133

GENERAL THEORY OF ELASTIC STABILITY*

by
CARL E. PEARSON
Harvard University

Summary. Some general topics in elastic stability are discussed. In particular,

attention is given to the relationship between adjacent-equilibrium-position and energy
techniques, to the effects of non-linearity, and to the sensitivity of certain stability
problems to the character of the loading.
1. Introduction. In analyzing the stability of an equilibrium state of a particular
elastic system, those terms which arise from the equilibrium condition eventually cancel;
consequently a number of writers have found it desirable to discuss stability in a general
manner, removing these terms once and for all, and directing their attention towards
the remaining terms. Usually, a certain equilibrium state is postulated, and one of two
criteria is then used to determine the stability of this state. The first criterion states that a
structure is unstable if an adjacent equilibrium state exists, whereas the second requires
for instability that the over-all potential energy not be a relative minimum.
In setting up these criteria in analytical form, it is recognized that some sort of
non-linearity is essential, in order for example to evade the uniqueness theorem of linear
elasticity. Such non-linearity may arise either from the geometry of the situation (large
displacements or non-linearized boundary conditions) or from the inclusion of higher
order terms in the stress-strain law. Although the above uniqueness theorem reason
may not be valid (on the grounds that the usual proof of this theorem contemplates the
same body configuration for the two supposedly different stress-strain states to be
proven identical, and so is not applicable to stability problems in any event), it is never-
theless clear that because of the cancellation of equilibrium terms it is well to include all
important higher order terms. The method of incorporating these terms varies widely,
as will be seen from a study of treatments by Bryan [1], Southwell [2], Biezeno and
Hencky [3], Trefftz [4], Biot [5], Neuber [6], Prager [7], Goodierand Plass [8], and others.
Most of the assumptions concerning non-linearity made in these treatments seem rather
artificial—for example, Trefftz and Goodier obtain non-linearities by regarding as funda-
mental a curvilinear coordinate system which moves with the material fibers of the body,
and there seems to be little basis for this method. Similarly, Prager uses techniques
of superposition which are questionable when dealing with non-linear effects.**
The role of non-linearity has been further complicated as a result of a paper by
Goodier [9], in which it is stated that the correct equations for the torsional buckling of
a bar can apparently not be obtained by conventional energy techniques. Goodier gives
a rather complicated analysis of this problem, using a Trefftz-type method, and his
incorporation of non-linearities again appears to be quite arbitrary. In the problem of
shell buckling, as discussed for example by Karman and Tsien [10], it has been suggested

*Received May 25, 1955; revised manuscript received September 22, 1955.
"Nevertheless, the treatment of Prager is probably the clearest available. He uses the adjacent
equilibrium position technique of Biezeno and Hencky, but obtains their results in a much more compact
manner. In addition, the effects of inelasticity and thermal gradients are considered, and the final eigen-
value problem is thrown into the form of a variational principle.
134 CARL E. PEARSON [Vol. XIV, No. 2

that some sort of non-linearity may be partly responsible for the discrepancy between
experiment and theory (besides the known effect of initial irregularity).
On the other hand, it is known that the effect of non-linearity in the stress-strain
laws governing the usual structural metals is only of the same order of magnitude as the
uncertainties in the ordinary elastic constants, and it does not seem physically reasonable
that such effects should materially influence practical stability problems. In the present
analysis, the stability problem is first analyzed without approximation; an engineering
approximation is then obtained by consistently neglecting terms of a certain order of
magnitude. Specifically, an arbitrary elastic body in equilibrium under certain loading
is considered. An arbitrary virtual displacement is assumed, and the discrepancy between
the work done by the loading and the increase in internal energy is calculated by means
of the exact stress-strain relationship of non-linear elasticity; the most convenient
form of this law is that due to Murnaghan [11], It is decided that a positive discrepancy
is the necessary and sufficient condition for instability, and the analytical consequences of
this are worked out. The appropriate engineering approximation is made in the final
result, and it is found that the result is different from and simpler than those usually
obtained. As a special case, the problem of Goodier [9] is considered and it is found
that the correct equations are obtained in a straightforward manner.
Recent discussion by Pfliiger [12] and Ziegler [13] have directed attention towards
certain fundamental problems in stability. Ziegler has shown by exemplification with
non-conservative systems that the result of examining for stability the equations of
motion of a perturbed mechanical system do not necessarily coincide with the familiar
energy or adjacent-equilibrium-position techniques. Since the equations-of-motion
method must be regarded as basic, Ziegler's work gives rise to some doubts as to the
usefulness of the other methods. However, for the case of conservative systems (to
which we restrict ourselves for the present; plastic buckling will be discussed elsewhere),
the equations-of-motion method and the energy method are equivalent (a proof will be
found in Whittaker [14]), and so the energy method may be used with confidence. How-
ever, the adjacent-equilibrium method and the energy method are certainly not equivalent
even for conservative systems. Consider for example an (always elastic) column com-
pressed beyond the buckling load but restrained from buckling; if the constraints are
removed the column will buckle despite the absence of an adjacent equilibrium position.
But we can perhaps obtain an equivalence by altering the problem somewhat.
Consider an elastic system which is in stable equilibrium under certain loading. As the
loading is increased in some manner, the system following an equilibrium path, a point
of instability (by the energy criterion) may be reached; does an adjacent equilibrium
position exist at this critical point? This is a question of considerable practical signifi-
cance. For example, in the problem of "tin-canning"—i.e., the problem of stability of
a fairly flat shell under lateral pressure, where at a certain critical load the shell tends
to snap suddenly through into an entirely different equilibrium position—there seems
to be no adjacent equilibrium position (in the conventional Euler-column sense) cor-
responding to the critical load. Can an adjacent-equilibrium-technique then (as used in
practice) give the correct answer to such problems?
It will be shown that the two techniques are indeed identical for the altered problem
of the last paragraph (so that, for example, the correct answer in tin-canning problems
is that conventionally obtained, and one must look elsewhere for the discrepancy between
theory and experiment). In particular, it will be found that the differential equations of
1956] GENERAL THEORY OF ELASTIC STABILITY 135

first variation of the general energy principle are precisely the same as the conditions
for existence of an adjacent equilibrium position, these conditions being again calcu-
lated without approximation.
Another stability topic of considerable interest relates to the precise character of
the loading applied to an elastic system. Such effects are not considered in previous
general stability analyses, yet have been shown to be important by Tsien [15]. A particu-
larly subtle example is given here in which, for any perturbation, the first-order work done
by two alternative types of loading is the same, yet the buckling loads are widely different.
Since the problem is a very practical one (buckling of a long cylinder under external
pressure), it is clear that the character of the loading must be included in any general
stability criterion. The appropriate analysis will be given for the two types of loading
of greatest importance, viz., dead loading and pressure loading.
2. Analytical condition or instability with dead loading. Consider an arbitrary
elastic body which is initially free from stress (state I). By the application of load or of
heat, the body alters position and shape (and achieves state II). The material particle
initially at the point (ax , a2 , a3) has now moved to (xi , x2 , x3), where subscripts are
used to distinguish between the usual three fixed Cartesian axes. The ith component
of the displacement vector is given by
Vi = a:,- - a, (1)

and the Lagrangian strain tensor is

Vii = hldVi/ddi + dVj/ddi + (dv,/da,)(dv,/da,)], (2)
where the summation convention is used (here and in the future) for repeated subscripts.
If U is the internal energy per unit mass, a symmetric function of the nine 77,,-, of
the absolute temperature T (or entropy S), and of position, then by Murnaghan's
treatment [11] the Eulerian stress t,-,- in state II is given by
Ta = p(d U/dO .(dx,/da„) (dXj/daQ), (3)
where p is the density in state II and the partial differentiation of U is to be carried
out at constant entropy. It is now required to analyze the stability of the body in its
deformed state II. From Sec. 1, the body will be considered stable if for each infinitesimal
displacement (compatible with the boundary conditions) the work that would be done
by the surface and body forces does not exceed that absorbed as an increase in internal
energy.* If this condition is not met, then for some virtual displacement excess energy
would be available for use as kinetic energy, and the appropriate displacement will
increase in magnitude.
The body force per unit mass, F, , will be assumed constant (e.g., gravitation). The
surface loading, T, per unit area in state II, is considered to be produced by fixed loads
which vary neither in total magnitude nor direction during the trial displacement.
Thus, under such "dead" loading, the material particles constituting a portion of the
surface in state II will always be subject to the same total surface vector force, irrespective
of their orientation or total area, throughout the trial displacement. Consequently, the

*In the equivalent potential energy form, this is the usual energetic stability criterion. We use the
above form because of its additional generality; as will be shown elsewhere, it can then be applied to
certain non-conservative systems also.
136 CARL E. PEARSON [Vol. XIV, No. 2

work done by the body and surface forces in a displacement w, from state II would be,
exactly,
W = J pFiUidV+ J T,MidS,
where the volume and surface integrals are calculated for state II. Altering to volume
integrals and using the equations of equilibrium gives

W = J Tuidujdx,) dV
which, upon substitution from Eq. (3), becomes

W = J (dU/dtiP<I)(dXi/dav)(dUi/daQ)p
dV. (4)

The increase in internal energy is, exactly,

fi = J (U' - U)PdV, (5)

where U' denotes the internal energy per unit mass following the displacement u{ ,
and depends on the temperature of that state as well as on «,• . Note that the volume
integral is still calculated for state II (this is allowable because the element of mass,
P dV, is invariant).
Consequently, the general condition for stability is that, for each allowable w, ,

/ [(dU/dvJXdXi/da^dUi/da,) - \U' - U\]pdV < 0 (6)

(in particular the integral vanishes for u{ = 0; if it vanishes for some u, ^ 0 but is
non-positive for all w, , the state will be called neutrally stable.)
It is now necessary to calculate U'. Because buckling is usually rapid, it is reasonable
to require the displacement u, to be of an adiabatic character, and we will make this
assumption. Had we at this point insisted on an isothermal motion, an entirely analogous
calculation (best carried out by use of the Helmholtz free energy function instead of U)
could have been made, and the same final results would be obtained in the sequel except
that the isothermal rather than the adiabatic elastic constants would appear. Experi-
mentally, the difference between these constants is negligible; then, using the fact that
in general the motion U; of the body would be somewhere between adiabatic and isother-
mal, it is seen that the particular thermal assumption at this point makes little difference.
In any event, we consider for definiteness an adiabatic motion, so that in the power
series expansion of U, viz.,
U' — U = (dU/dtiPa) br]VQ
+ i(d2U/dr)pa Stj,-,- -f • • • (7)
all partial derivatives are to be calculated for constant entropy and for state II. Using
by a = i[(dxr/da,){dur/da,) + (dxr/da,)(dur/dat) + (dur/da,){dur/da,)] (8)
in Eq. (7), and substituting the result into Eq. (6), gives as the condition for stability that

p dV[{dU/dtii,){dur/da^(dur/da,) + {d2U/drtii drj^) di\u ir]ra +

/
1950] GENERAL THEORY OF ELASTIC STABILITY 137

be greater than zero for all non-zero permissible it,- . Alternatively, use of Eq. (3) allows
the condition to be written as

dV[TVQ(dur/dx^{dur/dx„) + p{d2U/di]ii dr)VQ)Sriij 8ripa +•••]> 0. (9)

/
Except for pathological cases, only second-order terms need be considered, and the
criterion for stability becomes

J dV[Tpa(dur/dxp)(du./dxQ)
(10)
+ p{d2Ufdi)a dripa)(dxr/dai)(dx,/dap)(dur/dai)(du,/da,)] > 0.
Here, all quantities are calculated for state II. For adiabatic virtual displacements,
this criterion is exact, and must be used wherever non-linearity of the stress-strain law
is essential.
3. Engineering approximation. Isotropic media. The second term in Eq. (10)
may be calculated by means of a power series expansion in ij,-,-in terms of the various
derivatives of U evaluated for state I. Using the subscript "0" to indicate state I,
d^U/driij d-qpq= (d^U/dtia driP<,)0+ (d3U/dTin dt]pa dT)r,)0t]r, + • • • . (11)
For structural metals, the magnitude of the second term in Eq. (11) is generally smaller
than the uncertainty in the experimental value of the first term (the first term repre-
sents the usual elastic constants, and the second and following terms represent non-linear
elastic effects). Consequently, it is reasonable to approximate the second term coefficient
by
p0 (d2U/driij dr)PQ)0 ,

where the density in state II has also been replaced by the density in state I. This term
is recognized as the conventional (adiabatic) elastic coefficient and will be denoted by
c°ijPq. Then the second term may be written
[cliPXdxr/da^dx,/da?)(dx,/da^(dxJda^](dur/dx?)(du,/dxm). (12)
Now the deformation (although not the displacement) between states I and II is assumed
small; this means that the partial derivatives inside the square bracket of (12) represent,
within the approximation being made, a pure rotation. But the quantities c0iipaform a
Cartesian tensor, so that the quantity in square brackets reduces simply to the elastic
coefficients for the orientation of state II, i.e., to crt,m . Thus the stability condition,
Eq. (10), becomes
J dV[Tpq(dur/dxp)(dur/dxa) + crl.merle,m] > 0, (13)

where eu = \[(diii/dx,) + (du,/dXi)] and where the symmetry property of crt,m has been
used. For isotropic media,

Crt.m = + S,,8rm + I _ 2# (14)

where G is the shear modulus and a is Poisson's ratio. Use of this relation gives the
stability criterion as

fd¥[ Tva(dur/dxp)(dur/dxa) + 2Gje,me,m _ e,,e„ > 0. (15)

138 CARL E. PEARSON [Vol. XIV, No. 2

The only terms which have been neglected in the derivation of Eq. (15) are those
which are of higher order in powers of jj;,- .
4. Euler column. Before proceeding with the general theory, it is worthwhile to
consider a simple example. Let a slender column of length L and cross-sectional area A
be placed so that its neutral axis coincides with the x3 axis, and so that it may buckle
in the Xi — x3 plane only, the ends being restrained from lateral motion. A total load
P is applied to the end of the column, producing a stress of r33 = —(P/A), all other
Ta — 0. We now use Eq. (15), and see how large P must be, for certain trial displace-
ments, before the left-hand side of Eq. (15) becomes negative; such a situation would
correspond to buckling. Since the trial displacement will usually not be the exactly best
ones for this purpose, the buckling load obtained in this manner will always be too high;
this remark clearly holds in general also and is not restricted to the Euler column case
(see Ref. [8]). In fact, the buckling load Pc is given by

CP . J ~l~(o1/! — 2<r)ettemm } dV
(PJ2GA). mm ; (Su./BxMSu,^,) dV <16)
Whenever P exceeds Pc , Eq. (15) shows that the column is unstable.
Let us choose a trial displacement, being guided in our choice by the tendency of
plane cross sections to remain plane and perpendicular to the neutral axis during bending:
Mi = u(x3),

u2 = 0, (17)
u3 = —x{ll'(x3),

where u(x3) is an arbitrary function of x3 which vanishes at x3 — 0, L, and where u'(x3)

is its derivative. A straightforward calculation using Eq. (15) gives that

-{P/A^A fQ (u'j2 + I [ (u")2] + 2(?(p£™)/ £ (u"f < 0 (18)

for instability, where I is the appropriate moment of inertia of the cross section. Because
(P/A) « G, the second term in the square bracket is omitted and we obtain

Pc = 2Gl(j f) min ^ W
\1 - 2J rf (u'f (19)

This answer is too high, because

2g(} ~ j ) > E

so that the displacement (17) is deficient in some respect. The deficiency lies in the fact
that Ui and u2 do not contain terms allowing for lateral expansion of the column during
bending. Actually, instead of setting eu = e22 = 0 in Eq. (15), we should more correctly
have set en = e22 = —ae33 . A simple calculation shows that the incorporation of such
terms does not materially alter the first term in Eq. (15). A suitable altered displace-
1956] GENERAL THEORY OF ELASTIC STABILITY 139

ment would in fact be

Ui = u + | (x\ — x\)u",

u2 = <rXiX2u ,
(20)

u3 = —Xiu'.

If this displacement is inserted into Eq. (15), and the magnitude of various terms ex-
amined (which is most easily done by assuming that u is not far removed from that u
used to minimize Eq. (19), viz., sin (itx/L)), it is found that the first term of Eq. (15)
is essentially unaltered, whereas the second becomes (very closely) Ee233. Then a similar
calculation to that of Eq. (19) gives

Pc = EKtt/L)2.
It will be noted in Eq. (15) that the second term is the familiar strain energy term, so
that the first term must in a sense represent work done by the loading. It is therefore
not surprising that minor transverse alterations in the (these alterations incidentally
vanishing on the neutral axis) do not affect to any extent the value of the first term.
This is a rather useful point to note, because it means that simple displacements of the
type (17) may be used in many column problems, provided only that eu and e22 are
set equal to (—<re33)in Eq. (15).
5. Flexural buckling. Goodier [9] has examined the use of energy techniques in
the flexural buckling of a twisted bar. Since his analysis is geometrically complicated
and physically questionable it is worthwhile to show that the correct equations are
obtained by use of Eq. (15) in a routine manner. Since the only question is as to whether
or not certain terms occur, it is only necessary to consider a simple special case—that
of a circular cylinder. If the central line coincides with the £3-axis and if the angle of
twist per unit length is 0, the stresses are
t 13 — GOx2 , t23 = G6x 1 .

Assume a displacement of the form

Ui = u — (ix2 ,

u2 = v + 0x 1 ,

U3 = —Xiu' — x2v',

where u, v, ff are functions of x3 . (If the cylinder were non-circular, a term involving
the warping function multiplied by 0' should be adjoined to u%.) Then using the tech-
nique of Sec. 4, condition (15) requires for stability
,L
2GQ f {-u'v"h + v'u"h) + Gh [ (0T +E f [I2(u")2
+ Ify")2]> 0
Jo Jo Jo

and minimizing* this expression yields

EI2u"' + Mv" = 0, EIiv"' - Mu" = 0, 0" = 0.

*That minimization is the appropriate procedure will be shown subsequently.

140 CARL E. PEARSON [Vol. XIV, No. 2

With the appropriate boundary conditions, these coincide with the final results of Goodier.
(Note that I, and /2 are defined in an opposite way to that of Goodier. Here 7i is defined
as being about the a^-axis, i.e., / x\ dA.)
6. Curvilinear coordinates. Buckling of a cylinder under dead load. Very often,
the appropriate coordinate system is not Cartesian; in such cases it is useful to have
available a more general formulation of Eq. (15). Let the differential element of distance
be given by
ds2= hi dyl + hi dyl + hi dyl ,
where hi , h2 , h3 are functions of the three curvilinear coordinates 2/1,2/2,2/3. Then
denoting by t,-,- the curvilinear stress components and by u{ the curvilinear virtual
displacement component (i.e., in the parametric direction of 2/,), the first term of Eq.
(15) may be shown by direct calculation to become

£ \ft
p,Q,r,m LUprlq
^r.p^r.c
| U-pUq |
+ -J2~ KtrhQ
7

(21)
I (U'Um *1.1, I , _ 2m, , _ 2umuv \)
\h Jl p>m^J»»r l" nQ,mUQtV ^ Ur ,vHq,T

where a comma indicates differentiation with respect to the appropriate —thus ur,v
means (dur/dyv).
The form of the second term is unaltered, but eit must now be interpreted as a cur-
vilinear strain component, perhaps most conveniently given by

Bii= 2 ? [&Idfi (h) +1 Wi(t) + 2Siihti a^]' (22)

Consider for example a long thin circular shell of mean radius R, under the action
of an external pressure P of the present dead-loading (see Sec. 3) type. As cylindrical
coordinates, choose 2/1 = x along the axis of the shell, y2 = d, the polar angle, and y3 = r,
the radial distance from the central axis. Then
ds2 = dyl + 2/3dyl + dyl .
Choose
Ml = 0,

u2 = v + ft(y - w'),
u3 = w,
where v, w are functions of 6 representing the motion of the central surface of the shell,
and z = r — R. From shell theory, it is known that displacements of this type are suit-
able for calculating all strains except el3 , e23 , e33 . The former two are conventionally
negligible, and the latter is usually calculated by assuming the induced r33 stresses to
be much smaller than the bending stresses in the shell. Then the e„- to be used in Eq.
(15) are
en = <?i2 = ei3 = e23 = 0,

v' t w z r * m
€22 R R R2 ^

a
^33 — #22 )
1 — <7
1956] GENERAL THEORY OF ELASTIC STABILITY 141

where a minor approximation has been made. Using Eq. (21), the stability condition
(15) becomes that

/ dF{(~iP)[(w^ +-u[e"]} >0>

where t is the thickness of the shell. Because (PR)/t « E, the first term may be submerged
in the last to give, approximately,

/ dr{-TR
«•- ^ +1<«
- +rh K+1 - W-<■"
+»")]*}
>°-(23)
Taking a unit length of cylinder and integrating over the cross-sectional area gives

p _ E . t SI' (v' + w)2 + (t3/12R2) JT (w + w")2

(1 - a2)R mU1 („ - W'f

which is obtained (very closely) by setting v' = —w and v = sin 2d(v = sin 9 would
correspond to rigid body motion). Then

pc (1 - <x2)R3 ' C24)

{ )

where I — t3/12. Since the usual result is (3/4) of this, it is clear that the assumption of
dead loading has materially altered the critical load. Load-type sensitivity has been
remarked for this problem by Stevens [16] and more generally by Tsien [15]; we consider
it here to exemplify the manner in which the stability criterion will be generalized.
7. Pressure loading. We return now to the general theory of Sec. 2, and examine the
effects of different types of loading. Firstly, it is clear that forces exerted by fixed con-
straints (pin joints, etc.) are in general included in the theory of Sec. 2, for even though
such forces may alter in direction and magnitude during a trial displacement, the appro-
priate component of at this point of application is zero. If 'secondly, however, some
of the surface tractions are not of the dead-loading type, then additional terms must in
general be adjoined to Eq. (15). We consider here only the practically most important
such forces, viz., pressure-type forces, for which the force applied to a given portion of
the surface of the body varies in such a manner as to remain always perpendicular to
that portion and so as always to maintain the same magnitude per unit area. Further,
the system is still assumed conservative, so that the total work done by these pressure
forces is independent of the path. If then a pressure P acts on a portion SP of the surface,
the work done in the trial displacement ut can be calculated by allowing the intermediate
displacement to grow at a constant rate—i.e., if t is time, let the displacement at time t
be (Uit) and calculate the work done from t = 0 to t = 1. This work, Wi , is

Wi= fQdtJ (^.[l^.o^-P)-^),],

where the subscript t denotes evaluation at time t. But

/ \ / i ci \ 1 d(%j + U, t) d{xk + Ukt) JC,

(ni)t(dSp)t = ^ eiikerva ^ -Jl-^ nr dSP
142 CARL E. PEARSON [Vol. XIV, No. 2

so that

Wl= Jo dt/ + «*.«'}».■

J(-F«<)»
(25)
= —P J" dSpj^iiW,-+ | (nxukxku<— nkUk,<ut)+ | e,-,terMw/,put.(,w,J.

In this exact expression, the first term would already have been included if P had been
treated as a dead load; consequently the additional work done is that due to the remaining
terms. Again we omit terms of third order in m, [see Eq. (10)], and remembering that a
factor of —2 was incorporated into the derivation of Eq. (10), the term that must be
adjoined to Eq. (10) is

J dSPP[niUk,kUi- nkukiiui]. (26)

Considering again the problem of Sec. 6, the term to be added to Eq. (23) is easily seen
to be
/»2x
P / (w2 + wv' — vw' + v2) dd
Jo

and instead of Eq. (24) we obtain

p<= (T=75iF (27)

which is the conventional result.
8. Adjacent-equilibrium-position method. It is now proposed to set up in analytical
form the condition that an adjacent equilibrium position should exist. Using the notation
of Sec. 2, and denoting quantities in the perturbed state by primes, the stresses following
the virtual displacement ut will be

" wypJ dapda, '

where x • = x{ + w, . The partial derivatives of U will as before be calculated at constant
entropy. Expanding the energy term in a power series gives
(dU\ dU d2U „ . 1 d3u . .
la
WW
/ = OT)va ' di)v,
a adrjr. "I" o1 drjPQ
a H A
dr]r, dijIm
°r<" "T " " » (29)

where the partial derivatives on the right-hand side of Eq. (29) are evaluated for state II,
and where drin is given by Eq. (8). Inserting this result into Eq. (28), replacing x\ by
Xi + Ui , using Eq. (3), and neglecting all higher order terms (a process which by virtue
of Sees. 2 and 3 is considered legitimate) gives eventually
, _ p[ i _i i d2U (dXidXj^d^ dx„ \ . .
T,-,- p t»j + + TriUi,r + P ^ da^ da^ (30)

where a comma indicates partial differentiation with respect to xt . Then using

dx- ~ [ U,'i] dx.

1956] GENERAL THEORY OF ELASTIC STABILITY 143

(within higher order terms) the equation

£ «,) + p'F, - 0
becomes
_l_ _l T d2{7 dx, dXj dxt dxm 1 _
Tri , jU\ + + |^P ^ drj^ ^ ^ ^ ^ 0. (31)

In addition to Eqs. (31), a boundary condition must be satisfied. For the case of
dead loading, the condition is that
T[ dS' = T, dS
and using the fact that

n't dS' = (p/p')^nr dS

aXi

gives as this condition

/ . d2U dxr dx. dx„ dxt \ . .
[tpqu,.v + P ^ ^ ^ da^ u.^jn, - 0. (32)

Similarly, the boundary condition for that part of the surface where pressure forces act is

T\ dS' = —Pn'idS',
whence it is found that Eq. (32) should be altered for this portion of the surface by
adding to the left-hand side the term
P(S,ru,t, — u„,r)nQ . (33)

9. Relation between the two methods. It has been remarked in Sec. 1 that the
methods of Sec. 2 and Sec. 8 can at beSt be equivalent only for special situations, such
as at points where an originally stable structure first becomes unstable. Consider therefore
a structure which follows some stable equilibrium path as the load alters. The path
will remain stable as long as the second order variation in potential energy is positive
definite (vanishing only for zero displacement). Consequently, trouble can occur only
at points where this second order variation [essentially the left-hand side of Eq. (10)]
vanishes for non-zero displacements. Such a situation of neutral stability will in practice
be followed by unstable equilibrium states as the load is further increased (see Poincar6
[17]); we therefore investigate the condition under which the left-hand side of Eq. (10)
first vanishes for non-zero displacements. Since it always vanishes for zero displacements,
this condition is equivalent to requiring the minimum of Eq. (10) to be attained at
non-zero u{ as well as at zero , and this eigenvalue problem is that obtained by setting
the first variation of Eq. (10) equal to zero.
The result of doing this is easily seen to be the same as Eq. (31), with the natural
boundary condition (32). If pressure forces act on a portion of the surface, the result of
a variation of Eq. (26) must be adjoined to the natural boundary conditions. This is
not quite straightforward, because the condition that the pressure loading be conserva-
tive has not been explicitly stated (without such accessory conditions, the exact pressure
144 CARL E. PEARSON [Vol. XIV, No. 2

work would depend on the path). The appropriate condition turns out to be that on
the boundary of that portion of the surface where pressure forces act,

ervau„ 5uQ dxT = 0 (34)

/
for all variations SuQ . This condition would for example certainly be satisfied if the
pressure surface completely enclosed the body, for then the path length would vanish.
Alternatively the restraints on ut may ensure satisfaction of Eq. (34)—as in the case of
a hemispherical shell set on a rigid frictionless plane surface and subjected to external
pressure.
Using Stokes' theorem on Eq. (34) gives

dSP[(uk 8u,)_, — (u, 5uk),.]nk = 0

/
and if this is used in calculating the variation of Eq. (26), Eq. (33) is indeed obtained.
It thus follows that at points where Eq. (10) first attains the minimum value of zero
for non-zero u{, an adjacent equilibrium position exists, and in this sense the two methods
are equivalent.

References
1. G. H. Bryan, Camb. Phil. Soc. Proc. 6, 199, 287 (1886-89)
2. R. V. Southwell, Phil. Trans. Roy. Soc. 213, 187 (1913)
3. C. B. Biezeno and H. Hencky, Kon. Akad. von Wetensch. Amsterdam Proc. 31, 569 (1928); 32, 444
(1929)
4. E. Trefftz, Z.A.M.M. 13, 161 (1933); see also K. Kreutzer, Z.A.M.M. 12, 351 (1932)
5. M. Biot, Phil. Mag. (7) 27, 468 (1939)
6. H. Neuber, Z.A.M.M. 23, 321 (1943)
7. W. Prager, Quart. Appl. Math. 4, 378 (1947)
8. J. N. Goodier and H. J. Plass, Quart. Appl. Math. 9, 371 (1952)
9. J. N. Goodier,Quart. Appl. Math. 2, 93 (1944)
10. T. von K&rmdn and H. S. Tsien, J. Aeronaut. Sci. 7, 43 (1939)
11. F. D. Murnaghan, Finite deformations of an elastic solid, Wiley and Sons, N.Y., 1951
12. A. Pfliiger, Stabilitdtsprobleme der Elastostatik, Springer, Berlin, 1950
13. H. Ziegler, Z.A.M.P. 4, 89, 167 (1953)
14. E. T. Whittaker, Analytical dynamics, University Press, Cambridge, 1937
15. H. S. Tsien, J. Aeronaut. Sci. 9, 373 (1942)
16. G. W. Stevens, Quart. J. Mech. and Appl. Math. 5, 221 (1952); see also A. P. Boresi, J. Appl. Mech.
22, 95 (1955)
17. H. PoincarS, Acta Mathematica 7, 259 (1885)