21

NONLINEAR THEORIES FOR THIN SHELLS*

BY
J. LYELL SANDERS, JR.
Harvard University

Abstract. Strain-displacement relations for thin shells valid for large displacements
are derived. With these as a starting point approximate strain-displacement relations
and equilibrium equations are derived by making certain simplifying assumptions. In
particular the middle surface strains are assumed small and the rotations are assumed
moderately small. The resulting equations are suitable as a basis for stability investiga-
tions or other problems in which the effects of deformation on equilibrium cannot be
ignored, but in which the rotations are not too large.
The linearized forms of several of the sets of equations derived herein coincide with
small deflection theories in the literature.
Introduction. The literature is not devoid of papers in which some of the effects of
finite displacements on the deformation of thin shells are accounted for. This is most
obviously the case for papers dealing with the stability of shells, but these have been
concerned almost exclusively with cylinders, cones, and spheres. The differential equa-
tions governing the phenomenon have been derived specifically for these geometrical
shapes. Despite the potential usefulness of a general non-linear theory, the literature on
the subject is sparse. It is the purpose of the present paper to derive an exact theory for
large deflections of a thin shell with an arbitrary middle surface and then, by making
certain simplifying assumptions, to derive from this several approximate theories suitable
for applications.
Probably the earliest work of some generality is Marguerre's nonlinear theory of
shallow shells [1]. Donnell [2] developed an approximate theory specifically for cylinders
and suggested its extension for a general middle surface. The result, a theory for what
might be termed "quasi-shallow shells", has been worked out by a number of authors,
notably Mushtari and Vlosov [3]. The problem of symmetric deformations of shells
of revolution has been reduced to the solution of a pair of equations analogous to the
Reissner-Meissner equations by E. Reissner [4]. These several problems are adequately
formulated but the general problem presents difficulties not found in the special cases.
The earliest work of a completely general nature appears to be the paper by Synge
and Chien [5] followed by a series of papers by Chien [6, 7]. The intrinsic theory of
shells developed by Synge and Chien avoids the use of displacements as unknowns in the
equations. The theory of shells is deduced from the three-dimensional theory of elasticity
and then, by means of series expansions in powers of a small thickness parameter, ap-
proximate theories of thin shells are derived. A large number of problem types is found
classified according to the relative orders of magnitude of various parameters. Several
authors have discussed and criticized this work [8, 9, 10, 11].
An elegant and general formulation of the problem is to be found in the recent paper
by Ericksen and Truesdell [8]. In this paper there is a unified treatment of thin shells
and curved rods developed as two- and one-dimensional theories respectively without
an attempt to deduce them from the three-dimensional theory of elasticity. The consti-
*Received April 14, 1961; revised manuscript received July 16, 1962.
22 J. LYELL SANDERS, JR. [Vol. XXI, No. 1

tutive relations are purposely left out of consideration because they are unnecessary
for the description of strain and the establishment of equilibrium conditions. The authors
adopt the method of description of deformation originated by the Cosserats and are
able to account for transverse shear and normal strains and the rotations associated with
couple stresses. They find that in the general case eleven measures of strain are necessary
as compared to the six in the usual first approximation theory. The two-dimensional
approach to shell theory really evades the question of the approximations involved in
the descent from three dimensions, but this seems to be a virtue rather than a defect.
Such questions are effectively isolated and shown to belong to the part of the theory in
which constitutive relations are established.
An incomplete treatment of the general large deflection theory of thin shells has
been given by Novozhilov in [12]. He derives a theory for small middle surface strains,
but does not go into detail on further simplifications or discuss approximate equilibrium
equations. The author indicates that further simplifications would result if rotations
are assumed small. The expressions given for small strain do not vanish for rigid body
motions.
A general theory is developed in those chapters of the monograph [3] written by
Galimov. This author also begins with a theory based on the assumption of small middle
surface strains. The expressions given for the components of bending strain do not form a
tensor. In deriving the equilibrium equations the author occasionally neglects middle
surface shear strains but retains direct strains. This keeps the coordinate system orthog-
onal but is at best a questionable procedure. The first system of equations arrived at
which is not open to criticism is that belonging to the Donnell-Mushtari-Ylosov approxi-
mation.
The developments in the present paper begin with a derivation of an exact system
of equations with unrestricted displacements which is undoubtedly a special case of the
theory of Ericksen and Truesdell. There is some justification for starting afresh with a
simpler approach since we are willing to accept a result of less generality. The present
paper is not exclusively concerned with the exact theory but also with the problem of
deriving from it approximate systems of equations which may be suitable for application
in cases in which the displacements and rotations are restricted in magnitude. The
deformations herein are restricted by the Kirchhoff hypotheses and the loading does not
include couples distributed over the middle surface.
Since this paper was submitted, the author has learned that two others have arrived
independently at nearly the same results. Prof. W. T. Koiter presented his results at a
lecture during a recent visit to the United States and Dr. R. W. Leonard of the N.A.S.A.
submitted a thesis on the subject to the Virginia Polytechnic Institute [13].

Derivation op Exact Equations for Large Displacements

Geometrical preliminaries. Let the undeformed middle surface of the shell be
given by the equations
x'=x\n, (i= 1,2,3; a =1,2) (1)
where the x' are cartesian coordinates in space and the £" are curvilinear coordinates
on the surface. Let the displacements U* of material points on the middle surface of the
shell be resolved into components tangential and normal to the undeformed middle
surface as expressed by the following equation
1963] NONLINEAR THEORIES FOR THIN SHELLS 23

U* = u'®x*a + w^)n\ (2)

where x]a = dx'/d£a are tangent vectors to the coordinate curves on the undeformed
middle surface and n' is the unit normal to the undeformed middle surface. In this paper
the coordinates £" will be used to label material particles on both the undeformed middle
surface and the deformed middle surface.
Some of the important formulas of the theory of surfaces will be used repeatedly
and are reproduced here for convenient reference. For the undeformed middle surface
the formula for the squared element of arc in terms of the first fundamental form gaB is

ds2 = x\ax\fdi"d^ = gapdfdlf; (3)

the element of area is
da = gU2d£df, (4)
where g is the determinant of ga$ ; and the equations of Gauss, Weingarten and Codazzi
are
x',aP = — ba0nx, (5)

n'a = b&e, (6)

bafi.y = bayip, (7)
where a comma denotes covariant differentiation with respect to the metric gap and
where the second fundamental form bap , as here used, differs in sign from the usual
definition.
After the displacement U' given by eq. (2), the material particle originally at x'
will move to the point X' given by
X* = x* + U{ = a;1'+ wV'„ + wn\ (8)
This is the equation of the deformed middle surface in terms of the parameters £a.
Tangent vectors to the coordinate curves on the deformed middle surface are given by

X\a = \lx\y + nan\ (9)

where
= x'.aX',/, = gal! + ua,f + bapW, (10)

XI = gyS\5a , (11)
Ma = X]an' = W, a — baU0 . (12)

An expression for the unit normal to the deformed middle surface is found by taking
the cross-product of the tangent vectors, the result is

Ni = P-VV5(| + KwcxU), (13)

where p = (G/g)1/2. It is convenient to define the following quantities

va = x\aNx = — X?m«),
(14)
cos co = n'N' = | ~ X^f)-
24 J. LYELL SANDERS, JR. [Vol. XXI, No. 1

Indices on , and va will always be raised or lowered with the metric gaR . The
squared element of arc on the deformed middle surface is given by

(IS2 = X^Xifsdfdt = &VC^- (15)

In terms of \afi and na

G afl \a\y (1"1~ (16)

The element of area on the deformed middle surface is given by
dA = G1/2d?d? = p da. (17)
The equations of Gauss, Weingarten and Codazzi are

X\aft = -BatN4, (18)

N]a = BsaXU, (19)
Baff.y = Bay.p , (20)
where Bap is the second fundamental form of the deformed middle surface. A semi-colon
is used to denote covariant differentiation with respect to the metric GaS . An expression
for B«d in terms of , na ,va and cos a>can be derived as follows

Bap = XUNie = -Xi^N*

= —[(Xl.p+ bW)x\y + in*., - xIby^n'W
(Ma,^) COSCO (hya,/3 ""f~bpyH^)v . (21)
The following identity will prove to be useful

X]aN' = /Xpa + fia cos w = 0. (22)

Equilibrium equations. In the coordinate system of the deformed middle surface
the equilibrium equations of the shell are the same as in the linear theory and need not
be derived here. They are (see [11], for one): force equilibrium

N& + BiQl + P* = 0, (23)

Q°a ~ Ba0N0a3+ P = 0, (24)
moment equilibrium

MZl -Ql= 0, (25)

ea,{Nf + BayMl*)= 0, (26)
where N"/ is the membrane stress resultant, M*/ is the bending moment resultant, and
Q o is the transverse shear stress resultant, all defined with respect to the deformed shell.
The quantities P" and P are applied load intensities per unit of area of the deformed
middle surface, eafs is the covariant permutation tensor in the deformed coordinate
system.
The above equilibrium equations are exact but, of course, the ten stress quantities
entering into them do not furnish a complete description of the state of stress throughout
the thickness of the shell. However, in thin shell theory it is always assumed that the
state of stress is adequately described in terms of these quantities.
1963] NONLINEAR THEORIES FOR THIN SHELLS 25

Finite strains. The strain quantities entering into a thin shell theory are a matter
for definition. The literature of the subject shows a wide variety of choices of strain-
displacement relations, particularly for the bending strains. Some choices have been
shown to be better than others (see [14]) but at the present time no set of conditions
sufficient to render the choice unique has been generally agreed upon. In the present
paper the choice has been guided by two considerations, the first of which was the desire
to derive a theory which admits a principle of virtual work. This requirement forces a
close relation between the equilibrium equations and the strain-displacement relations.
The second consideration was simplicity. The resultant choice will be shown to furnish
an adequate description of the deformation of the shell provided the Kirchhoff hypotheses
are accepted as adequate descriptions of the displacements.
Let A be a simply connected region on the deformed middle surface enclosed by
the curve C. The following identity follows from eqs. (23) to (26).

/ w;?a + BiQl + P>)SUf+ (Qo:a- + P)8W

+ {Mo+ ea,(Nf + B'Ml") 8<t>]dA

= 0. (27)
By application of the divergence theorem for a curved surface (27) may be transformed
into the following identity which is the preliminary form of the principle of virtual work
and all subsequent derivations will proceed from it.

j) (WSUp + Q°05W
+ MZ*8<t>e)r,adS
+ J (JP'tU,+ P8W)dA

L [No\SUp;a + Bap8W — eap8<f>)

+ Q"{8W-a — B„8Up + &£a)
+ Mo (Sipn-a — eyfBya8<t>)]dA,
(28)
where the virtual displacements 5Ua , 8W and rotations 8<pa, 8<t> refer to components
in the directions of the tangents and normal to the deformed middle surface. In (28)
the terms on the left hand side are interpretable as the external virtual work of edge
loads and surface loads respectively. The right hand side of (28) might be interpreted
as internal virtual work if the coefficients of N"/, Q"0 and were identified with
strain increments. Such an identification will be postponed. First these coefficients will
be written in a different form.
By definition
8V* = 8UaXia + SIFiV' = 8uax\a + 8wn\ (29)
Now
SZJie= - Bap8Ua)Ni + (SUIt + Bap8W)X'a=.;5C/:3
= (8w,p — bafl8ua)n* + (8u"p + bap8w)x',a. (30)

From this, it follows that

iU„. + B ap8W — \y$ 8\ya + (31)

8W;a — BUUy = vy8\^ cos«5yua. (32)

26 J. LYELL SANDERS, JR. [Vol. XXI, No. 1

The rotation around the normal 50 is given in terms of displacements by the formula

50 = |e0/,5t/„:o. (33)

By the use of (31) and the fact that Bap is symmetric (33) becomes

50 = |ea*(XT,5Xl + M,«M.). (34)

Also,

e«^50 = | (X7(35X^+ tipSna — XT„5XJ — naSn0). (35)

From (31) and (35) the coefficient of NZ" in (28) is

| (\yp5\l + ne&Va+ Xto5XJ + na8nfi) — | 8(\yf\ya + najus) = | 8Gap. (36)

The natural definition of the finite membrane strain is thus

Ea? = 2 (@<>0
— 0 <«/>)• (37)

The coefficient of Q" in (28) is

5ya = 5TF;a — B„8Up -f- 50a = vy8\7,+ cos u8fia + 50„. (38)
Since the intention is to derive a theory in which transverse shear strains are neglected,
set 8ya = 0 which gives
50« = —vy8\ya — cos uSfia, (39)
which serves to relate rotations to displacements. From (9), (13), (14) and (39)

N'8X]a = VyfiXaH~ coseo8n„ = —50a (40)

or, since N'X]a = 0, it follows that
50„ = X\a5N\ (41)
A finite transverse shear strain ya consistent with (38) and the requirement ya = 0
may be defined as follows
7a = NlX)„ = \yavy + fla cos to (42)

The coefficient of in (28) may be found in terms of Bafl and Gan as follows. From (41)

S0P:„ = X'^SN' + X%8N:a = -B^N'SN' + X'^SN'.,, = Xie8Nia (43)

since N'SN' = 0. Now recall that
BaP = NiaX% ,
so that

8Baf>= X^SNU + N'.aSXie

= 8<fi„.,a
+BlX\8X\e (44)
1963] NONLINEAR THEORIES FOR THIN SHELLS 27

This gives
S<t>l3;a
— SBap — B^X'^ySX'^ . (45)'
From (9) and (35), it follows that

ea,Sct>
= | (x:„sx:a - Xi.SX'J. (46)

From (45) and (46) the coefficient of M"0 in (28) reduces to

8Ba„- \BlKX\eX\) = SBa„- |BltG„ = SBaP- BlSE,y . (47)

Using the foregoing results the right hand side of (28) may be written

J [NZ08Eae
+ M^(SBas - ByJEM)]dA. (48)

There is obviously some difficulty in defining a finite bending strain tensor because the
coefficient of Mo" in this expression is not the exact variation of anything. However, a
way to proceed suggests itself if (48) is rewritten in the following form

£ [{No"- ByMl°)8Ea0 + Ma0e&Bae\dA. (49)

Define a finite bending strain tensor by

Kaf = Ban - bap . (50)

Define a modified membrane stress tensor by

N"0 = Nf - ByMo". (51)

Since Bap is symmetric there will be no loss in generality by defining a modified bending
moment tensor by

| {Mf + Mi"). (52)

Note that the third moment equilibrium equation (26) is equivalent to the statement
that N"" is symmetric. In terms of the newly defined quantities (49) becomes

f (NafSEttll+ M"SKaj) dA. (53)

The details will not be shown here but this expression for the internal virtual work may
be derived from the three dimensional theory by integration through the thickness of
the shell and without approximation provided the displacements are restricted by the
Kirchhoff hypotheses.
Modified tensors similar to, or identical to, N"1' and M" ' have been introduced by
several authors. We emphasize the importance of these quantities more than has been
done by using them to replace No" and in all the subsequent equations. The fact
that the six unknowns Nafl and Ma" are sufficient for the theory in place of the eight
unknowns No" and M„0 simply means that the equilibrium eqs. (23) to (26) have a
slightly more general form than is necessary for a theory with three displacement quanti-
28 J. LYELL SANDERS, JR. [Vol. XXI, No. 1

ties. The reduction in number of unknowns is not possible in a theory which admits
couple stresses and additional modes of deformation.
The two strain tensors Eaf, and Kaf, (or equivalently the two tensors Gafl and
furnish an adequate description of the deformation of the shell as shown by the following
argument. In the first place the deformation of the shell is completely described in terms
of the displacements of points on the middle surface provided the displacements through-
out the thickness are restricted by the Kirchhoff hypotheses. Secondly, from the theory
of surfaces, a knowledge of Gaf, and Bal, as functions of £" and subject to the Gauss and
Codazzi integrability conditions (which in the present case are equivalent to com-
patibility conditions) completely determines the deformed middle surface together
with a coordinate system (the deformed £" system) except for a rigid body motion.
Modified equilibrium equations. Since new stress quantities have been introduced
the equilibrium equations (23) to (26) are no longer quite appropriate for the theory
being developed here. Appropriate equations can be derived from the expression (53)
for internal virtual work.

J CNa'8Eal>
+ Ma0SKae)
dA = J [Na\SUfi;a+ Ba„ 8W)
+ Ma\-8Wiafi + B}.aiUy + 2By„8Uy;a+ ByaByS8W)]
dA

= f [(Na*+ BayM"y)8Ua
+ Mfa8W+ dS
C

- f Wfa + 2B"aM°J
+ By;aMya)8Uf,
+ (M'af - BapNa$ - ByaByl)Ma$)8W]
dA. (54)
The line integral around C is the external virtual work of the edge forces and moments
If a principle of virtual work is required to hold, and if the portion of the shell within C
is in equilibrium, then the internal virtual work must equal the external virtual work
for arbitrary virtual displacements. Thus the condition of equilibrium is that the last
integral in (54) vanishes. This leads to the following equilibrium equations (surface
forces have been omitted in this derivation for simplicity).
N°Z + 2BsaM"y+ By;aMya = 0, (55)
Mfafi- B^N"* - BIB^M"' = 0, (56)
Equations (55) and (56) are in fact identical to the equations (23) to (25) with Q" elimi-
nated. Equation (26) is accounted for by the symmetry of Nal>. If Q" defined by
Qa = M"* (57)

is introduced as an approximation to Qa0, then the equilibrium equations, in an expanded

form, may be written
(Nafi + B'yMay),a + BiQa = 0, (58)

- Ba,{Na" + ByMay) = 0, (59)

Mfa - QB= 0. (60)

1963] NONLINEAR THEORIES FOR THIN SHELLS 29

The following equations equivalent to eqs. (58), (59) and (60) express equilibrium of
forces and moments in the directions of the tangents and normal to the undeformed
middle surface.
[P\l(N°' + BasM's)]., + p^)>a(iV<"^, + BasM") + (pvyQa),a

+ pbl coscoQ"+ Vy = 0, (58)'

(P cos coQa),a - PbayQa + [pfia(N°$ + B\M»)],t

- pb0y\l(Na* + B"SM'S)+p = 0, (59)'

(pM°% + pCZ,M- PQ" = 0, (60)'

where C\p is the difference between the Christoffel symbols of the deformed and unde-
formed coordinate systems. It is expressed in terms of displacements by

Ca0 = GSl[\l,p\yf + Ma.lSMe+ — XT„Me)]-

The load terms in the preceding equilibrium equations are forces per unit of area of the
undeformed middle surface.

Small Strain Approximations

Strain-displacement and equilibrium equations. If the shell bends without extension
then the metric of the deformed middle surface is the same as the metric of the unde-
formed middle surface and eqs. (58), (59) and (60) can be simplified by replacing a
semi-colon by a comma or eqs. (58)', (59)' and (60)' can be simplified by setting p = 1
and (71(5 = 0. The expression for Eaf, does not simplify but the expression (21) for Bafi
simplifies by setting p = 1 in the expression (14) for cos o>and va . The system of equa-
tions thus obtained could serve as an approximate theory for small strains provided it is
legitimate to neglect the covariant derivative of Eap as well as Eai3. In some applications
this might not be true in which case another method of approximation may be sub-
stituted. One of the groups of terms in the expression for Ba$ (see [3]) may be transformed
as follows
\ya cos co- pavy = (X'.XyWn') - (XUn'XN'xU). (61)
An application of Lagrange's identity shows this to be equal to

p-V'cjXiG., = p"Yp€8tA'(<7„p
+ 2EJ. (62)
If we neglect Ea^ and set p = 1, then an approximation to Bafl is

Bap = bafl\a — byp\.ya— —Ma,^(X^Xp

— XpXj)— — \",Hy). (63)

Equilibrium equations appropriate to this approximation can be found by use of the

principle of virtual work.
Constitutive relations. Consistent constitutive relations for the linear small strain
theory of thin elastic shells have been derived in [14, 15 and 16]. These derivations
require only minor modifications in the case of finite displacements and small strains
so they will not be reproduced here. For a thin shell of uniform thickness h composed of
30 J. LYELL SANDERS, JR. [Vol. XXI, No. 1

an isotropic hookean material, the constitutive relations are the same as in Love's first
approximation, namely the linear relations

EhEat, = (1 + v)gayg^NyS - vgaegySNyS,

, (64)
Eh3Kap = (1 + v)gayg0sMyS ~ vga0gyiMy>.

According to [14] these relations may be used even if the definition of Kap in terms of
displacements is altered by the addition of terms of the form ByaEyli. A similar argument
to that in [14] shows that Nafi may be altered by addition of terms of the form ByMay.
In the case of small strain the indices on N"p and Maf>may be raised and lowered with the
metric ga0 instead of Ga$ with negligible error.
If the material of the shell is not elastic and isotropic the relations (64) must be
replaced by others appropriate for the material. However, the strain-displacement
relations and the equilibrium equations given previously are unaffected by the material
so long as transverse shear and normal strains can be neglected.

Approximation of Small Strains and Moderately Small Rotations

The exact theory was somewhat simplified by the assumption that the middle surface
strains are small, but the equations are still very complicated. Considerable additional
simplification can be achieved if the rotations are assumed small also. This simplification
will be carried out in the following.
Approximate strains. For infinitesimal displacements and rotations it is evident
from (33) and (38) that the rotations are given by the formulas

(65)
and
= —w.a + bsau,p= —fia. (66)

For small but finite rotations it is convenient to think of the expressions in (65) and (66)
as rotations (just as in the linear theory of shells). Purely for convenience, suppose that
the coordinates £" have the units of length so that <j>,<f>aand Eais are dimensionless. The
following order of magnitude assumptions will lead to a theory for small strain and
moderately small rotations
<t> or | (ua,B - us,0) = 0(e) (67)

4>a = 0(e) (68)

%(ua,e + u„<a) + baflw = 0(e2) (69)

where e is a number small compared to unity. Write Aa(3in the form

= gap — €<*80+ KUa.P + Up.a + 2 bapW) (70)

From (16) and (37) we find that Ea0 is 0(e2) and is given approximately by the expression

Eat = h(Ua.P + u0.a) + b altW+ + |<7 (71)

The order of magnitude assumption (69) was made so that those terms would not domi-
nate the expression for Eafi ; otherwise the linear theory would result.
1963] NONLINEAR THEORIES FOR THIN SHELLS 31

From G.j — gaP = 0 (e2) it follows that p = 1 + 0 (e2); then from (67) to (70) and
(14) it follows that
vs = + 0(e2) (72)

cos co = 1 + 0(e2) (73)

From the foregoing and (21) and with due regard for symmetry, Kafi = Bafl — ba/>is
given to 0(e) by
Kaft = ifa.a + + hitayb} + 6(74)
(assuming <j><a= 0(e)). Similarly the expression for ya becomes
7« = <t>«+ ~ K fly = 0, (75)
It is conceivable that the above expression for Kap could become 0(e2), in which case the
approximation would be invalid, but in this case the shell would be acting essentially
as a membrane and errors in Kap would be immaterial. The expression given by (74) for
the bending strain is linear and this simplifies the theory considerably. The definition of
bending strain given above is the same as the one derived in references [14] and [17].
Approximate equilibrium equations. Approximate equilibrium equations corre-
sponding to the approximate strain-displacement equations may be found by the same
method used to derive the equilibrium equations (55) and (56). Since the strains are
small dA may be replaced by da and we have

f {Na^[8ua,e + bafl8w — 4>a(Sw^ — b}8uy) + 8us,y]

Ja

+ Ma^[84>clil3+ %ba(5uy,p — 5U(jiT)]+ Qa(8<t>a+ 8w,a — bya8uy)} da

=f - [ - K<t>aN"y
+ W'itNb.y+ i(biM"y),y
- W.M"),y
c

+ blQa]8up + [Q- - - (<t>aNa^),p]8w

+ [M% - Q"]^a( da. (76)
By inspection the equilibrium equations are

N* - Jfo.N"1' + ha\m.a + - WaMa').y + btQ" +P' = 0, (77)

Q:« - ba„Na' - (<f,aNa"),„ + p = 0, (78)
M# - Qa = 0, (79)

where the load terms p" and p have been supplied. These equations express equilibrium
of forces and moments in directions parallel to the tangents and normal of the unde-
formed middle surface. In the left-hand side of (76) 8<j>acould be expressed in terms of
displacements and the term Qa8ya could be omitted. The result for the equilibrium
equations would be (77) and (78) with Q" eliminated by means of (79).
Boundary conditions. The Kirchhoff boundary conditions may be obtained from
the boundary integral in (76) which when written out reads

j) {[AT" - ^a^Nyy + ib^M7' - \b*May]8ua

+ (Q" - (t>aNaP)8w+ Ma^„} % da. (80)

32 J. LYELL SANDERS, JR. [Vol. XXI, No. 1

Let ta be the unit tangent to the curve o, then % = epyty is the unit normal to c in the
surface a. Let

0a "l" j (^1)
where </>,and <pnare scalars. From (81)

= <M" = (~w.a + bluy)ta = -^ + bluyta (82)

Obviously <j>,is not independent of w and ua on c. The last term in (80), namely,

Ma"5<t>anp
ds = j) AT^(- ~ + b\8uyts)ta + 50„naJniSds (83)
c c

becomes, upon integrating by parts,

j) |j| (Maf>tan^bw + My0b"tstyn^Sua + ds (84)

c

assuming o has a continuously turning tangent. Altogether (80) becomes

</>|[N"fi - h"^Nyy + \VyMy0 - ib*May + byM^ty ts]np8ua

+ — 4>aNafnp+ — (M "'Vt^) J oio + MaPnan^8<j>n^

ds. (85)

From this the boundary conditions on c may be read off. They are: prescribe
[N" - h"%Nyy+ lbayMy0- \byMay + b;M"ft,]n, or ua (86)

Q\ - cj>aNa\ + js (I"V«) or w (87)

Ma0nanf, or (88)

Further Approximations
Small rotation about the normal. If the rotation about the normal can be neg-
lected compared to the other two rotations, then the equations can be simplified further.
The importance of the rotation about the normal is not entirely established at the
present time. Several linear theories for thin shells have been constructed which differ
from Love's first approximation only by terms in the bending strain proportional to the
rotation about the normal. The differences between these theories and Love's are tabu-
lated in [14] where the general validity of these theories is questioned. That the rotation
about the normal can sometimes be neglected is evidenced by the fact that these theories
lead to very nearly the same results as more accurate theories in some specific applica-
tions. See, for example, [18]. On the other hand these theories lead to erroneous results
in other applications. See [19 and 20],
For those cases in which the approximation is valid the strains Ea& and Ka/1 given by
equations (71) and (74) can be simplified to read

Eap — + %,«) + bapw -J- %4>a<t>fi

, (89)
1963] NONLINEAR THEORIES FOR THIN SHELLS 33

Kap = + 4>0,a)- (90)

The corresponding approximate equilibrium equations (obtained via the virtual work
principle) are
N%>+ bW - ba^yN'y + Va = 0, (91)
Q\ ~ ba,Na" - (,KNa*)., + v = 0, (92)
- Q" = 0, (93)
and the boundary conditions are to prescribe

[N" + btt'tyMyfi}nf or ua , (94)

Q*n„ - 4>aNa% + js (MaPtanff) or w, (95)

Maf>nanp or 4>n• (96)

When these equations are linearized they reduce, essentially, to those given in [11].
The Donnell-Mushtari-Vlasov approximation. A further simplification of the
above equations is possible under assumptions discussed in [3 and 16]. This consists in
neglecting the term containing ua in the expression for with the following results for
strains,
= i(Wa./j + %,«) + bafiW + \w,aw,B , (97)

Kap = + 03. a) = ~W.afl , (98)

7 a = <£a + W,a (99)

for equilibrium equations,

2\r/ + va = o, (loo)
Q% - batN"" + (w,aNa"),e + V = 0, (101)
MaJ - Qa = 0, (102)

and for boundary conditions prescribe

Na% or ua , (103)

(Q0 + w.aNa")n, + | (Ma"tan,) or w, (104)

Maf>narip or </>„. (105)

Marguerre's shallow shell equations. If applied to a shallow shell the preceding

equations can be further simplified because of the geometry. Suppose that the shell is
nearly flat and parallel to the x3 = z = 0 plane, and that the squares of the slopes of
the shell with respect to the z = 0 plane may be neglected. Then, approximately:

bap = —z,a» . (106)

Since the displacements ua are considered small compared to w, the horizontal dis-
34 J. LYELL SANDERS, JR. [Vol. XXI, No. 1

placements Ua and the vertical displacements W are given approximately in terms of

ua and w by
Ua + z,aW, w£H]W (107)
In terms of Ua and W the membrane strain Eaff , eq. (97)_becomes:

Ea„ = + Uf.„ + e.aW., + S.,W.S+ W,aW,f) (108)

The strains Ka$ and ya are as in eqs. (98) and (99). The conditions of equilibrium in the
horizontal and vertical directions are
N:£ + qa = 0, ' (109)

+ [(«.. + + q = 0, (110)
- Q' = 0, (111)
where q" is the horizontal load intensity and q is the vertical load intensity. The boundary
conditions are to prescribe
Nafnf or Ua , (112)

[Qf + (z.a + WmJN"fr, + Js {MttfStane) or W, (113)

Ma"narii, or <t>n. (114)

These are Marguerre's shallow shell equations in tensor'form [1],

Concluding remarks
Several nonlinear theories for thin shells have been derived in increasing stages of
approximation. The linearization of these equations, which is more or less obvious, has
been omitted but in most cases the resulting linear equations are essentially the same
as shell equations already given in the literature. In all cases the theories are first approxi-
mation theories in the sense that transverse shear and normal strains are neglected.
In each of the theories derived in this paper the equilibrium equations and strain-
displacement relations are related by a principle of virtual work and hence the usual
variational principles may be formulated and proved.
The author is grateful to the Office of Naval Research for supporting this work.

Appendix
Shell equations in ordinary notation. In the ordinary notation with lines of curvature
for coordinates (as used in [10] and [17]) the expressions (71) for middle surface strains
and (74) for bending strains are:

«n = (<*iO!j)-1[a^tti,i + ,2«2 + a^R^W + l/2a1a14>l + l/2ala2<p2*], (-4-1)

e12 = l/2(a1a2)~1[a2W2,i + aiWi,2 — ai,2«i — a2ilu2 + , (A-2)

Kh = (ai«2)_I [<*20i.i + ali2^2], (A-3)

k12 = l/2(a1a2)_1[a202,1 + ax<t>ll2 ~ <*i,20i — a2,i$2 + <*ia^R^1 — ftf1^*], (A-4)

where a comma means partial differentiation with respect to or |2 as the subscript

following the comma indicates.
1963] NONLINEAR THEORIES FOR THIN SHELLS 35

The rotations are given by

0, = -afVi + , (A-5)
<j>= l/2(ctia2) ,i (<^iWi)i2]. (^4-6)
Here, as in the following, missing equations may be obtained by interchanging sub-
scripts 1 and 2 and changing the sign of 4>.The equilibrium eqs. (77), (78) and (79) now
read
(a^iVu),! "I" (c^l-A^12) ,2 "t" Ol, 2-^12 ®2,1-^22 "I" OtlCl2Ri Qi**

+ 1 /2CL&R? - R?)M12]*2 - a1a2R:\<l>1N11 + <I>2N12)**

— l/2ai[<j)(Nii + iV22)]* + ai«2Pi = 0, (A-7)

(®2Qi),i (&1Q2).2 ^1^2(^1 -A^11 H- R2 N22) (a2<t>iNn -)- ol2<I>2N12)

— (arfiNu + 0ii<i>2N22),
2+ = 0, (A-8)
(oC2M 11) tl "1" 12) ,2 ~4~ CK
1,2-^12 a2,1-^22 = 0. (.A-9)

The boundary conditions on an edge = constant are to prescribe

N„ or Mi,
JV12+ 1/2(3R;1 - RT'WV + l/2(iV„ + N22)4>* or u2 ,
(A-10)
Qi "I" ^2 M\i,i 01-A^n 02-A^i2 or t#,
Mn or
The aforementioned rule for interchanging subscripts 1 and 2 applies here as welL
The terms in the preceding equations which drop out when rotations around the
normal are neglected have been marked with a single asterisk. The terms which drop
out in the Donnell-Mushtari-Vlosov approximation are those with either a single or a
double asterisk.
References
1. K. Marguerre, Zur Theorie der gekriimmten Platte grosser Formanderung, Proc. of Fifth International
Congress of Appl. Mech., Wiley and Sons, 1938, pp. 93-101
2. L. H. Donnell, Stability of thin-walledtubesunder torsion, N. A. C. A. TR 479, 1934
3. Kh. M. Mushtari and K. Z. Galimov, Non-linear theory of thin elastic shells, Israel Program for
Scientific Translations, 1961
4. E. Reissner, On axisymmetrical deformations of thin shells of revolution, Proc. of Symposia in Appl,
Math., vol. 3, McGraw-Hill, 1950, pp. 27-52
5. J. L. Synge and W. Z. Chien, The intrinsic theory of elastic shells and -plates, Th. von Kdrmdn Anniver-
sary Volume, Cal. Inst, of Tech., 1941, p. 103
6. W. Z. Chien, The intrinsic theory of thin shells and plates, Part I■—General theory, Q. Appl. Math. 1,
297-327 (1944)
7. W. Z. Chien, The intrinsic theory of thin shells and plates. Part III—Application to thin shells, Q.
Appl. Math. 2, 120-135 (1944)
8. J. L. Ericksen and C. Truesdell, Exact theory of stress and strain in rods and shells, Archive for
Rational Mech. and Anal. X,295-323 (1959)
9. A. E. Green and W. Zerna, The equilibrium of thin elastic shells, Q. Jour, of Mech. and Appl. Math. 3,
9-22 (1950)
10. F. B. Hildebrand, E. Reissner and G. B. Thomas, Notes on the foundations of the theory of small
displacements of orthotropic shells, N. A. C. A. TN 1833, Mar. 1949
36 J. LYELL SANDERS, JR. [Vol. XXI, No. 1

11. A. E. Green and W. Zerna, Theoretical elasticity, Oxford Press, 1954, pp. 375-394
12. Y. V. Novozhilov, Foundations of the nonlinear theory of elasticity, Grayloek Press, 1953, pp. 186-198
13. R. W. Leonard, Nonlinear first approximation thin shell and membrane theory, Thesis, Virginia
Polytechnic Inst., June 1961
14. W. T. Koiter, A consistent first approximation in the general theory of thin elastic shells, Proc. of
I. U. T. A. M. Symposium on the Theory of Thin Elastic Shells, North-Holland Pub. Co., 1959,
pp. 12-33
15. A. E. H. Love, The mathematical theory of elasticity, Dover, Fourth edition, 1944, pp. 528-531
16. V. V. Novozhilov, The theoryof thin shells,P. Noordhoff Ltd., 1959
17. J. L. Sanders, An improved first approximation theory for thin shells, N. A. S. A. Report 24, June
1959
18. P. M. Naghdi and J. F. Berry, On the equations of motion of cylindrical shells, Jour, of Appl. Mech.
21, 160-166 (June 1954)
19. J. K. Knowles and E. Reissner, Note on the stress-strain relations for thin elastic shells, Jour. Math,
and Phys. 37, 269-282 (Oct. 1958)
20. J. W. Cohen, The inadequacy of the classical stress-strain relations for the right helicoidal shell, Proc.
■ofthe I. U. T. A. M. Symposium on the Theory of Thin Elastic Shells, North-Holland Pub. Co.,
1959, pp. 415-433