IN I I HNA I IONAI J O I I H N A I , t O R NlIMtHI(’AI M E T H O D S IN E N < i I N t E R I N ( ; . VOL 12.

1853-1871 (1978)

LIMIT ANALYSIS OF STRUCTURES FORMED

FROM RIGID BLOCKS

R. K . LlVESLEYt
University of Cambridge. Cambridge, England

SUMMARY
In this paper a technique previously developed for the analysis of rigid-plastic structural frames is
adapted to provide a formal procedure for finding the limit load of any structure formed from rigid
blocks. In this procedure the load factor is maximized subject to the equilibrium equations of the
structure and linear constraints imposed by criteria of failure at the block interfaces.
In part I of the paper it is assumed that the limiting shear force associated with sliding at a block
interface is independent of the normal component of force across the interface. This assumption means
that the normality rule is satisfied, so that the upper- and lower-bound theorems of classical limit analysis
apply. This part includes a description of a computer program for the collapse analysis of masonry arches
with joints incapable of carrying tension.
In part I1 the limit on the shear force at a block interface is assumed to be that associated with
Coulomb friction. It is not difficult to extend the computational algorithm described in part I to deal with
this situation. However, the failure mechanism computed by the algorithm will not necessarily satisfy the
normality rule. The corresponding limit load may therefore be an over-estimate of the true failure load,
even though it is computed by a lower-bound (equilibrium) approach. A criterion is established for
testing the validity of a failure load computed in these circumstances.

PART I
A general formulation of linear limit analysis problems
In 195 1 Charnes and Greenberg’ showed that if the ‘equilibrium’ and ‘mechanism’ formu-
lations of a limit analysis problem are linearised they give rise to dual linear programming
problems. A systematic procedure for setting up the relevant equations for any specific
structure may be based on either approach-a procedure based on an ‘equilibrium’ formula-
tion may be set out as follows
(a) The structure, whether skeletal or continuous, is divided into elements, in much the
same way as for elastic analysis.
(b) For each element an approximation to the stress state is defined in terms of a finite set of
parameters, here denoted by r.
(c) For each element the equations of equilibrium are written in such a way as to express the
stress-resultants p, acting on the inter-element boundaries in terms of r and any external loads
acting on the element. These stress-resultants are statically equivalent to the actual boundary
stresses, as far as the overall equilibrium of the element is concerned. They may be distributed
(for example, a linear variation of stress along an edge or face) or may consist of concentrated

t Lcclurcr in Engineering and Fellow of Churchill College, Cambridge

0029-5981/78/ 1212-1853$01.00 Received 29 July 1977

01978 by John Wiley & Sons Ltd. Revised 30 March 1978
1853
1854 R. K. LIVESLEY

forces or moments at discrete boundary nodes. These equations are written as

PI = H,r - (Pequivlir

where i scans the boundary segments or node points and @equiv)i represents a set of boundary
stress-resultants statically equivalent to the external loads on the element.’
(d) The expressions for the boundary stress-resultants (1) are substituted into the inter-
element equilibrium equations. These will be nodal equilibrium equations if the pi represent
nodal forces, or boundary equilibrium equations if the pi represent distributed boundary
stress-resultants. Since these equations relate stress-resultants rather than stresses they only
ensure, in general, that equilibrium is satisfied in an overall sense. The complete set of
equations is written as
p=Hr.
The load vector p consists of the equivalent element loads (pequiv);, plus any external loads
which may act at nodes or on boundaries, while r represents the set of r’s for all the elements.
(e) The stress limits associated with each element are linearised and expressed as constraints
on the set of element stress parameters r. By a suitable choice of r, followed if necessary by the
introduction of additional (auxiliary) variables, it is possible to express these constraints in the
form
- r L < r < r .U
(3)
The complete set of constraints for all the elements may be written as
-rL< r s r”. (4)
(f) The limit analysis problem for proportional loading is now written as ‘Maximize the load
factor A, subject to the equilibrium equations Ap = Hr and the yield constraints - r L S r < r U . ’
This statement of the problem may easily be converted to standard linear programming form,
or to a dual (mechanism) form.
Details of the above procedure, as it appears when applied to the collapse analysis of
rigid-jointed frameworks, have been given elsewhere.’ Computer programs for such analyses
may use a standard linear programming algorithm or an algorithm3 specially tailored to the
form of the problem set out in (f) above. (The second alternative has certain practical
advantages.) Whichever approach is chosen, the collapse mechanism may be obtained as a
by-product of the analysis (see Reference 2, p. 99).
The work described in the rest of this paper was stimulated by two papers by Heyman4*’,
who showed that the collapse load of a masonry arch or dome could be formulated as a
problem in limit analysis. That work led the present author to consider the possibility of
converting an existing computer program for plane skeletal frame collapse, based on the
procedure outlined above, into a program for the collapse analysis of plane masonry arches of
arbitrary shape.

The collapse analysis of masonry arches

A typical arch of N trapezoidal blocks is shown in Figure 1. Each block is assumed to be of
constant thickness, the faces of the blocks being perpendicular to the plane of the arch. At first
sight it seems self-evident that the blocks correspond to the ‘elements’ of the procedure set out
in the previous section. However, this superficial identification ignores the fact that the yield
constraints (3) operate on vectors r which are associated with elements, not element boun-
 LIMIT ANALYSIS OF STRUCTURES FORMED FROM RIGID BLOCKS 1855

Live

Figure 1 . A plane arch of N rigid blocks

daries. In the general procedure elements are simply systems of stresses with which are
associated yield constraints, while nodes or boundaries are points, or sets of points, at which
equilibrium must be satisfied. It follows that to apply the general ideas set out in the previous
section we must regard the block interfaces as our elements, and treat the blocks simply as
extended nodes connecting the elements.
Figure 2 shows a typical block interface. The stress distribution over the contact area may be
represented, as far as overall equilibrium is concerned, by the stress-resultants q, s and t. These
three self-equilibrating quantities make up the vector r for the element. Associated with r is a
deformation vector E whose components E ~ E,, , E , ‘correspond’ to q, s and t in a virtual work
sense. These three deformation components are associated with the three types of relative
joint movement shown in Figure 3.
/
/
/

/
/
/

Figure 2. Sign conventions for the components of r

1856 R. K. LIVESLEY

(Cl
Figure 3. Sign conventions for the components of E

The various interface vectors r are related by the equations of equilibrium of the blocks.
Figure 4 shows a typical block J connecting ‘elements’ (i.e. interfaces) J and J + 1. The block
has dead-weight W, and may also support one or more live loads

I:[ m
i

acting at specified points xi,y,. The three scalar equilibrium equations (force equilibrium in the
x and y directions and moment equilibrium about G j )may be written in the form
w,+p/ =Cjr j+H j r i + , , ( j = 1,. . . , N). (5)
Details of the vectors w, p and the matrices C, H are given in the Appendix.
Equations (5) may be written out in matrix form as

+
0
 LIMIT ANALYSIS OF STRUCTURES FORMED FROM RIGID BLOCKS 1857

Figure 4. The forces acting on a typical block

or as
w+p=Hr. (7)
Note that (7) has three more scalar unknowns than scalar equations, corresponding to the
three degrees of statical indeterminacy of the arch in Figure 1.
The constraints on the individual components of the vector r associated with each element
are assumed to be as follows,
(a) - ql-sq 4 q I1. If q = - qL then eq s 0, corresponding to a hinge on the inner surface of
thc arch (i.e. at point R in Figure 3(a)). The limit q = q' corresponds to a joint failing by
crushing, with E~ 3 0.
(b) sL4 s sub. If s = -sL then E, s 0, corresponding to a hinge on the outer surface of the
~

arch (i.e. at point A in Figure 3(b)). The limit s = su again corresponds to failure by crushing of
material.
(c) - t L s t s t". For the present we assume that the limits tL and are independent of the
normal stress-resultants q and s. If I = 1'' or I = - I ~then E , can be non-zero, corresponding to
joint slip in a positive or negative sense.
The constraints for all the blocks may be combined and written as
- rl*s r s r". (8)
We can now use the equilibrium equations (7) and the constraints (8) to formulate two
limit-analysis problems. The first is concerned with whether the arch can support its own
weight. Putting p = O in (7) we look for a solution of the equation w = Hr which satisfies the
constraints (8). If this solution exists we ask what load factor A associated with the live loads
1858 R. K. LIVESLEY

will cause collapse of the arch. This involves maximizing A subject to the equilibrium equations
Ap + w = Hr and the same constraints on r as before.
Both these problems may be reduced to the problem formulated in the previous section.
The dead-load problem is equivalent to the maximization of a parameter A,, where h,w = Hr
and - r L s r s r U , the process being terminated as soon as a value A, = 1 is reached. If the
maximum value of Aw is less than unity then clearly the arch cannot support its own weight. If a
valid solution exists and r, is the value of r associated with this solution, the live-load problem
can be written in the form ‘Maximize A subject to Ap= H(r-r,) and constraints -rL-rw<
r - r, S rU - r,’. This is essentially the same problem as before in the ‘shifted’ variables r - r,.
In the detailed account of the plastic collapse algorithm’ it is shown that the final pivotal row
of the modified matrix H defines the collapse mechanism. This row gives (to within an arbitrary
multiplier) the values of the variables which correspond in a virtual work sense to the various
components of r. In the present case these variables are the deformations E ~ E-, E,. A simple
extension of the algorithm gives the values of the variables which correspond, agdin in a virtual
work sense, to the components of the load vectors p and w. These variables represent the
translations of the centres of gravity of the blocks and the rotations of the blocks about an axis
perpendicular to the plane of the arch.
Modification of the original frame-collapse program in accordance with the above analysis
was carried out without difficulty. A typical example of graphical output from this program is
shown in Figure 5. Figure 5(a) shows the approximate line of thrust in the arch. This is
approximate in the sense that the only points on the line which are computed are the points
where it intersects t h e block faces, these points being joined by straight lines within the blocks.
Figure 5(b) shows the collapse mechanism. Note that the mechanism produced by the analysis
is essentially an ‘infinitesimal displacement’ mechanism, since it comes from a matrix H based
on the undeformed geometry of the arch. It is scaled up in Figure 5(b) to give a reasonable
visual impression when plotted, although there is of course a difference between scaled
infinitesimal displacements and a truly compatible set of finite displacements. The procedure
used by Marchand6 to trace the effect of finite geometry changes on the collapse load of a
rigid-plastic frame could be used equally well in the present case.t

Extensions to continuum problems

The procedure described in the previous sections can easily be extended to multi-span
arches or more general plane continuum problems. The basic elements are still the block
interfaces, the only difference being that some nodes (i.e. blocks) now connect three or more
interfaces. Such nodes will give rise to equilibrium equations (5) involving three or more r’s,
and the structure of the matrix H in (7) will be correspondingly less regular.
If failure can only occur by sliding at the block interfaces then there is no need to divide the
normal stress-resultant on an interface into the two components q and s. It is sufficient to
define . . vector r for each interface with only two components-a normal and a tangential
stress-resultant. A typical triangular block relating three r’s is shown in Figure 6. Even though
it does not constitute an ‘element’ in the present analysis, it has an obvious similarity to the
triangular membrane elements introduced by Thomann’ and de Veubeke.’
An extension to three dimensions changes the basic element, the block interface, from a line
segment to a region of a plane. The stresses acting on such a plane can easily be represented by

t Note, however, that there is one important difference between the two problems. A plastic hinge becomes elastic
again as soon as the associated moment reduces in value. In a masonry arch a joint which has opened has to close
completely before the line of thrust can move from the edge of the interface.
I.IMIT ANALYSIS O F STRUCTURES FORMED FROM RIGID BLOCKS 1859
1860 R. K. LIVESLEY

Fig. 6. Triangular block element in a plane continuum

a finite number of stress-resultants, and to that extent the essentials of the analysis are
unchanged. However, the tangential stress-resultant now has two components, say t and t,,,
s
and the constraint associated with yield by sliding has the non-linear form - t L S J ( t , + 1;)s
I". Since the algorithm used in the analysis of the previous section is based on a linear
programming approach, it is necessary to replace all such non-linear constraints by constraints
which are piecewise-linear. The non-linearity which occurs here is very similar to the non-
linear stress constraints which occur in a finite element approach to the limit analysis of
rigid/perfectly-plastic continua.

PART I1
A paper describing a computing technique will usually begin by establishing a sound theoreti-
cal basis, then move on to discuss points of computational detail and finally conclude with
illustrative examples. The actual processes of research do indeed often follow this sequence,
and purists may argue that they should always do so. However, other less logical patterns do
occur and it seems to the present author that there is value in occasionally following the actual
chronological development of an idea. This belief influenced the arrangement of the remain-
ing sections of this paper. As the reader will discover, the analysis of the following two sections
involves an unjustified assumption, and a computer program based on it will, on occasion,
produce incorrect results. However, it was the obvious falsity of some of these results which
provided the stimulus for the development of the analysis presented in the final sections of the
paper.

Yield constraints involving Coulomb friction-a naive approach

We now return to the problem of the general plane arch discussed in Section 'The collapse
analysis of masonry arches'. If the condition at a block interface is one of simple Coulomb
friction with no cohesion, the constraints o n the components of r are
(a) 0 5q s q",
(b) O S S S S " ,
(c) - p(q + s)s t S p(q + s). 11 As before, q U and s u are limits
set by crushing of material.
(Note that q + s 3 0 )

If the associated limit analysis problem is solved by using a standard linear-programming

algorithm the nature of the constraint (c) imposes no particular problems, since the algorithm
assumes the constraints to be linear but otherwise treats them quite generally. If the 'bounded
variables' algorithm mentioned in Section 'A general formulation of linear limit analysis
problems' is used, the constraints can be put in the required simple form (3) by the intro-
 I.IMIT ANALYSIS O F STRIJC'TURES FORMED FROM RIGID BLOCKS 1861

duction of two additional variables at cach interface,

11 = /L(q +s)- 1. u = p(y +s)+ 1. (9)
The vector r for an interface is now regarded as consisting of t h e tivet quantities q, s, f, 14, u, the
constraints being
osqsq".
0 s s s sl',
I unconstrained,
0s u,
0 s u.
These constraints are in the form -rl s r s r". Equations (9) can be written as
O = Br,
where

Similar equations hold for each interface.

T h e homogeneous equations ( 1 0 ) may be combined with the equations of block equilibrium
(A) t o give

WI PI

WN PN
t
0 0 1.
\
\
0
\
\
,
\
\
0 \
\
0 0 'BN

(Note that, since each r now has 5 components, columns of zeros must be added to the G and
H matrices in the positions associated with the additional variables u and v . ) Equation (1 1)
can be written in the same form as (7),
w + p = Hr ,
provided that the quantities w, p, H and r are appropriately re-defined. The dead-weight and
live-load limit analysis problems now have exactly the same algebraic form as before.
t The three linearly independentvariables q, s and f still detine the state of stress at an interface. T h e variables 14 and JI
iire introduccd simply for computational convenience.
1862 R. K. LIVESLEY

The computer program described in Section ‘The collapse analysis of masonry arches’ was
easily modified in accordance with the above analysis. However, one of the first test pro-
blems-a simple two-block arch with a sliding failure mechanism, produced a surprising result.
The value of the limit load was computed correctly, and the internal stress-resultants satisfied
the equilibrium equations and the yield constraints. However, the computer plot of the failure
mechanism, shown in Figure 7, was clearly incorrect.

Figure 7. Computer plot of incorrect failure mechanism for a system of two blocks with Coulomb friction

Correcting the computer plot

The reason for this failure to produce the correct mechanism lies in the fact that the solution
algorithm computes the failure mechanism as a set of deformations which ‘correspond’ in a
work sense to those elements of r which have attained limiting values. For example, if a
particular q is at its lower (zero) limit the corresponding E~ gives the amount by which the joint
opens out in the failure mechanism. Similarly, if failure occurs by sliding with Coulomb
friction, the stress-resultant which reaches its limiting value is either u or u, and the computed
non-zero deformation variable is the ‘corresponding’ E , or E,. This has displacement
components both normal and tangential to the interface, and thus produces a failure
mechanism such as that shown in Figure 7.t To put it more concisely, the computer algorithm
assumes that t h e deformation mechanism obeys the normality rule, but this rule is not obeyed
by a mechanism involving Coulomb friction-a fact originally pointed out by Drucker.’
An example which illustrates this point is the block on a rough surface shown in Figure B(a).
The interface stress-resultants are simply the quantities n and r, and the ‘corresponding’
deformation variables E,, E , are associated with movements of the block in the directions
shown in the figure (the sign conventions are the same as in Figures 2 and 3). If we write
u = p n - I, u = p n + I then the region of the n, f plane for which equilibrium states exist is the
region u a 0, u 2 0 shown in figure 8(b). Since n = W and r = P, the limiting value of P is that
associated with the point A in the figure. Applying the normality rule will give a failure

t This phenomenon is not a consequence of the introduction of u and u as algebraic quantities. An analysis by a
standard linear-programming algorithm will produce a similar result if an attempt is made to obtain the failure
mechanism from the dual variables, since the algebraic concept of duality corresponds to the existence of a work
relationship between ‘stress-resultant’ and ‘deformation’ variables.
 LIMIT ANALYSIS OF STRUCTURES FORMED FROM RIGID BLOCKS 1863

7777

Figure 8 . (a) Forces acting o n a block on a rough surface, (b) Permissible region in n, r space showing correct Failure
mechanism E and incorrect mechanism E,,,,

mechanism enor,,,in which E. = -pe,, while the correct mechanism E has en = O . This is
essentially the phenomenon which appears in Figure 7.
This example may also be used to show how the correct mechanism can be obtained. Once
the solution has been determined (point A in Figure 8(b)) the active constraint u C 0 may be
replaced by the constraint t c f A without altering the solution as far as the values of t and n are
concerned. With the constraint I S t A active, an application of the normality rule gives the
correct deformation mechanism, as shown in Figure 8(b).
In t h e general case an incorrect failure mechanism can be corrected in exactly the same way.
When the process of maximizing A terminates, the algorithm will have produced the set of
stress-resultants associated with collapse, including values of all the shear stress-resultants f,.
(Note that there are no direct constraints imposed on these variables.) If a particular interface
k is in a state of limiting friction (i.e. uk = 0 or u k = 0) then the Coulomb friction constraint at
this interface may be replaced by a simple constraint --tk C t k or t h S t y , where - fk or I: is the
numerical value of fk which has just been computed. Making this constraint ‘active’ in place of
the constraint on uk or t)h involves a Gauss-Jordan transformation? of the coefficient matrix H
which sets the element of the pivotal row of H corresponding to fk to the correct value of the
deformation variable (&,)k. From a programming point of view it is a simple matter to scan all
the interface stress-resultants and carry o u t this process where necessary.

The validity of limit analysis in the presence of Coulomb friction

The incorrect failure mechanism shown in Figure 7 was referred to earlier as ‘a surprising
result’. However, the real surprise lies in the correctness of the associated load factor and
stress-resultant vector. For it is clear that the method of determining the failure load set out in

t In linear programming terms this involves converting r, from a ‘basic’ to a ‘non-basic’variable (see Reference 1,
Sec. 6.5).
1864 R. K. LIVESLEY

Section 'Yield constraints involving Coulomb friction-a naive approach' depends on the
lower bound theorem, and the paper by Drucker mentioned earlier shows that in general this
theorem is not true when yield is associated with Coulomb friction.
Drucker's disproof of the theorem is by means of a counter-example. The existence of this
counter-example, together with other examples for which the method of Section 'Yield
constraints involving Coulomb friction--a naive approach' gives the correct load factor,
implies that problems involving Coulomb friction are of two types-those where limit analysis
gives the correct failure load and those where it does not. In this section we show why this
division occurs.
In the following discussion of t h e lower bound theorem we assume that the stress-resultant
vector r at each interface has three components q, s and 1, with constraints

The yield surface for a single interface is shown in Figure 9. This surface is associated with
cohesionless Coulomb friction-the introduction of cohesive terms into (1 2) alters the pro-

\
s=su

\
q =q'

\
p ( q + s l + t=O 8'
Figure Y. Three-dimensional yield surface for an interface with Coulomb friction
 LIMIT ANALYSIS OF STRUCTURES FORMED FROM RIGID BLOCKS 1865

portions of the yield surface but does not affect the argument we shall develop. Figure 9 also
shows the directions of the deformations which can exist under various conditions of limiting
equilibrium. On each face the deformation has a unique direction, while on each edge it must
lie within a sector of a plane, as shown in the figure. Note particularly that deformations
associated with points on the planes OABC, OA'B'C' correspond to sliding at the interface
and do not obey the normality rule. At the vertex 0 the deformation must lie in the region
, s 0. The complete set of constraints associated with all the interfaces generates a closed
E ~ E~
yield surface in r-space which takes the form of a convex polyhedral cone, as shown in Figure 10.

Figure 10. The yield surface for an assembly of blocks with Coulomb friction

The computer algorithm used in this paper (or the equivalent linear programming algorithm)
maximizes A with respect to all stress-resultant systems r which satisfy both the equilibrium
equations and the yield constraints, attaining a solution such as the point C in Figure 10. We
have to consider whether failure can occur at some point on the yield surface such as A, where
AA Ac.
If normality holds, such a failure is impossible. The proof may be set out as follows. Assume
that a failure mechanism defined by a set of compatible deformations € A exists at A. If the
r-space has M dimensions and there are N equations of equilibrium Ap + w = Hr, then for any
given A these N equations define a hyperplane of M - N dimensions in the r-space. The
hyperplane of constant A passing through A is AAp+w=H(rA+rX),where rx is a stress-
resultant vector of arbitrary magnitude satisfying the homogeneous equilibrium equations
Hr = 0. Since rx defines a self-equilibrating stress state the principle of virtual work implies that
rk E A = 0,and hence that the vector € A is normal to the hyperplane A = A A . If € A is also normal
1866 R. K. LIVESLEY

to the yield surface (or lies within the cone of outward normals if the surface has no uniquely
defined normal at A) it follows that the hyperplane A = A A is tangent to the yield surface, and
hence that A A = Ac. This is the usual proof of the lower bound theorem.
Without normality the argument follows a somewhat different course. Again our hypothesis
is that a failure mechanism exists at a point A on the yield surface, where O < A A < Ac. (If
normality does not hold, this mechanism must involve sliding on at least one interface.) It is
clear that the hyperplane A = A A is parallel to the tangent hyperplane A = Ac. Since the yield
surface is convex, and since the hyperplane A = 0 contains a point W (the dead loading state)
interior to the yield surface, it follows that the hyperplane A = A A must also contain points
interior to the yield surface, as shown in Figure 10. Since this hyperplane has the equation
A A p + w = H(rA+ rx), where Hrx= 0, at least one non-zero vector rx must exist such that
r = r A + r X lies within the yield surface. Thus we deduce that, if failure is to occur at A with
stress state r A , a non-zero solution of the homogeneous equations H r = O must exist which
satisfies (as inequalities) those constraints which are active at A. In this situation a maximizing
algorithm will be able to find an increment to r A which corresponds to movement along the
yield surface, with A increasing, until a new yield constraint is encountered. Further progress
will depend on the existence of a self-equilibrating rx which satisfies this new constraint, and
the process will terminate (at C) when no rx can be found.
As ansxample we consider the system of two blocks shown in figure 11. We assume that any
failure mechanism is such that the blocks do not rotate, so that the quantities q and s can be

Figure 11. Interface forces in a system of two blocks

combined into a single normal stress-resultant n, with upper limit nu. The sign convention for
positive n and t is that used in the rest of the paper. We assume that there is no friction at
interface 2, and that the angles of friction at interfaces 1 and 3 are a l and a3 where a1> a3.
Since there is no friction at interface 2 the only self-equilibrating force system which can exist
is the force rx shown in the figure.
First consider the case where a > a l > a 3 . Physical intuition tells us that the failure
mechanism is that shown in Figure 12(a), and it is a matter of elementary statics to obtain an
expression for the value of P at which failure occurs. The yield surface for the system can be
represented by the two triangles shown in Figure 13(a),the stress state at failure being indicated
by the points marked A.
 LIMIT ANALYSIS OF STRUCTURES FORMED FROM RIGID BLOCKS 1867

Figure 12. (a) Failure mechanism for a > a1> a3, (b) Failure mechanism for a,> a > a3 with rx at its maximum
permissible value, (c) Failure mechanism for a I>a,> a,with rx at its maximum permissible value

Application of a limit analysis algorithm will reach this stress state. Since a is greater than
both a1 and a3 there is no self-equilibrating rx satisfying the yield constraints at A and the
algorithm therefore terminates. Application of the mechanism correction process described in
Section ‘Correcting the computer plot’ gives the deformations c l , ~3 shown in Figure 13(a),
which correspond to the mechanism shown in Figure 12(a). Thus for this case limit analysis gives
the correct load factor and the correct mechanism, even though normality does not hold.
Next consider the case a,> a > a3.The failure mechanism Figure 12(a) can still occur, the
associated value of P being given by the same expression as before. However, the yield surface
is now represented by Figure 13(b). It is clear from the figure that although it is not possible to
find an rx which satisfies all the yield constraints when P = W = 0, one can be found as an
increment to the stress state denoted by the two points A. This increment is shown by the two
lines AB in Figure 13(b). Application of a limit analysis algorithm will cause movement to the
points marked C. The mechanism correction process will produce the deformations € 1 , ~3 shown
1868 K. K . LIVESLEY

Figure 13. Permissible stress-resultants and deformations for (a) a > a l > a3,( b ) al > a > u3,(c)a l > a3> a

in the figure, the complete mechanism being shown in Figure 12(b).It is clear from Figure 13(b)
that the self-equilibrating force rx does positive work in this mechanism. This can only occur if at
least one of the boundaries 1 and 3 moves inwards, as shown in Figure 12(b).
The case a l > a j > a has similar characteristics. An rx exists which satisfies all the yield
constraints when P = W = 0; it therefore satisfies the yield constraints at the points marked A
in Figure 13(c). Thus a limit analysis algorithm will move from this state to the points marked
C. The (corrected) deformations E , , F~ associated with this limit state are shown in the figure,
and the corresponding mechanism is shown in Figure 12(c). Once again we see that the
mechanism is such that the self-equilibrating force rx does work in the deformation.
It is apparent that in the last two cases the value of P obtained by a limit analysis approxh
will only be obtained if some external agency is available to raise rx to its limiting value. In
practice equilibrium will first be broken at some point below C corresponding to the unknown
initial value of rx. A very small amount of deformation in mechanism Figure 12(b) or 12(c) will
cause rx to drop to the value associated with mechanism Figure 12(a) and deformation will
then continue at the value of P associated with that mechanism.
 LIMIT ANALYSIS OF STRUCTURES FORMED FROM RIGID BLOCKS 1869

In the general case the same phenomenon occurs. If a compatible failure mechanism € A
exists with A A < Ac then, as wc have seen, a self-equilibrating system rx must exist which
satislies all t h e yield constraints active at A. Application of a limit analysis algorithm will reach
the load factor Ac, but the corrected mechanism E~ will be such that r;EC>O, the positive
work done by rx implying some movement of supposedly fixed surfaces.? The result may be
stated as follows,
'If the maximizing algorithm produces a corrected failure mechanism E~ in which the
boundaries do positive work then a failure mechanism exists at some A A < Ac and hc will not,
in general, be attained.'
The converse is,
'If the maximizing algorithm produces a corrected failure mechanism E~ in which the
boundaries do no work then the load factor hc is correct, even if the mechanism does not
satisfy the normality rule.'
These two statements separate problems involving Coulomb friction into the two classes
mentioned at the start of this section.
Note that the existence of a self-equilibrating rx which does not violate the yield conditions
is not, of itself, sufficient to ensure failure at a load lower than that obtained by limit
analysis. It is the fact that rx does positive work in the computed failure mechanism which
is the essential feature. For example, consider a block held between two rough vertical walls
and acted o n by a vertical force P. The walls exert lateral forces o n the block, forming a
self-equilibrating pair of forces which constitute rx. The maximizing algorithm will increase
t h e limiting value of P by increasing rx until a crushing constraint is reached. If P acts centrally
on the block then the increase in rx implies an external agency capable of doing work, and in
the corrected failure mechanism one of the walls will move inward. If, however, P acts
off-centre so that the block jams then in the real system rx will rise quite naturally until a
crushing limit is reached. The limit analysis approach will produce this limit, and its correctness
will be indicated by the fact that in the computed mechanism the walls do not move inward and
rx no work.
In this section we have considered the validity of the lower bound theorem in the presence
of Coulomb friction. The validity of the upper bound theorem has been considered by
Collins,"' and the general problem has also been discussed by Prentis." Collins' approach has
an interesting link with the analysis of the previous section in that he derives an extremum
principle by considering failure mechanisms (or, alternatively, velocity distributions) which do
not necessarily satisfy all the displacement boundary conditions. Such mechanisms, as we have
seen, may be produced by the lower bound procedure (see Figure 7, for example).

CONCLUSION
The results deduced in the previous section show that limit analysis fails t o give a 'correct'
answer only in those friction problems where the failure load is affected by boundary move-
ments. A lower bound algorithm will over-estimate the failure load in such cases, but the
existence of the over-estimate can be detected immediately by inspecting the (corrected)
failure mechanism for positive boundary work. The computed failure state is a 'passive
Rankine state' in soil mechanics terminology.
The previous section also shows that an over-estimate of the failure load-factor requires the
existence of a set of self-equilibrating stress resultants rx. Although strictly such a set need

t The general result follows from the convexity of the yield surface in Figure 10.
1870 R. K . LIVESLEY

only satisfy the yield constraints at the lower failure load ( A A in the notation of the previous
section) an indication of possible trouble may be obtained by examining the homogeneous
equations Hr = O for non-zero solutions satisfying all the constraints o n r (this may be
converted into the problem of finding a basic feasible solution in linear programming). In the
case of an arch it is usually obvious whether such a solution exists. For example, in the arch
shown in figure 5 there is clearly no set of self-equilibrating forces which can exist when
p = w = 0, whatever the value of p.
The first part of this paper described a straightforward application of limit analysis prin-
ciples, in which the theory was translated into a computer program which behaved exactly as
expected. In the second part, however, the direction of the work was determined firstly by a
desire to understand computer results which appeared to be wrong and secondly by a need to
justify correct results which seemed to be based on an unsound method.

APPENDIX
Derivation of the block equilibrium equations
We consider the block shown in Figure 14. The block geometry is assumed to be defined by
the co-ordinates of the vertices A, B, C and D. The co-ordinates of the mid-points of the faces
AB and CD are denoted by x l ryl; XZ, y2 respectively. For brevity we write sin exl = S1,
cos a l= C1,etc.
The area of the quadrilateral ABCD is given by
+ AOBC+AOCD+ AoDA,
A = AOAB

A
C

Figure 14. (Appendix) Notation for block equilibrium equations

 LIMIT ANALYSIS OF STRUCTURES FORMED FROM RIGID BLOCKS 1871

where 0 is the origin of co-ordinates and

AoAB = ( x A Y B - x B Y A ) / ~ , etc.
The weight of the block is W = pAh, where h is the dimension of the block perpendicular to
the plane of the arch. The co-ordinates of the centre of gravity G are
xG=[AOAB(XA+XB)+ ... + . .. + * . *]/3A,
YG + YB)+
= [AOAB(YA . . . + . .. + . . ]/3A,
*

The three equations of plane equilibrium for the block may be written
c pxt = - G q , - CI SI + Sl t l +c 2 q 2 +G S 2 - SZt29

xpyt- w = - s,91 - SI - S1 1I ]
c + S2qz + S Z S 2 + c2 t 2 ,
2: [pyix: -pxiy: + mi I = [CIYX - St XX 191+ [CIYL - SIXL ]S I - [SIY ; + C,x’l I l l
-[CzyL -S2xL]q2-[Czy.L -szxL-]S2+[S2y; + c2x;1t2,
where x i = xA-xG, etc. and 1 denotes summation over the live loads on the block. These
equations may be set out in matrix form as

+[ 2
S*xL - C2yL s2x;
c
2
s2

- C*y& s2y;
2+ 1
c2x;
or as
w + p = Grl + Hr2.
This is the form used in equation (5).

REFERENCES

1. A: Charnes and H. J. Greenberg, ‘Plastic collapse and linear programming’, Summer meeting Amer. Marh. Soc.
( I 95 1 ).
2. R. H. Gallagher and 0. C. Zienkiewin (Ed.), Opfimurn Structural Design, Wiley, (London, 1973), Chap. 6.
3. R. K. Livesley, ‘A compact Fortran sequence for limit analysis’ Inr. J. num. Meth. Engng. 5,446-9 (1973).
4. J . Heyman, ‘The stone skeleton’, Inr. J. Solids Srrucr. 2, 249-279 (1966).
5 . J . Heyman, ‘On shell solutions for masonry domes’, Inr. J. Solids Srrucr. 3, 227-241 (1967).
6. R. K. Livesley, Manix Merhods of Sfrucfural Analysis, 2nd edn., Pergamon, Oxford, 1975, p. 207.
7. G. E. A. Thomann, ‘Aeroelastic problems of low aspect ratio wings’, Aircrafr Engng, 28, 36-42 (1956).
8. B. Fraeijs de Veubeke (Ed.), Marrix Merhods of Sfrucrural Analysis, AGARDograph No. 72, Pergamon, Oxford,
1964, p. 170.
9. D. C. Drucker, ‘Coulomb friction, plasticity and limit loads’, J. Appl. Mech. 21. 71-74 (1954).
10. 1. F. Collins, ‘The upper bound theorem for rigid/plastic solids generalised to include Coulomb friction’, 1.Mech.
Phys. Solids. 17, 323-338 (1969).
I I . J. M. Prenlis. ‘Use of upper and lower bounds in evaluating friction elfects in mechanisms’, SYROM Conk
Bucharest (1977).