Struct Multidisc Optim (2011) 43:181–190

DOI 10.1007/s00158-010-0557-z

RESEARCH PAPER

A 99 line code for discretized Michell truss optimization

written in Mathematica
Tomasz Sokół

Received: 17 April 2010 / Revised: 15 July 2010 / Accepted: 25 July 2010 / Published online: 14 September 2010

c The Author(s) 2010. This article is published with open access at Springerlink.com

Abstract The main purpose of the paper is to provide an 1 Introduction

easy-to-use code for topological optimization of the least
weight trusses, written in the Mathematica programming The optimum design problem: find the lightest, plane,
language. The main idea of the presented approach consists pin-jointed frame of a bounded compliance, transmitting
in using a fixed ground structure and the linear program- a given loading to a part of the boundary of the given
ming formulation of the optimization problem. The solver feasible domain—turns out to be equivalent to the prob-
is based on the fast interior point method. The strong effort lem of finding the lightest pin-jointed frame of a bounded
is done to create the effective generator of the computa- stress level, with equal stress limits in tension and compres-
tional model utilizing the high regularity of the ground sion, transmitting the loading to the prescribed boundary
structure and the high sparsity of the geometric matrix. The of the feasible domain, see Hemp (1973) and Achtziger
efficiency and reliability of the algorithm is confirmed in (1997). The analytical solutions to this problem must sat-
several numerical tests. Due to a linear programming formu- isfy the conditions of the theory of Michell trusses; see
lation of the optimization problem the method presented in Hemp (1973) and Rozvany et al. (1995). These solutions
the paper assures finding the global optimum, hence it may exceed the class of trusses: they are discrete-continuous
be considered as the useful tool for verification of results structures composed of the fibrous domains of orthogonal
obtained in other ways. The appended complete Mathemat- microstructure reinforced by bars of finite cross sections.
ica code of the program developed will be supplied by the These solutions can be approximated from within the sub-
Publisher on SpringerLink. class of trusses (i.e. pin-jointed frames of finite number of
bars) or from within the continuum description thus giving
Keywords Topology optimization of trusses · up prediction of the reinforcing bars and excluding possi-
Ground structure · Linear programming · bility of considering the point loads. Having at our disposal
Interior point method · Sparse matrix representation these two approximate methods to attack the initial prob-
lem it is thought appropriate yet to choose the discrete
approximation—as free of the two drawbacks mentioned.
The natural incorporation of the point loads seems here cru-
cial; note that almost all available Michell solutions concern
this class of loadings.
Electronic supplementary material The online version of this article
(doi:10.1007/s00158-010-0557-z) contains supplementary material Note, however, that the sole knowledge of the Michell’s
(the complete computer code with an example), which is available to theory does not deliver hints of how the optimal layout
authorized users. looks like. Only having an impression of the correct lay-
T. Sokół (B) out one can construct the Hencky net and then endow it
Department of Structural Mechanics and Computer with appropriate mechanical properties. The complete solu-
Aided Engineering, Faculty of Civil Engineering, tions are rare. It is sufficient to stress here that the static part
Warsaw University of Technology,
Al. Armii Ludowej 16, 00-637 Warsaw, Poland of the theory of Michell’s cantilever supported on a circle
e-mail: [email protected] has only been put forward in the relatively new paper by
182 T. Sokół

Graczykowski and Lewiński (2005), where a proof of van- been appended to the paper. This reference to the title of
ishing the duality gap between kinematic and static formu- the popular paper by Sigmund (2001) has been suggested
lations of this problem has been published. Note also, that by anonymous Reviewer. The code listed here can be freely
due to hyperbolicity of the governing equations of Michell’s downloaded from the internet web page connected with this
trusses the layouts are usually composed of many subdo- paper, see Online Resource 1.
mains with border lines being the lines of discontinuity of The software developed applies to the case of plas-
some stress components; see e.g. Fig. 3 in Graczykowski tic design with: (a) equal permissible stresses in tension
and Lewiński (2007). and compression: σT = σC , (b) unequal stress bounds
Having in view the above mentioned properties of for tension and compression: σT = σC . This numerical
Michell trusses it is not easy to predict a new solution cor- tool makes it possible to construct new optimal layouts for
rectly: many layouts can be imagined as correct for the same which the analytical solutions are unknown, and to check
load and the same feasible domain geometry. It is really not already published analytical solutions. In particular, the
clear which of them should be discussed and which of them present paper confirms that the analytical results published
should be rejected a priori. The new layouts are not the sim- in Lewiński et al. (1994a, b) for the case of σT = σC and in
ple composition of the known layouts. Some suggestions are Graczykowski and Lewiński (2007) for the case of unequal
very convincing, but may be misleading. Thus a theoretician stress limits—are correct. In particular, this software has
needs an unbiased hint from the purely numerical side. delivered a numerical confirmation of the huge family of
The really valuable numerical predictions of Michell-like benchmarks being published in Graczykowski and Lewiński
structures do not draw upon the properties of the known (2010).
analytical solutions. The genuine method of their construc- The software developed has also made it possible to pre-
tion was developed in Dorn et al. (1964); it is based on the dict a correct layout of the unsolved till now problem of
concept of the ground structure composed of all bars con- two symmetric parallel forces to be optimally transmitted
necting given regular set of nodes. Most of these bars are to fixed nodes, in the case of the feasible domain being the
not necessary to equilibrate the applied loading and, con- half-plane. This numerical prediction has paved the way to
sequently, disappear during the optimization process. This find the analytical solution. The optimal layout turns out
prediction, however, cannot be done in advance. The cross to compare favourably with the numerical results for the
section areas are the main but not the only design variables. ground structures of high density; the reader is referred to
The crucial point of this approach is that the cross section Sokół and Lewiński (submitted) for the details.
areas may assume nonnegative values. Thus the zero values Concluding, the software developed delivers numerical
are admissible, which paves the way to find the catenary predictions of the characteristics of the optimal layouts with
structures, capable of carrying only one set of loads. Let the volume predictions approaching the optimal volume
us remind here the funicular structures transmitting systems with arbitrary accuracy. Since the analytical predictions are
of parallel forces to fixed supports. Such a specific class of usually found by the kinematic method, while the numerical
structures should be encompassed by the numerical meth- predictions follow from static consideration, this consis-
ods. The ground structure methods satisfy this condition, tency of both the results—showing that the duality gap
provided that the global stiffness matrix is not required to is slowly vanishing with increase of the density of the
be invertible. ground structure - simultaneously confirms correctness of
The present paper is close in spirit to the paper by Gilbert the analytical results.
and Tyas (2003), where the initial optimization problem
has been rearranged to the so called plastic design formula-
tion assuming the form of the linear programming problem. 2 Minimum volume truss optimization problem
Gilbert and Tyas (2003) solved this problem with using the
interior point method. They significantly improved the tra- The goal of the present paper is to find a truss of minimal
ditional ground structure approach by the iterative member volume subject to the compliance constraint, as formulated
adding technique. bellow; cf. Achtziger (2007)
The same numerical problem is solved in the present
min LT A
paper by other implementation of the interior point method, A∈R M , q∈R N
announced in Sokół and Lewiński (2009). The code has s.t. K q = P (1)
been written in Mathematica 7 symbolic language, seek P T q ≤ W0
Wolfram (2003). The method as such would be use-
A≥0
less, without taking advantage of the geometric matrix
being sparse. The details of the program are explained in Here L = [L 1 , L 2 , ..., L M ]T represents the vector of bar
Section 3, while the whole 99-line optimization code has lengths, A = [A1 , A2 , ..., A M ]T represents the vector of
A 99 line code for discretized Michell truss optimization written in Mathematica 183

cross section areas, while P and q are the vectors of nodal truss topology optimization problem may be written in the
loads and displacements. K is the stiffness matrix and W0 is standard form of linear programming as follows
a given positive constant. The last inequality is understood 
component-wise i.e. Ai ≥ 0. The cross section areas Ai are min vi = eT v
v∈R 2M
the main design variables. If the ground structure is viewed i
s.t. H v = P (5)
as fixed, the vector L of bar lengths is prescribed. Note that
the equilibrium equations are treated as equality constraints. v≥0
The displacements q j are state variables, independent of Ai .
with all cost coefficients in the objective function equal to
Thus the number of unknown variables is N + M. Finding
one (e is a vector of length 2M, whose entries are all equal
numerical solution to problem (1) requires expensive meth-
to 1).
ods of the nonlinear programming. Therefore, it is rather
The objective functions in (2) or (5) correspond to—but
not applicable for large tasks corresponding to the ground
are not exactly—the optimum volume determined by (1).
structures of high density, with big N and M numbers.
The problem consists just in simple scaling but to make the
To make the problem less complex we shall pass to the
paper self-contained the basic formulas will be outlined bel-
formulation with forces in members as the only state vari-
low. For easier comparison of results obtained in different
ables, see Achtziger (2007) and Gilbert and Tyas (2003).
ways it is worth to introduce a normalized, non-dimensional
The problem (1) can be reformulated to equivalent formula-
volume as well as other auxiliary quantities. The static prob-
tion (Achtziger 2007)
lem considered in the paper is linear. Consequently, if two
times bigger loading is applied two times bigger displace-
min LT (T + C) ments and axial forces will appear. Similarly the lengths
T∈R M , C∈R M
(2) of bars are proportional to the size of the structure. It is
s.t. BT (T − C) =P
worth to introduce the non-dimensional quantities (denoted
T ≥ 0, C≥0
by upper dash) in the following manner:

where T and C are the vectors of tension and compression P = PP, S = PS, L = hL, q = Phq, etc. (6)
forces in bars and B is the geometric matrix of components
representing directional cosines of bars. The latter problem, where the scalar P is a referential load intensity and h is an
being still equivalent to the problem (1), involves 2M design arbitrary chosen size of a structure (height, length, width,
variables but now it becomes a linear programming prob- etc.). The optimal volume subject to constrained compliance
lem that can be solved by well-developed numerical tools, may be written as
applicable for large number of unknowns. Thus the ground  2
|Si | L i P 2 h 2 
2 P 2h2
structures of high density may be successfully analyzed. VW = = Si L i = VW,
The problem (2) may be rearranged to a standard form E W0 E W0 E W0
of a linear programming problem by introducing the new (7)
design variables: where V W is a non-dimensional compliance controlled
volume, given by
vi = Ti L i and v M+i = Ci L i for i = 1, . . . , M. (3) 
2
VW = Si L i , (8)
and by introducing the matrix
Correspondingly, the optimal volume subject to con-
  strained stresses is expressed by
H = BT , −BT  , where

 = diag[1/L 1 , 1/L 2 , . . . , 1/L M ]. (4) |Si | L i Ph 
Ph
Vσ = = S i L̄ i = Vσ, (9)
σ0 σ0 σ0
Computation of this matrix may be performed by divid- where V σ is a normalized, non-dimensional stress con-
ing every column of matrix BT by the length of the cor- trolled volume, given by
responding bar, but H may also be derived directly by

computing the quotients xi /L i2 and yi /L i2 . Note that for Vσ = Si L i . (10)
integer increments xi and yi the components of matrix
H are rational numbers and may be preserved in exact form Note that equations linking stress and compliance controlled
(using appropriate software). For currently available opti- volumes are quadratic; for example:
mization method it is not any advantage, but it may be 
important for future numerical treatments. To conclude, the Vσ = VW (11)
184 T. Sokół

or the primal affine scaling method and the primal–dual inte-

σ02 rior point method are presented in detail in the book by
VW = Vσ2 . (12) Bhatti (2000). The second version requires more memory
E W0
but is significantly faster.
These formulas are important for proper interpretation The next, very important topic in developing the whole
and comparison of different results presented in the liter- algorithm is the generator of the computational model. It
ature, see Gilbert and Tyas (2003) and Achtziger (2007). should be noted that the fixed ground structure has to be
Further, we shall write V σ = V . dense but at the same time very regular. It is composed of
The plastic layout optimization problem (2) can be gen- groups of identical elements (neglecting unessential trans-
eralized to the case of unequal stress limits for tension lations in XY plane or XYZ space). The stiffness matrices
and compression: σT = σC . Note that in fully-stressed as well as directional cosines are equal in these groups and
trusses the volumes of the bars under tension are equal to may be calculated only once for a given group of bars. For
Ti L i /σT and the volumes of the bars under compression a dense ground structure, thousands of bars in one group
are equal to Ci L i /σC . Thus we can formulate the following may exist so the final profit of the single calculation is sig-
optimization problem nificant. Moreover, due to high regularity of the mesh there
min LT T + κ LT C is no need of creating nodes and elements like in a typical
T∈R M , C∈R M FEM program—they are optionally needed only for graphi-
s.t. BT (T − C) =P , (13) cal presentations. In the program presented in this paper the
T ≥ 0, C≥0 patterns of elements are introduced to preserve the mem-
ory. The pattern includes the necessary and condensed data
in which the ratio of limiting stresses is denoted by κ = for the whole family of bars; they are: nodal increments,
σT /σC . Passing from (2) to (13) requires only a “cosmetic” indices for loops, lengths and directional cosines of bars
change in the objective function. Note, however, that the (see Appendix 1). To solve the optimization problem (2)
latter problem is not equivalent to the problems previously all what is needed is creating the vectors L, P and the geo-
discussed in this paper, see Rozvany (1996). metric matrix B. The last one requires, however, the further
attention. It is clear that this matrix is very sparse. Inde-
pendently of the problem size every row of it contains only
3 Details of numerical implementation four nonzero elements (or six for space trusses). This spar-
sity must be utilized to make the algorithm efficient. First
The choice of the adequate optimization method directly of all, the sparse matrix representation allows to preserve a
depends on the type of the problem to be solved. In the huge amount of memory and to deal with large-scale prob-
case of optimization problem (1) the general or specialized lems, that otherwise would be intractable. Additionally, the
methods of nonlinear programming have to be applied. Usu- great benefit of the computation time may be achieved by
ally they are expensive and limited to relatively small tasks. avoiding unnecessary algebraic operations, especially mul-
Contrary, the problems (2), (5) and (13) may be solved tiplication by zero. Obviously it requires applying some
using linear programming methods. The most popular, the specialized procedures (libraries) for sparse matrices, but in
simplex or revised simplex methods are possible but not ade- the present day they are commonly available. Concluding,
quate for large-scale problems. In practice, such tasks with the sparse matrix representation is essential for the over-
thousands of unknowns are almost intractable using these all efficiency of the algorithm, especially for large-scale
methods due to the exponentially growing time of compu- problems. The larger problem size is—the greater benefit
tation regarding to the problem size. However, the newer of memory and processor time may be achieved.
and much more effective linear programming methods are Let us consider the plane truss with the ground struc-
becoming available. For today, the interior point method is ture restricted to the rectangular area. Three demonstrative
one of the most reasonable choices. This method was devel- examples of such structures are shown in Fig. 1. Note that
oped by Karmarkar (1984) and later improved by Mehrotra the mesh density is described by four numbers: N X , NY ,
(1992), Wright (1997) and Nocedal and Wright (1999). The D X , DY . The first two denote the numbers of divisions
method has even been successfully generalized to a class of of the rectangle in X and Y directions, and correspond to
nonlinear convex optimization problems. The basic idea of the “external” density. (Every node in the mesh is uniquely
the method consists of a barrier function used to encode the identified by a pair of integers, numbered from 0 to N X and
convex set of constraints. Contrary to the simplex method from 0 to NY for X and Y directions, respectively.) The next
which goes along the boundary, it moves through the inte- two numbers D X and DY describe the “internal” density—
rior of the feasible region and reaches the optimal solution it is a depth (in the Manhattan distance manner) of possible
asymptotically. Two most popular versions of this method: connections between nodes. In general these numbers may
A 99 line code for discretized Michell truss optimization written in Mathematica 185

a Thanks to many functions ready for use, like global oper-

ations on the whole vectors or matrices, the code is much
stricter than in any classical programming languages (like
C/C++ or similar). Moreover the Mathematica has an
internal support for sparse matrices and a powerful ver-
sion of the interior point method (see SparseArray[...]
and LinearProgramming[..., Method → “InteriorPoint”]
for details). Description of the internal implementation of
the interior point method applied in Mathematica is included
e.g. in Champion and Strzebonski (2008). The program has
been divided into separate procedures (modules) that have
thoroughly been tested and optimized for speed and memory
usage. For example, to create a dynamic list, the natural but
b
relatively slow functions Append or AppendTo are replaced
by more efficient pair of functions Reap-Sow. Many of
auxiliary arrays were packed by ToPackedArray function
to preserve the memory. Thanks to a compact coding in
Mathematica the program is short. Its listing is included in
Appendix 1. The program is adjustable and offers the full
control of the size and density of the ground structure as
well as any loading and supports. The typical call to the pro-
gram and its output are shown in Fig. 2. The block of input
data is simple and takes only first four lines. The example of
the data input presented below corresponds to a well known
Michell cantilever; cf. Fig. 2.
c
Xmax = 3; Ymax = 2;
NX = 60; NY = 40; DIST = 20;
supports = {{{0, NY/4}, {1, 1}},
{{0, NY 3/4}, {1, 1}}};
loads = {{{NX, Round[NY/2]}, {0, -1}}};
r = OptimalTruss[Xmax, Ymax, NX, NY,
supports, loads, DIST]

Mesh 60x40:20x20, Nodes 2501,

Elements 800076, DOF 4998
Matrix H 4998eqs x 1600152dvs in 73MB
(59.6GB full)
Fig. 1 The examples of ground meshes with different depth of internal Objective S.L = 13.0049,
connections: a 4 × 3:1 × 1–55 elements, b 4 × 3:4 × 2–115 elements,
c 8 × 6:5 × 5–1,054 elements CPU time = 225.4s

also be defined separately for X and Y directions. Never-

theless, if possible, they should be equal. It results from
the fact that the ground structure should form the mesh of
mutually orthogonal families of bars. Therefore, it is also
strongly recommended that the basic cell of the mesh is
square, rather than rectangular.
The code for truss topology optimization has been
written in Mathematica 7 (Wolfram 2003). It is the pow-
erful mathematical program with the comfortable envi-
ronment for advanced computations and visualizations. Fig. 2 The example of calling of the OptimalTruss and its output
186 T. Sokół

Fig. 7 Mesh 12 × 4:2 × 2; V = 13.9321 P h/σ0

b

Fig. 3 Long cantilever problem and the exact optimal truss layout by
Lewiński et al. (1994a) Fig. 8 Mesh 60 × 20:3 × 3; V = 13.6953 P h/σ0

As the first example, consider a long cantilever beam

shown in Fig. 3a. Assume that the domain of possible mate-
rial is restricted to a rectangular panel with the ratio of
length to height equal to 3:1. The left side of the panel
is clamped while the right side is loaded by the vertical
Fig. 4 Mesh 6 × 2:1 × 1; V = 15.0 P h/σ0 force applied at the middle-central point. The problem is
classical and may be viewed as one of the most popular
benchmark tests. The exact-analytical solution was obtained
For today only plane trusses were implemented and for the first time in Lewiński et al. (1994a); see Fig. 3b. This
tested in the program. Theoretically the similar algo- layout consists of two Michell circular fans of origins at
rithm would be used for space trusses but the problem the corner-supports and then of four orthogonal-curvilinear
size increases dramatically, thus the smarter strategies are Hencky nets: Michell “shield” (Michell 1904), two Chan
required in this case. It will be the topic of the further fans (Chan 1967), and the “shield” found in Lewiński et al.
research and the next versions of the presented program. (1994a). The exact value of the optimal stress controlled
volume is equal to 13.5972 Ph/σ0 . It may be observed that
middle supports as well as material in some regions are
4 Examples not necessary and disappear in the optimal structure. It is
worth to note that the analytical solution is hard to obtain
The program described in the previous section has been (it requires the advanced mathematical tools for solving
thoroughly examined in many numerical tests. A few of hyperbolic differential equations and is based on Bessel and
them are presented bellow.

Fig. 9 Mesh 60 × 20:5 × 5; V = 13.6439 P h/σ0

Fig. 5 Mesh 60 × 20:1 × 1; V = 15.0 P h/σ0

Fig. 6 Mesh 6 × 2:2 × 2; V = 14.5 P h/σ0 Fig. 10 Mesh 60 × 20:20 × 20; V = 13.6343 P h/σ0
A 99 line code for discretized Michell truss optimization written in Mathematica 187

Fig. 11 Mesh 120 × 40:10 × 10; V = 13.6126 P h/σ0

Fig. 13 Half-plane with two forces and two fixed supports

10% worse than exact one. The improvement of the solution

Fig. 12 Mesh 120 × 40:20 × 20. V = 13.6120 P h/σ0
may be achieved only after including bars with new direc-
tions; compare the results in Figs. 6–10. The influence of
Lommel special functions). Contrary, the numerical solu- the internal density is evident. For the densest mesh 120 ×
tion may easily be obtained using the algorithm presented 40 with internal density 20 × 20 (see Fig. 12), the result
in this paper. is very close to the exact solution; the relative error of the
To make the investigation more extensive many different volume is equal to 0.11%. Topology and shape of the final
ground structures are tested. The truss layouts obtained for truss correspond also very well to the exact layout shown in
different densities of the ground structure are presented in Fig. 3b.
Figs. 4, 5, 6, 7, 8, 9, 10, 11, and 12. The thickness and the The efficiency of the presented algorithm has been thor-
gray scale of lines are proportional to square root of cross oughly examined using the internal Mathematica’s func-
section areas. Dark-thick lines correspond to thick bars tions like ByteCount[], Timing[], etc. The results are col-
and light-thin lines—to thin bars, respectively. Obviously lected in Table 1. The first column describes the applied
the denser ground structures give better results regarding mesh density. The next two columns include the total num-
to the optimal volume as well as the distribution of mate- bers of nodes and elements. The RAM capacities needed
rial. Density should however be understood not only as the for storing the matrix H defined in (4) are given in two next
external division of the domain into the number of rectan- columns, respectively for sparse and full matrix representa-
gular cells. The density of internal connections is of the tions. The next (sixth) column includes the processor time
same importance because it allows transferring of forces in used to solve the optimization problem. The next column
different directions. It is clearly visible on results presented presents the normalized, non-dimensional stress volume
in Figs. 4 and 5. They refer to the ground structures with defined by (10) and the last column includes the relative
internal density 1 × 1. It means that only horizontal, ver- error corresponding to the exact solution given by Lewiński
tical and 45◦ -slope bars are included, like in Fig. 1a. It is et al. (1994a). Note that the speed and the numerical stability
no matter what the external density is used—in all cases the of the program are very good. The sparse matrix represen-
optimal volume is equal to 15.0 Ph/σ0 ; this result is over tation allows preserving a lot of memory. For example, in

Table 1 Results of the first

example regarding to efficiency Mesh density No. No. bars RAM for H in [MB] CPU Volume V σ Relative
nodes time [s] error [%]
Sparse Full

60 × 20:1 × 1 1,281 4,880 0.34 188 0.3 15.0000 10.32

60 × 20:2 × 2 1,281 9,520 0.76 366 1.0 13.8671 1.98
60 × 20:3 × 3 1,281 18,328 1.56 705 1.8 13.6953 0.72
60 × 20:4 × 4 1,281 26,672 2.32 1,026 3.2 13.6580 0.45
60 × 20:5 × 5 1,281 42,448 3.75 1,632 5.4 13.6439 0.34
60 × 20:10 × 10 1,281 113,912 10.23 4,380 10.7 13.6350 0.28
60 × 20:20 × 20 1,281 280,136 25.29 10,772 43.3 13.6343 0.27
120 × 40:10 × 10 4,961 532,872 48.17 80,009 147.0 13.6126 0.11
120 × 40:20 × 20 4,961 1,745,496 158.68 262,080 768.0 13.6120 0.11
188 T. Sokół

Fig. 15 Exact-analytical solution from Sokół and Lewiński

(submitted)
b

Fig. 16 Michell cantilever problem for σT = σC ; see Graczykowski

and Lewiński (2010)

d
different densities of the mesh are shown in Fig. 14. The
results are collected in Table 2. The layouts of Fig. 14 have
recently played the role of hints of the analytical solution.
This solution is being put forward in the paper by Sokół and
Fig. 14 Optimal layouts for different densities of the ground structure Lewiński (submitted); see Fig. 15, where the optimal layout
of the problem of Fig. 13 is reported.
All trusses shown in Fig. 14 are structurally unstable.
the last densest mesh almost 3.5 millions of design vari- The same property characterizes the Michell continuum
ables were used (two times the number of bars). This task in Fig. 15. Despite this instability the solutions are cor-
requires more than 250 GB RAM just for storing matrix H rect, because the virtual work of the forces P on the
in full form and only 159 MB of RAM using sparse matrix zero-energy modes vanishes. This is discussed in detail in
representation. The profit is evident. The calculations were Sokół and Lewiński (submitted). The great advantage of
performed on Intel Core 2 Duo E8400/3 GHz based PC with the numerical method proposed is that it does not exclude
4 GB of RAM. This fact clearly demonstrates the power such unstable structures from the algorithm. This complies
and efficiency of the presented approach. The last problem with the Michell remark no III, p. 591 in Michell (1904).
took only about 13 min, which is an excellent result for so It should also be noted that due to direct solving of the lin-
large-scale problem. ear programming problem (2) the method terminates with a
The second example is shown in Fig. 13. It is a struc- guaranteed correct result.
ture transmitting two symmetrically located vertical forces The last example concerns the reliability of the pro-
to two fixed supports. The forces are applied at 1/4 and 3/4 gram in the case of different stress limits for tension and
of the span length. The feasible domain is the half plane compression: σT = σC . The corresponding optimization
over the line linking the supports. The layouts obtained for problem for this case is given in (13). The exact results of

Table 2 Results of the second

example regarding to efficiency Mesh density No. No. bars RAM for H in [MB] CPU Volume V σ Relative
nodes time [s] error [%]
Sparse Full

20 × 10:5 × 5 231 5,998 0.5 41 0.2 3.83790 1.78

40 × 20:10 × 10 861 73,172 6.6 1,900 5.0 3.80594 0.93
40 × 20:20 × 20 861 169,796 15 4,400 21.3 3.80593 0.93
80 × 40:20 × 20 3,321 1,115,216 102 112,000 290.0 3.78171 0.29
a Two last results were computed 80 × 49:20 × 20a 4,050 1,441,349 131 177,002 599.7 3.77643 0.15
for the half-symmetric part of 100 × 61:20 × 20a 6,262 2,406,373 219 457,510 1,312.4 3.77509 0.11
the structure
A 99 line code for discretized Michell truss optimization written in Mathematica 189

a volumes obtained in numerical way agree very well with

the exact volumes reported in Graczykowski and Lewiński
(2010). The layouts fit also very well. For example, the
angle of flare of the upper fan matches well the theoretical
√
value of this angle which is equal to arctan( κ). The com-
parison of the results is given in Table 3. One can notice
that the relative error of the volume as well as CPU time
b increase for growing κ. Thus the tasks with κ = 1 are harder
to solve. The program presented in the paper is capable of
solving them successfully.

5 Conclusions
c
The computational program developed should serve as a
convenient tool for predicting new exact solutions to the
Michell problems. The listing of the program can be found
in Appendix 1. It requires the Mathematica, version 6 or
higher. The program is based on the concept of the ground
structure and that is why it produces results which are not
d built on any knowledge of properties of the optimum lay-
outs. This is the essential virtue of the method, since the
complete theory of Michell structures has not been devel-
oped till now and we cannot unconditionally say that the
properties noted in the most cited reference sources hold
good in all specific cases.
The correct prediction of the solution of problem in
Fig. 17 Numerical solutions for: a κ = 1, b κ = 3, c κ = 9, d Fig. 13 seems to be a success, since the pseudo truss in
κ = 100 Fig. 15 is structurally unstable and this solution is highly
sensitive to the position of forces P, in contrast to the
better known cantilever solutions (e.g. Graczykowski and
Lewiński 2010; Lewiński et al. 1994a, b) in which the
the problem shown in Fig. 16 are given by Graczykowski Hencky net is independent of the loading applied.
and Lewiński (2010). The following ratios κ = σT /σC have
been examined: 1, 3, 5, 7, 9, 100. The selected optimal- Acknowledgments The paper was prepared within the Research
numerical layouts are presented in Fig. 17. All of them have Grant no N506 071338, financed by the Polish Ministry of Science
been executed for the ground structure of density 110 × and Higher Education, entitled: Topology Optimization of Engineer-
ing Structures. Simultaneous shaping and local material properties
40:20:20, with 4,551 nodes and 1,587,926 elements. The determination.

Open Access This article is distributed under the terms of the Cre-
ative Commons Attribution Noncommercial License which permits
Table 3 Comparison of analytical and numerical solutions
any noncommercial use, distribution, and reproduction in any medium,
√ provided the original author(s) and source are credited.
κ arctan( κ) Exact Num. vol. Relative CPU
volumea Vσ error [%] time [s]

1 45◦ 11.8273 11.8406 0.112 579

3 60◦ 23.2900 23.3176 0.118 720 References
5 66◦ 34.6664 34.7083 0.121 731
7 69◦ 46,0208 46,0777 0,124 753 Achtziger W (1997) Topology optimization of discrete structures: an
introduction in view of computational and nonsmooth aspects. In:
9 72◦ 57.3665 57.4380 0.125 788 Rozvany GIN (ed) Topology optimization in structural mechan-
100 84◦ – 573.8080 – 889 ics. Springer, Heidelberg, pp 57–100
Achtziger W (2007) On simultaneous optimization of truss geometry
a Exact-analytical results from Graczykowski and Lewiński (2010) and topology. Struct Multidisc Optim 33:285–304
190 T. Sokół

Bhatti MA (2000) Practical optimization methods with Mathematica Lewiński T, Zhou M, Rozvany GIN (1994a) Extended exact solutions
applications. Springer, New York for least-weight truss layouts—part I: cantilever with a horizontal
Champion B, Strzebonski A (2008) Constrained optimization. axis of symmetry. Int J Mech Sci 36:375–398
Wolfram Mathematica Tutorial Collection. http://www.wolfram. Lewiński T, Zhou M, Rozvany GIN (1994b) Extended exact solutions
com/learningcenter/tutorialcollection/ConstrainedOptimization/ for least-weight truss layouts—part II: unsymmetric cantilevers.
ConstrainedOptimization.pdf. Wolfram Research, Inc Int J Mech Sci 36:399–419
Chan HSY (1967) Half-plane slip-line fields and Michell structures. Mehrotra S (1992) On the implementation of a primal-dual interior
Quart J Appl Mech 34:433–448 point method. SIAM J Optim 2:575–601
Dorn WS, Gomory RE, Greenberg HJ (1964) Automatic design of Michell AGM (1904) The limits of economy of material in frame
optimal structures. J Mec 3:25–52 structures. Phil Mag 8:589–597
Gilbert M, Tyas A (2003) Layout optimization of large-scale pin- Nocedal J, Wright S (1999) Numerical optimization. Springer, New
jointed frames. Eng Comput 20:1044–1064 York
Graczykowski C, Lewiński T (2005) The lightest plane structures of a Rozvany GIN (1996) Some shortcomings in Michell’s truss theory.
bounded stress level transmitting a point load to a circular support. Struct Optim 12:244–250
Control Cybern 34:227–253 Rozvany GIN, Bendsøe MP, Kirsch U (1995) Layout optimization of
Graczykowski C, Lewiński T (2007) Michell cantilevers constructed structures. Appl Mech Rev 48:41–119
within trapezoidal domains – part IV: complete exact solu- Sigmund O (2001) A 99 line topology optimization code written in
tions of selected optimal designs and their approximations by Matlab. Struct Multidisc Optim 21:120–127
trusses of finite number of joints. Struct Multidisc Optim 33:113– Sokół T, Lewiński T (2009) Application of the interior point method
129 to truss topology optimization. In: Proceeding of int. conf. on
Graczykowski C, Lewiński T (2010) Michell cantilevers constructed “Lightweight Structures in Civil Engineering” (XV LSCE 2009).
within a halfstrip. Tabulation of selected benchmark results. Struct Micro-Publisher-Consultant-Project, Warsaw, pp 162–168
Multidisc Optim (in press) Wolfram S (2003) The Mathematica book, 5th edn. Wolfram Media,
Hemp WS (1973) Optimum structures. Clarendon, Oxford Champaign
Karmarkar N (1984) A new polynomial time algorithm for linear Wright S (1997) Primal-dual interior-point methods. SIAM,
programming. Combinatorica 4:373–395 Philadelphia, PA