Hambly’s paradox: why design Proc. Instn Civ.

Engrs,
Civ. Engng, 1996,

calculations do not reflect real 114, Nov., 161-166

Paper 11082
behaviour Written discussion closes
15 December 1996
J. Heyman, FSA, FICE, FEng

Past ICE president Edmund Hambly used to use the simple example of a
four-legged stool to show that engineering design calculations are fre-
quently wide of the mark when it comes to analysing real-world structures.
This paper explores these ideas, which highlight the fact that conventional
elastic and plastic design processes—the basis of most contemporary com-
puter design software—still cannot accurately quantify many real loading
conditions. Static design equations for a redundant structure have an infi-
nite number of solutions, whereas simple plastic analysis does not account
for instability. With increasing reliance on computers, a new set of design
rules is needed.

The paradox
Edmund Hambly, who died in 1995
while holding the office of
President of the Institution of Civil
Engineers, would sometimes set a
simple problem to a class studying
the theory of structures. With a
slightly different formulation, the
problem is as follows.

A man weighing 600 N sits on a

three-legged stool. For what basic
force (incorporating a suitable fac-
tor of safety) should each leg be
designed? (Fig. 1).

There is no catch to this problem—

the stool is supposed to be symmet-
rical, the man sits in the centre of
the seat, and so on. The answer is,
of course, that each leg should be
designed to carry a force of 200 N.
It is the next question that is the
tricky one.

The same man now sits on a

square stool with four legs, one at
each corner, and again the stool
and the loading are symmetrical.
For what force should each leg of
the stool be designed?
Jacques Heyman is
Fig. 1. Man on a three-legged The answer of 150 N is wrong, and professor emeritus and
stool—one of the few situations this is the point that Hambly former head of the
that engineers can analyse with wished to drive home. A robust, engineering department
certainty (‘An illuminator’, Jost nearly-rigid stool, standing on a at Cambridge
Amman, 1568) firm, nearly-rigid floor, will rock; University

161
 HEYMAN

Fig. 2. Computers do
not absolve design
engineers from consid-
ering the most critical
loading conditions
(‘Marital quarrel’,
Hans Sebald Beham,
c. 1535)

three of the legs will be in contact, supporting the provides material for discussion of the whole
weight of the man, but the fourth will be clear of process of structural design.
the floor. If this fourth leg is clear by only a frac-
tion of a millimetre, then it is certain that the The design process
force it is carrying is zero. By simple statics (as There are two stages of structural design or
noted below) the force in the leg diagonally analysis. The first stage consists of finding the
opposite will also be zero, even if this leg appears values of the external forces which act on a
to be in contact with the floor. The weight of the structure (including among those forces the
man is in fact supported symmetrically by the reactions supplied by the external environment
other two legs of the stool, and each must in response to the given loading) together with
therefore be designed to carry a force of 300 N. the values of the internal stress resultants,
Now the stool may be imagined to be placed such as bending moments, axial forces and so
arbitrarily on a randomly rough floor and, as on. Thus the weight of the man induces reac-
there is no way of deciding which leg will be off tions on the legs of the stool from the floor,
the floor, all legs must therefore be designed to and the values of those reactions must be
carry a force of 300 N. There is thus a paradox found. In Hambly’s problem, the forces in the
apparent in the problem of Hambly’s stool—the legs of the stool are the internal stress resul-
addition of a fourth leg to a three-legged stool tants and they happen, of course, in this ver y
increases, rather than decreases, the force for simple structure, to be equal to the reactions
which each leg must be designed. This paradox from the floor.

162
 HAMBLY’S
PARADOX

If both the external reactions and the internal tion. Two other statements must be made before
stress resultants can be found by the use of the a solution can be achieved—something must be
equations of statics alone, then the structure is, said about the material properties of the struc-
by definition, statically determinate. For the three ture, and some geometrical conditions must be
legged-stool (and ignoring ideas of symmetry of imposed. The material may be assumed to be lin-
loading) three equations of overall equilibrium ear elastic, for example, in which case an ‘elastic
may be written—vertical resolution of forces, and solution’ will be derived.
moments about two horizontal axes. These three The geometrical statements are sometimes
equations suffice to determine the three support referred to as compatibility conditions. If the
reactions, and hence the three values of the structural elements are analysed individually,
forces in the legs. then the deformations of the elements must be
The four-legged stool is statically indetermi- compatible—the elements are required to fit
nate; only the same three equations may be writ- together in a specified way. Most importantly,
ten, and there are four reactions to be found. If something must be stated about the way the
these reactions are R1, R2, R3 and R4, and the man structure as a whole is connected to its founda-
sits centrally but no assumption is made about tions.
any consequent symmetry in the values of the All of this information, resulting from the equa-
reactions, then at least it is known that the sum tions of statics, the material properties, and the
of the four reactions is 600 N. The two further geometrical statements (including the ‘boundary
equations obtained by taking moments show that conditions’), must be used in order to solve the
diagonally situated legs carry equal loads, that is prime structural problem—the determination of
R1 = R3 and R2 = R4. Thus R1 + R2 = 300, but no the values of the internal stress resultants. Since,
further information results from the equilibrium traditionally, the material has been taken to be
equations. However, the physical problem elastic for the second stage, that of calculating
requires that both R1 and R2 should be positive, stresses, the whole process leads to a convent-
so that 0 R1, R2 ≤ 300. ional elastic design.
It may be remarked that the statement that In practice, the elastic problem of the
only three equations of overall equilibrium are four-legged stool is really rather difficult. If there
available for analysis of the stool results from a is no symmetry to the loading (the man is off cen-
preliminary simplification in the modelling of the tre) then, as has been seen, the equations of stat-
problem. For example, the designer will have ics give only three equations between the values
assumed that the legs make point contact with of the four reactions supporting the legs—the
the floor—the leg-ends are rounded, perhaps— problem has a single redundancy. To make
and that the coordinates of each point of contact progress, elastic information must be introduced;
are known. Further, it will have been assumed that is, the flexural properties of the slab forming
that the reactions on the legs from the floor are the seat of the stool must be specified, as must
vertical—the floor is smooth. If in fact contact is the axial compressibility of the legs. The flexure
rough, then horizontal forces may act at the feet of a (square) slab supported at the four corners is
of legs; extra equations of equilibrium may be not a simple problem; fortunately, finite-element
written, but the degree of structural redundancy programs exist, and the designer can pass the cal-
rises sharply. Such considerations may or may not culations to a computer.
be of importance in the final assessment of the All of this will appear unnecessarily complex to
design requirements for a leg of the stool, but in the experienced engineer, who is accustomed to
practice they are liable to be ignored altogether. using skill and judgement in making approxima-
tions so that a simpler but still valid solution can
Elastic analysis be obtained. In the case of the stool, for example,
Thus the equations of statics, which form only the designer may well decide (correctly) that the
a part of the theory of structures, will determine slab forming the seat is virtually rigid, and the
the external forces and internal stress resultants corresponding part of the finite-element package
in a statically determinate structure. The engineer can be eliminated. Equally, the designer may
can then proceed with the second stage of struc- decide (also correctly) that the four legs are
tural design—the internal stress resultants must essentially incompressible, and that the whole
lead to acceptable values of stress. The problem problem should be regarded as that of a rigid
of finding the stresses in a structural element structure supported by a rigid floor.
which correspond to the values of the stress The computer package will not, however, allow
resultants falls into the field of strength of materi- the designer to do this—it must be fed with val-
als; for example, known values of bending ues of elastic constants before it will go to work
moment and shearing force in a beam or frame on the problem. The designer will perhaps retain
will lead to calculable stresses. the approximation of a rigid slab, but will intro-
The theory of structures proper is concerned duce some elastic compressibility for the legs; the
with the solution of structures for which the stati- computer will then print out the values of the
cal equations alone do not give enough informa- forces in those legs. The program may be run

163
 HEYMAN

again for different values of the elastic constants, ter loads, it will be found from a series of experi-
and the designer will be reassured that the initial ments that the load in a given leg may have a
common-sense assumption—that the legs could value lying anywhere between 0 and 300 N, and
be regarded as rigid—is apparently justified; no that a fair number of experiments record the load
matter what value is taken for the elastic com- as exactly 0 or 300 N. This situation is by no
pressibility, the computer determines that the means new or artificial.
force in each of the four legs is 150 N. In the 1930s the Steel Structures Research
The designer may then be satisfied that the Committee3, 4 made tests on steel-framed build-
problem has been solved, although, as has been ings then under construction in London; they
noted in the introductory statement of Hambly’s found that measured stresses bore almost no rela-
paradox, the wrong solution has been obtained. It tion to the elastic values predicted by the design-
is clear, of course, where the analysis has erred— ers. The conclusion was that, in practice, the host
the designer (and the computer package) has of very small imperfections inherent in such
assumed tacitly that the floor is rigid and level, structures made elastic analysis the wrong tool
and that all legs are of exactly the same initial for design. It was these observations which led to
length so that all four make the same contact with the development5,6,7 of the plastic method of
the floor. The boundary conditions are apparently design for steel structures (or indeed for a struc-
so simple that no thought has been given to their ture made of any ductile material).
specification. If the boundary conditions are The essence of the words ‘plastic’ and ‘ductile’
thought about, then it is at once clear that they is that a limiting stable behaviour is implied. For
are, in essence, unknown and unknowable. The the simple and familiar example of a transversely
surface of the floor could perhaps be specified, loaded steel beam, an imagined slow increase in
but the man will place his stool at an unknown load will lead to the development of a plastic
location. hinge. Rotation can take place at this hinge lead-
If the point is appreciated at all, then it may ing to large structural deflections, but the process
well be concluded (once again correctly) that triv- is quasi-static, with the applied loads remaining
ial and accidental irregularities of the floor cannot constant. The important feature of this plastic col-
really have any significant influence on the prob- lapse is that a plot of load against displacement
lem whose solution is required—namely, the for the structure, as the applied load is increased,
forces for which the legs of the stool should be will show, after an initial elastic response, a final
designed. However, in a conscientious way, the flat-topped characteristic.
program may be run several times with one leg Reinforced concrete shows the same kind of
initially clear of the floor by say 0·01, 0·1 and behaviour, at least for the range of displacements
1 mm. Only one leg need be considered; a tripod experienced by usual structures. In fact any mate-
is located exactly on a rough floor, and it is only rial (e.g. aluminium alloy or wood) used in practi-
clearances of a fourth leg that can affect the cal design is ductile in this way, as opposed to
analysis. brittle materials (e.g. glass or cast iron) which
The computer output will disagree profoundly will crack if over-strained and which would lead to
with the common-sense approach of the design- catastrophic collapse.
er—very small differences in initial clearance will If then, a plastic analysis is to be made of the
give wide variations in the values of the elastic stool, it must be ensured that final failure of the
forces in the legs. This unexpected and legs is ductile. This implies that each leg should
counter-intuitive result, that small, almost trivial be ‘stocky’, and capable of sustaining a limiting
imperfections can lead to apparently greatly dif- ‘squash load’ without exhibiting unstable behav-
ferent states of the stool, should indicate to the iour. The loading history of such a stool may now
designer that elastic theory is the wrong basis of be followed as a point load at the centre (the
analysis.1,2 Unfortunately, the designer may never man) is slowly increased.
reach this state; without an awareness that bound- Initially, in general, only two legs will be carry-
ary conditions must be introduced specifically ing load. If the squash load of each leg is P, then
before the program is run, the values of 150 N for the two legs will yield, and compress stably, when
each leg may be accepted without question. the point load has a value 2P. The compression of
the yielded legs will allow the unloaded legs to
Simple plastic analysis make firm contact with the floor, and the point
A sophisticated elastic analysis of the four- load may then be increased above the value 2P—
legged stool will fail once the designer realizes in fact, to the value 4P. At this final stage all four
that the theory (or the computer package) cannot legs are carrying their maximum loads, and a
handle essentially unknowable boundary condi- large displacement of the seat of the stool will fol-
tions. A straightforward ‘conventional’ elastic low.
analysis, on the other hand, will give results (all The plastic design of the stool to carry a man
legs at 150 N) which will be contradicted in a sim- of 600 N would therefore require the value of P to
ple way by experimental evidence. If tests are car- be 150 N, and this is the force for which each leg
ried out, with gauges attached to the legs to regis- should be designed—of course, with a suitable

164
 HAMBLY’S
PARADOX

factor. It should perhaps have been stated at the The question has in fact been addressed, and a
outset that the stool is actually being designed for partial answer may be found, for example in the
the sole use of a child, the man’s son, who weighs ‘weak-beam strong-column’ philosophy of design
200 N. The man, who weighs 600 N, is a figment for multi-storey frames.8 The beams in such a
of the designer’s imagination, introduced into the structure are assumed to be stable elements
discussion in order to provide a (load) factor of 3 (involving restraint against lateral buckling), so
in the calculations. that they can be designed to have some sort of
This plastic solution—each leg to be designed minimum section—that is, a plastic design would
for 150 N—has been obtained by following a par- be appropriate. The columns of the multi-storey
ticular loading history, but other histories are pos- frame, however, are given cross-sections which
sible. For example, one leg may be initially only may be more than adequate from the point of
just clear of the floor, and application of the load view of strength, but which ensure that unstable
may bring this leg into contact due to elastic buckling cannot occur.
shortening of the two loaded legs. However, fur- It is the word ‘ensure’ that causes difficulties.
ther increase of loading will eventually, in the If the end conditions on a column are known,
symmetrical case, cause two legs to yield, and the then, despite the fact that buckling is sensitive to
load can be increased until collapse occurs, again imperfections, to residual stresses, and so on,
at a value of 4P. Simple plastic design is indepen- then stability can indeed be assured. In any case
dent of the loading path, and powerful theorems of doubt, a column can be replaced by one slightly
(the upper and lower bound theorems, and the larger, and certainly stable, at almost no penalty
uniqueness theorem) allow the collapse state to of cost or weight. It is, however, the determina-
be examined without reference to loading history. tion of the end conditions that is not easy. Neither
the elastic designer, relying on a computer output,
Design with unstable elements nor the plastic designer, taking known bending
Hambly’s paradox does not exist either for the moments applied to the columns from collapsing
unsophisticated elastic designer or for the simple beams, may be aware of the problem, but neither
plastic designer—both are united in their view is absolved from the duty of trying to assess the
that the legs of the four-legged child’s stool worst end conditions (Fig. 2). These are not given
should be designed for 150 N to support a ficti- by any of the standard computational procedures,
tious man of 600 N (who is really a 200 N child and the pursuit of the consequences of Hambly’s
with a load factor of 3). However, plastic design, paradox leads to largely uncharted territory.
and its comforting ‘safe’ theorem (the lower- Much investigation is needed to establish rules
bound theorem), is valid only if the structural ele- for designers, even for relatively simple building
ments (the legs) are stable. Should a leg buckle, structures.
the load/deflection characteristic for that leg will An answer can be given for the problem of
fall instead of being flat-topped, and the leg will design of Hambly’s four-legged stool. The stool
not possess the required ‘ductility’. itself will not be aware of the thought processes
For example, the simple plastic designer, of its designer. Whether those processes have
aware of the buckling danger but relying mistak- been based on elastic or on plastic methods, the
enly on the safe theorem, may design each leg to stool will in fact, at some time in its existence
buckle at 160 N. If the stool is placed carefully under the weight of a fictitious 600 N man (actual-
and exactly on a flat floor, then the 600 N man will ly a 200 N child), be in danger of experiencing
induce leg forces of 150 N, and all will be well. On maximum loads of 300 N (actually 100 N) in any
a rough floor, however, one leg will not be in ini- of its legs. If the legs are ductile and stable, then
tial contact, and the man’s weight will cause (at the designer can achieve a load factor of 3 by lim-
least) one leg to buckle and to deform permanent- iting the carrying capacity of each leg to 150 N. If
ly, leading to collapse of the stool. Put another this guarantee of stability cannot be given, then
way, the assumed design load factor against col- each leg must be designed not to buckle under a
lapse has been reduced substantially from 3 to load of 300 N.
1·6.
The arguments apply in exactly the same way Conclusion
to the unsophisticated elastic designer, who is Hambly’s four-legged stool stands, of course,
also designing against instability but who believes for the general problem of design of any redun-
the forces in the legs all to be 150 N. The fact is dant structure. It has long been recognized that,
that elastic analysis cannot be used since the in order to calculate the ‘actual’ state of a struc-
boundary conditions are unknowable, and the ture under specified loading, all three of the
resulting design may be unsafe; equally, plastic basic structural statements must be made—equi-
design is invalid, and known to be unsafe, if there librium, material properties and deformation
are unstable elements in the structure. Hambly’s (compatibility and boundary conditions).
question appears not to have been answered: For However, the calculations do not in fact lead to a
what force should each leg of the stool have been description of the actual state. Boundary condi-
designed? tions are, in general, unknown and unknowable;

165
 HEYMAN

Fig. 3. A small settlement in the

24 4.8
footings (a) is more critical for
designing this portal frame’s
columns than the conditions gener-
ated by computer (c) and (d),
which assume perfect fixity 1

Mp = 6

Plastic design of a uniform portal frame.

Full plastic moment 6 units at unit load factor.

6 6

6 > M > –1.2

(4.8 –M) M
Analysis at collapse. One redundancy remains—
M can have any value between 6 and –1.2.

6 6 6 6

6 1.2 1.2 6
(a) Worst condition for stability of (b) Similarly, best conditions
columns, corresponding to slight for columns.
settlement of left-hand column.

6 6 3.3 6.3
7.2

0.6 5.4 1.5 3.3

(c) Column moments given by elastic- (d) Similarly, wholly elastic
plastic analysis, assuming perfect computer analysis.
fixity at column feet.

an imperfection in assembly, or a small settle- References

ment of a footing, will lead to a state completely 1. HEYMAN J. Plastic design and limit state design.
different from that calculated (Fig. 3). This is Structural Engr, 1973, 51, 127–31.
not a fault of the calculations, whether elastic or 2. HEYMAN J. Elements of the theory of structures.
not—it is a result of the behaviour of the real Cambridge, 1996.
3. STEEL STRUCTURES RESEARCH COMMITTEE. First
structure.
Report/Second Report/Final Report. HMSO,
The equations of statics are reliable, but they London, 1931/1934/1936.
have, for a redundant structure, an infinite num- 4. BAKER J. F. The steel skeleton. Vol. 1: Elastic behav-
ber of solutions. There is no correct solution to iour and design. Cambridge, 1954.
the equations, but one solution will lead to the 5. BAKER J. F., HORNE M. R. and HEYMAN J. The steel
greatest economy in material. This, although it skeleton. Vol. 2: Plastic behaviour and design.
may seem to be in disguise, is the solution sought Cambridge, 1956.
by the simple plastic designer, and it is safe and 6. BAKER J. F. and HEYMAN J. Plastic design of frames.
valid provided that no instability is inherent in the Vol. 1: Fundamentals. Cambridge, 1969.
structure. If, however, there are unstable ele- 7. HEYMAN J. Plastic design of frames. Vol. 2:
Applications. Cambridge, 1971.
ments (say columns in a building frame), then
8. JOINT COMMITTEE OF THE WELDING INSTITUTE AND
safety can no longer be assumed unless the worst THE INSTITUTION OF STRUCTURAL ENGINEERS.
conditions for those elements have been investi- Fully-rigid multi-storey welded steel frames.
gated. No conventional design process addresses Report/Supplement/Second Report. Institution of
this problem. Structural Engineers, London, 1964/1968/1971.

166