Sunteți pe pagina 1din 6





S O M M A R I O : Si affronta il problema dell'analisi limite dei

mezzi continui costituiti di materiale rigido perfettamente plastico aventi snperficie di snervamento di tipo poliedrico. Si definiscono in forma unitaria, avvalendosi di noti concerti di programmazione lineare, i principi di estremo riguardanti la valutaz.ione della polenza specifica di dissipazione. Si deduconopod i
principi fondamentali che governano il problenm del collasso
p/astico attraverso una semplice formulazione basata sul ben
noto teorema cinemalico dell'analisi lim#e. Si sn&gerisce inflne
mt procedimento approsrimato tatt'a~ratlo generate per la valu/azione del carico di col/asso di mt qualsiasi mezzo continuo.
S U M M A R Y : The problem of limit anal-ysis of continuaJormed
b-y rigid-perfectO, plaslic material with piecewise .yield stwface
is discussed. IVith the use of the known concepgs of linear programming, the extremum principles concerning /he determination
of specific dissipation power are defined in unitary form. The
basic principles governing the problem of plastic collapse are then
expressed through a simple formulation based on the well known
kinemaUcal theorem of limit anal),sis. Finally, an approximate
general procedtwe is suegestedfor the calcMation of the collapse
load of continuous media.

1. Introduction.
The limit analysis of rigid-plastic continua with associated flow rules is based on the two known theorems of
structure plastic collapse [1], [2]. However, the systematic
application of such theorems to real structural problems
is generally quite complicated and cannot be generalized
to determine unitary procedures for numerical calculation.
On this subject, many important old and recent studies
[3 through 11 ] refer such problems to the known principles
of linear programming, by connecting the appropriate
linearization of the yield surface with the discretization
of the continuum, thereby offering a systematic procedure
for the numerical solution of such problems. However,
in our opinion, the simultaneous application of both
discretization procedures prevents a full evaluation of
such procedures for the general formulation of the problem.
Continuum discretization - - which can be achieved by
several m~_thods (finite differences, finite elements, series
expansions etc.) - - only represents a mean of numerical
calculation to solve an actual structural problem, while
the linearization of the yield surface implies the possibility
of introducing, in autonomous form, theoretical basic

* Istituto di Scienza delle Costruzioni, Universitb di Bologna.

MARCH 1971

principles in addition to the well known general principles.

The purpose of this study is to formulate a general approach for the limit analysis of rigid-plastic bodies having
piecewise linear yield surface.
The first part of this work establishes the extremum
principles governing the problem of determining the specific dissipation power. In the second part, the fundamental
theorems of limit analysis are discussed in unitary form.
Finally, the third part offers some appropriate numerical
procedures for the actual and systematic application of
such principles to the solution of structural problems.

2. E x t r e m u m principles for the calculation of spe-"

cific dissipation power.

Consider a yield surface, determined in the stress space
o~s by m linear conditions (Fig. 1):
ntso'i.s = cr

(a = 1..... m)


where nts= are the direction cosines of the ctth hyperplane,

and a ~ is the distance from the origin of such hyperplane.
A generic stress vector a = [cris] is within the yield
limits If:
o - n ==alsn~s ~<a ~

( a = l ..... m)


that is, if it is contained by the convex dominion defined

by the yield surface.


Fig. 1.

On the other hand, the known principle of maximum

dissipation states that, for standard materials: the specific

With such rules, the dual minimum program (2.5), or

(2.3), becomes:

dissipation power associated with a given system of strain rates

o is the greatest qf all fictitious powers computed with all possibk states of stresses (70 within the yield limits.
Therefore, in our case, denoting by D(k,~) the specific

D(}o) = rain {~)f],,~2 =/> },j, - - ate2=/> - - ~q, ff/> 0}

dissipation power, in accordance with (2.2), the formulated

principle furnishes:
D(~',~) = max (m~},~]crom
~ ~< (7=}.

for which we use the common notation of tensorial summation also for index a. However, since the two conditions:


Thus, we have a typical problem of linear programming;

i. e. we must find the maximum value of linear function
(Tq~,~ in the convex dominion delimited by m linear inequalities (2.2).
We also know that in linear programming every maximum problem can be associated with a minimum problem
(and viceversa) related to the first one by various duality
properties [12], [14].
The formulation of the <Muab) problem is derived automatically from the ~primab~ problem with a transposition
of the latter written on an appropriate tableau. The transpositron should make some changes in the primal problem so that all its variables are non-negative. In our case
we can write:

mean that:
n,~2~ = i~j

(7,~ >t 0 ,

(7~ >t 0 .

D(}q) = min {~ffln,~2 = = ~,j, 2 =/> 0}.




D(;,~) = max {((7,~-- o',~)e,j,((7,~

>1 0





)~ > 0.

~,j = a,~2~'




. a


n U


n U





0 = ( 7 =


and the primal program can be read with the following

rules :
the inner product of the first and last rows gives the
objective function;
>/ 0 on the left of first row applies to all variables in
such row;
0 on the right of first row must be considered as
inserted between the last column vector and that obtained
by multiplying the first row vector to the partitioned
matrix enclosed in the solid hne rectangular frame.
When the dual variables 2~ (a = 1 ..... m) are introduced
in the first column, the dual program can be read in tableau
(2.6) applying the above mmtioned rules, but replacing
"row" with "column", "left" with "top", "right" with

nqao < (7~

However, since the opnmal set must also satisfy the

limitations of program (2.9), we obtain:


The transposition tableau then becomes:



non o = (7=

(7,j),,~j< ,~, (7, >i O,

(77j >i 0}.



With such position, Eq. (2.3) can be written:



We must take into consideration that, if (7~j and ~.0

respectively represent two optimal sets of the primal (3)
and dual (9) programs, for the known properties of the
linear programming, we must have:


program (2.7) can be also written:

),~ -----0

~r,~ = a~'~-- (77~,


Taking into account Eqs. (2.1) of the yield planes,

we find that Eqs. (2.11) and (2.12) are just the plastic
flow rules associated with the chosen yield surface. The
2 = therefore become <~plastic multipliers~ associated with
each yield plane.
Considering any strain rate system ~s and assuming
to be "kinematically admissible" all the plastic multipliers systems 2~ which satisfy the constraints of problem
(2,9), this latter leads to the formulation of the following
principle of minimum: the specific power of dissipation corre-

sponding to a given set of strain rates k is is /be smallest of

all fictitious powers computed for all possible systems oJ" nonnegative kinematical/9, admissible plastic mui/ipliers.
Such a general principle for bodies with piecewise
linear surface, can be proved in an autonomous form,
assuming the validity of Eqs. (2.11) and (2.12), that is
the generahzed normality rule. ]n fact, with such relations,
we have :

~j) = (Tl~,e~j = (70nlszo = (7 20.


Denoting by ;t= any system of plastic multipliers satisfying

the condition :

Taking into consideration that, as:

a~s = mj(x,, x=, xs),

n71;t~ = } , j ,

2 = >/ 0

from (2.12) we have:

(a ~ -



o',jn,j) (2 - - 2;) i> 0.


2 = = 2=(xl, x.., xs)

are respectively any stress state within ymld limits and

any system of non-negative kinematically admissible plastic
multipliers, the fundamental inequalities"

From Eq. (2.11) and from first Eq. (2.14) we obtain:


n , ] ( f f - - 2,,) = 0


and, correspondingly:

0 .

a=2 = > a=ag

are satisfied point by point, we can write"



a~n~(2 - - 20) = O.


Comparing Eq. (2.17) with Eq. (2.13) and in consideration

of (2.9), we obtain the inequality:





a~)f >i a~).o = D(~,j)

which further confirms, in autonomous form, Eq. (2.9).


Remarks on the calculation of the total dissipation


However, the actual calculation, through Eq. (3.8),

of dissipated power P is very complicated. Although, we
can calculate a lower bound of P through the first Eq.
(3.8), by dividing the volume V into a fimte number of
sub-volumes Vn (h = 1. . . . . M), and assuming for each

Let us consider a body having volume V delimited

by S. We refer this body to a system of cartesian orthogonal
axes x, (i = 1, 2, 3). We shall now apply to each point
of such continuum a generic system of displacement
rates :
t}, = ,},(x,, x.,, xa)

(i = 1, 2, 3)


continuous with piecewlse continuous derivatives in the

volume V.
Denoting by:

1 (o;,,
['u = ~

\ c-~xj

a x , / = c9";0~


the strain rates associated with the displacement rates (3.1),

we want to determine the total power of dissipation:

P(it,) = f v D('e's)dV"

For this purpose, denoting by:


a,i = a i j ( x l ,

a~j(xt, x.,, xa) = au = const,




On this assumption, being:

.j'" = f ,;,




and denoting by ah = the values of the limit stresses for

t h e generic plastic plane a, constant in each sub-volume
(but possibly variable from one to another), the first
Eq. (3.8) yields as best lower bound P , of total power
P, the quantity:
h .h I ~


P,, ---- max taoet,,im/r,v ~< a,,i


where we adopt the common notation of tensorial summation with regard to index h.
B,, dualizing the program (3.11), which is a typical
linear program, we obtain the dual principle:
P,, = min al,#n n~lt~, = ei) ,

x 2 , xa),


#h >1 0 .


By comparison with second Eq. (3.8), the dual variables

Ftn= can be obtained from relations:

)~ = 2a(x~, x~., xs)

,u,, =

the internal stresses and the plastic multipliers which

can be associated, point by point, with the strain rates
(3.2) through the extremum principles of previous section,
we obtain:
P(;l~) =

f a ,F u d V

MARCH 1971







keeping in mind that Eq. (3.12) can be considered equivalent to second Eq. (3.8) if and only if:
P,('u,) = P(u,)


that is, if the class of functions (3.9) contains at least one

real solution of the optimization problem expressed by
first Eq. (3.8).
Logically, as the subdivisions of volume V increase,
Eq. (3.14) can be satisfied with an ever smaller difference
and, a this reasoning, we can expect for any distribution
of total plastic multipliers /zh~ satisfying the conditions:
x o;
llU~h :


(i,j = 1, 2, 3; h = l ..... M)



I~1~>~ 0

..... m; b = l

multiplier ~ can be calculated using the condition:

6 = rain (}l,) max (mj) f

~r,,,e.,/dV O)


subject to constraints:
i 0 -= cgta'O~, m/m1
~< 0~ in



..... A)
,1f V



Xt;MV + f 8 T Tt}~dS = 1

to satisfy the fundamental inequality:

or even using the condition:

cr=tz,~, >1 P('uO

6 = min (t}0 rain (2=)

which will be used for the approximate numerical calculation of the continua static collapse load.


Fundamental theorems of the l i m i t analysis.

Let us consider a body having volume V and boundaries

S subject in V to body forces A~, in the portion S r of
the boundary to surface forces T~ and in the complementary portion 3",, of S to constraints annulling all the components of displacement u,.. Denoting by:
1/, = t}i(:q, x o, xa)


any system of displacement rates, the loads multiplier k

defined by relation:

fv D(elj)dV

k --


f ,, X,i,,dV + f sp T,i,,dS
is called kinema/ica//y admissible if k,~ represent the strain
rates associated with the displacement rate (4.1) and these
latter comply with the external compatibility conditions:





f ,. A'~i,,dV + f s r T/mdS = l


f v a=2=dV


subject to constraints:

2 ~>/ 0

ue = 0





f v A~it,dV + f s T T,i,idS = 1 .
Eqs. (2.4) and (4.8) are both analytical formulations of
the kinematical theorem. It is easy to demonstrate that
they also include, as particular aspect, the static theorem.
In fact, if we denote by s a generical multiplier of external
loads and restrict the class of stresses a~j appearing in
Eq. (4.6) to the class which satisfies the equilibrium conditions:
c)f~ri/+ sA'j = 0


~rqni = sT/





where nl (i = 1, 2, 3) are the direction cosines of the normal to the external surface in ST, we obviously have:




for whichever system of velocities u~ chosen with respect

to conditions :
e,~ = O(,}ti},



ut = 0




Eq. (4.2) becomes:

f ,, X[u,dV + f %, T[mdS = 1 .
k = f D('e,/)dV.


The kinematically admissible multiplier therefore coincides with the total power of dissipation P (see (3.3)),
associated with strain rates (4.1). The kinem,tic theorem
of limit analysis states however that the collapse multiplier 6 is the minimum kinematically admissible coefficient.
Taking into consideration the equivalence between the
kinematically admissible coefficient and the total power
of dissipation, from Eq. (3.8) we deduce that the collapse

To the formulation expressed by Eqs. (4.6) and (4.7)

we therefore replace the condition:

mix (~ro) s


(l) The symbols in parenthesis indicate the variables for

which the relative extremum condition is valid.
(8) In fact, the solution of the problem undoubtedly belongs
to the class satisfying Eq. (4.10) with s = &

Denoting by 6= the approximate value of the collapse

multiplier, the formulation (4.6) and (4.7) becomes:

subject to constraints:
O,a~ + sA'~ = O,
m~atj <~ a =


atjnt = s T j




& a = m i n ( q ~ ) m a x ( mns ) { a t,j.q~ net~

~ mjo'~
..h ~< a~,,
'~ ~O k'~
q = 1}


which is the analytical formulation of the static theorem.


5. Approximate determination of the collapse multiplier.

The approximate definition of the collapse multiplier
6 can be based indifferently on the formulation expressed
by Eqs. (4.6) and (4.7), by Eqs. (4.8) and (4.9) or by Eqs.
(4.13) and (4.14). In our opinion, of the three alternatives
the first is the most advantageous. In that expression the
following functions appear as variables:

g = ,}t(.~,, .,:o_, m)
( i , j = 1, 2, 3)


atj = a~(x~, x~, xa).

We can assume with some approximation that the first

ones can be expressed in terms of a discrete number of
generalized coordinates qe (k = 1. . . . . N ) in the form:


.k k


u; = 0





..... N). (5.3)

On the other hand, the continuity in the whole volume

of the strain components:

( i , j = 1, 2, 3; k = 1..... N )

e u = O,lO~


will not be strictly required and piecewise continuity

will be sufficient. For the stress components a~j(x,, x2,
x3), once divided volume V into a finite number of subvolumes Vn (h =-- 1 ..... M), we will assume that for each

~,j(xl, x.,_, xa) = a'S = const,

max (a,j){atjq etj n,W,j ~ an} = min (/*n){~h/~hlmjplt =



( X I ~ X 2 , X3)

it being understood that the common notation of tensorial

summation is extended to index k. The functions u~~ shall
be linearly independent and will be chosen in such a way
that each one satisfies the internal compatibility, that is
respects the material continuity of the body, and satisfies
the external compatibility with the constraints, so that:

the generalized loads corresponding to functions u, ~, and

having imposed index h on the limit values o~ to indicate
that they can vary from one sub-volume to the other,
although keeping constant within each single sub-volume.
The minimax principle (5.8), as its equivalent in the
continuum expressed by Eqs. (4.6) and (4.7), can lead
to a minimum principle of the type (4.8) and (4.9) or
even to a maximum principle of the type (4.13) and (4,14).
The minimum principle can be reached by observing,
as in section 3 above, that, upon introduction of dual
variables /,t, = (total plastic multipliers relative to subvolume Vh), in accordance with the known duality theorems of linear programming, we have:



.l~ h k

f o,,,';,,v


and in consideration of Eqs. (5.2), satisfies the relation:


MARCH 1971


h .k hk
j~MV = a~q




.k hk



it being understood that the minimum condition refers

to q~ as welt as to #n =.
On the other hand, we can determine the maximum
condition by noting that, if in Eq. (5.8) we limit the field
of search for atfl' to the class satisfying the conditions:
h hk

atjeo = ~


where s is an arbitrary coefficient, we have:

t~ , k h k

afjq ets = ~


q = s


then Eq. (5.8) furnishes:


and ao n will be assumed as unknowns.

On this assumption, the integral appearing in Eq.(4.6), if:


~. = rain {a,,m,l,,,~m, = q ~,J. ~,,,/-- 0, ~ q = 1}

h hk



Assuming as independent variables the parameters qX

and the total plastic multipliers /~=, Eq. (5.8) becomes:

6, = max {s aoeb = ~


q eli,



, ,tj crti ~< ah}


since the maximum is valid obviously only for the sole

remaining variables, that is atj n. The two principles (5.11)
and (5.14), both included in the linear programming procedures, correspond one to another as primal and dual
program and therefore make it possible to determine,
in addition to the approximate value of the collapse multiplier, the collapse mechanism(s) and corresponding
static conditions.



This work has demonstrated how the rigid-plastic

materials with piecewise linear yield conditions can be
studied in an autonomous form, based on known mathematical principles of linear programming. Of particular
importance are the unitary formulations of the extremum
principles concerning the specific power of dissipation

and of the principles concerning the static collapse multiplier of the continua, on which is based the proposal of a
numerical approach for a strictly general calculation. The
application of this procedure to actual cases of technical
interest will be the subiect of other studies to be published
in the near future.
Received 15 June 1970.


[1 ]


D.C. DRuc~zzR,W. PaaGEa and H. S. GREENBERG,Extended

limi! design theoremsfor continuous media, Quart. Appl. Math.,
Vol. IX, 1952.
A.A. GVOZDEV, The determination of lhe value of lhe collapse
load for statical/), indeterminate O,stems tmdergoing plaslic deformation, Int. Jour. Mech. Sci,. Vol. I, 1960.
W. S. DORN and H. S. GREENBEaG, Linear programming
and plaslic limil anaO,sis of slructures, T.R. No. 7, Dept.
of Math., Carnegie Inst. of Technology, 1955.
Virlnal work, linear programming and plastic limit analysis,
Proc. Roy. Soe., A, Vol. 251, 1959.
W. Pa^G~rt, Linear Ungleichungen in der Bauslalik, Schweizerishe Bauzeitung No. 19, 1962.
C. CEaAI~INZand C. GavaazNz, Calcolo a rotlura e programmagione lineare, Giornale del Genio Civile, No. 1, 1965.
D.C.A. KOOVMaNand R. H. L,~NCE, On linearprogrammin~



andplaslic lira# anal,sis, Journ. Mech. Phys. Solids, Vol. 13,

C. GAVARZNZ, Plastic analysis of strnctnres and dualiO, in
linear programming, Meccanica, Vol. 1, 1966.
R.H. LANCE, Duali O, in the finite-difference method of plaslic
limO analysis, T.R.NSF GK 687/2, Dept. Theor. Appl.
Mech., 1967.
C. CERADINZand C. GAVARIrqX,Calcolo a rolttn'a e programma~ione lineare. Conlintd bie lridimensionali. Nora I. Fondamenli teorici, Giornale del Genio Civile, No. 2-3, 1968.
C. CErtADINXand C. GXVARZNI,Caleolo a rollura e programma~ione lineare. Contimd b i e lridimensionali, lola II. A p p/ica~ioni a piastre e a vo/te di rivo/tt~ione, Giornale del Genio
Civile, No. 4, 1968.
G. H^r~LEY, Linear programming, Addison-Wesley, 1962.
G. B. DANTZIG,Linear programming and extension, Princeton
University Press, 1963.
L. MtSRACCmNL Programma~ione matemalica, Ed. UTET,
Torino, 1969.