International Journal of Solids and Structures 97–98 (2016) 137–149

Contents lists available at ScienceDirect

International Journal of Solids and Structures

journal homepage: www.elsevier.com/locate/ijsolstr

On a consistent finite-strain shell theory based on 3-D nonlinear

elasticity
Zilong Song, Hui-Hui Dai∗
Department of Mathematics, City University of Hong Kong, 83 Tat Chee Avenue, Kowloon, Hong Kong

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents a general finite-strain shell theory, which is consistent with the principle of station-
Received 24 May 2016 ary three-dimensional (3-D) potential energy. Based on 3-D nonlinear elasticity and by a series expansion
Revised 18 July 2016
about the bottom surface, we deduce a vector shell equation with three unknowns, which preserves the
Available online 28 July 2016
local force-balance structure. The key in developing this consistent theory lies in deriving exact recursion
Keywords: relations for the high-order expansion coefficients from the 3-D system. Appropriate 2-D boundary con-
Shell theory ditions and associated 2-D weak formulations are also proposed, including various practical cases on the
Nonlinear elasticity edge. Then, to demonstrate its validity, axisymmetric deformations of spherical and circular cylindrical
Soft materials shells are investigated, and comparisons with the exact solutions are made. It is found that the present
Asymptotic analysis shell theory produces second-order correct results for the general dead-load case and internally pressur-
ized case. The advantages of the present shell theory include consistency, high accuracy, incorporating
both stretching and bending effects, no involvement of higher-order stress resultants and its applicability
to general loadings.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction nished some well-known shell theories with famous names at-
tached to them, such as Koiter (1966), Reissner (1974) and Donnell
Shells are ubiquitous structures in nature and engineering, (1976). For detailed reviews of early developments, we further re-
which have inspired extensive research interests on shell theory fer to Koiter and Simmonds (1973), Pietraszkiewicz (1979) and
and its applications (see, e.g., Timoshenko et al., 1959; Ventsel and Wan and Weinitschke (1988). Also based on assumptions, it is
Krauthammer, 2001; Pietraszkiewicz and Gorski, 2013). Shell the- worth mentioning the hierarchic method (Naghdi, 1972) and con-
ories with small strains can be dated back at least 120 years ago, straint method (Podio-Guidugli, 1990), where admissible displace-
and the literature is vast, including both direct and derived shell ment fields are restricted. Although these theories have found a
theories. For direct theories, we refer to Altenbach et al. (2010) for huge success in engineering applications, due to the hypotheses
a recent review. Here, we only give a brief review on some selected involved their consistency with the three-dimensional (3-D) theory
derived shell theories, which are more relevant to the present cannot be expected under general loadings. Present research efforts
work. concerning such theories are to justify them mathematically and to
Early attempts on shell theories relied on a priori hypotheses of find out the range of applicability.
a geometrical or mechanical nature, mostly motivated by engineer- Some mathematical approaches have been utilized to derive
ing intuition. Based on Kirchhoff’s assumptions (Kirchhoff, 1850) consistent shell theories or to justify existing shell theories un-
and thin shell approximation, the classical shell theory was de- der certain circumstances. Gol’denveizer (1963) first developed the
veloped by Love (1888) to provide the leading approximation. For idea of applying formal asymptotic expansions to the 3-D differen-
the “best” equivalent variants of first approximation theory, we re- tial formulation to derive shell theory, and Hamdouni and Millet
fer to Sanders (1959) and Koiter (1960). Later works relaxed some (20 03a; 20 03b) extended it to more general cases. An alternative
assumptions, with an attempt to include certain nonlinear and/or way of applying asymptotic method is based on the 3-D weak for-
rotational effects, e.g., incorporating geometrically nonlinear effect mulation. This approach was initiated by Ciarlet and Destuynder
and postulating a linear distribution for displacements. These fur- (1979a; 1979b) for plates and was first attempted by Destuynder
(1980) for shells, and we refer to Ciarlet (20 0 0) and Ciarlet and
Mardare (2008) for a comprehensive account. A welcome feature
∗
corresponding author. Fax: +852 27888561.
is its capability of providing convergence results for some linear
E-mail address: [email protected] (H.-H. Dai). shell theories. Convergence results, even for some nonlinear theory,

http://dx.doi.org/10.1016/j.ijsolstr.2016.07.034
0020-7683/© 2016 Elsevier Ltd. All rights reserved.
138 Z. Song, H.-H. Dai / International Journal of Solids and Structures 97–98 (2016) 137–149

can also be achieved by Gamma convergence method, which is to parametrize the base surface  of the shell in the reference con-
concerned with the limiting 2-D variational problem of vanish- figuration, where a generic point is denoted by r = r(θ α ). The co-
ing thickness (Friesecke et al., 2003; Le Dret and Raoult, 1996). variant base (tangent) vectors along the coordinate lines are given
Although quite impressive in terms of rigorous derivations and by gα = r,α = ∂ r/∂ θ α , and their contravariant counterparts are de-
technical sophistication, these approaches do rely on a priori scal- noted by gα . Then, the unit normal n to  is well-defined accord-
ings, assuming applied loads (or deformation) be of certain orders ing to n = g1 ∧ g2 /|g1 ∧ g2 | to form a right-handed basis, where
of shell thickness. Nevertheless, applied loads are external, which ∧ means the cross product. In component forms, we will identify
should not be directly related to the thickness in various situations. g3 = g3 = n, and components with subscripts and superscripts are
As a consequence of scalings, these methods do not lead to a sin- respectively associated with bases gi and gi . Subsequently, the 3-D
gle shell model incorporating both bending and stretching effects, reference position is decomposed as
which is probably most needed in applications.
X = r(θ α ) + Zn(θ α ), 0 ≤ Z ≤ 2h, (1)
Based on more general expansions of displacements, some
higher-order (refined) shell theories have also been derived by where Z is the shell thickness coordinate perpendicular to . A
2-D variational principle method, see Reddy and Liu (1985) and distinct feature from plates is that the normal n depends on the
Reddy and Arciniega (2004) for linear case and Meroueh (1986) for coordinates of the base surface. By the Weingarten equations, the
nonlinear case. Recently, starting from 3-D nonlinear elasticity, variation of n is described by the curvature map κ
Steigmann (2008; 2010; 2013) derived a notable shell theory which
dn = −κdr, dn = n,α dθ α , dr = gα dθ α . (2)
extends Koiter’s shell theory and dictates an optimal third-order
approximation for the 3-D potential energy. However, his deriva- Equivalently, we may write curvature tensor as
tion is restricted to the nearly traction-free case.
κ = −n,α  gα = καβ gβ  gα = κβα gβ  gα . (3)
In this work, we aim to generalize the previous plate theory
(Dai and Song, 2014; Song and Dai, 2016; Wang et al., 2016) to Then, the mean and Gaussian curvatures are defined by
a finite-strain shell theory, which is consistent with the 3-D en- 1 1 β
ergy principle and with no special restrictions on applied loads. H= tr(κ ) = καα , K = det(κ ) = [(2H )2 − κα κβα ]/2. (4)
2 2
Specifically, we start with a general series expansion for deformed
By choosing orthogonal principal directions as the base vectors, the
positions about the bottom surface, and make asymptotically con-
nonzero components κ11  κ1 and κ22  κ2 are called the two prin-
sistent truncations on the 3-D differential system. The key in the
cipal curvatures, and consequently H = (κ1 + κ2 )/2 and K = κ1 κ2 .
derivation lies in obtaining the recursion relations for all high-
Accordingly, the differential of the reference position is given
order expansion coefficients, leading to a final vector shell equa-
by
tion for the leading coefficient, which preserves the local force bal-
ance in all three dimensions. Suitable 2-D boundary conditions, as dX = dr + Zdn + ndZ = μdr + ndZ  gˆ α dθ α + ndZ,
well as the associated 2-D weak form, are also consistently derived, μ = ( 1 − Z κ ), (5)
accommodating various practical situations. The specific forms of
shell equations for cylindrical and spherical shells are also pro- where 1 is the rank-2 identity tensor within the tangent plane, and
vided, and concrete problems concerning these structures with ax- we have defined the covariant and contravariant base vectors at an
isymmetric deformations are investigated to verify the theory. The arbitrary point inside the shell
distinctive features of the present shell theory, not all enjoyed by gˆ α = μgα , gˆ α = μ−T gα , (6)
existing ones, include: (i) finite strains; (ii) consistency with 3-D
energy principle; (iii) general loadings; (iv) incorporating bending which are also orthogonal to n. Note that previous assumption on
and stretching effects simultaneously; and (v) without any ad hoc geometry implies that |2hκ α | < 1, and hence the inverse μ−1 is
hypotheses or any a priori scaling assumptions. In short, it is an well-defined. The volume element is calculated as
“all-purpose” shell theory. dV = μ(Z )dAdZ, μ(Z ) = det(μ ) = 1 − 2HZ + KZ 2 , (7)
The manuscript is arranged as follows. In Section 2, we recall
the kinematics of a shell and the 3-D formulation. Section 3 de- with the area element dA on the base surface defined by
√
rives the finite-strain shell theory, which is consistent with the 3- dA = |g1 ∧ g2 |dθ 1 dθ 2 = gdθ 1 dθ 2 , (8)
D weak formulation. Subsequently in Section 4, the associated 2-D
weak formulation is discussed in detail, which also yields various where g is the determinant of the metric tensor (gαβ ) induced by
natural boundary conditions. Section 5 presents the shell equations the base vectors gα .
and some concrete problems for two common shell structures, to The current position is written as x = x(θ α , Z ), then the defor-
ease application and to substantiate the consistency. Finally, some mation gradient is given by
concluding remarks are made. ∂x ∂x
F = x,α  gˆ α + n= (∇ x )μ−1 +  n, (9)
∂Z ∂Z
2. Kinematics and the 3-D formulation where ∇ = gα ∂ /∂ θ α is the in-plane 2-D gradient on the base sur-
face (∇ x = x,α  gα ). Keep in mind that the tensor μ and its de-
We consider a homogeneous thin shell of constant thickness terminant μ depend on Z and the curvatures of the shell, which
composed of a hyperelastic material. A material point in the ref- degenerate to 1 and 1 for plates.
erence configuration  × [0, 2h] is denoted by X, where the thick- For a hyperelastic material with strain energy function (F),
ness 2h of the shell is assumed to be small compared with the the nominal stress tensor is defined by
other two dimensions of the bottom (base) surface  as well as  
the local radius of curvature. The deformed position in the current ∂ ∂
S= , Si j = , (10)
configuration is denoted by x. In the following notations, Greek let- ∂F ∂ Fji
ters (α , β , γ ...) run from 1 to 2, whereas Latin letters (i, j, k...) run
and the associated first and second-order elastic moduli are de-
from 1 to 3, repeated summation convention is employed and a
fined by
comma preceding indices means differentiation.
Following Ciarlet (2005) and Steigmann (2012), we first recall ∂2 ∂3
A1 ( F ) = , A2 ( F ) = . (11)
the kinematics of the shell. Curvilinear coordinates θ α are utilized ∂ F∂ F ∂ F∂ F∂ F
 Z. Song, H.-H. Dai / International Journal of Solids and Structures 97–98 (2016) 137–149 139

It is assumed that the strain energy function for the deformations The principle of stationary potential energy requires the first
concerned satisfies the strong-ellipticity condition variation of E to be zero, which is expressed as
 
a  b : A 1 ( F )[ a  b ] > 0 ,
2h
for all a  b = 0, (12)
δE = − (Div S ) · δ x μ(Z )dZdA
where the colon between second-order tensors means the trace of   0
their tensor product A : B = tr(AB ), and the square brackets after − ST n + q− · δ x(θ α , 0 )dA
Z=0
a modulus tensor represent the operations  
 i j  i j + ST n − q+ · δ x(θ α , 2h ) μ(2h )dA (20)
A1 [A] = (A1 )i jkl Alk , A2 [A, B] = (A2 )i jklmn Alk Bnm . (13)  Z=2h
  2h
For the case of dead-loading and in the absence of body forces, + ST N · δ x(s, Z )da
the 3-D potential energy E is given by ∂ 0 0
  2h 
  2h + ST N − q · δ x(s, Z )da = 0.
E = ¯ + V̄ , ¯ = (F )μ(Z )dZdA, ∂ q 0
 0
   It follows that the 3-D field equations together with boundary con-
V̄ = − q− (θ α ) · x(θ α , 0 ) + μ(2h )q+ (θ α ) · x(θ α , 2h ) dA (14) ditions are

  2h Div S = 0, in  × [0, 2h],
− q(s, Z ) · x(s, Z )da,
∂ q 0
T
S n = −q , −
in ,
Z=0

where q±
are the applied tractions on the top and bottom surfaces T
S n +
=q , in , (21)
Z=2h
of the shell, q is the applied traction on the lateral surface (the
x = b(s, Z ), on ∂ 0 × [0, 2h],
edge), and da (in distinction from dA in (8)) is the area element
on the edge. The boundary ∂  of the base surface is divided into ST N = q(s, Z ), on ∂ q × [0, 2h],
two parts, the position boundary ∂ 0 and traction boundary ∂ q .
where b is the prescribed position on part of the lateral surface.
On ∂ , let s be the arclength variable, and τ and ν be the unit tan-
gent and outward normal vectors, in a way that ν = τ ∧ n. By these
3. The 2-D shell theory
definitions, we easily get dr = τ ds along the boundary. Then, it fol-
lows from (5) that the local differential consists of dXτ = μτ ds
In this section, we intend to reduce the previous 3-D system to
along tangential direction and dXn = ndZ along normal direction.
a consistent 2-D shell theory, consisting of a fourth-order vector
The outward normal of the lateral surface is denoted as N, and we
shell equation and appropriate 2-D boundary conditions. The same
obtain
consistency criterion as in Dai and Song (2014) is employed, i.e.,
Nda = dXτ ∧ dXn = (μτ ) ∧ ndZds, each of the five terms in the 3-D weak formulation (20) should be
of a required asymptotic order, say O(h4 ), separately for the shell
μτ = (1 − Z κ )τ = (τ β − Z τ α καβ )gβ ,
approximation. The starting point is a generic series expansion of
|(μτ ) ∧ n|2 = |μτ|2 = τ · τ − 2Z τ · (κτ ) + Z 2 (κτ ) · (κτ ) (15) the current position vector, followed by proper truncations of the
= 1 − 2Z τ α καβ τ β + Z 2 τ α κασ κσ β τ β 3-D differential system (21) carried out with the consistency crite-
rion in mind.
 1 − 2κτ Z + cτ Z 2 ,

where the coefficients of Z in last expression are connected with 3.1. Series expansions
the three fundamental forms of the base surface (note cαβ 
κασ κσ β are the components of the third fundamental form). Im- Suppose the current position x(X) and the strain energy (F)
mediately, we deduce are both C5 in their arguments. Then, the current position is ex-
 panded about the bottom surface Z = 0 in the form
√
da = 1 − 2κτ Z + cτ Z 2 dZds  gτ dZds, 1 2 (2 ) α 1
√ (16) x(X ) = x(0) (θ α ) + Zx(1) (θ α ) +
Z x ( θ ) + Z 3 x (3 ) ( θ α )
gτ N = (μτ ) ∧ n = ν + Z ν1 , ν1  −κτ ∧ n, 2 6 (22)
1 4 (4 ) α
where gτ represents the preceding quadratic form. + Z x (θ ) + O(Z ), 0 ≤ Z ≤ 2h,
5
24
Remark. If principal directions are utilized and the edge is along where (· )(n ) = ∂ n (· )/∂ Z n |Z=0 (n = 1, · · · , 4 ). Accordingly, the defor-
θ 1 , then mation gradient has a similar expansion
√ √ √
τ = g1 /|g1 | = g1 / g11 , ds = g11 dθ 1 , gτ = 1 − Z κ1 , (17) 1 2 (2 ) α 1
F = F(0) (θ α ) + ZF(1) (θ α ) + Z F ( θ ) + Z 3 F ( 3 ) ( θ α ) + O ( Z 4 ).
2 6
and the vector ν1 has the same direction as ν, which implies N = ν.
(23)
If we follow Steigmann (2012) and locally adopt a different repre-
sentation for curvature tensor like It follows from the definition (9) that

κ = κτ τ  τ + κν ν  ν + κτ ν (τ  ν + ν  τ ), (18) ∂x ∂x
F = (∇ x )μ−1 + n= ( ∇ x )[ 1 + Z κ + Z 2 κ 2 + · · · ] +  n.
then the area element is computed as ∂Z ∂Z
(24)
Nda = (μτ ) ∧ ndZds = [(1 − Z κτ )ν + Z κτ ν τ ]dZds,
 (19) Substituting (22) into (24) and comparing with (23), we obtain the
da = (1 − Z κτ )2 + (Z κτ ν )2 dZds. relations
m
Comparing this representation with (16), we can identify ν1 = F (m ) =
m!
(∇ x(i) )κm−i + x(m+1)  n, m = 0, 1, 2, 3, (25)
κτ ν τ − κτ ν and cτ = κτ2 + κτ2ν . i=0
i!
140 Z. Song, H.-H. Dai / International Journal of Solids and Structures 97–98 (2016) 137–149

which indicates that F(m) depends linearly on x(m+1 ) . recast as

Based on the smoothness of , the nominal stress S can be ex-
Bx(2) + f(2) = 0,
panded as 
1
Ba  (A1 [a  n] )T n, ∀ a, Bi j = A13i3 j , (33)
S (F ) = S (0 )
(θ α ) + ZS(1) (θ α ) + Z 2 S(2) (θ α )
2 f ( 2 ) = ∇ · S ( 0 ) + ( A 1 [ ( ∇ x ( 0 ) )κ + ∇ x ( 1 ) ] )T n,
1
+ Z 3 S ( 3 ) ( θ α ) + O ( Z 4 ). (26) where both the acoustic tensor B and the vector f(2) only involve
6
x(0) and x(1) . The strong-ellipticity condition ensures that B is in-
By the chain rule and using the elastic moduli in (11), we obtain vertible, and the above equation yields that
S ( 0 ) = S ( F ( 0 ) ), S ( 1 ) = A 1 ( F ( 0 ) )[ F ( 1 ) ] ,
(27) x(2) = −B−1 f(2) . (34)
(2 ) (0 ) (2 ) (0 ) (1 ) (1 )
S = A (F
1
)[ F ] + A ( F 2
)[ F , F ] .
Similarly, from the second Eq. (32)2 it follows that
The formula for S(3) is omitted for brevity since it is not needed
in the derivations to follow. An observation is that S(m) (m = 1, 2) x(3) = −B−1 f(3) , (35)
also depends linearly on x(m+1 ) , which is one key point of the with the vector
reductions in the next subsection. 
f (3 ) = κT gα · S,(α0) + ∇ · S(1)
3.2. Derivation of the vector shell equation  T
2
2 (36)
+ A1 (∇ x(i) )κ2−i + A2 [F ( 1 ) , F ( 1 ) ] n.
By substituting the preceding series expansions into the differ- i!
i=0
ential system (21), it seems that there are too many unknowns
x(i) (i = 0, . . . , 4) involved. But they are necessary in the sense that The coefficient x(4) is an intermediate quantity, whose explicit ex-
they form a closed system by consistently truncating the 3-D sys- pression is not needed, however the third Eq. (32)3 as a whole
tem to O(h4 ). Some equations in (21) will serve to deduce re- will be utilized to eliminate it later. In brief, by repeatedly sub-
cursion relations for the high-order coefficients x(i) (i = 1, . . . , 4), stituting the recursion relations, all x(i) (i = 1, 2, 3) can essentially
leading to a final shell equation for x(0) . Note that some works be expressed by x(0) and its derivatives.
(e.g. Naghdi, 1972) in the literature often treat these expansion co- Finally, the top traction condition (21)3 states
efficients as independent unknowns. The approach adopted here is
4 3 ( 3 )T
different. We observe that these coefficients are correlated if one S ( 0 )T n + 2 hS ( 1 )T n + 2 h2 S ( 2 )T n + h S n + O ( h4 ) = q + . (37)
substitutes them into the 3-D differential system (21). In the fol- 3
lowing, we manage to explore these intrinsic relations by using the Alternatively, here we take into consideration the factor μ(2h ) =
3-D pointwise information. 1 − 4hH + 4h2 K as in the third term of (20), then the above equa-
First of all, we analyze the recursion relation for x(1) . Substitut- tion is equivalent to (with a slightly different remainder)1
ing (26) into the bottom traction condition (21)2 leads to
 T μ(2h )S(0)T n + 2hμ(2h )S(1)T n + 2h2 (1 − 4hH )S(2)T n
S (0 ) ( ∇ x (0 ) + x (1 )  n ) n = −q− . (28) (38)
4 3 ( 3 )T
+ h S n + O ( h4 ) = μ ( 2 h )q + .
The strong-ellipticity condition, together with the implicit function 3
theorem, guarantees that x(1) can be uniquely solved in terms of Clearly this equation and (28,32) form a closed system for the five
∇ x(0) (cf. Steigmann, 2013). unknown vectors x(i) (i = 0, . . . , 4). Substituting (28,32) as a whole
Next, the 3-D field Eq. (21)1 dictates into (38) and neglecting the remainder, we deduce the vector shell
∂ T equation
Div S = gˆ α · S,α + ( S n ) = 0, (29)
∂Z
2
where gˆ α is given in (6) and has the expansion (1 − 4hH + 4h2 K )∇ · S(0) + h(1 − 4hH )∇ · S(1) + h2 ∇ · S(2)
3
gˆ α = [1 + Z κ + Z 2 κ2 + · · · ]T gα . (30) + h(1 − 4hH )[κT gα ] · S,(α0) +
4 2 T α
h [κ g ] · S,(α1)
3
Substituting the expansions (26,30) into (29), the vanishing of the 4 2
coefficients of Zn leads to + h [(κ2 )T gα ] · S,(α0)
3
m
m! = −(1 − 4hH + 4h2 K )(q+ + q− )/(2h ).
[(κm−i )T gα ] · S,(αi ) + S(m+1)T n = 0, m = 0, 1, 2, . . . (31)
i! (39)
i=0

In particular, the first three are explicitly expressed by This 2-D vector shell equation preserves the 3-D force bal-
ance/equilibrium in three directions in the sense of through-
∇ · S ( 0 ) + S ( 1 )T n = 0, thickness average, since formally we can directly deduce this equa-
 T α
κ g · S,(α0) + ∇ · S(1) + S(2)T n = 0, (32) tion by an integration of (29) multiplied by μ(Z) over Z (with the
  aid of top and bottom traction conditions). By some delicate ma-
2 (κ2 )T gα · S,(α0) + 2 κT gα · S,(α1) + ∇ · S(2) + S(3)T n = 0, nipulations (see Appendix A), the above shell equation can be re-
cast in a compact form
where ∇ · S(0 ) = gα · S,(α0 ) denotes the 2-D divergence. For plates 
κ = 0, these reduce to two-term relations as in Dai and Song ∇ · 1(S˜ + hκSˆ ) = −q˜ , (40)
(2014).
In principle, the three relations in (32) can be utilized respec-
tively to solve x(i) (i = 2, 3, 4), since S(m+1 ) (m = 0, 1, 2) only de- 1
We have tacitly assumed that the radius of curvature has similar order as or
pends linearly on x(m+2 ) . For example, the first Eq. (32)1 can be larger than planar length scale, i.e., Rα = 1/|κα | ≥ O(L ), or O(h/Rα ) ≤ O(h/L).
 Z. Song, H.-H. Dai / International Journal of Solids and Structures 97–98 (2016) 137–149 141

where satisfied (in a pointwise manner), the local form of the 3-D mo-
8 2 ment equilibrium equations is T = TT in the referential descrip-
S˜ = (1 − 2hH )S(0) + h(1 − hH )S(1) + h2 S(2) , tion, where T = SF−T is the second Piola-Kirchhoff stress tensor.
3 3
4 The principle of frame indifference requires the strain energy func-
Sˆ = S(0) + hS(1) , (41) tion to be a function of the Green-Lagrangian strain tensor E =
3
2 (F F − I ), which is symmetric. Thus, by definition, T = ∂ /∂ E is
1 T
1  
q˜ = (1 − 4hH + 4h2 K )q+ + q− . also symmetric, which implies that the 3-D moment equilibrium
2h
equations are automatically satisfied. For the present shell theory
For plates (i.e., κ = 0 ), S˜ + hκSˆ reduces to S̄ (averaged stress ten- based on general expansions, it is easy to check that the symme-
sor over the thickness) defined previously in Dai and Song (2014). tries of E and T are retained up to O(Z3 ). So, the 3-D moment equi-
Therefore, the quantity S˜ + hκSˆ is considered as the averaged librium equations in local form are satisfied up to O(Z3 ). Note that
stress, and q˜ is the weighted average of surface tractions (the the shell equations only have three unknowns, but they are sup-
weight depends on the surface area). The component form is given plemented with recursion relations (from which one can recover
by the 3-D displacement field).
[S˜α j + hκδα Sˆδ j ];α = −q˜ j , (42)
Remark 3. It may be worth recalling two previous famous formu-
where the covariant derivatives, indicated by a semicolon, are de- lations on shell theory. Naghdi (1972) also started with a general
fined in Appendix B. The shell equation only entails up to x(3) , series expansion of 3-D current position vector, but treated all the
and by recursion relations, results in a fourth-order differential expansion coefficients as independent unknowns. The system of
equation for x(0) . Although only the leading coefficient x(0) ap- 2-D shell equations can be obtained by direct integration of 3-
pears in above shell equation, the other expansion coefficients x(k) D field equations and moment equilibrium equations after multi-
(k = 1, . . . , 4) can be calculated from recursion relations once x(0) plication by Zn (Naghdi, 1972, p. 519). The final shell system has
is solved. Therefore, the distribution of 3-D current position vector a nice structure, but contains many independent unknowns since
can be recovered. the higher-order coefficients (unknowns) cannot be expressed in
Now we briefly analyze the asymptotic order concerning above terms of lower-order coefficients by using the 2-D shell equations
truncations to derive (40). Since the three equations in (32) are uti- there. It should be noted that Naghdi’s approach does not rule out
lized, the error for the 3-D field equation is Div S = O(Z 3 ). Also, the that the high-order expansion coefficients can be represented in
two traction conditions (28,38) are kept correct to O(h3 ). There- terms of the lower-order ones directly from the 3-D equations in a
fore, the present shell equation, together with the recursion rela- pointwise manner, since in deriving his shell theory the 3-D equa-
tions, guarantees an O(h4 ) error for each of the first three terms tions were only used in the manner of through-thickness averages.
in the variation δ E in (20). In the next subsection, we shall intro- In the present formulation, we manage to express all higher-order
duce proper shell boundary conditions to make the error of the expansion coefficients in terms of leading coefficient x(0) by utiliz-
two edge terms in (20) also O(h4 ). ing 3-D pointwise information, including the bottom traction con-
Remark 1. Now, we show that the shell equations are actually lo- dition and the expansions of 3-D field equations. This leads to a
cal force equilibrium equations for a shell element. Take a shell simpler vector shell equation for only x(0) . Under the assumption
element with a base surface R ⊂  and thickness 2h. The force bal- of smoothness, the derivation is rigorous. The idea of simplifying
ance requires that higher-order coefficients by 3-D information has also been utilized
   in recent papers of Steigmann (e.g. Steigmann, 2013). The differ-
ST Nda = − q− + μ(2h )q+ dA. (43) ence is that his derivation was restricted to the nearly traction-free
∂ R×[0,2h] R
case with the purpose of obtaining an optimal O(h3 ) 2-D energy,
The left-hand side can be written as while the present work deals with the general case and intends to
 
derive a theory consistent with the 3-D energy principle.
ST Nda = ST [(1 − Z κτ )ν + Z κτ ν τ ]dZds
∂ R×[0,2h] ∂ R×[0,2h] In the mixed approach (between a direct shell theory and a de-
 rived one) adopted in Libai and Simmonds (1998), averaged 2-D
= ST [(1 − 2HZ )1 + Z κ]νdZds (44) quantities including displacement and stress are suitably defined
∂ R×[0,2h]
 and treated as variables instead of being expanded in series. One
 N νds, elegance of such an approach is that the balances of translational
∂R and rotational momentums are exactly formulated for the shell
where use has been made of (19)1 and (18), and N is the so-called structure. In this theory, like a direct shell theory, the constitutive
stress resultant tensor defined in Libai and Simmonds (1998, p460) relations for the 2-D variables need to be specified, which may not
and given by be an easy task. In the present work, the constitutive relations for
 2h S(i) are naturally inherited from the original 3-D one and need not
N = ST [(1 − 2HZ )1 + Z κ]dZ. (45) be postulated again. Also, once x(0) is solved, the distribution of 3-
0 D current position can be recovered by recursion relations and the
One can easily check by substituting the expansions of S that rotation can then be calculated.
1 T
N = 1(S˜ + hκSˆ ) + O(h3 ). (46) Remark 4. The above shell equation can be easily adapted to
2h
some non-dead loading cases, when the base surface is sub-
The local form of the shell element force equilibrium equations can
jected to a pressure or sitting on a foundation. For instance,
be obtained by using the divergence theorem, also see (F.2) of Libai
when the pressure P is given in the current configuration, the
and Simmonds (1998, p460). Then, dividing the factor 2h, one ob-
traction q− now depends on the deformation, in the form q− =
tains exactly (40) except the remainder O(h3 ).
P det(F(0 ) )(F(0 ) )−T n. In this case, as long as the derivative of
Remark 2. In the 3-D formulation, one only needs 3 force equilib- S(0 )T n + q− with respect to x(1) is invertible, one can similarly
rium equations with three unknowns (displacement components), solve x(1) from the condition (28). The final shell Eq. (40) is still
and the 3 moment equilibrium equations are automatically satis- valid. This case will be mentioned in the examples of spherical and
fied. Here is the reason. When the force equilibrium equations are cylindrical shells in Section 5.
142 Z. Song, H.-H. Dai / International Journal of Solids and Structures 97–98 (2016) 137–149

Remark 5. The explicit formulas for x(2) and x(3) involve moduli 0, 1, 2, 3). In this case, we define the averaged traction q0 (s) as
A1 and A2 , which make the formulas inconvenient to use directly. 
1 2h
√
Thus, these formulas have more theoretic value than practical. In q0 = q(s, Z ) gτ dZ
many concrete examples, one can readily get the explicit expres- 2h 0
sion of S(i) (from the strain energy function), and directly derive 1
 2h  1 2 (2 )

= q (0 ) + Z q (1 ) + Z q + ··· 1 − 2κτ Z + cτ Z 2 dZ
the recursion relations from the Eq. (32) without calculating such 2h 2
moduli.
0
 
= q(0) + h q(1) − κτ q(0)
2 2  (2 ) 
+ h q − 2κτ q(1) + κτ2ν q(0) + O(h3 ).
3.3. Boundary conditions 3
(51)
In this subsection, we aim to reduce the 3-D boundary condi- And, we can adopt the following two conditions2
tions to appropriate ones for the preceding 2-D vector shell Eq. 2 2 (2 )
(40). Since the shell equation is of fourth-order, two conditions re- p̄  p(0) + hp(1) + h p + O ( h3 ) = q 0 ,
3
garding x(0) and its derivatives are needed, either on the position
S ( 0 )T ν = q ( 0 ) , on ∂ q , (52)
boundary ∂ 0 or on the traction boundary ∂ q .
Case 1. Prescribed position in the 3-D formulation where p(i)
are defined in (49). Alternatively, we may define a stress
On ∂ 0 × [0, 2h], the position b is prescribed as in (21)4 . In resultant (or moment about the middle line) as
this case, we define  2h
1 √
m (s ) = q(s, Z ) gτ (Z − h )dZ
 2h 0
1  
2h
1 2 2 (2 )
x̄(s ) = x(s, Z )dZ = x(0) + hx(1) + h x + O ( h3 ), = h2 q(1) − κτ q(0) (53)
2h 0 3 3
(47) 1  
+ h3 q(2) − 2κτ q(1) + κτ2ν q(0) + O(h4 ),
3
and adopt the following two conditions
and the second condition in (52) can be suitably replaced by

x ( 0 ) = b ( 0 ) ( s ), x̄ = b̄, on ∂ 0 , 1 2h
1 1
(48) (Z − h )pdZ = h2 p(1) + h3 p(2) + O(h4 ) = m(s ).
2 1 2h 3 3
x ( 0 ) = b ( 0 ) ( s ), x(1) + hx(2) + O(h2 ) = (b̄ − b(0) ),
0
⇔
3 h (54)
The conditions in (52,54) involve up to third-order derivatives of
where b(0 ) = b|Z=0 and b̄ is defined in the same way as x̄. The sec-
x(0) by virtue of recursion relations.
ond condition contains up to the second-order derivatives of x(0)
To check the consistency, we examine the asymptotic order of
upon using the recursion relations.
the fifth term in (20). For convenience, we denote
To check the consistency, we should examine the asymptotic or- √ √
der of the fourth term in (20). For simplicity (see (16) for da), we qˆ = gτ ( S T N − q ) = p − gτ q , (55)
define and use qˆ (i ) (i = 0, 1, 2 . . .) to represent the coefficients of its Taylor
expansion over Z. Then, the inner integral of that term in (20) is
√ 1
p = g τ S T N = [ S ( 0 ) + Z S ( 1 ) + Z 2 S ( 2 ) + · · · ] T [ ν + Z ν1 ] given by
2   
1 2h 2h 2h  
=S ( 0 )T
ν + Z[S ν + S ν1 ] + Z 2 [S(2)T ν + 2S(1)T ν1 ] + O(Z 3 )
( 1 )T ( 0 )T qˆ · δ xdZ = qˆ · δ x(0) dZ + Z qˆ (0) + Z qˆ (1) · δ x(1) dZ
2 0 0 0
1 2 (2 )  2h
(0 ) (1 )
p + Z p + Z p + O ( Z ) 3 1 2 (0 )
2 + Z qˆ · δ x(2) dZ + O(h4 ).
0 2
(49)
(56)
Recall that ν1 = κτ ν τ − κτ ν in the remark following (16), thus we On the right-hand side, the first term is zero due to the condition
get N = ν for the general case. Nevertheless, when the edge is (52)1 . By simple manipulations of the two conditions in (52) (or
along the principal direction, we indeed have ν1 = −κτ ν, N = ν (52)1 and (54)), one can show that qˆ (0 ) = O(h2 ) and qˆ (1 ) = O(h ),
√ and consequently the remaining terms are of O(h4 ).
and gτ = 1 − Z κτ . Based on the definition of p, we get
To sum up, the 2-D vector shell Eq. (40) together with recursion
  relations and boundary conditions (48) and (52) (or (48), (52)1 and
2h
√ 2h
ST N · δ x gτ dZ = p · δ xdZ (54)), is consistent with the 3-D weak formulation, with an O(h4 )
0 0
 2h  2h
error. Note that no higher-order stress resultants are involved in
= p(0) · δ xdZ + Zp(1) · [δ x(0) + Z δ x(1) ]dZ (50) above boundary conditions, in contrast to some higher-order shell
0 0 theories. The derivation of the above 2-D boundary conditions di-
 2h
1 2 (2 ) rectly from 3-D ones is very natural, since these boundary con-
+ Z p · δ x(0) dZ + O(h4 ). ditions satisfy the proposed consistency criterion. Other types of
0 2
common 2-D boundary conditions in existing shell theories are of-
Obviously the first term in (50) is zero due to the condition (48)2 , ten based on 2-D variational principle. In the present formulation,
and the third term is zero as δ x(0 ) = 0. Also, the second term is of one can similarly propose such 2-D boundary conditions based on
O(h4 ) since it is easily seen from (48) that δ x(1 ) = O(h ). Therefore, a 2-D weak form, as to be done in the next section.
the overall error of this integral, or the fourth term in δ E, is O(h4 ).
Case 2. Prescribed traction in the 3-D formulation 2
Note that since N depends on Z, from the 3-D condition ST N = q, in general
On ∂ q × [0, 2h], the traction q is specified as in (21)5 . De- we can not derive S(1)T N = q(1) or S(1)T ν = q(1) , which is true only for principal
note the first four coefficients of its Taylor expansion by q(i) (i = directions.
 Z. Song, H.-H. Dai / International Journal of Solids and Structures 97–98 (2016) 137–149 143

   
4. Associated weak formulations 2
(W2 − W3 )dA = h2 (S(1) − 2HS(0) + κS(0) )T ν · η
 3 ∂
In this section, we present the associated 2-D weak formula-
+ (S0T ν ) · x(2) ds (65)
tion for the previous shell system, for the ease of numerical cal- 
2
culations. Another direct consequence of such a weak form is that = h2 p(1) · η + (S0T ν ) · x(2) ds,
one can propose suitable boundary conditions for various practical 3 ∂
cases when the 3-D edge conditions are unclear. where
First of all, multiplying the 2-D shell Eq. (40) by ξ = δ x(0 ) and
integrating over  lead to 2  
W3 = − h2 (S(1) − 2HS(0) + κS(0) ) : ∇η + [∇ · (1S0 )] · x(2) .
     3
∇ · 1(S˜ + hκSˆ ) · ξ dA = − q˜ · ξ dA. (57) (66)
 
In summary, coupling (60) and (65) furnishes the following 2-D
To tackle the left-hand side, we need to utilize the 2-D divergence weak form
theorem. It is well-known that for a vector w = wα gα on the base 
surface , the divergence theorem dictates (W1 + W3 − q˜ · ξ )dA
  
 (67)
∇ · wdA = w · νds. 2 2  (1 )
(58) = p̄ · ξ − h p · η + (S0T ν ) · x(2) ds.
 ∂ ∂ 3
Applying this to w = 1Sξ , we readily get In the following, we will adapt this to two distinct types of bound-
   ary conditions, i.e., the previous cases in Section 3.3 based on 3-D
[∇ · (1S )] · ξ dA = ∇ · wdA − (1S ) : ∇ξ dA information, and some practical cases when the 3-D edge condi-

   tions are unclear.
= [(1S )T ν] · ξ ds − (1S ) : ∇ξ dA (59)
(1) 2-D weak form for the previous cases of boundary condi-
∂   
tions
= (S ν ) · ξ ds − S : ∇ξ dA.
T
On ∂ 0 , it is easy to deduce from (48) that ξ = δ x(0 ) = 0
∂ 
and η = δ x(1 ) = O(h ), which, together with (63), further im-
And, with the above S replaced by S˜ + hκSˆ , it follows from (57,59) ply that ∇ξ = O(h ) and S0 = O(h ). As a result, the boundary
that integral on ∂ 0 in (67) is of O(h3 ). In this context, we can
    readily replace ∂  by ∂ q in (67).
(W1 + W2 − q˜ · ξ )dA = S˜ T ν + h(κSˆ )T ν · ξ ds While on ∂ q , it follows from conditions (52,54) that S0T ν =

∂  (60) δ [S(0)T ν] = O(h2 ). Thus, the third term inside the boundary
= p̄ · ξ ds, integral can be neglected. Also, replacing p(1) by the condi-
∂
tion in (54) only causes a higher-order correction. Therefore,
where the terms in square brackets correspond to exactly p̄ by the 2-D weak form (67) reduces to
comparing (18,41) with (49,52) and neglecting the O(h3 ) remain-    
 2 2 (1 )
der. The two scalar functions are defined by W1 + W3 − q˜ · ξ dA = p̄ · ξ − h p · η ds
   ∂ q 3
8  
W1 = (1 − 2hH )S(0) + h(1 − hH )S(1) + hκSˆ : ∇ξ
3 = q0 · ξ − 2m · η ds,
∂ q
2 
+ h2 A2 [F(1) , F(1) ] : ∇ξ + A1 [F1(2) ] : ∇ξ , (61) (68)
3
2 2 1 (2 ) 2 where we have used conditions (52)1 and (54) in the last
W2 = h A [F2 ] : ∇ξ = h2 A1 [∇ξ ] : F2(2) ,
3 3 equality. If the two conditions in (52) are adopted, we
should replace m by a combination of q0 and q(0) .
where we have used the decomposition F(2 ) = F1(2 ) + F2(2 ) , see (2) 2-D weak form for some practical edge conditions
Appendix C for the explicit expressions. In a number of practical situations, one does not know the
In principle, the weak formulation for the fourth-order Eq. edge traction distribution (e.g., a pinned edge) or displace-
(40) should only involve the second-order derivatives of x(0) (espe- ment distribution (e.g., a clamped edge). In these cases, one
cially regarding the functional space in finite-element calculations). should resort to the weak formulation to propose the so-
However, the term F2(2 ) contains the third-order derivative of x(0) , called 2-D natural boundary conditions. To this end, we
which we intend to eliminate by conducting divergence theorem would like to recast the boundary terms in (67) as a com-
once more. To this end, we further recast W2 as bination of ξ and its normal derivative ξ , ν .
For convenience, we introduce a third-order (moment) ten-
2   
W2 = h2 S0 : ∇ x(2) + η · ∇ · (1S(1) − 2H 1S(0) + κS(0) ) , sor M through
3
(62) 2 2  (1 )
− h p · η + S0T ν · x(2)
3
where 2   (69)
= h2 A1 [B−1 p(1) + (A1 [x(2)  ν] )T n  n
3
η = −B−1 (A1 [∇ξ ] )T n, S0 = A1 [∇ξ + η  n]. (63) 
−x(2)  ν] : ∇ξ  (M[ν] ) : ∇ξ ,
T

In fact, it can be proved by (28) and (27)1 that

where p(1) also contains ν as in (65). Furthermore, we intro-
η = δ x ( 1 ) , S0 = δ S ( 0 ) . (64) duce the decomposition

Subsequently, integration by parts leads to ∇ξ = ξ,s  τ + ξ,ν  ν, (70)

144 Z. Song, H.-H. Dai / International Journal of Solids and Structures 97–98 (2016) 137–149

where subscripts s, ν preceded by a comma are tangen- From the assumptions of the present theory on geometry and cur-
tial and normal derivatives on ∂ . Substituting (69,70) into vature, we need the condition that 2h/A  (B − A )/A is small.
(67) followed by an integration by parts provides the 2-D Now we illustrate the procedure of writing down the shell
weak form equations for a practical spherical shell problem. First, the defor-
  mation gradient F follows from (9), and the components Fij are
W1 + W3 − q˜ · ξ dA obtained by expressing x in terms of the base gi . Once the strain

 (71) energy function is given, the nominal stress S is calculated ac-
= {p̄ − (M[ν]τ ),s } · ξ + {M[ν]ν} · ξ,ν ds, cording to (10). Often for isotropic materials, S can be easily ex-
∂
pressed in terms of base (eR , e , e ), and one can rewrite it in
where the terms before ξ and ξ , ν are respectively the gener- terms of the base gi to get the components Sij . The vector x(1) or
alized traction and moment on the edge. Here, the smooth- its components xi(1 ) are solved in terms of x(0) by (28). Note that
ness of ∂  is tacitly assumed. Based on (71), one can pro-
with known explicit expression of S, the recursive relations for x(2)
pose boundary conditions regarding the following conjugate
and x(3) will be obtained directly from (32)1, 2 instead of using the
variables
moduli A1 and A2 . With the above data in (77,78), the final shell
primary variables : x (0 ) , x,(ν0) , equations in (42) take the form
(72)
secondary variables : p̄ − (M[ν]τ ),s , M[ν ]ν . 1 13
S̄,αα1 + S̄11 cot  + (S̄ + S̄23 ) − cos  sin S̄22 = −q˜1 ,
One might specify the primary or secondary variables or A
1 (79)
a mixture of them. As in the variational approach, various S̄,αα2 + (2S̄12 + S̄21 ) cot  + (S̄13 + S̄23 ) = −q˜2 ,
boundary conditions such as clamped and simply-supported A
S̄,αα3 + S̄13 cot  − A(S̄11 + S̄22 sin ) = −q˜3 ,
2
conditions can be suitably proposed (see Dai and Song, 2014,
for details).
where
h αj
5. Examples S̄α j  S˜α j − Sˆ . (80)
A
In this section, we examine in detail two commonly used struc- In the following, we consider a concrete example with axisym-
tures, spherical shell and circular cylindrical shell, in order to facil- metric deformations, defined by
itate applications and to validate the present theory. For axisym-
metric deformations, some exact solutions are available in the lit- x ( X ) = r ( R )eR , A ≤ R ≤ B. (81)
erature, which will be utilized to compare with the approximate
As in Ogden (1997), we adopt the strain-energy function
solutions obtained from the present shell theory.
1
(I1 , I2 , I3 ) = (2ν − μ )I13 − ν I2 + μI3 + ν, (82)
5.1. The spherical shell 27
where√I1 , I2 , I3 are the three principal invariants of the stretch
Conventionally for a spherical shell, the spherical coordinates tensor FT F, and ν ≥ 0, μ ≤ 0 are two material constants. By
(R, , ) are used with the domains the 3-D field equation, the exact general solution found in Ogden
A ≤ R ≤ B, 0 ≤  ≤ π, 0 ≤  ≤ 2π , (73) (1997) takes the form

and they are related to rectangular cartesian coordinates by r (R ) = C1 R + C2 /R2 , (83)

X1 = R sin  cos , X2 = R sin  sin , X3 = R cos . (74) where the two constants C1 and C2 are to be determined by
boundary conditions on the inner and outer surfaces of the spher-
In the notations of current shell theory, we have the corresponding ical shell. On the two surfaces, we consider the dead-load case
relations
q− = q− eR , q+ = q+ eR . (84)
θ 1 = , θ 2 = , Z = R − A. (75)
As an illustrative example for comparison, we set μ = q− = 0, and
The base vectors on the base surface  (i.e., the inner surface of as a result the constants are explicitly given by
spherical shell) are calculated as  
A3 B3 q+ 1 B3 ( 2q+ + ν ) − A3 ν
g1 = A(cos  cos e1 + cos  sin e2 − sin e3 ) = Ae = A g , 2 1
C2 =  , C1 = 1+  . (85)
2 B3 − A3 ν 2 B3 − A3 ν
g2 = A sin (− sin e1 + cos e2 ) = A sin e = A sin g , 2 2 2

g3 = sin  cos e1 + sin  sin e2 + cos e3 = eR = g3 = n, We point out that the exact solution given by (83,85) is valid pro-
vided that q+ > 12 ν (A3 /B3 − 1 ), which corresponds to the condi-
(76) tion q > −3hν /A asymptotically as h/A tends to 0.
where ei (i = 1, 2, 3) are cartesian base vectors and e , e , eR By the present shell theory, we write r = r (Z ) (recall that R =
are the commonly adopted physical base vectors along the co- A + Z in (75)) and expand it as
γ β
ordinates. The associated nonzero constants for αβ , καβ , κα in
1 1
Appendix B are given by r ( Z ) = r0 + r1 Z + r2 Z 2 + r3 Z 3 + · · · , 0 ≤ Z ≤ 2h = B − A,
2 6
 =  = cot ,
2
12
2
21  = − cos  sin ,
1
22
(86)
1 (77)
κ11 = −A, κ22 = −A sin2 , κ11 = κ22 =− , where for clarity subscripts rather than superscripts are used in
A the unknown coefficients ri . The nonzero components of Fij are
where the second line implies given by

A2 r sin 
2
1 1 1 A2 r
κ = − 1, H = − , K = . (78) F11 = , F22 = , F33 = r (Z ), (87)
A A A2 A+Z A+Z
 Z. Song, H.-H. Dai / International Journal of Solids and Structures 97–98 (2016) 137–149 145

and the three principal stretches of the stretch tensor are see Appendix D. Therefore for this problem, the shell theory pro-
r duces pointwise O(h2 )-correct results for the deformation r(Z).
λ1 = λ2 = , λ3 = r (Z ). (88)
A+Z Remark. Although we only choose an illustrative example for q− =
The nonzero stress components are calculated as 0, the final conclusion that r(Z) is correct to O(h2 ) holds for the
general dead-load case (82,84). Another widely investigated case
∂ 1 1 1
S11 = S22 sin  = [ (2ν − μ )I12
2
= (e.g., Chung et al., 1986) is that the inner surface of the spherical
∂λ1 A2 9
A2 shell is subjected to a pressure P, i.e.,
− ν (λ2 + λ3 ) + μλ2 λ3 ], (89)
q− = P λ1 λ2 = P (r0 /A )2 . (97)
∂ 1
S33 = = (2ν − μ )I12 − ν (λ1 + λ2 ) + μλ1 λ2 ,
∂λ3 9 Although this is not a dead-load case, the above procedure can be
readily applied to this case with only slight modifications and also
where I1 = λ1 + λ2 + λ3 . Then, by expanding them with respect to
produces O(h2 )-correct results.
Z, we easily get the explicit expressions for components S(0)ij , S(1)ij ,
S(2)ij , which are given in Appendix D.
Now, we present the recursion relations and final equation for 5.2. The circular cylindrical shell
the general case (82,84). By the bottom traction condition (28), r1
can be solved as For this structure, the cylindrical coordinates (R, , X3 ) are re-
 
lated to the cartesian coordinates by
−2Ar0 (μ − 2ν ) − 3 A2 (μ − 2ν ) q− A2 − 2r0 ν A + r02 μ X1 = R cos , X2 = R sin , X3 = X3 , (98)
r1 = ,
A2 ( μ − 2ν ) with the domains
(90)
A ≤ R ≤ B, 0 ≤  ≤ 2π , 0 ≤ X3 ≤ L. (99)
and the other unreasonable solution is discarded. For the recursion
relation of r2 , the Eq. (32)1 reduces to According to the present shell theory, we identify them as

S(0)11 κ11 + S(0)22 κ22 + 2Hq− + S(1)33 = 0, (91) θ 1 = , θ 2 = X3 , Z = R − A. (100)

which yields that The base vectors on the inner surface of cylinder are utilized
 g1 = A(− sin e1 + cos e2 ) = Ae = A2 g1 ,
(μ − 2ν )r12 + 9q− A2 − 2r0 (r1 (μ − 2ν ) + 9ν )A + r02 (μ + 16ν )
r2 = − .
A2 (2r0 + Ar1 )(μ − 2ν ) g2 = e3 = g2 , (101)
(92) g3 = cos e1 + sin e2 = eR = g = n. 3

Similarly, one can derive the recursive formula for r3 (see σ , κ , κβ

Appendix D). The nontrivial shell Eq. (79)3 has the form
The associated nonzero constants for αβ αβ α in
Appendix B are given by
−2AS˜11 + 2hSˆ11 = −q˜3 , (93)
1
which, by substituting all the recursion relations, provides an alge- κ11 = −A, κ11 = − , (102)
A
braic equation for r0 , see (D.3) in Appendix D. For the illustrative
which implies H = −1/2A and K = 0. For this structure, the final
example with μ = q− = 0, it reduces to
 shell equations in (42) have the form
A2 q+ ( A + 2 h ) 2 − 4 h 3 A2 + 4 h2 ν r0
h 11 1 13 h 13
√  √ (94) S˜,αα1 − Sˆ + (S˜ − Sˆ ) = −q˜1 ,
+ 4 Ah 3 A 2 + 4 h 2 ν r 0 = 0 , A ,1 A A
α h 12
which yields that ˜ 2 ˆ
S,α − S,1 = −q˜ , 2 (103)
A
 2
h
1 A ( A + 2h )2 q+ S˜,αα3 − Sˆ,13 − AS˜11 + hSˆ11 = −q˜3 .
r0 = A 1 + 1+  . (95) A 1
4 4 h3 + 3 A2 h ν
In parallel with the previous subsection, we consider the ax-
Finally we make a comparison between the exact result by isymmetric deformations defined by
(83,85) and the approximate one from the shell theory. By setting r = r ( R ), θ = , x3 = λX3 , (104)
R = A in the solution (83,85), we get the exact r0 . Then, expanding
the exact and approximate r0 in terms of h provides the same first where (r, θ , x3 ) denote the current coordinates. And we still adopt
three terms the strain-energy function in (82). The exact general solution for
√  r(R) found in Ogden (1997) takes the form
1
 √   1 3
r0 = A 6 + q + 2 3 q + 3 + q 1 +  h 1
12 3 q+3 r (R ) = (C1 − λ )R + C2 /R. (105)
(96) 2
 √
q 2 ( q + 3 ) 3/2 − 3 ( q − 6 ) h2 where the integrating constants C1 and C2 are to be determined
+ + O ( h3 ),
9 A ( q + 3 ) 3/2 by the boundary conditions on the inner and outer surfaces of the
cylindrical shell. And, the parameter λ should be determined by
where we have utilized the scale q+ = qhν /A (note that this scale
the resultant axial load (the quantity q0 in (51) on the edge in our
already gives finite deformations since the thickness is small).
notations), but for simplicity we set λ = 1 in following illustrative
Clearly, the present r0 in (95) is correct to O(h2 ). From the recur-
examples. As the first example similar to the previous subsection,
sion relations, the other coefficients ri (i = 1, 2, 3) can be recovered.
we set
By comparing them with those from the exact solution (83), we
conclude that all ri (i = 1, 2, 3) have the correct terms up to O(h2 ), μ = 0, λ = 1 , q − = q− eR = 0, q+ = q+ eR . (106)
146 Z. Song, H.-H. Dai / International Journal of Solids and Structures 97–98 (2016) 137–149

The constants in the exact solution (105) are given by From the present shell theory, the recursion relations are sim-
 ilar to the previous example (see Appendix D), and the nontriv-
9 3 B2 (32q+ + 25ν ) − 25A2 ν A2 B2 q+ ial shell
 equation leads to the following algebraic equation for
C1 = +  , C2 =  . r˜0 = (ν − P )r0 + Aν ,
8 8 B2 − A2 ν B2 − A2 ν
 
(107) A2 (A + 2h )P − 4h A2 − hA + 2h2 ν r˜02
√ √  √
By the present shell theory, we also write r = r (Z ) and expand + 3 2 Ah A2 − hA + 2h2 ν (ν − P )r˜0 (115)
it as in (86). Accordingly, the nonzero components Fij and principle 
+ Aν −PA + 2hν A − 2h (P + ν )A + 4h (P + ν ) = 0,
3 2 2 3
stretches are given by
which gives
A2 r
F11 = A2 λ1 = , F22 = λ2 = λ, F33 = λ3 = r (Z ). (108) 1
A+Z r˜0 = −  
The nonzero stress components can be computed according to 2 A2 (A + 2h )P − 4h A2 − hA + 2h2 ν
 √ √  √
1 ∂ ∂ ∂ 3 2 Ah A2 − hA + 2h2 ν (ν − P )
S11 = , S22 = , S33 = . (109)
A2 ∂λ1 ∂λ2 ∂λ3   2 (116)
For the case (106), the explicit expressions of their expansion co- + 18Ah2 A2 − hA + 2h2 ν ( ν − P )2
efficients S(k)ij (k = 0, 1, 2) are given in Appendix D. With a similar  
− 4Aν A2 (A + 2h )P − 4h A2 − hA + 2h2 ν
procedure as the previous subsection, the recursive formulas for ri
(i = 1, 2, 3) can be obtained, see Appendix D. Finally, the nontrivial  1 / 2 
−PA3 + 2hν A2 − 2h2 (P + ν )A + 4h3 (P + ν ) .
shell Eq. (103)
 3 provides an algebraic equation for r0 or equiva-
lently r˜0 = r0 + A, By Taylor expansion, one obtains the three-term approximation
√  √  
− 4 2h A2 − hA + 2h2 ν r˜02 + 6 Ah A2 − hA + 2h2 ν r˜0 A(2 p + 3 p˜ + 1 ) 3 p 8 p − 37 p + (5 p − 2 ) p˜ + 74 h
2

√  (110) r0 = −
+ 2A q+ A3 + 2h(q+ + ν )A2 − 2h2 ν A + 4h3 ν = 0, ( p − 4) 2 ( p − 4 )3 p˜
ph2 
which yields the solution + ( p(70 p − 47 ) + 40 ) p˜3 (117)
⎛ ⎞ A( p − 4 )4 p˜3
√ + 3( p( p( p( p(63 p − 598 ) + 2389 ) − 4308 )
A 8q+ A3 + h(16q+ + 25ν )A2 − 25h2 ν A + 50h3 ν
r˜0 = √ ⎝   + 3⎠. (111) 
4 2 h A2 − hA + 2h2 ν + 2689 ) + 2600 ) + O(h3 ),
where we have used the scale P = phν /A and denoted p˜ =

By Taylor expansion, we get the three-term approximation for
2( p − 6 ) p + 25. Again this expansion (117) is the same as that
r0
 from the exact solution (105,114), and all the coefficients ri (i =
1   3
 3 1, 2, 3) are all correct to O(h2 ) (omitted for brevity). Therefore
r0 = A 4q + 3 8q + 25 + 1 + q 1 +  h for this internally pressurized case, the shell theory also produces
16 4 8q + 25
(112) pointwise O(h2 )-correct results for the deformation.

q (8q + 25 )3/2 − 30q + 75 h2 In summary, the illustrative examples have convinced us that
+ + O ( h3 ), the present theory is asymptotically correct, and can produce high-
4A(8q + 25 )3/2
order results in some cases. Besides the dead-load case, the theory
where we have also made use of the scale q+ = qhν /A. The expan- also has the flexibility to deal with the pressurized case, and is
sion (112) is exactly the same as that from the exact solution in expected to have wide applications.
(105,107). Also, one can easily verify that the other coefficients ri
(i = 1, 2, 3) are all correct to O(h2 ), therefore we conclude that the 6. Conclusions
shell theory produces an O(h2 )-correct result for the deformation.
We remark in passing that this conclusion is also true for the gen- In this paper, a consistent finite-strain shell theory for a gen-
eral dead-load case. eral shell of arbitrary geometry is developed, with no special re-
In Fourney and Stern (1968), the stability of an inflated and ex- strictions on external loadings (like scalings with thickness). It in-
tended cylindrical shell has been studied within finite deforma- cludes the previous plate theory (Dai and Song, 2014) as a special
tion theory for incompressible materials. The equilibrium states case, and is also consistent with the 3-D weak formulation with
are constructed by a finite membrane state and a superposed lin- an O(h4 ) error for each term. The derivation procedure is similar
earized deformation. Here, as the second example, we similarly to that of a previously-derived plate theory in that recursion rela-
consider the case subjected to an internal pressure but for above tions are derived from the 3-D differential system, but is not trivial
compressible material (82). We set considering the complicated effects due to the presence of curva-
tures. The present shell equation naturally preserves the local force
μ = 0, λ = 1, q− = q− eR = (Pr0 /A )eR , q+ = 0, (113)
balance. Proper 2-D boundary conditions and associated weak for-
where P represents the pressure inside the cylindrical shell. For mulations are proposed, accommodating either the reduced cases
this case, the constants C1 and C2 in (105) are calculated as from the 3-D edge conditions or various practical cases. The va-
3
 lidity of the shell theory is substantiated by the concrete exam-
C1 = −  3 ( P − ν )A2 − 3B 2 P + 3B 2 ν ples regarding axisymmetric deformations of spherical and circu-
8ν A2 + B 2 ( P − ν )
  lar cylindrical shells. In recent years, soft materials (e.g., gels, elas-

+ 9P 2 − 2ν P + 25ν 2 A4 − 2B2 P 2 − 26ν P + 25ν 2 A2 + 25B4 (P − ν )2 , tomers and polymers) have found broad applications. A basic char-
acteristic of those materials is their ability to undergo finite-strain
A2 B2 ((C1 − 1 )P + 2q+ )
C2 = −  . deformations. The present shell theory is established in the frame-
2 ν A2 + B 2 ( P − ν ) work of finite-strains with a general strain energy function, which
(114) incorporates bending and stretching simultaneously, thus it may
 Z. Song, H.-H. Dai / International Journal of Solids and Structures 97–98 (2016) 137–149 147

provide a general mathematical framework for studying deforma- On the base surface, the equations of Guass and Weingarten read
tions of thin structures composed of soft materials.

∂ gα σ g + κ n, with  σ = ∂ gα · gσ ,
Acknowledgments = αβ σ αβ αβ
∂θ β ∂θ β (B.2)
∂n β β
The work described in this paper was supported by a GRF grant = −καβ g = −κα gβ ,
(Project No.: CityU 11303015) from the Research Grants Council of ∂θ α
Hong Kong SAR, China and a grant from the National Nature Sci-
ence Foundation of China (Project No.: 11572272). which can be universally written as

Appendix A. Simplification of the shell equation ∂ gi β β

= ipβ g p , with 3α = −κα , αβ
3
= καβ , 33β = 0.
∂θ β
Here we intend to simplify the vector shell Eq. (39) to the (B.3)
form (40), by utilizing (28,32), Cayley–Hamilton (C-H) formula
κ2 = 2H κ − K1 and the following identities Then, the covariant derivative for a tensor under base gi is defined
∇ · S(i) = ∇ · (1S(i) ) − 2HS(i)T n, as follows

[κT gα ] · S,(αi ) = [κT gα ] · (1S(i ) ),α − (gα · κ2 gα )S(i )T n βj

1(1S ),α = S,α gβ  g j + Sβ j βα
δ g  g + Sβ j  p g  g
δ
(A.1) j jα β p

= [κ T
gα ] · (1S(i ) ) − ( 4H − 2K )S
2 ( i )T
n,
 
,α βj β
= S,α + Sσ j σ α + Sβ k kjα gβ  g j (B.4)
C ∇ · (1S(i ) ) = ∇ · (C1S(i ) ) − S(i )T ∇ C,
βj
 S ;α g β  g j ,
for i = 0, 1, 2 and any scalar function C. More precisely, the left-
hand side of (39) is calculated as
which immediately implies
 8

C−H
LHS = 1 − 4hH + h2 K ∇ · S(0) + h(1 − 4hH )∇ · S(1)
3 ∇ · (1S ) =S;ααj g j , [κT gα ] · (1S ),α = κδα gδ · (1S ),α = κδα S;δαj g j .
2 2 4
+ h ∇ · S + h(1 − hH )[κT gα ] · S,(α0)
(2 ) (B.5)
3 3
4 With S replaced by S˜ + hκSˆ in (B.5)1 , one obtains the component
+ h2 [κT gα ] · S,(α1)
3 form of the first term in the shell Eq. (40).
 8
  8

(A.1 )1 ,(32 )2 Similarly, we define the covariant derivative for a tensor with
= 1 − 4hH + h2 K ∇ · S(0) + h 1 − hH ∇ · S(1) β
3 3 mixed base, like κ = κγ gβ  gγ , through
2 2 4 2 T α
+ h ∇ · (1S ) + h[κ g ] · S,α + h [κ g ] · S,(α1)
(2 ) T α (0 )
β β β γ

3

3 1(κ ),α = κγ ,α gβ  gγ + κγ βα
δ g  gγ − κ  g  gδ
δ γ αδ β
(A.1 )1 ,(32 )1 8 2
= C0 + h (K − 2H ) ∇ · S + hC1 ∇ · (1S(1) )
2 (0 ) β γ
+ κγ κα gβ  n
3
  (B.6)
2 2 4 β γ
+ h ∇ · (1S(2) ) + h[κT gα ] · S,(α0) + h2 [κT gα ] · S,(α1) = κγβ,α + κγδ δα
β β δ
− κδ αγ gβ  gγ + κγ κα gβ  n
3 3
β β γ
(A.1 )2
= C0 ∇ · S + hC1 ∇ · (1S ) + h ∇ · (1S(2) )
(0 ) (1 ) 2 2  κγ ;α gβ  gγ + κγ κα gβ  n,
3
T α (0 ) 4 2 T α where we have used ∂ gα /∂θ β = −βσ
α gσ + κ α n. Then, it follows
+ h[κ g ] · S,α + h [κ g ] · (1S(1) ),α β
3 that
(A.1 )1,2 ,(28 ) 2
= C0 ∇ · (1S ) + hC1 ∇ · (1S(1) ) + h2 ∇ · (1S(2) )
(0 )
3 ∇ · κ =κγα;α gγ + κγα καγ n. (B.7)
T α 4 2 T α
+ h[κ g ] · (1S ),α + h [κ g ] · (1S(1) ),α
(0 )
3 The Mainardi–Codazzi equations can be written as (Steigmann,
+ [2HC0 + h(4H 2 − 2K )]q− 2012)
(A.1 )3
= ∇ · (1S˜ ) + 2hSˆ T ∇ H + h[κT gα ] · (1Sˆ ),α + (2H − 2hK )q− κ11;2 = κ21;1 , κ12;2 = κ22;1 , (B.8)
=∇ · (1S˜ + hκSˆ ) + h(1Sˆ )T [2∇ H − ∇ · κ] + (2H − 2hK )q−
(B.9 )
then one can easily check
= ∇ · (1S˜ + hκSˆ ) + (2H − 2hK )q− ,
(A.2) 1[2∇ H − ∇ · κ] = [(2H ),β − κβα;α ]gβ = [(2H ),β − καα;β ]gβ = 0.
where (B.9) can be found in Appendix B, C0 = 1 − 2hH, C1 = 1 − (B.9)
8 ˜ ˆ
3 hH, and the two stresses S and S are defined in (41). Then, the
Eq. (40) follows by rearranging the terms.

Appendix C. Decomposition of F(2)

Appendix B. Covariant derivatives
It follows from (25,35) that
The stress tensor S has the decomposition like

S =Si j gi  g j = 1S + n  ST n = gα  ST gα + n  ST n F(2) =2(∇ x(1) )κ + 2(∇ x(0) )κ2 + ∇ x(2) + x(3)  n,

    (B.1) (C.1)
= Sαβ gα  gβ + Sα 3 gα  n + S3α n  gα + S33 n  n . x(3) = − B−1 f(3) .
148 Z. Song, H.-H. Dai / International Journal of Solids and Structures 97–98 (2016) 137–149

To get a decomposition of F(2) , we first decompose f(3) in (36) as The recursion relation for r3 is given by
follows 1 
 r3 = − 2r22 (μ − 2ν )A4
f (3 ) = κT gα · S,(α0) + ∇ · S(1) 2A3 (2r0 + Ar1 )(μ − 2ν )
 (D.2)
 T − 12r1 r2 (μ − 2ν )A3 + 45q− − 2r12 (μ − 2ν ) A2
2
2 
+ A1 (∇ x(i) )κ2−i + A2 [F ( 1 ) , F ( 1 ) ] n + 4r0 (4r1 (μ − 2ν ) − 9ν )A + 2r02 (32ν − 7μ ) .
i!
i=0
The algebraic equation for r0 for the general dead-load case takes
=∇ · (κS(0) ) − (∇ · κ ) · S(0) + ∇ · (1S(1) ) − 2HS(1)T n the form
 T  
2 (q− + q+ )t (μ − 2ν )A5 + 4h(μ − 2ν ) − 3q− ν A4 + 3r0 2ν 2 + q− μ A3
2   
+ A1 (∇ x(i) )κ2−i + A2 [F ( 1 ) , F ( 1 ) ] n (C.2) + −9μν r02 + q− t + q+ t A2 + 3 r03 μ2 − r0 t ν A + 3r02 t μ A2
i!
i=0  
  + h2 − 72q− r0 μ(μ − 2ν )A4 − 4 2q− t (4μ + ν )
= ∇ · [κS(0) + 1S(1) − 2H 1S(0) ] + (A1 [∇ x(2) ] )T n 
 −(μ − 2ν ) 36μν r02 + q+ t A3
T
+ A1 [2(∇ x(1) )κ + 2(∇ x(0) )κ2 ] + A2 [F(1) , F(1) ] n 2 
− 72r0 μ r0 μ(μ − 2ν ) − t ν A2 − 72r02 t μ(μ − ν )A
  
− (∇ · κ ) · S(0) − 2HS(1)T n + ∇ · (2H 1S(0) ) + 4h3 − 7q− (μ − 2ν )ν A4 + r0 (μ − 2ν ) 8ν 2 + 79q− μ A3

+ 2 q− t (19μ − 2ν ) − 78r02 μ(μ − 2ν )ν A2
f2(3) + f1(3) , 
+ 4r0 19r02 (μ − 2ν )μ2 + t ν (2ν − 19μ ) A
where (A.1)1 has been used in the second equality and the two 
+ 4r02 t μ(19μ − 20ν ) = 0,
braces define respectively f2(3 ) and f1(3 ) . In the above derivation, the
(D.3)
term ∇ · (2H1S(0) ) is added and subtracted to get the desired form.
Then we define where
 
(2 ) (2 ) A2 (μ − 2ν ) q− A2 − 2r0 ν A + r02 μ .
F(2) =F1 + F2 , t= (D.4)
(2 )
 −1 (3 )

F1 =2(∇ x(1) )κ + 2(∇ x(0) )κ2 − B f1  n, (C.3) The asymptotic expansions for ri (i = 1, 2, 3) from both the ex-
(2 ) (2 )
 (3 )
 act solution and present shell theory are given by
F2 =∇ x − B −1
f2  n.
√ 
1

√  qh
 3
The purpose of such a decomposition is to make sure that all the r1 = 3−q+ 3 q+3 +  −2
third-order derivatives of x(0) (after using recursion relations) are 6 3A q+3
kept in the part F2(2 ) .   √  √ 
q 4 q + 3q + 3q + 12 q + 3 − 6 3 h2
− + O ( h3 ),
9 A2 ( q + 3 ) 3/2
Appendix D. Expressions in the examples
q 2qh 4qh2
r2 = + 2 + + O ( h3 ),
D1. The spherical shell 2A A 3A3
2q 8qh 16qh2
The expanded coefficients of nonzero stresses in (89) are given
r3 = − 2 − 3 − + O ( h3 ).
A A 3A4
by (D.5)

−r1 (r1 (μ − 2ν ) + 9ν )A2 + r0 (5r1 μ + 8r1 ν − 9ν )A − 4r02 (μ − 2ν )

S(0)11 = , D2. The circular cylindrical shell
9 A4
1 
S(1)11 = 5 (5μ + 8ν )r12 + (−2Ar2 μ + 4Ar2 ν − 9ν )r1 − 9Ar2 ν A2 The expanded coefficients of nonzero stresses in (109) are given
9A
 by
+ r0 (−13r1 μ + 5Ar2 μ + 8r1 ν + 8Ar2 ν + 9ν )A + 8r02 (μ − 2ν ) , 
1   (0 )11
(r1 + 1 )(2r1 − 7 )A2 + 4r0 (r1 + 1 )A + 2r02 ν
S(2)11 = 6 (−18μr12 + −2r3 (μ − 2ν )A2 + 3r2 (5μ + 8ν )A + 18ν r1 S = ,
9A 9A4
  
+ A A −2(μ − 2ν )r22 − 9r3 ν − 9r2 ν )A2 − 24r02 (μ − 2ν ) A2 (4r0 + A(4r1 − 5 ))r2 − 4(r0 − Ar1 )(r1 A + A + r0 ) ν
  S(1)11 = ,
+ r0 r3 (5μ + 8ν )A2 − 18r2 μA + 6r1 (7μ − 8ν ) − 18ν A , 9A5
ν 
−9μr02 + 18Aν r0 + (2r0 + Ar1 )2 (μ − 2ν ) S(2)11 = 4r22 + (4r1 − 5 )r3 A4 + 4(3r1 r2 + r2 + r0 r3 )A3
S(0)33 = − , 9A6

9 A2
−4(r1 (r1 + 2 ) + r0 r2 )A2 − 8r0 (r1 − 1 )A + 12r02 ,
2 
S(1)33 =− r1 (2r1 (μ − 2ν ) + Ar2 (μ − 2ν ) + 9ν )A2
9 A3 (r0 + A(r1 − 2 ))(2r0 + A(2r1 − 1 ))ν
 S(0)22 = ,
+ r0 (2Ar2 (μ − 2ν ) − 9ν − r1 (7μ + 4ν ))A + r02 (5μ + 8ν ) , 9A2
2   (4r0 + A(4r1 − 5 ))(A(r1 + Ar2 ) − r0 )ν
S(2)33 = − (−9μr12 + r3 (μ − 2ν )A2 + 6r2 (μ − 2ν )A − 18ν r1 S(1)22 = ,
9 A4 9A3
+ Ar2 (Ar2 (μ − 2ν ) + 9ν ))A2 − 3r02 (5μ + 8ν )
ν 
S(2)22 = 4r22 + (4r1 − 5 )r3 A4 + ((12r1 − 5 )r2 + 4r0 r3 )A3
  9A4
+ r0 2r3 (μ − 2ν )A − 9r2 μA + 18ν + 24r1 (μ + ν ) A ,
2 
−2(r1 (2r1 − 5 ) + 2r0 r2 )A2 − 2r0 (4r1 + 5 )A + 12r02 ,
(D.1) 
(0 )33
(2r1 (r1 + 2 ) − 7 )A2 + r0 (4r1 − 5 )A + 2r02 ν
and S(k )22 = S(k )11 / sin2 , (k = 0, 1, 2).
S = ,
9A2
 Z. Song, H.-H. Dai / International Journal of Solids and Structures 97–98 (2016) 137–149 149


(1 )33
4(r1 + 1 )r2 A3 + (r1 (4r1 − 5 ) + 4r0 r2 )A2 + 5r0 A − 4r02 ν Hamdouni, A., Millet, O., 2003a. Classification of thin shell models deduced from the
S = , nonlinear three-dimensional elasticity. part i: the shallow shells. Arch. Mech. 55
9A3 (2), 135–176.
(2 )33 ν   Hamdouni, A., Millet, O., 2003b. Classification of thin shell models deduced from
S = 4 + r1 r3 + r3 A + ((12r1 − 5 )r2 + 4r0 r3 )A3
r22 4
the nonlinear three-dimensional elasticity. part ii: the strongly curved shells.
9A4
 Arch. Mech. 55 (2), 177–220.
−2(r1 (2r1 − 5 ) + 2r0 r2 )A2 − 2r0 (4r1 + 5 )A + 12r02 . (D.6) Kirchhoff, G., 1850. Über das gleichgewicht und die bewegung einer elastischen
scheibe. Journal für die Reine und Angewandte Mathematik 40, 51–88.
The recursion relations for ri (i = 1, 2, 3) are given by Koiter, W.T., 1960. A consistent first approximation in the general theory of
√  thin elastic shells. Theory of Thin Elastic Shells. North-Holland, Amsterdam,
−2ν A2 − 2r0 ν A + 3 2 A3 ν (r0 ν + A(ν − q− )) pp. 12–33.
r1 = , Koiter, W.T., 1966. On the nonlinear theory of thin elastic shells. Koninklijke Neder-
2A2 ν
 −  2 landse Akademie van Wetenschappen Proc. Ser. B 69 (1), 1–54.
9q − 2r1 + 7 ν A2 + r0 (4r1 − 1 )ν A + 6r02 ν Koiter, W.T., Simmonds, J.G., 1973. Foundations of Shell Theory. Springer.
r2 = , Le Dret, H., Raoult, A., 1996. The membrane shell model in nonlinear elasticity: a
4A2 ( r1 A + A + r0 )ν variational asymptotic derivation. J. Nonlinear Sci. 6 (1), 59–84.
1  (D.7) Libai, A., Simmonds, J.G., 1998. The Nonlinear Theory of Elastic Shells. Cambridge
r3 = 3 − 4r22 ν A4 − 4(3r1 + 1 )r2 ν A3 University Press.
4A ( r1 A + A + r0 )ν Love, A.E.H., 1888. The small free vibrations and deformation of a thin elastic shell.
 Philos. Trans. R. Soc. London. A 179, 491–546.
+ 2r12 + 4r1 + 4r0 r2 + 7 ν − 9q− A2 Meroueh, K., 1986. On a formulation of a nonlinear theory of plates and shells with
 applications. Comput. Struct. 24 (5), 691–705.
+ r0 (4r1 − 3 )ν A − 14r02 ν . Naghdi, P., 1972. The Theory of Plates and Shells. Handbuch der Physik, Springer,
vol. VIa/2, 425–640.
where q− = 0 and q− = P r0 /A respectively for the two illustrative Ogden, R.W., 1997. Non-linear Elastic Deformations. Courier Corporation.
examples in Section 5.2. Pietraszkiewicz, W., 1979. Finite rotations and Lagrangean Description in the Non–
Linear Theory of Shells. Polish Scientific Publishers Warszawa-Poznan.
Pietraszkiewicz, W., Gorski, J., 2013. Shell Structures: Theory and Applications,
References
Vol. 3. CRC Press.
Podio-Guidugli, P., 1990. Constraint and scaling methods to derive shell theory from
Altenbach, J., Altenbach, H., Eremeyev, V.A., 2010. On generalized cosserat-type the- three dimensional elasticity. Riv. Mat. Univ. Parma 16, 78–83.
ories of plates and shells: a short review and bibliography. Arch. Appl. Mech. 80 Reddy, J., Arciniega, R., 2004. Shear deformation plate and shell theories: from
(1), 73–92. stavsky to present. Mech. Adv. Mater. Struct. 11 (6), 535–582.
Chung, D.-T., Horgan, C., Abeyaratne, R., 1986. The finite deformation of internally Reddy, J., Liu, C., 1985. A higher-order shear deformation theory of laminated elastic
pressurized hollow cylinders and spheres for a class of compressible elastic ma- shells. Int. J. Eng. Sci. 23 (3), 319–330.
terials. Int. J. Solids Struct. 22 (12), 1557–1570. Reissner, E., 1974. Linear and nonlinear theory of shells. Thin-Shell Structures: The-
Ciarlet, P.G., 20 0 0. Mathematical Elasticity, Volume III: Theory of shells, vol. 3. Else- ory, Experiment, and Design, Prentice-Hall, 29–44.
vier. Sanders, J.L., 1959. An Improved First-Approximation Theory for Thin Shells. NASA
Ciarlet, P.G., 2005. An introduction to differential geometry with applications to Report 24.
elasticity. J. Elast. 78 (1–3), 1–215. Song, Z., Dai, H.-H., 2016. On a consistent dynamic finite-strain plate theory and its
Ciarlet, P.G., Destuynder, P., 1979a. A justification of a nonlinear model in plate the- linearization. J. Elast. 1–35. doi:10.1007/s10659- 016- 9575- 4.
ory. Comput. Methods Appl. Mech. Eng. 17, 227–258. Steigmann, D.J., 2008. Two-dimensional models for the combined bending and
Ciarlet, P.G., Destuynder, P., 1979b. Justification of the 2-dimensional linear plate stretching of plates and shells based on three-dimensional linear elasticity. Int.
model. J. Mécanique 18 (2), 315–344. J. Eng. Sci. 46 (7), 654–676.
Ciarlet, P.G., Mardare, C., 2008. An introduction to shell theory. Differ. Geom. 9, Steigmann, D.J., 2010. Recent developments in the theory of nonlinearly elastic
94–184. plates and shells. In: Pietraszkiewicz, Kreja (Eds.), Shell structures: Theory and
Dai, H.-H., Song, Z., 2014. On a consistent finite-strain plate theory based on three- Applications, vol. 2. CRC Press Taylor and Francis Group, pp. 19–23.
-dimensional energy principle. Proc. R. Soc. London A. 470 (2171), 20140494. Steigmann, D.J., 2012. Extension of koiter’s linear shell theory to materials exhibit-
Destuynder, P., 1980. Sur une Justification des Modeles de Plaques et de Coques par ing arbitrary symmetry. Int. J. Eng. Sci. 51, 216–232.
les Methodes Asymptotiques. Universite Pierre et Marie Curie, Paris Ph.D. thesis. Steigmann, D.J., 2013. Koiter’s shell theory from the perspective of three-dimen-
Donnell, L.H., 1976. Beams, Plates and Shells. McGraw-Hill Companies. sional nonlinear elasticity. J. Elast. 111 (1), 91–107.
Fourney, W., Stern, M., 1968. Stability of a finitely deformed thin cylindrical shell. Timoshenko, S., Woinowsky-Krieger, S., Woinowsky-Krieger, S., 1959. Theory of
Int. J. Eng. Sci. 6 (11), 661–683. Plates and Shells, vol. 2. McGraw-hill New York.
Friesecke, G., James, R.D., Mora, M.G., Müller, S., 2003. Derivation of nonlinear bend- Ventsel, E., Krauthammer, T., 2001. Thin Plates and Shells: Theory, Analysis, and Ap-
ing theory for shells from three-dimensional nonlinear elasticity by gamma– plications. CRC Press.
convergence. Comptes Rendus Mathematique 336 (8), 697–702. Wan, F., Weinitschke, H., 1988. On shells of revolution with the Love-Kirchhoff hy-
Gol’denveizer, A., 1963. Derivation of an approximate theory of shells by means of potheses. J. Eng. Math. 22 (4), 285–334.
asymptotic integration of the equations of the theory of elasticity. J. Appl. Math. Wang, J., Song, Z., Dai, H.-H., 2016. On a consistent finite-strain plate theory for
Mech. 27 (4), 903–924. incompressible hyperelastic materials. Int. J. Solids Struct. 78, 101–109.