing with a discrete physical lengths l = > e = (w – l) we may write the forces s also
realisation of the so far abstract concept ‘net’ we have in the form(2.3)
to first identify what, in this context, we do understand
by the above introduced general variables. Thus we 2.3
achieve the solution in three distinct steps
Combining (2.2) with (2.3)
1st step: identification
(2.4)
General Name Actual identification
denomination of
variable
Node variables x spatial coordinates of nodes we finally have succeeded in finding the required
Branch variables t x-,y-,z- components of relationship between node and branch variables from
(tension) forces acting in the where we can find the Jacobian necessary to solve the
bars problem using equ.(13)
Differences of u differences of x-,y-,z- Summarising:
node var. coordinates between The relationship needed t = f(u)

78 International Journal of Space Structures Vol. 14 No. 2 1999

K. Linkwitz

is expressed by equ.(2.4), and the Jacobian to be and W and have to be introduced into them.
determined is
(b) “Non-linear Approach” = 2nd order theory
We start by defining the new variables

which then has to be used in equ.(13)! (2.8)

Putting together equations (2.1),(2.2), and (2.3) we which, as it turns out, are the “force densities”.
get However, in the present context they serve only to
(CT*UT*W–1)*H*L–1*(w–l) = r (2.5) facilitate the determination of the Jacobian we need in
(13).
3rd step: finding the ‘appropriate crucial Jacobian’
Substituting the forces via Hooke:
(a) “Linear Approach” = 1st order theory
We assume U T*W –1 = const. i.e. only small we can write the new variables also in the equivalent
deformations occur in the net. In this case the Jacobian form of
of (2.4) becomes – as t: = UT*W–1*s – , vide above (2.9)

Remembering the above form for writing the

branch variables t and using the newly established
variables q we can now write the function of which we
Applying the chain rule have to find the Jacobian

(2.10)
From (2.3) s = H*L–1*(w–l) we get

Reminding ourselves that the variables of

differentiation u are contained both in U and in q we
now have to employ the product rule while
determining the Jacobian of (2.10); thus

Combining the last four lines of equations into one

equation we get

(2.6) Evaluating part ||1|| we find

and using this, our general equation (13) of above now

(2.11)
becomes

and the two Jacobians necessary in (2.11) are

(2.7)
*

In the right hand side of (2.7) the s0 are estimated

forces, and UT*W–1*s0 = t0 are estimated components
of forces in x-,y-,z-direction. The estimated
appertaining coordinates x0 – they define the direction
cosinus of the forces – are embodied in the matrices U

International Journal of Space Structures Vol. 14 No. 2 1999 79

 Formfinding by the "Direct Approach"

as may be verified by recomputing both expressions. fixed anchoring points with coordinates xf do exist.
Thus for part ||1|| we get Therefore, we distinguish now in the vector x between
the fixed coordinates xf and the new points xN of the
(2.12) figure of equilibrium to be created. Thus we write now
x = (xN,xf), and after partitioning also C = (C,C f)
accordingly, we find for (2.15) now the system
Turning now to part ||2|| we have

(2.13) (CT*Q*C)*xN +(CT*Q*C f)*xf = r < = >

(CT*Q*C)*xN = r – (CT*Q*Cf)*xf (2.17)

where Q is a diagonal matrix with as q elements.

The complete Jacobian then is exhibiting now clearly the new coordinates xN of the
figure of equilibrium as the solution-vector of a linear
system of equations the right sides of which consist of
loads in the nodes and linear functions of the
and the generalised system of solution (13) now coordinates xf of the anchoring points.
becomes
(2.14) 4. COMPARISON OF THE DIRECT AND
INDIRECT SOLUTION: C OMP L E TE
or, written slightly differently AGR E E ME NT
Comparing the equations of equilibrium (4) of the
direct and (2.5) of the indirect solution we see
=(2.14) immediately the absolute correspondence of
*

(2.18)
(c) “Method of force densities” = ridgid linear
approach
Repeating equ. (2.1) C T *t = r, representing
equilibrium in each node, and repeating also (2.9)
t = UT *q, namely the relationship between branch
and thus may formulate the solution algorithm of the direct
variables t and the newly introduced variables q we
approach corresponding to (2.14) of the indirect approach
get immediately

(2.19)
CT*UT*q = r <=> CT*Q*u = r (2.15)

As according to (7) u = C*x

we arrive at the result
and since s0 = H*L–1*(f(x0)–1)
(CT*Q*C)*x -r (2.16) we get finally, by substituting the s0 in (2.19), the equ.
(2.20):
Equ. (2.16) represent a linear system of equations
permitting, after the “force densities” q contained as (2.20)
elements in Q have been estimated, the direct solution
for the unknowns x ( = figure of equilibrium) in one
direct step by solving the linear system (2.16) *
As (2.16) is still written rather generally we will
elaborate it a little further. Naturally, to create a figure
of equilibrium always demands as a prerequisite that

80 International Journal of Space Structures Vol. 14 No. 2 1999

K. Linkwitz

Interpre tation, discussion and consequent (2.17)

strategies induced by the “Direct Approach” and its
principal formulas.
To begin with, and for the sake of easy reading the
principal formulas are summarised again:
The essential contents of the non-linear equations
1.) Non-linear equations of equilibrium describing of equilibrium (3) and their equivalent (3a) (3) (3a)
exhaustively a figure of equilibrium
*s = g (CT *UT *W-1)*s = r
Direct formulation
Non-materialised form Materialised form
(3) (3a)
(3) (4)
is: Any set of 3h coordinates x = (x1, y1, z1.......,
xh,yh,zh) and m forces s = (s1...., sm) satisfying these
equations constitutes a figure of equilibrium.
= Indirect formulation However, equations (3) do not embody - and therefore
Non-materialised form Materialised form do not demand (!) - any formulation concerning the
(3a) (2.5) materialisation of the figure of equilibrium: neither the
lengths of the bars nor their elastic behaviour under
tension forces are contents of these equations! Also
the (later) materialised constructural realisation of the
2.) 2nd order theory numerical solution nodes themselves are not - contained in them. Thus (3)
Direct formulation and (3a) are the purest form in which a figure of
Non-materialised form Materialised form equilibrium can be described as an abstract entity.
(2.19) Now, the equations (2.17) are absolutely equivalent
to the non-materialised description (4) of a figure of
equilibrium, as can be seen form the sequence of
equations (2.15) ..(2.17) Thus, any set of 3h
= Indirect formulation
coordinates x = (x1, y1, z1,......., xh, yh, zh) found as the
Non-materialised form Materialised form
solution of (2.17), where only the topology of the net
(2.14) C and an arbitrary set of m force densities q =
(q 1,q2,...,qm) are given - constitutes a perfect,
consistent figure of equilibrium.
(2.20) The essential new fact, however, is - comparing
(2.17) with (3) - that now the figure of equilibrium is
described as the solution of a set of linear equations,
permitting the rapid calculation of a great number of
or, using (2.9), figures of equilibrium and not demanding any pre-
Direct formulation knowledge about the approximate coordinates x0.!
Figures of equilibrium thus determined may be
materialised completely arbitrarily: Once the
coordinates x and the appertaining force densities q of
3.) Equations of “force densities” a figure of equilibrium are known, also the forces s
Definition can immediately be calculated, namely

(2.8) (2.9)

*
Linear equations for solution

International Journal of Space Structures Vol. 14 No. 2 1999 81

 Formfinding by the "Direct Approach"

Then the appertaining cutting pattern can be 5. INTERPRETATION, DISCUSSION,

determined for any material, be its material-law linear AND CONSEQUENT STRATEGIES IN
or non-linear: The slack length l = li,k between DESIGN AND FORMFINDING
adjacent nodes Pi and Pk must be cut according to the
prescription that wi,k–li,k=ei,k, where ei,k is the elastic
5.1 Essential Properties of the Solution
We will now discuss equations (4) and their
elongation of the material chosen under the force si,k.
appertaining numerical solutions (2.19),(2.20).
Coming back in this context to our prerequisites These equations describe and determine the unique
(and limitations !) formulated on page 73 above, we figure of equilibrium x compatible with the slack
have to state, that for the method and application of lengths l of all the bars. If h is the number of free
force densities the limitations quoted are not valid: As nodes and m the number of bars=cable pieces in the
we are dealing with the non-materialised state of net, then (4) consists of 3h equations. For h³2
equilibrium no limitations with respect to material- “always” (for all practical purposes) 3h³m is true.
laws do exist nor need we consider at this stage any Still - strange enough -equations (4) are regular and
question of the later constructural materialisation of can be solved uniquely by (2.19) This is due to the
the nodes. This makes the application of force fact, that (2.19) contains implicitly the prerequisite
densities extremely general and powerful. that the net be prestressed. Furthermore it is
worthwhile noting and remembering:

(1) The solution vector x^ of (4) minimises the

quadratic form eT*(H*L–1)*e corresponding
to the potential energy of the net generated by
the elastic elongations in the bars.

(2) The elastic elongations e may be of arbitrary

length; the net is not confined to small
Figure 4. Net of appr. 200 nodes in its initial deformations (2nd order theory). (4) is valid
position in a horizontal plane for the deformed state x^ of the net.

(3) The derivatives in the Jacobian

^
have to be taken for the deformed state x.

Especially in the numerical solution according to

(2.19) – where here only the 1st step of an iteration
process of r steps has been noted – the Jacobian has
to be updated in each successive iteration step
Figure 5. Non-materialised net in equilibrium according to the foregoing, last actual solution-
= result of solution of 3x200 lin.(!) equ. vector x p

(4) Non-linear material behaviour can be

accomodated by updating also in each
iteration step the material constant H to its
then actual value.

(5) If the slack lengths l and the anchoring points

belong to a net, in which all members remain
in tension (“prestressed net”), then the solution
Figure 6. Deformation of non materialised
net under static point load
of the equations (4) is unique and – after
having chosen appropriate starting values x0 –
Fig. 4,5,6: Experiments of formfindig with force- leads in (2.19) to a numerically stable iteration
densities process.

82 International Journal of Space Structures Vol. 14 No. 2 1999

K. Linkwitz

5.2.2 Extensions of the basic equations.

We arrive at more practical solutions and applications if
we subdivide the net into its interior part J and its
boundary part B by partitioning the Jacobian accordingly

Subdividing also the bars in the interior part with

Figure 7. Timber shell Bad Duerrheim; formfinding executed by equal and boundary part with irregular slack lengths, and
methods of “Direct Approach” additionally the forces also in to those of the interior and
of the boundary zone

our equ. (4) become

(4b)

permitting - after analogous 2nd linearisation for an

iteration process as above- to compute figures of
equilibrium, if we prescribe approximate coordinates x0 of
Figure 8. Small Sail, belonging to project of the whole net, slack (equal) lengths lJ for the interior part
membrane structure and forces sB for the boundary part.
5.2 Subsequent Strategies for Practical Also (4b) has as a unique solution that figure of
Applications equilibrium, which in its interior part has equal slack
lengths and in its boundary part forces as prescribed.
5.2.1 Limitations of the unaltered basic
However, the total shape of the net, the arrangement of the
equations
bars, the forces in the interior may deviate from those
Although equ. (2.19) are ‘exact’ and - if we dispose
wanted in a good design, and again many trial and error
of good starting values x0 - steer to a unique solution,
calculations become necessary until feasible shapes finally
their applicability in the practical process of
have been found.
formfinding in this form is rather limited. This is
mainly due to the fact, that we need not only the
starting values for the coordinates but additionally,
and this a more serious obstacle, also all the slack
lengths of the bars. Even presuming that we want to
prefabricate a net with equal slack lengths in its
interior part, it is impossible to know at the outset of
the calculation the slack lengths in the always
irregular boundary zone. A formfinding process
limiting itself to the computational potential of (2.19)
only will soon become a tedious process of trial and Figure 9. Interior and boundary part of a net
error. We will show the remedy further down.
An efficient strategy consists of working outward from
In one application, however, (4) and (2.19) can be the equal mesh interior towards the boundary zone and thus
used advantageously, this is the post-calculation of an confining all changes to the latter. This strategy keeps also
already existing cutting pattern as a means of a the topology of the interior part at each trial change
thorough and independent check: The coordinates x undisturbed and only the boundary topology must be
and slack lengths l must be compatible and thus fit amended.
exactly into equ. (4), and , if not, the application of
5.2.3 A further extension: the approach via
(2.19) will yield just that figure of equilibrium
Least Square s
associated exactly with the cutting pattern in question.
A real breakthrough can be achieved if we now

International Journal of Space Structures Vol. 14 No. 2 1999 83

 Formfinding by the "Direct Approach"

interpret (4b) differently: Contrary to our reasoning

where we considered only the coordinates x as • spatial distances between nodes, and thus
unknowns we now consider the coordinates x and the prescribe individual forces,
forces sB in the boundary zone as unknowns. As we • forces or relation between forces,
have 3h unknown coordinates and (sm – sb) unknown • force densities (!),
forces º slack lengths but only 3h equations of • Hooke’s law itself in the form given above and
equilibrium, the system (4b) in this interpretation is thus prescribed slack lengths.
(sm – sb) - fold underdetermined. Expanding our
interpretation still further we now read (4b) as a The appertaining figures of equilibrium – including
system of equations of conditions which have to be all the constraints prescribed – are exact and uniquely
exactly fulfilled by the coordinates, the slack lengths found by minimising the squared sum of the
and the forces in the boundary zone. differences between the “observations” (=starting
To find a unique solution we use the technique of values of coordinates and certain entities from above)
least squares and introduce as additional constraints to and the “corrected observations”.
find that figure of equilibrium which minimises As a last example we quote the formfinding of the
shape of a hanging net. This is achieved by – in the context
of least squares – formulating as additional constraints
Here Barch is a matrix of weights for the “architect”
and Bing for the “engineer”. This approach has the • slack lengths = elastically elongated lengths by
invaluable advantage that, after approximation increasing the Hooke’s factors towards infinity, and
coordinates x0 have been taken from a model (or the • spatial differences between adjacent nodes=slack
results of studies with force densities are taken as the lengths=elongated lengths.
approximate co-ordinates for the consequent
“materialised” computations) and also approximate After only one step of iteration the corresponding
forces s0 for the boundary zone have been estimated , figure of equilibrium is found, whereas the
an adequate choice of the two weight matrices permits formulation with (4) and (2.19) approaches non-
the generation of a figure of equilibrium being as close convergence due to the fact that under dead loads only
as possible to the wanted shape and the wanted forces. the elastic elongations in the bar are very small and
And , compared to the strategies outlines above , this lead to a near singularity of equ. (4), (2.19).
can be achieved in a much smaller number of
computational steps. CO NCL USIONS
Still, also this approach has certain weaknesses. As The “Direct Approach” discerns itself by a number of
experiments have shown, the influence of in- or properties from approaches of other schools and
decreasing the boundary forces fades away rapidly allows a wide spectrum of strategies for the process of
towards the interior and it becomes difficult to predict formfinding as well as for a subsequent structural
the results in detail. analysis. It is especially characterised by:

5.2.4 Least Squares and introduction of 1.) Starting with a direct formulation of
additional constraints compatibility between elastically elongated
However, the least squares interpretation can be further bars and the coordinates of the nodes and
extended by allowing a whole set of additional including then the condition of equilibrium in
constraints to be included in the analysis and each point, a set of non-linear equations
computations. This is achieved by the technique of “non- describing the appertaining non materialised
linear condition equations with additional constraints”. figure of equilibrium results.
Applying this principle, forces in the interior of a net can
be exactly prescribed, or connecting angles of net and 2.) By introducing into these equations the new
boundary cables can be systematically influenced. entities of “force densities” and distinguishing
Translating into an associated least squares between “fixed nodes” (corresponding later to
approach we may introduce upto r # h additional, anchoring points), and “new, unknown nodes”
exactly to be satisfied constraints, alone or in (constituting the figure of equilibrium) a system
meaningful combinations, i.e. of linear equations results, which is totally

84 International Journal of Space Structures Vol. 14 No. 2 1999

K. Linkwitz

equivalent to the non-linear system describing 6. APPENDIX: MAT HE MAT I C A L

the figure of equilibrium. The latter is found NOTAT ION
easily by solving the system of linear equations l = (l1,l2,.......ln)T slack lengths of bars,
for the unknown nodes after force densities s = (s1,s2,.......sn)T tension forces acting in bars,
have been “estimated” and then introduced into e = (e1,e2,.......en)T elastic elongations of bars,
the equations as fixed parameters. Abundant induced by forces s,
experience and experiments have shown, that e = (1,1,.......1)T unit vector (in equ. 2.8
almost “trivial” estimates of the force densities ...(2.14) !
result in feasible shapes of figures of w = (w1,w2,.......wn)T distances in space between
equilibrium, which then can be iteratively adjacent nodes,
improved by “playing” with different sets of calculated from their
parameters for the force densities. coordinates, and length of
elastically elongated bars !)
3.) Only in a second, distinctive step following g = (g1,g2,.......gn)T loads at nodes,
the process of non-materialised formfinding x = (x1,y2,.......xh,yh,zh))T
coordinates of nodes
the figure of equilibrium selected may be Fi Cross section area of bar
arbitrarily materialised. h = (E1*F1,.......En*Fn)T
= (hl.......hn)T Hooke—factors of bars
4.) Using graph theory the “Direct Approach” can L,H,W,.....S diagonal matrices with
be formulated indirectly yielding sets of again elements of vectors
equivalent equations. Also here the force l,h,w,.....s.
densities can be derived. This type of
formulation, by emphasising the topological 7. LITERATURE
description of the net, leads to data structures [1] Linkwitz, K.: Fehlertheorie und Ausgleichung von
which can be used advantageously in Streckennetzen nach der Theorie elastischer Systeme.
(Theory of Errors and Least Squares Analysis of
computer implementations of the method.
Trilateration Networks by Using the Theory of
Elastomechanics”); Doctor Thesis,Deutsche Geodätische
5.) After extending the “Direct approach” by Kommission bei der Bayerischen Akademie der
incorporating methods of least squares, Wissenschaften, Reihe C, Heft 46, München 1961.
additional powerful formfinding strategies are
[2] Linkwitz, K. und Schek, H.J.: Einige Bemerkungen zur
made accessible, allowing especially the Berechnung von vorgespannten
effective treatment of supplementary Seilnetzkonstrukti onen. (“Remarks concerning the
constraints. This reduces the trial and error Analysis of Prestressed Cable-Structures”) Ingenieur –
process of formfinding substantially. Archiv 40 (1971), S.145-158; manuscript received from
authors Febr. 1970

6.) A discrete “net” of quadrilateral or triangular [3] John H. Argyris und Dieter W. Scharpf: Berechnung
vorgespannter Netzwerke (“Computation of prestressed
meshes is the initial object for the description
networks”); Bayerische Akademie der Wissenschaften,
of a structures whose form has to be studied Mathem.- Naturwiss. Klasse, Sonderdruck 4 aus den
and found and which later has to be analysed. Sitzungsberichten 1970, Verlag C.H. Beck, München
The appertaining mechanical-mathematical 1970
model is a net of discrete bars pin-jointed in
[4] Linkwitz, K.:New Methods for the Determination of
the nodes. In an extension of the approach - Cutting Pattern of Prestressed Cable Nets and their
not treated here - the bars may also be flexible Application to the Olympic Roofs Munich; Proc. 1971
beams. Continuous membrane-type structures IASS Pacific Symposium, TOKYO and KYOTO ,
of arbitrary material are discretised by spatial Architectural Institute of JAPAN
nets of triangular meshes. As has been shown
[5] Linkwitz, K., Schek,H.J.: A New Method of Analysis of
in [20] and [30] the method of force densities Prestressed Cable Networks and ist Use on the Roofs of
may be extended to triangular nets and the Olympic Game Facilities at Munich; IABSE,
membranes. (also not treated here). Amsterdam 1972

[6] Linkwitz, K.: Die Ermittlung des Zuschnitts für Dächer

International Journal of Space Structures Vol. 14 No. 2 1999 85

 Formfinding by the "Direct Approach"

der Olympiasportstätten München. weitsgepannender Flächentragwerke (“A non-linear

(“The Determination of the Cutting Patterns of the theory of of space trusses and ist application to wide
Olympic Roofs Munich 1972”)ZfV 1972, Heft 9, 10. spanning surface stsructures”), Doctor-Thesis,
Sonderforschungsbereich 64, Mitteilung 52, Universität
[7] Brinkmann, G.: Untersuchung zur Geometrie und Statik Stuttgart 1980
gleichmaschiger Netze. (“Investigation about the
Geometry and Statics of equally meshed networks”) [16] Bahndorf, J., Gründig, L. und Hangleiter, U.: Memcut –
Schriften-reihe des SFB 64 Weitgespannte A strategy for cutting computation of structural
Flächentragwerke, Mitteilung 2/1972, Universität membranes. Proc. of the 3rd Int. Symposium on Wide
Stutgart Structures, Stuttgart März 1985.

[8] Schek, H.J.: The force densities method for form finding [17] Stark, W.: Untersuchungen zur Lösung und Inversion
and computation of general networks. Computer schwach besetzter, großer geodätischer
Methods in Applied Mechanics and Engineering, 1974, Normalgeichungssysteme (Investigations about the
S. 115-134. solution and inversion of sparse, large systems of
geodetic normal-equations”), Doctor-Thesis Deutsche
[9] Gründig, L.: Die Berechnung vorgespannter Seil- und Geodätische Kommission, Reihe C, Heft 301,
Hängenetzen unter Berücksichtigung ihrer topologischen München 1985
und physikalischen Eigenschaften und der [18] Gründig, L., Bahndorf, J., Neureither, M.:
Ausgleichungsrechnung. (“The Analysis of prestressed Minimalflächen zur Formfindung von Membranen.
Cable- and Hanging-Nets with regard to their topological ARCUS (Journal für Architektur), München, Heft 1,
and physical properties and to Least Squares Theory”) 1987.
Doctor_Thesis; Deutsche Geodätische Kommission bei
der Bayerischen Akademie der Wissenschaften, Reihe C, [19] Linkwitz, K., Gründig,L., Bahndorf,J., Ströbel, D.:
Heft 246, München 1976. Geodätische Methoden als Mittel zur Form- und
Zuschnittsberechnung von Flächentragwerken. Beispiel
[10] Linkwitz, K.: Die Berechnung vorgespannter Seilnetze Dach Bad Dürrheim. X. Int. Kurs für
als Aufgabe der Ausgleichungs-rechnung.(“ The Analysis Ingenieurvermessung, München 1988.
of Prestressed Cable-Nets as an Application of Least
Squares Theory” Internationaler Kurs für [20] Singer, Peter: Formfindung und statische Berechnung
Ingenieurvermessungen hoher Präzision. Darmstadt von Membranen mit der Kraftdichtemethode;
1976. (“Formfinding and statical analysis of membranes using
the method of force densities”), Diplomarbeit, Institut für
[11] Schek, H.J.: Über Ansätze und numerische Methoden zur Anwendungen der Geodäsie im Bauwesen, Universität
Berechnung großer netzartiger Strukturen. (“About Stuttgart, 1988
Theories and Numerical Methods for the Analysis of
large, netlike Entities”) Thesis of Habilitation, [21] Gründig,L.: Datenverwaltungskonzepte für typische
Universität Stuttgart 1977. Aufgaben aus der Ingenieurgeodäsie. Habilitation
Thesis, Deutsche Geodätische Kommission, Reihe C,
[12] Linkwitz, K.: Über eine neue Methode der Gauß’schen München 1988.
Methode der kleinsten Quadrate. Die Formfindung und
statische Analyse von räumlichen Seil- und [22] Linkwitz, K., Gründig, L., Bahndorf, J., Neureither, M.,
Hängenetzen. (“About a New Method within the Theory Ströbel, D.: Geodätische Methoden als Mittel zur Form-
of Least Squares: Formfinding and Analysis of Spatial und Zuschnittsberechnung von Flächentragwerken.
Prestressed and Haning Nets”) Braunschweigische Beispiel Dach Olympiastadion Montreal/Kanada.
Wissenschaftliche Gesellschaft, Sonderveröffentlichung Ingenieurvermessung 1988, Bonn 1988.
zum 200. Geburtstag C.F. Gauß, Braunschweig 1977.
[23] Bahndorf, J., Neureither, M., Hangleiter, U.: Arbeits- und
[13] Linkwitz, K.: Zur Systematik der Analyse von Netzen. Ergebnisbericht des Teilprojektes D3 im SFB 230.
(“On a Systematic Representation of the Analysis of Soderforschungsbereich 230 Natürliche Konstruktionen,
Nets”) Special Lecture Intern. Summer School Ettore Universität Stuttgart 1989.
Majorana, Trapani/Sizilien 1978.
[24] Linkwitz, K., Krause, C.: Formfindung eines
[14] Schwenkel, D. : Mathematisch-numerische Methoden Computermodells. (“Formfinding of a Computer
zur Approximation und Abbildung der Systemgeometrie Model”) Deutsche Bauzeitung, April 1989, Stuttgart
weitgespannter Fläachenntragwerke (“Mathematical- 1989.
numerical methods for the approximation and mapping
of wide spanning surface structures”); Deutsche [25] Bahndorf, J.: Zur Systematisierung der
Geodätische Kommission bei der Bayerischen Akademie Seilnetzberecnung und zur Optimierung von Seilnetzes.
dier Wissenschaften, Reihe C, Heft 243, München 1978 (“About a Systematic Approach to the Analysis and
Optimisation of Cable Nets”) Doctor Thesis; Deutsche
[15] Nguyen-Tong, B.: Eine nichtlineare Stabwerkstheorie Geodätische Kommission, Reihe C, Heft 373, München
und ihre Anwendung zur Lösung von Problemen 1991.

86 International Journal of Space Structures Vol. 14 No. 2 1999

K. Linkwitz

[26] Neureither, M.: Modellierung geometrisch-topologischer [31] Linkwitz, K.:Holzschalendächer Hölderlinhaus der
Daten zur Beschreibung und Berechnung netzartiger und Anthroposophia, Maulbronn (“Timber Shells on the
flächenhafter Strukturen.(“Modeling of Geometrical- Hölderlinhaus der Anthroposophia , Maulbronn”)
Topological Data for the Analysis of Netlike Entities”) Bautechnik 4, 1996, Ernst & Sohn, Berlin 1996
Doctor Thesis Deutsche Geodätische Kommission,
Reihe C, Heft 387, München 1992 [32] Linkwitz, K., Ströbel.D.,Singer P.: Die Analytische
Formfindung (“Analytical Formfinding”), in Prozess und
[27] Linkwitz, K.: Formfinding of lightweight surface Form Natuerlicher Konstruktionen , Ernst & Sohn,
structures by geodeticmethods. Proceddings of the Berlin1996, ISBN 3-433-02883-4
International Symposium on the application of Geodesy [33] Ströbel, D.: Die Anwendung der Ausgleichungsrechnung
t Engineering, Stuttgart 1991, Heidelberg, New York auf elastomechanische Systeme (“Application of the
1992. Theory of Least Squares on the Analysis of
Elastomechanical Systems”) Doctor Thesis, Bayerische
[28] Linkwitz, K., Bahndorf, J. Hoerdt, R. Singer,P., Akademie der Wissenschaften, Reihe C, Heft 478,
Stroebel,D.: Least Squares Methods in Non-Linear München 1997
Formfinding and Analysis of Prestressed and Haning
Nets,Seiken IASS Symposium Tokyo 1993 [34] Röder, R.: Zur Analyse und Optimierung von
[29] Linkwitz, K., Bahndorf, J.: Separation of Topological Transportnetzen (“On the Analysis of Transportation
and Physical Properties in the Non-linear Analysis of Networks”) Doctor Thesis, Bayerische Akademie
Surface Lightweight Structures,International Conference der Wissenschaften, Reihe C, Heft 484 , München
on Computational Methods in Structural and 1997
Geotechnical Engineering, Hong Kong 1994
[35] Linkwitz, K. : Students’ Training in Formfinding and
[30] Singer,S.: Die Berechnung von Minimalflächen, Analysis of Tension Structures, NEWSLETTER
Seifenblasen, Membranen und Pneus aus geodätischer December 1997 Structural Morphology Group,
Sicht (“Analysis and Computation of soap-films, Working Group 15 of IASS, Delft University Printing
membranes, and pneus from the view point of the Office 1997
geodetic sciences”) Doctor Thesis, Deutsche Geodätische
Kommission Reihe C, Heft 448, Muenchen 1995 [36] Linkwitz, K.:About Formfinding of Double-Curved
Structures, Engineering Structures 21, 1998 Elsevier
Scince Ltd. England

International Journal of Space Structures Vol. 14 No. 2 1999 87