Applications in Engineering Science 6 (2021) 100046

Contents lists available at ScienceDirect

Applications in Engineering Science

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

New refined model for curved linear anisotropic rods with circular cross
section
Erick Pruchnicki a,∗, Xiaoyi Chen b, Hui-Hui Dai c,1
a
Université et Unité de Mécanique de Lille EA 7512, F 59655 Villeneuve d’Ascq, France
b
Division of Science and Technology, BNU-HKBU United International College, Zhuhai, China
c
Department of Mathematics and Department of Materials Science and Engineering, 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

Keywords: An asymptotic reduction method is introduced to construct a curved rod theory for a general anisotropic linearized
Curved rod theory elastic material. For the sake of simplicity, the cross section is assumed to be circular. The starting point is Taylor
Reduction method expansions about the mean-line in curvilinear coordinates, and the goal is to eliminate the two spatial variables
Anisotropic linearized elasticity
in the cross section in a pointwise manner in order to obtain a closed system for the displacement coefficients.
Rod variational formulation
We achieve this by using a Fourier series for the lateral traction condition together with the use of cylindrical
Fourier series
Classical rod theories coordinates in the cross section and by considering exact tridimensional equilibrium equation. We get a closed
differential system of ten vector unknowns, and after a reduction process we obtain a differential system of
the vector of the mean line displacement and twist angle. Six boundary conditions at each edge are obtained
from the edge term in the tridimensional virtual work principle, and a unidimensional virtual work principle is
also deduced from the weak forms of the rod equations. Through one example, we show that our theory gives
more accurate results than the ones of both classical Euler-Bernoulli rod theory and Timoshenko rod theory. The
displacement field is computed for two types of material symmetry : isotropy and transverse isotropy.

1. Introduction
Rods are very important engineering structures. For straight or
curved rods, the dimension of the cross section is much smaller than
the third one, its length. Due to this relative smallness, one may model
the behaviours of these thin structures by one-dimensional rod the-
ory through certain dimensional reduction processes. The most pop-
ular approach is to introduce some kinematic assumptions such that
the displacement has certain particular forms. Classical rod theories,
such as Euler-Bernoulli rod theory, Timoshenko rod theory and Reddy’s
third-order rod theory Reddy (2006) are obtained in this way. Due
to its simplicity, Carrera and Giunta (2010) and Carrera and Petrolo
(2012) use various forms of this approach. But if one substitutes the
assumed displacement form into the equilibrium equation and traction
condition, some contradictions will arise. Another approach is to ex-
pand the displacement in terms of a series of basis functions and to
truncate the potential energy Pruchnicki and Dai (2019). It is interest-
ing to note that this method gives an error estimation between the ex-
act three dimensional plate problem and the approximate bidimensional
one (Schneider and Kienzler 2020). Only for isotropic material, the num-
ber of main unknowns can be reduced by using the pseudo-reduction
method (Schneider et al., 2014) and then a plate (symmetric transverse
loading) model with three (due to considered symmetry) boundary con-
ditions (for transverse shear, twisting moment and bending moment)
can be constructed.
Another theory is introduced by Ciarlet and Destuynder (1979) for
plate through use of asymptotic expansions in variational formulation.
This theory was adapted by Bermudez and Viano (1984) for rods;
for further works on rods in this direction see Trabucho and Viano
(1996). Another approach is through the use of asymptotic expan-
sion in the minimization of potential energy. The method was first
introduced by Pantz (2000) for homogeneous plates, and by Meunier
(2008) and Marigo and Meunier (2006) for rods and extended by
Pruchnicki(2009); (2011) for heterogeneous plates. This problem is also
addressed through the use of a rigorous Gamma convergence tool, which
generates one-dimensional models when the small transverse dimen-
sion goes to zero (Giovanna Mora and Müller, 2003). Different scal-
ings lead to a hierarchy of different models, but those approaches do
not yield a single one dimensional model which contains extension,
bending and torsion effects together. Also, a priori scalings adopted,
assumes that the loads depend on the thickness but in reality applied
loads are external and thus are not linked to the geometry of the thin

∗
Corresponding author.
E-mail addresses: erick.pruchnicki@gmail.com (E. Pruchnicki), xiaoyichen@uic.edu.cn (X. Chen), mahhdai@cityu.edu.hk (H.-H. Dai).
1
Dedicaced to the memory of Professor Hui Hui Dai

https://doi.org/10.1016/j.apples.2021.100046
Received 27 December 2020; Received in revised form 24 March 2021; Accepted 7 April 2021
Available online 14 April 2021
2666-4968/© 2021 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
E. Pruchnicki, X. Chen and H.-H. Dai Applications in Engineering Science 6 (2021) 100046

structure. Moreover it is difficult to apply this method to obtain error where O is the chosen origin. If L is the length of this curve in the refer-
estimates. ence configuration, we have s ∈ [0, 𝐿]. The orthonormal Frenet frame
To overcome these drawbacks, Dai and Song (2014); Song and Dai at a point G is denoted by (𝐚1 , 𝐚2 , 𝐚3 ):
(2016 a,b); Wang et al. (2019); Yu et al. (2020) introduced for plate
𝐚3 = 𝑟, 𝑠 , (1)
a novel method of dimension reduction deduced from tridimensional
equilibrium equation and boundary condition on the upper and lower
𝐚3, 𝑠
face of the plate. This approach was used in Chen et al. (2018) to ob- 𝐚1 = ,
tain a plane-stress rod model for a linearized isotropic elastic material 𝑘
with pointwise error estimates. In this way Pruchnicki (2018) obtains a
unidimensional model for a rod with rectangular cross section. 𝐚2 = 𝐚3 ∧ 𝐚1 .
Obtention of a rod theory is more complicated than plate or shell
where 𝐚1 is the unit vector tangent to Γ ̃𝑐 and 𝑘(𝑠) = |𝐚 (𝑠)| is the cur-
ones. For rod with circular cross section, we present a one dimensional 3, 𝑠
vature at point G.
model as already done in a previous work for straight rod (Chen et al.
The variation of (𝐚1 , 𝐚2 , 𝐚3 ) is described the Frenet-Serret formulae:
(2020)).
The paper is arranged as follows. We first give specific used nota-
tion in Section 2. Then in Section 3, we define the mean line of the rod 𝐚1, 𝑠 = −𝑘 𝐚3 + 𝜏 𝐚2 , (2)
(Subsection 3.1) and for the rod the curvilinear coordinate and covari-
ant and contravariant basis (Subsection 3.2). Then, we start the deriva- 𝐚2, 𝑠 = −𝜏 𝐚1 , (3)
tion process. In Section 4, we give the three-dimensional equilibrium
equations together with boundary conditions in both rectangular and
curvilinear coordinates. In Section 5, we derive a system of differential 𝐚3,𝑠 = 𝑘 𝐚1 . (4)
equations for the rod. In order to do, we begin with Taylor expansions
where 𝜏(𝑠) = |𝐚2,𝑠 (𝑠)| denotes the torsion of the curve Γ̃ 𝑐 at point G.
in curvilinear coordinates for displacement, strain and Cauchy stress,
together with the relations among their coefficients. Next the Fourier se-
3.2. Geometric and kinematic formulae for rod
ries is used to represent the given traction on the lateral surface, and the
expansion for the Cauchy stress is represented in terms of polar coordi-
In a neighbourhood of Γ̃ 𝑐 , the geometric points will be defined by
nates. Then, we carefully truncate the traction conditions on the lateral
the position vector X depending on the curvilinear coordinates 𝑦 =
boundary in order to keep bending and twisting terms. After consider-
(𝑦1 , 𝑦2 , 𝑦3 ) (Sanchez-Hubert and Sanchez-Palencia 1999):
ing three dimensional equilibrium equation in a pointwise manner, a
( ) ( ) ( )
calculation leads to a closed ten equations with ten unknown vectors. 𝐗(𝐲) = 𝐫 𝑦3 = 𝑠 + 𝑦1 𝐚1 𝑦3 = 𝑠 + 𝑦2 𝐚2 𝑦3 = 𝑠 , (5)
In Section 6 a number of refinements based on truncation of equations
The mapping X is injective for sufficiently small 𝑦1 , 𝑦2 .
at the same order of magnitude, leads to asymptotically-consistent four
For the sake of simplicity, we assume that the cross section Σ (curvi-
scalar rod equations with four unknowns (the three components of the
linear parametrization) of the rod is circular with radius𝑅0 :
mean line displacement and the averaged twist angle). In Section 7, rod
{ ( ) }
boundary conditions are derived from the edge term of the tridimen- Σ = 𝐲′ = 𝑦1 , 𝑦2 , 𝑦21 + 𝑦22 < 𝑅20 .
sional virtual work principle and then the rod virtual work principle is
Indeed the rod is defined by the set of points such that:
deduced from the weak forms of the rod equations. In the appendix,
{ ] }
we give an example for a straight rod with two clamped ends and 𝐵̃ = 𝐗(𝐁) = 𝑃 , 𝑂𝑃 (𝐲) = 𝐗(𝐲) 𝑠𝑢𝑐ℎ 𝑡ℎ𝑎𝑡 𝐲′ ∈ Σ 𝑎𝑛𝑑 𝑦3 ∈ 0, 𝐿[ ,
loaded by a uniform line force at the top of lateral surface along
in which 𝐵 = Σ ×]0, 𝐿[ .
the negative y1-axis. We compare the obtained system of differ-
Next we record a number of formulae pertaining to the geometry
ential equations with Euler-Bernoulli rod theory and Timoshenko
of the curved rod and to the kinematics of the material in a three-
rod theory and we see that our theory gives refined results. Finally,
dimensional neighbourhood of the mean line.
the displacement field is computed with Mathematica software for
At each point of the rod, we define the covariant basis 𝐠𝑖 = 𝑋, 𝑖 as
two types of material symmetry: isotropy and transverse isotropy.
In Section 8, we conclude and give perspective. well as the contravariant basis 𝐠𝑖 (𝐠𝑖 . 𝐠𝑗 = 𝛿𝑖𝑗 ,where 𝛿𝑖𝑗 are the Kronecker
symbols).
2. Notation From formulae (4)-(5), we easily obtain:
𝐠1 = 𝐚1 , (6)
Boldface letters represent vector-valued functions, tensors or spaces.
Let (𝐞1 , 𝐞2 , 𝐞3 ) denotes an orthonormal basis of the space ℝ3 . (𝐚1 , 𝐚2 , 𝐚3 )
is the local orthonormal covariant basis. Also, we adopt Einstein con- 𝐠2 = 𝐚2 , (7)
vention summation for repeated indices. The Euclidean scalar product,
vector product and tensorial product of vectors 𝐚, 𝐛are denoted by 𝐚 . 𝐛, ( )
𝐠3 = 1 − 𝑘 𝑦1 𝐚3 − 𝜏 𝑦2 𝐚1 + 𝜏 𝑦1 𝐚2 , (8)
𝐚 ∧ 𝐛and 𝐚 ⊗ 𝐛 respectively. The double dot product of two tensors is
A:B. The latin index take value 1, 2 and 3. Greek index take value 1, 2.
𝜏 𝑦2
𝐵𝑖𝑗 (respectively 𝐵̃𝑖𝑗 ) is the component of a second order tensor B with 𝐠1 = 𝐚1 + 𝐚 , (9)
1 − 𝑘 𝑦1 3
respect to 𝐞𝑖 ⊗ 𝐞𝑗 (respectively 𝐚𝑖 ⊗ 𝐚𝑗 ), this notation also holds for a
general k-th order tensor. 𝜏 𝑦1
𝐠2 = 𝐚2 − 𝐚 . (10)
1 − 𝑘 𝑦1 3
3. The curved rod
1
𝐠3 = 𝐚 , (11)
̃𝑐 of the rod
3.1. The mean line Γ 1 − 𝑘 𝑦1 3
We note that 𝐠𝑖 (0, 0, 𝑠) = 𝐠𝑖 (0, 0, 𝑠) = 𝐚𝑖 .
We begin by defining a general (not necessary plane) curve Γ ̃𝑐
The determinant of the metric tensor of components 𝑔𝑖𝑗 = 𝐠𝑖 . 𝐠𝑗 is:
parametrized by its curvilinear measures 𝑠in the reference configuration
( )2
and its running point in this configuration will be denoted by OG = r(s), 𝑔 = 1 − 𝑘 𝑦1 . (12)

2
E. Pruchnicki, X. Chen and H.-H. Dai Applications in Engineering Science 6 (2021) 100046

From (12), we see that the volume element of the rod is given in Then we can write the three-dimensional field equations together
terms of curvilinear coordinates by: with boundary conditions (17) in curvilinear coordinates:
√ ( )
𝑑𝐁̃ = 𝑑 𝐗 = 𝑑 𝑋1 𝑑 𝑋2 𝑑 𝑋3 = 𝑔 𝑑 𝐲 = 1 − 𝑘 𝑦1 𝑑 𝐲. (13) 𝑑𝑖𝑣 𝜎 + 𝐟 = 0, 𝐢𝐧 𝐁,
𝜎.𝐧𝑙𝑎𝑡 = 𝐓̄ , 𝐨𝐧 Γ𝑙𝑎𝑡 = [0, 𝐿] × 𝜕Σ,
̃ of the rod 𝐵
̃ is composed (23)
Also we can observe that the boundary 𝜕 𝐵 𝜎. 𝐚3 = 𝐭̄ 𝐨𝐧 Σ𝐿 ,
of the cross sections at 𝑠 = 0 denoted byΣ̃0 , 𝑠 = 𝐿 denoted byΣ ̃𝐿 , and the 𝐮 = 𝐮̄ 𝐨𝐧 Σ0 ,
lateral boundary Γ ̃𝑙𝑎𝑡 (the boundaries of all cross sections Σ=
̃ X(Σ)) .
𝜕𝜎(𝐲)
in which 𝑑𝑖𝑣 𝜎(𝐲) = 𝜕 𝑦𝑚
. 𝑔 𝑚 = 𝜎𝑖𝑗‖𝑗 𝐚𝑖 ,
𝐰𝐢𝐭𝐡 𝛼
𝝈 𝑖𝑗‖𝑗 = 𝜎𝑖𝑗, 𝛼 𝑔𝑗 + (𝜎𝑖𝑗, 3 + 𝜎1𝑗 (𝜏 𝛿𝑖2 − 𝑘 𝛿𝑖3 ) − 𝜏 𝛿𝑖1 𝜎2𝑗 + 𝑘𝛿𝑖1 𝜎3𝑗 ) 𝑔𝑗3 +
4. The three-dimensional formulation
(−𝑘 𝑔33 + 𝜏𝑔23 )𝜎𝑖1 + (−𝜏 𝜎𝑖2 + 𝑘𝜎𝑖3 )𝑔13 ,
∼ ∼
It is supposed that the rod is composed of a general linearized 𝐟 (𝐲) = ̃
𝐟 (𝐗(𝐲)), 𝐧𝑙𝑎𝑡 (𝐲) = 𝐧̃ 𝑙𝑎𝑡 (𝐗(𝐲)) , 𝐓̄ (𝐲) = 𝐓̄ (𝐗(𝐲)), 𝐭̄ (𝐲) = 𝐭̄ (𝐗(𝐲)) 𝐚𝐧𝐝
anisotropic elastic material, for which the constitutive relation in carte- ∼
sian coordinate is given by: 𝐮̄ (𝐲) = 𝐮̄ (𝐗(𝐲)).

𝜎( ̃ ∶ 𝑔𝑟𝑎𝑑
̃ 𝐮̃ (𝐗)) = 𝐄 ̃ (𝐮̃ ), (14) 5. The one-dimensional rod equations

in which 𝜎̃ is the Cauchy stress tensor, 𝐄̃ is the fourth-order elastic mod-

Based on the tridimensional formulation given in Section 4, we in-
ulus tensor and𝜀̃ is the linearized Green-Lagrange strain tensor and ̃ 𝐮 is tend to establish a rod theory which eliminates the cross-section vari-
the displacement vector in cartesian coordinates. ables 𝐲′ . This will be achieved through Taylor series expansions in terms
The modulus tensor 𝐄 ̃ in cartesian coordinates (𝐞𝑖 ⊗ 𝐞𝑗 ⊗ 𝐞𝑘 ⊗ 𝐞𝑙 , i, j,
of the curvilinear coordinates 𝐲′ , while on the lateral surface the polar
k and l =1, 2, 3) possesses the symmetry relations : coordinates will be used. We shall assume sufficient smoothness such
̃𝑖𝑗𝑘𝑙 = 𝐸
𝐸 ̃𝑗𝑖𝑘𝑙 = 𝐸
̃𝑖𝑗𝑙𝑘 = 𝐸
̃𝑘𝑙𝑖𝑗 , that the displacement field u(y) defined by (19) has the following ex-
pansion:
thus there are at most twenty one free components in 𝐄̃. 𝑛
∑5 ∑
𝑢(𝑘 , 𝑛−𝑘) ( ) 𝑘 𝑛−𝑘 ( )
It is well-known that the three-dimensional field equations together 𝑢 (𝐲 ) = 𝑦3 𝑦1 𝑦2 + 𝑂 𝑅60 , (24)
with boundary conditions are : 𝑛=0 𝑘=0
𝑘 ! ( 𝑛 − 𝑘) !
( ) 𝜕𝑛 𝐮 ( )
𝐷𝑖𝑣 𝜎̃ + ̃
𝐟 = 0, 𝐢𝐧 ̃ ,
𝐁 (15) 𝐢𝐧 𝐰𝐡𝐢𝐜𝐡 𝐮(𝑘, 𝑛−𝑘) 𝑦3 = 0, 0, 𝑦3 . (25)
𝜕 𝑦𝑘1 𝜕 𝑦𝑛2−𝑘
∼ This approach provides under regularity assumption, the order of
̃ 𝐧̃ 𝑙𝑎𝑡 = 𝐓̄ ,
𝜎. 𝐨𝐧 Γ̃ 𝑙𝑎𝑡 , (16) the remainder in every equations. This is the least order expansion such
that rod equations are of leading order correctness.
∼ If we assume that 𝑘 𝑦1 < 𝑘 𝑅0 << 1, we can write the Taylor-Young
̃ 𝐧̃ = 𝐭̄ ,
𝜎. 𝐨𝐧 ̃𝐿 ,
Σ (17) expansion of the contravariant basis:
( ) ∑ 4
( ) ( )
∼
̃0 , 𝑔 1 (𝑦) = 𝐚1 𝑦3 + 𝑔 1 (𝑛) 𝑦3 𝑦𝑛1−1 𝑦2 + 𝑂 𝑅50 , (26)
𝐮̃ = 𝐮̄ 𝐨𝐧 Σ (18)
𝑛=1

in which 𝐷𝑖𝑣 𝜎̃ = 𝜎̃ 𝑖𝑗,𝑗 𝐞𝑖 , 𝐧̃ 𝑙𝑎𝑡 (respectively𝐧̃ ) is the unit outward normal ( ) ∑ 4

( ) ( )
∼ ∼ 𝑔 2 (𝑦) = 𝐚2 𝑦3 + 𝑔 2 (𝑛) 𝑦3 𝑦𝑛1 + 𝑂 𝑅50 , (27)
vector on the boundary Γ ̃𝐿 ),𝐓̄ (respectively 𝐭̄ ) is the
̃𝑙𝑎𝑡 (respectivelyΣ 𝑛=1
prescribed traction on the lateral boundaryΓ ̃𝑙𝑎𝑡 (respectively Σ
̃𝐿 ), at the
∼ ( ) ∑ 4
( ) ( )
̃ ̃
edges Σ0 ,the displacement 𝐮̄ is imposed and finally 𝐟 is the volumic force. 𝑔 3 (𝑦) = 𝐚3 𝑦3 + 𝑔 3 (𝑛) 𝑦3 𝑦𝑛1 + 𝑂 𝑅50 , (28)
Now we want to work in curvilinear coordinates y, we introduce the 𝑛=1

components of the displacement field as follows: in which

𝑢̃ (𝑋 ) = 𝑢̃ 𝑖 (𝑋 ) 𝑒𝑖 = 𝑢𝑖 (𝑦) 𝐚𝑖 = 𝐮(𝑦). (19) 𝐠1 (𝑛) = 𝜏 𝑘𝑛−1 𝐚3 , 𝐠2 (𝑛) = −𝜏 𝑘𝑛−1 𝐚3 , 𝐠3 (𝑛) = 𝑘𝑛 𝐚3 , 𝐟 𝐨𝐫 1 ≤ 𝑛 ≤ 4 (29)

From formula (19), we see that (14) can be written as follows: From (21), (24)-(28), we obtain the expansion of the gradient of the
displacement field (24):
𝜎(𝐮(𝐲)) = 𝜎( ̃ ∶ 𝑔𝑟𝑎𝑑 (𝐮(𝐲)),
̃ 𝐮̃ (𝐗)) = 𝐄 (20)
∑4 ∑ 𝑛
𝑔 𝑟𝑎𝑑 (𝑢)(𝑘 , 𝑛−𝑘) ( ) 𝑘 𝑛−𝑘 ( )
in which by using chain rule and formulas (2)-(4), we see that 𝑔 𝑟𝑎𝑑 (𝑢(𝐲)) = 𝑦3 𝑦1 𝑦2 + 𝑂 𝑅50 , (30)
𝑛=0 𝑘=0
𝑘 ! (𝑛 − 𝑘 ) !
𝑔 𝑟𝑎𝑑 (𝐮) = 𝐮, 𝑙 (𝐲) ⊗ 𝐠𝑙 = 𝑔 𝑟𝑎𝑑 (𝐮)𝑖𝑗 𝐚𝑖 ⊗ 𝐚𝑗 , (21) in which 𝑔 𝑟𝑎𝑑(𝐮)(𝑘, 𝑛−𝑘) = 𝑔 𝑟𝑎𝑑(𝐮)(𝑖𝑗𝑘, 𝑛−𝑘) 𝐚𝑖 ⊗ 𝐚𝑗 ,

𝑘
∑ +1
( ) 𝐰𝐡𝐞𝐫 𝐞 𝐠𝐫 𝐚𝐝(𝑢)(𝑖𝑗𝑘,𝑛−𝑘) = 𝑢(𝑖𝑘+1,𝑛−𝑘) 𝛿1𝑗 + 𝑢(𝑖𝑘+1−𝑝,𝑛−𝑘−1) 𝑔𝑗1(𝑝) + 𝑢(𝑖𝑘,𝑛−𝑘+1) 𝛿2𝑗
𝐰𝐢𝐭𝐡 𝑢𝑖 = 𝐮 . 𝐚𝑖 , 𝒈𝑚 𝑚 𝜶
𝑙 = 𝐠 . 𝐚𝑙 , 𝑔 𝑟𝑎𝑑 (𝐮)𝑖𝑗 = 𝑢𝑖, 𝜶 𝑔𝑗 + 𝑢𝑖 |3 𝑔𝑗 ,
3
𝑝=1

𝑢1 |3 = 𝑢1, 3 − 𝜏 𝑢2 + 𝑘 𝑢3 , 𝑢2 |3 = 𝑢2, 3 + 𝜏 𝑢1 , 𝑢3 |3 = 𝑢3, 3 − 𝑘 𝑢1 𝐚𝐧𝐝 𝑘

∑ 𝑘
∑
+ 𝑢(𝑖𝑘−𝑝,𝑛−𝑘+1) 𝑔𝑗2(𝑝) + 𝑢(𝑖|3
𝑘,𝑛−𝑘)
𝛿3𝑗 + 𝑘−𝑝,𝑛−𝑘) 3(𝑝)
𝑢(𝑖|3 𝑔𝑗 ,
𝜎𝑖𝑗 = 𝐸𝑖𝑗𝑘𝑙 𝑔𝑟𝑎𝑑 (𝐮)𝑘𝑙 𝐚𝑖 ⊗ 𝐚𝑗 .
𝑝=1 𝑝=1

From formulae (20)-(21), we see that: with 𝑢(1𝑙,𝑚

|3
)
= 𝑢(1𝑙,𝑚
,3
)
− 𝜏 𝑢(2𝑙,𝑚) + 𝑘 𝑢(3𝑙,𝑚) , 𝑢(2𝑙,𝑚
|3
)
= 𝑢(2𝑙,𝑚
,3
)
+ 𝜏 𝑢(1𝑙,𝑚) , 𝑢(3𝑙,𝑚
|3
)
= 𝑢(3𝑙,𝑚
,3
)
−
( )( ) 𝑘 𝑢(1𝑙,𝑚) , for 𝑙 ≥ 0 and 𝑚 ≥ 0.
̃𝑖𝑗𝑘𝑙 𝑔𝑟𝑎𝑑 (𝐮)𝑟 𝑠 𝐚𝑘 𝐚𝑙 𝐞𝑖 ⊗ 𝐞𝑗 ,
𝝈(𝐮(𝐲)) = 𝜎𝑖𝑗 (𝐮(𝐲)) 𝐚𝑖 ⊗ 𝐚𝑗 = 𝐸 𝑟 𝑠
( ) ( ) ( )( ) The expansion of the stress tensor is obtainable from (26) and (30):
=𝐸̃𝑖𝑗𝑘𝑙 𝑔𝑟𝑎𝑑 (𝐮)𝑟 𝑠 𝐚𝑖 𝐚𝑗 𝐚𝑘 𝐚𝑙 𝐚𝑜 ⊗ 𝐚𝑝 ,
𝑜 𝑝 𝑟 𝑠
∑ 𝑛
4 ∑ 𝑦𝑘1 𝑦𝑛2−𝑘
= 𝐸𝑖 𝑗 𝑘 𝑙 𝑔𝑟𝑎𝑑 (𝐮)𝑘 𝑙 𝐚𝑖 ⊗ 𝐚𝑗 , (22) ( )
𝜎= 𝜎 (𝑘, 𝑛−𝑘) + 𝑂 𝑅50 , (31)
𝑛=0 𝑘=0
𝑘!(𝑛 − 𝑘)!
where 𝐸𝑜 𝑝 𝑟 𝑠 ̃𝑖𝑗𝑘𝑙 (𝐚𝑖 )𝑜 (𝐚𝑗 )𝑝 (𝐚𝑘 )𝑟 (𝐚𝑙 )𝑠 .
=𝐸

3
E. Pruchnicki, X. Chen and H.-H. Dai Applications in Engineering Science 6 (2021) 100046

̃ ∶ 𝑔𝑟𝑎𝑑 (𝐮)(𝑘, 𝑛−𝑘)

𝐰𝐡𝐞𝐫𝐞 𝝈 (𝑘, 𝑛−𝑘) = 𝐄 order𝑅30 involves six vectorial equation:
( )( ) ( ) ( )
=𝐸̃𝑖𝑗𝑘𝑙 𝑔𝑟𝑎𝑑 (𝐮) 𝑛−𝑘) 𝐚𝑘 𝐚𝑙 𝐞𝑖 ⊗ 𝐞𝑗 ,
(𝑘,
𝑟𝑠 𝑟 𝑠 𝑑𝑖𝑣(𝑘, 𝑛−𝑘) 𝝈 𝑦3 + 𝐟 (𝑘, 𝑛−𝑘) 𝑦3 = 0, 𝐟 𝐨𝐫 𝑛 = 0, 1, 2 𝐚𝐧𝐝 𝑘 = 0, 𝑛, (36)
𝑘, 𝑛−𝑘)
= 𝐸𝑖𝑗𝑙𝑚 𝑔𝑟𝑎𝑑 (𝐮)(𝑙𝑚 𝐚𝑖 ⊗ 𝐚𝑗 , (32) in which the volumic force f is assumed to have expansion of the type
We point out that the stress coefficients (32) for n = 4 is not needed (24) of which the vector coefficient are 𝐟 (𝑘, 𝑛−𝑘) and
for the end results so in the above equation we take 0 ≤ n ≤ 3. ( ) ( ) ( ) ( ) ( ) ( ) ( )
𝑑𝑖𝑣(0, 0) 𝝈 𝑦3 = 𝜎 (1, 0) 𝑦3 . 𝐚1 𝑦3 + 𝜎 (0, 1) 𝑦3 . 𝐚2 𝑦3 + 𝜎,(03, 0) 𝑦3 . 𝐚3 𝑦3 ,
An observation is that 𝜎 (𝑘, 𝑛−𝑘) depends linearly on higher-order coeffi- ( ) ( ) ( ) ( ) ( ) ( ) ( )
𝑑𝑖𝑣(1, 0) 𝝈 𝑦3 = 𝝈 (2, 0) 𝑦3 . 𝐚1 𝑦3 + 𝝈 (0, 1) 𝑦3 . 𝐠2 (1) 𝑦3 + 𝝈 (1, 1) 𝑦3 . 𝐚2 𝑦3
cients 𝑢(𝑖𝑘+1, 𝑛−𝑘) , 𝑢(𝑖𝑘−𝑝, 𝑛−𝑘+1) for p = 0, k and on lower order coefficients
(0, 0) ( ) 3 (1) ( ) (1, 0) ( ) ( )
+ 𝝈 , 3 𝑦3 . 𝐠 𝑦3 + 𝝈 , 3 𝑦3 . 𝐚3 𝑦3 ,
𝑢(𝑖𝑘+1−𝑝, 𝑛−𝑘−1) for p = 1,
( ) ( ) ( ) ( ) ( ) ( ) ( )
k +1. For this reason the dimension reduction procedure works as in 𝑑𝑖𝑣 (0, 1)
𝝈 𝑦3 = 𝝈 (1, 1)
𝑦3 . 𝐚1 𝑦3 + 𝝈 (0, 2) 𝑦3 . 𝐚2 𝑦3 + 𝝈 (, 03, 1) 𝑦3 . 𝐚3 𝑦3 ,
the similar approach for plate and shell theories. ( ) ( ) ( ) ( ) ( ) ( ) ( )
𝑑𝑖𝑣(2, 0) 𝝈 𝑦3 = 𝝈 (3, 0) 𝑦3 . 𝐚1 𝑦3 + 2 𝝈 (0, 1) 𝑦3 . 𝐠2 (2) 𝑦3 + 2 𝝈 (1, 1) 𝑦3 . 𝐠2 (1) 𝑦3
To satisfy the lateral traction condition in a pointwise man- ( ) ( ) ( ) ( ) ( ) ( )
+ 𝝈 (2, 1) 𝑦3 . 𝐚2 𝑦3 + 2 𝝈 (, 03, 0) 𝑦3 . 𝐠3 (2) 𝑦3 + 2 𝝈 (, 13, 0) 𝑦3 . 𝐠3 (1) 𝑦3
ner, we use Fourier series expansion for the applied traction
( ) ( )
̂̄ (𝑅 cos 𝜃 , 𝑅 sin 𝜃 , 𝑦 ) (polar coordinate is used on 𝜕Σ:𝑦 =
𝐓̄ (𝐲) = 𝐓 + 𝝈 (, 23, 0) 𝑦3 . 𝐚3 𝑦3 ,
0 𝑎 0 𝑎 3 1
( ) ( ) ( ) ( ) ( ) ( ) ( )
̂ ⃖⃖⃖⃖⃖⃖⃖⃗) is the azimuthal angle on 𝑑𝑖𝑣(1, 1) 𝝈 𝑦3 = 𝝈 (1, 0) 𝑦3 . 𝐠1 (1) 𝑦3 + 𝝈 (2, 1) 𝑦3 . 𝐚1 𝑦3 + 𝝈 (0, 2) 𝑦3 . 𝐠2 (1) 𝑦3
𝑅0 cos 𝜃𝑎 , 𝑦2 = 𝑅0 sin 𝜃𝑎 and𝜃𝑎 = (𝐺𝑦1 , 𝐺𝑀
( ) ( ) ( ) ( ) ( ) ( )
𝑦1 𝑦2 -plane), then we assume that 𝐓 ̂̄ is a sufficiently regular function of + 𝝈 (1, 2) 𝑦3 . 𝐠2 (1) 𝑦3 + 𝝈 (1, 2) 𝑦3 . 𝐚2 𝑦3 + 𝝈 (, 03, 1) 𝑦3 . 𝐠3 (1) 𝑦3
𝜃𝑎 to ensure that the Fourier series expansion is convergent : ( ) ( )
+ 𝝈 (, 13, 1) 𝑦3 . 𝐚3 𝑦3 ,
∑
∞ ( ) ( ) ( ) ( ) ( ) ( ) ( )
̂̄ (𝜃 ) = 𝐓
𝐓 ̂̄ + ̂̄ cos 𝑝 𝜃 + 𝐓
𝐓 ̂̄ sin 𝑝 𝜃 , (33)
𝑑𝑖𝑣(0, 2) 𝝈 𝑦3 = 𝝈 (1, 2) 𝑦3 . 𝐚1 𝑦3 + 𝝈 (0, 3) 𝑦3 . 𝐚2 𝑦3 + 𝝈 (, 03, 2) 𝑦3 . 𝐚3 𝑦3 ,
𝑎 0 𝑐𝑝 𝑎 𝑠𝑝 𝑎
𝑝=1
in which from formulae (2)-(4) and (22)2 , we obtain:
̂̄ = 1 ∫ 𝜋 𝐓 ̂̄ (𝜃 ) 𝑑 𝜃 , ̂̄ 1 𝜋 ̂ ̄
where 𝐓 0 2𝜋 −𝜋 𝑎 𝑎 𝐓 𝑐𝑝 = 𝜋 ∫−𝜋 𝐓 (𝜃𝑎 ) cos(𝑝 𝜃𝑎 ) 𝑑 𝜃𝑎 and 𝜎,(3𝑙, 𝑚) = 𝐸𝑖𝑗𝑘 𝑟 𝑔𝑟𝑎𝑑 (𝐮)(𝑘𝑙,𝑟,𝑚3 ) 𝐚𝑖 ⊗ 𝐚𝑗
̂̄ =
𝐓 1 𝜋 ̂
∫ 𝐓̄ (𝜃𝑎 ) sin(𝑝 𝜃𝑎 ) 𝑑 𝜃𝑎 .
𝑠𝑝 ( ) (̃) (̃)
𝜋 −𝜋
+𝐸 ̃𝑖𝑗𝑘𝑡 𝑔𝑟𝑎𝑑 (𝐮)(𝑙, 𝑚) 𝐴𝑘 𝑟 ̃ 𝐚𝑡 𝑠 𝐚𝑖 𝑜 𝐚𝑗 𝑝 𝐚𝑜 ⊗ 𝐚𝑝
Since 𝐧𝑙𝑎𝑡 = cos(𝜃𝑎 ) 𝐚1 + sin(𝜃𝑎 ) 𝐚2 , we obtain: 𝑟𝑠
( ) ( ) (̃)
∑ 𝑛
3 ∑ 𝑅𝑛0 +𝐸 ̃𝑖𝑗𝑘𝑡 𝑔𝑟𝑎𝑑 (𝐮)(𝑙, 𝑚) ̃ 𝐚𝑘 𝑟 𝐴𝑡 𝑠 ̃𝐚𝑖 𝑜 𝐚𝑗 𝑝 𝐚𝑜 ⊗ 𝐚𝑝 . (37)
( )𝑘+1 ( )𝑛−𝑘 𝑟𝑠
𝜎. 𝐧𝑙𝑎𝑡 = cos 𝜽𝑎 sin 𝜽𝑎 𝜎 (𝑘, 𝑛−𝑘) . 𝐚1
𝑘!(𝑛 − 𝑘)!
𝑛=0 𝑘=0 in which𝐴𝑘 𝑟 = (𝑘𝛿𝑘3 − 𝜏 𝛿𝑘2 )(̃
𝐚1 )𝑟 + 𝜏 𝛿𝑘1 (̃
𝐚2 )𝑟 − 𝑘 𝛿𝑘1 (̃
𝐚3 )𝑟 ,
∑ 𝑛
3 ∑ 𝑅𝑛0 ( )𝑘 ( )𝑛−𝑘+1 (𝑘, 𝑛−𝑘) 𝑙+1 (
+ cos 𝜽𝑎 sin 𝜽𝑎 𝜎 . 𝐚2 ∑ )
𝑛=0 𝑘=0
𝑘!(𝑛 − 𝑘)! 𝑔𝑟𝑎𝑑 (𝐮)(𝑘𝑙,𝑟,𝑚3) = 𝑢(𝑘,𝑙+1
3
, 𝑚)
𝛿1 𝑟 + 𝑢(𝑘,𝑙+1−
3
𝑝, 𝑚−1) 1(𝑝)
𝑔𝑟 + 𝑢(𝑘𝑙+1−𝑝, 𝑚−1) 𝑔𝑟,1(3𝑝)
( ) 𝑝=1
+ 𝑂 𝑅40 cos 𝜽𝑎 , 𝑅40 sin 𝜽𝑎 , 𝑅40 cos 3𝜽𝑎 , 𝑅40 sin 3𝜽𝑎
( ) 𝑙 ( )
∑
+ 𝑂 𝑅50 , 𝑅50 cos 2𝜽𝑎 , 𝑅50 sin 2𝜽𝑎 , 𝑅50 cos 4𝜽𝑎 , 𝑅50 sin 4𝜽𝑎 . (34) + 𝑢(𝑘,𝑙,3𝑚+1) 𝜹2𝑟 + 𝑢(𝑘,𝑙−3 𝑝, 𝑚+1) 𝑔𝑟2(𝑝) + 𝑢(𝑘𝑙−𝑝, 𝑚+1) 𝑔𝑟,2(3𝑝)
𝑝=1
Then, from Eqs. (23) 2 , (33) and (34), by setting equal the coefficients 𝑙 (
∑ )
of cos(𝑚𝜃𝑎 ), 𝑠𝑖𝑛(𝑚𝜃𝑎 )for 1 ≤ m ≤ 3 on both sides, one can obtain seven + 𝑢(𝑘𝑙,‖33
𝑚)
𝜹3 𝑙 ′ + 𝑢(𝑘𝑙− 𝑝, 𝑚) 3(𝑝)
𝑔𝑙′ + 𝑢(𝑘𝑙− 𝑝, 𝑚) 3(𝑝)
𝑔𝑙 ′ ,3 ,
‖33 |3
equations (in vector form) for the stress coefficients. However, since 𝑝=1
Eq. (34) is not exact, for the coefficients of cos(𝑚𝜃𝑎 )(0 ≤ m ≤ 1) and
𝑠𝑖𝑛(𝜃𝑎 ), all terms are kept, while for other coefficients only leading-order with 𝑢(1𝑙,‖33
𝑚)
= 𝑢(1𝑙,, 33
𝑚)
− (𝜏, 3 𝑢(2𝑙, 𝑚) + 𝜏 𝑢(2𝑙,,3𝑚) ) + 𝑘,3 𝑢(3𝑙, 𝑚) + 𝑘 𝑢(3𝑙,, 3𝑚) , 𝑢(2𝑙,‖33
𝑚)
=
terms are kept (Chen et al. (2020)). So we obtain: 𝑢(2𝑙,, 33
𝑚)
+ 𝜏,3 𝑢(1𝑙, 𝑚) + 𝜏 𝑢(1𝑙,,3𝑚) , 𝑢(3𝑙,𝑚
‖33
)
= 𝑢(3𝑙,, 33𝑚) − (𝑘,3 𝑢(1𝑙, 𝑚) + 𝑘 𝑢(1𝑙,,3𝑚) ), for 𝑙 ≥
2 ̂̄ 0 and 𝑚 ≥ 0.
𝐓 = 𝜎 (1, 0) . 𝐚1 + 𝝈 (0, 1) . 𝐚2 Next we modify Eqs. (35)-(36) in order to obtain a system of ten vec-
𝑅0 0
tors equations with ten vectors displacement field unknows. By consid-
𝑅20 ( ) ( )
+ 𝝈 (1, 2) . 𝐚1 + 𝝈 (2, 1) . 𝐚2 + 𝝈 (3, 0) . 𝐚1 + 𝝈 (0, 3) . 𝐚2 + 𝑂 𝑅40 , ering (36) (for (k, n) = (0, 0), (2,0), (0,2)), we can eliminate 𝜎 (1, 0) . 𝐚1 +
8 𝜎 (0, 1) . 𝐚2 and 𝜎 (1, 2) . 𝐚1 + 𝜎 (2, 1) . 𝐚2 + 𝜎 (3, 0) . 𝐚1 + 𝜎 (0, 3) . 𝐚2 from (35)1 :
𝑅2 𝑅2 ( ) ( )
̂̄
𝐓 = 𝝈 (0, 0) . 𝐚1 + 0 𝝈 (1, 1) . 𝐚2 + 0 𝝈 (0, 2) + 3𝝈 (2, 0) . 𝐚1 + 𝑂 𝑅40 ,
𝑐1 2 ̂̄
4 8 𝐓 + 𝐟 (0, 0)
𝑅0 0
𝑅2 𝑅2 ( ) ( )
̂̄
𝐓 = 𝝈 (0, 0) . 𝐚2 + 0 𝝈 (1, 1) . 𝐚1 + 0 3𝜎 (0, 2) + 𝜎 (2, 0) . 𝐚2 + 𝑂 𝑅40 , 𝑅20 ( )
𝑠1
4 8 = −𝜎,(03, 0) . 𝐚3 − 𝜎 (0, 1) . 𝐠2 (2) + 𝜎 (1, 1) . 𝐠2 (1) + 𝜎,(03, 0) . 𝐠3 (2) + 𝜎,(13, 0) . 𝐠3 (1)
2 ̂̄ ( ) 4
𝐓 = 𝝈 (1, 0) . 𝐚1 − 𝝈 (0, 1) . 𝐚2 + 𝑂 𝑅20 , 𝑅20 ( ) ( )
𝑅0 𝑐2 − 𝜎,(23, 0) . 𝐚3 + 𝜎,(03, 2) . 𝐚3 + 𝐟 (2, 0) + 𝐟 (0, 2) + 𝑂 𝑅40 . (38)
2 ̂̄ ( ) 8
𝐓 = 𝝈 (0, 1) . 𝐚1 + 𝝈 (1, 0) . 𝐚2 + 𝑂 𝑅20 ,
𝑅0 𝑠2 Finally Eqs. (35) 2 - 7 , (36) (for (k, n) = (0, 0), (1,0), (0,1))) and (38)
4 ̂̄ 1( ) ( ) is a system of ten vector equations with ten unknown vectors 𝐮(𝑘, 𝑛)
𝐓𝑐3 = −𝝈 (1, 1) . 𝐚2 + 𝝈 (2, 0) − 𝝈 (0, 2) . 𝐚1 + 𝑂 𝑅20 ,
𝑅20 2 (for (k, n) = (0,0), (1,0), (0,1), (2,0), (1,1), (0,2), (3,0), (2,1), (1,2),
4 ̂̄ 1( ) ( ) (0,3)). However, the errors in different equations are not of the same
𝐓𝑠3 = 𝝈 (1, 1) . 𝐚1 + 𝝈 (2, 0) − 𝝈 (0, 2) . 𝐚2 + 𝑂 𝑅20 . (35) order of magnitude. Eqs. (38) and (35)2 - 3 are truncated at 𝑂(𝑅40 ) while
𝑅 2 2
0
Eqs. (35) 4 - 7 are truncated at 𝑂(𝑅20 ). Therefore in what follows, we shall
By using (32), we see that the fifteen unknown coefficients of perform a number of refinements.
the expansion of the displacement field (24) intervenes in the seven
Eqs. (35). Next considering equilibrium equations, it is possible to ob- 6. Reduction process
tain a closed system of ten equations with ten unknown coefficients of
the displacement field (24). The method of Section 5 presents a number of undesirable features.
By using expansion of stress tensor (31) and expansions (26)-(28) There are too many unknowns (ten vector unknowns). It would be better
for the contravariant basis vectors, equilibrium Eq. (23) 1 truncated at if the rod equations contains only terms with clear physical meanings

4
E. Pruchnicki, X. Chen and H.-H. Dai Applications in Engineering Science 6 (2021) 100046

𝑙, 𝑚) 𝑙, 𝑚
(stretching, bending and twisting). This system of ten vector equations 𝜎,(33 = 𝐸𝑖𝑗𝑘 𝑟 𝑔𝑟𝑎𝑑 (𝐮)(𝑘 𝑟,33) 𝐚𝑖 ⊗ 𝐚𝑗
requires in total twenty four scalar boundary conditions which cannot ( ) (̃) (̃)
be imposed physically. Indeed, for a force edge, one can only know six + 2𝐸 ̃𝑖𝑗𝑘𝑡 𝑔𝑟𝑎𝑑 (𝐮)(𝑙, 𝑚) 𝐴𝑘 𝑟 ̃ 𝐚𝑡 𝑠 𝐚𝑖 𝑜 𝐚𝑗 𝑝 𝐚𝑜 ⊗ 𝐚𝑝
𝑟𝑠,3
scalar boundary conditions given by the components of the torsor re- ( ) ( ) (̃)
+ 2𝐸 ̃𝑖𝑗𝑘𝑡 𝑔𝑟𝑎𝑑 (𝐮)(𝑙, 𝑚) ̃ 𝐚𝑘 𝑟 𝐴𝑡 𝑠 ̃𝐚𝑖 𝑜 𝐚𝑗 𝑝 𝐚𝑜 ⊗ 𝐚𝑝
sultant. In this section, we reduce the number of main unknows to four 𝑟𝑠,3
(̃) (̃) (̃)
which represent stretching, two bendings and twisting. +𝐸 ̃𝑖𝑗𝑘𝑡 𝑔𝑟𝑎𝑑 (𝐮) (𝑙, 𝑚 )
𝐵𝑘 𝑟 𝐚𝑡 𝑠 𝐚𝑖 𝑜 𝐚𝑗 𝑝 𝐚𝑜 ⊗ 𝐚𝑝
𝑟𝑠
By considering linear elastic relation (32), one can see that (35)4,5 ( ) (̃)
and (36) (for k = 0, n = 0) provide 3 linear algebraic vectorial equations + 2𝐸 ̃𝑖𝑗𝑘𝑡 𝑔𝑟𝑎𝑑 (𝐮)(𝑙, 𝑚) 𝐴𝑘 𝑟 𝐴𝑡 𝑠 ̃ 𝐚𝑖 𝑜 𝐚𝑗 𝑝 𝐚𝑜 ⊗ 𝐚𝑝
𝑟𝑠
for 3 unknown vectors 𝐮(𝑘, 𝑙) (for (k, l) = (2,0), (1,1), (0,2)) and we ob- (̃) (̃) (̃)
+𝐸̃𝑖𝑗𝑘𝑡 𝑔𝑟𝑎𝑑 (𝐮) (𝑙, 𝑚 )
𝐚𝑘 𝑟 𝐵𝑡 𝑠 𝐚𝑖 𝑜 𝐚𝑗 𝑝 𝐚𝑜 ⊗ 𝐚𝑝 . (42)
tain: 𝑟𝑠

⎡𝐮(2, 0) ⎤ ⎧⎡ −𝐀(1, 3) 𝐮(1 |3

, 0) ⎤ ⎡ 𝐀(2, 3) 𝐮(0|3
, 1) ⎤⎫ where 𝐵𝑘 𝑟 = (−(𝜏 2 + 𝑘2 )𝛿𝑘1 + 𝑘, 3 𝛿𝑘3 − 𝜏, 3 𝛿𝑘2 ) (̃
𝐚1 )𝑟 + ((𝑘𝛿𝑘3 − 𝜏 𝛿𝑘2 )𝜏 +
⎢𝐮(1, 1) ⎥ = 𝐊 −1 ⎪ ⎢ (2, 3) 𝐮(1, 0) ⎥ ⎢ (1, 3) 𝐮(0, 1) ⎥⎪ ( ( ) )(̃)
⎨⎢ − 𝐀 |3 ⎥ ⎢
+ − 𝐀 |3 ⎥⎬+ ̃
𝜏,3 𝛿𝑘1 )(𝐚2 )𝑟 − 𝑘 𝑘𝛿𝑘3 − 𝜏𝛿𝑘2 + 𝑘,3 𝛿𝑘1 𝐚3 𝑟 ,
⎢ (0, 2) ⎥ ⎪⎢−𝐀(3, 1) 𝐮 ⎥ ⎢−𝐀(3, 2) 𝐮 ⎥⎪
⎣𝐮 ⎦ ⎩⎣
(1 , 0)
− 𝐀 (1, 3) 𝐮 (1 , 0)
⎦ ⎣
(0 , 1)
− 𝐀 (2 3) 𝐮 (0 , 1)
⎦⎭
,3 |3 ,3 |3 𝑚, 𝑛)
⎧⎡ 2 ∧ ( 2) ⎤ ⎫ with grad(𝑢)(𝑘𝑙, 33
= 𝑢(𝑘,𝑚33
+1, 𝑛)
𝛿1𝑙
̄
⎪⎢ 𝑅 𝐓𝑐2 + 𝑂 𝑅0 ⎥ ⎡𝐀(1,3,1) ⎤ ⎪
⎡𝐀(1,2,1) ⎤ ⎡ 0 ⎤ 𝑚
∑ ( )
⎪ 0
⎪ +1
𝐊−1 ⎨⎢ 2 ̄ ( 2 ) ⎥ − ⎢𝐀(2,3,1) ⎥𝐮(0,0) − ⎢𝐀(2,2,1) ⎥𝐮(0,1) − ⎢ 0 ⎥𝐮(1,0) ⎬, 𝑚+1−𝑝, 𝑛−1) 1𝑝 𝑚+1−𝑝, 𝑛−1) 1𝑝 𝑚+1−𝑝, 𝑛−1) 1𝑝
∧
+ 𝑢(𝑘,33 𝑔𝑙 ( ) + 2𝑢(𝑘,3 𝑔𝑙,(3 ) + 𝑢(𝑘 𝑔𝑙,(33)
⎪⎢⎢ 𝑅0 𝐓𝑠2 + 𝑂 𝑅0 ⎥⎥ ⎢⎣ 𝐀(3) ⎥⎦ ⎢ (2 ) ⎥ ⎢ (1 ) ⎥
|3
⎣ 𝐀 ⎦ ⎣ 𝐀 ⎦ ⎪ 𝑝=1,𝑛≥1
⎪⎣−𝐀(3, 3) 𝐮(0, 0) − 𝐟 (0, 0) ⎦ ⎪
⎩ ‖3 3 ⎭ 𝑚 (
∑ )
𝑚, 𝑛+1) 𝑚−𝑝, 𝑛+1) 2𝑝 𝑚−𝑝, 𝑛+1) 2𝑝 𝑚−𝑝, 𝑛+1) 2𝑝
(39) + 𝑢(𝑘,33 𝛿2𝑙 + 𝑢(𝑘,33 𝑔𝑙 ( ) + 2𝑢(𝑘,3 𝑔𝑙,(3 ) + 𝑢(𝑘 𝑔𝑙,(33)
𝑝=1
where 𝑚 (
∑ )
𝑚, 𝑛 𝑚−𝑝, 𝑛) 3𝑝 𝑚−𝑝, 𝑛) 3𝑝 𝑚−𝑝, 𝑛) 3𝑝
+ 𝑢(𝑘‖333) 𝛿3𝑙 + 𝑢(𝑘‖333 𝑔𝑙 ( ) + 2𝑢(𝑘‖33 𝑔𝑙,(3 ) + 𝑢(𝑘‖33 𝑔𝑙,(3 )
⎛𝐀(1, 1) 𝐀(1, 2) − 𝐀(2, 1) −𝐀(2, 2) ⎞ 𝑝=1
𝐊 = ⎜𝐀(2, 1) 𝐀(1, 1) + 𝐀(2, 2) 𝐀(1, 2) ⎟, ( )
⎜ (1, 1) ⎟
⎝𝐀 𝐀(1, 2) + 𝐀(2, 1) 𝐀(2, 2) ⎠ 𝐚𝐧𝐝 𝑢(1𝑚, 𝑛)
‖333
= 𝑢(1𝑚, 𝑛)
, 333
− 𝜏, 33 𝑢(2𝑚, 𝑛) + 2𝜏,3 𝑢(2𝑚,
,3
𝑛)
+ 𝜏 𝑢(2𝑚,
,33
𝑛)
+ 𝑘,33 𝑢(3𝑚, 𝑛)
(( ) ) ( )( ) ( )
(𝑟 )
𝐀 =𝐸 ̃𝑖𝑗𝑘𝑙 𝑘𝛿𝑘3 − 𝜏 𝛿𝑘2 𝐚1 + 𝜏 𝛿𝑘1 𝐚2 − 𝑘 𝛿𝑘1 𝐚3 ∗ 𝐚𝑙 𝐚𝑖 ∗ 𝐚𝑗 𝐚𝑖∗ ⊗ 𝐚𝑘∗
( ) (( )
𝑘 𝑟
) ( ) ( )
𝑖 3 + 2 𝑘,3 𝑢(3𝑚,
,3
𝑛)
+ 𝑘 𝑢(3𝑚, 𝑛)
, 33
,
+𝐸 ̃𝑖𝑗𝑘𝑙 𝐚𝑘 ∗ 𝑘𝛿𝑙3 − 𝜏 𝛿𝑙2 𝐚1 + 𝜏 𝛿𝑙1 𝐚2 − 𝑘 𝛿𝑙1 𝐚3 𝐚𝑖 𝑖∗ 𝐚𝑗 3 𝐚𝑖∗ ⊗ 𝐚𝑘∗
𝑢(2𝑚,‖333
𝑛)
= 𝑢(2𝑚, 𝑛)
+ 𝜏,33 𝑢(1𝑚, 𝑛) + 2𝜏,3 𝑢(1𝑚, 𝑛)
+ 𝜏 𝑢(1𝑚, 𝑛)
𝑘 𝑟
, 333 ,3 ,33
,
𝐀(𝑖, 𝑗 ) = 𝐸𝑜𝑖𝑝𝑗 𝐚𝑜 ⊗ 𝐚𝑝 , 𝐀(𝑖, 𝑗, 𝑘) = 𝐸𝑜𝑖𝑝𝑛 𝑔𝑛𝑗 (𝑘) 𝐚𝑜 ⊗ 𝐚𝑝 . ( )
𝑢(3𝑚,‖333
𝑛)
= 𝑢(3𝑚, 𝑛)
, 333
− 𝑘,33 𝑢(1𝑚, 𝑛) +2𝑘,3 𝑢(1𝑚,,3
𝑛)
+ 𝑘 𝑢(1𝑚,
,33
𝑛)
, 𝐟 𝐨𝐫 𝑙 ≥ 0 𝐚𝐧𝐝 𝑚 ≥ 0.
We recall that we keep 𝑂(𝑅20 ) terms in (38) and (35)2 3 for two rea-
sons: they may give bending or torsion effects and they may become As for straight rod (Chen et al. 2018), the term linked to bend-
leading-order terms. If the first term on the right hand side of (38) can 𝑅2
ing along 𝑦2 (− 40 𝐚3 . 𝜎,(133, 0) . 𝐚3 in the right hand side of (41)) may be the
be eliminated, then the 𝑂(𝑅20 ) term becomes the leading one. Based on
leading-order term when k and 𝜏 are sufficiently small (due to presence
this argument, we multiply (38) by 𝐚1 and then add it to the derivative
of term 𝐚3 .𝜎 (0, 0) .(−𝑘 𝐚3 + 𝜏 𝐚2 )).
(with respect to y3 ) of (35)2 multiplied by 𝐚3 . Then we take into ac-
In the same way, by taking into account of (2), (3), (29) and (36)
count of (2), (3), (29) and (36) for k = 0 and n = 1 to obtain one scalar
for k = 0 and n = 1, then we multiply (38) by 𝐚2 and add it to the deriva-
equation:
tive (with respect to y3 ) of (35)3 multiplied by 𝐚3 to obtain one scalar
̂̄ 2 ̂̄ ( ) equation:
𝐓 𝑐1, 3 . 𝐚3 + 𝐓 . 𝐚 = 𝐚3 .𝜎 (0, 0) . −𝑘 𝐚3 + 𝜏 𝐚2
𝑅0 0 1 ( )
̂̄ = −𝐚 . 𝐟 (0, 0) − 𝜏 𝐚 .𝜎 (0, 0) . 𝐚 + 𝑅0 𝐚 . 𝜎 (1, 1) . 𝐚 + 𝜎 (0, 2) . 𝐚
2
̂̄
𝐓
2
𝑅20 ( ) 𝑠 1,3 . 𝐚3 + 𝐚2 .𝐓
𝑅0 0 2 3 1
4 3 ,3 1 ,3 2
− 𝐟 (0, 0) . 𝐚1 − 𝐚1 . 𝐟 (2, 0) + 𝐟 (0, 2)
8 𝑅20 ( ) 𝑅2 ( ) 𝑅2 ( )
𝑅20 ( ) + 𝐚3 .𝜎 (1, 1) . −𝑘 𝐚3 + 2𝜏 𝐚2 − 0 𝐚2 . 𝐟 (2, 0) + 𝐟 (0, 2) − 0 𝜏 𝐚3 . 3𝜎 (0, 2) + 𝜎 (2, 0) . 𝐚1
4 8 8
− 𝐚1 . −𝜏 𝑘 𝜎 (0, 1) . 𝐚3 − 𝜏 𝜎 (1, 1) . 𝐚3 + 𝑘2 𝜎,(03, 0) . 𝐚3 + 𝑘 𝜎,(13, 0) . 𝐚3
4 𝑅2 ( ) ( )
− 0 𝐚2 . −𝜏 𝑘 𝜎 (0, 1) . 𝐚3 + 𝑘2 𝜎,(03, 0) . 𝐚3 + 𝑘 𝜎,(13, 0) . 𝐚3 + 𝑂 𝑅40 . (43)
𝑅20 ( ) 𝑅2 4
1, 1 2, 0
+ 𝐚3 . 𝜎,(3 ) . 𝐚2 + 𝜎,(3 ) 𝐚1 − 𝜏 0 𝐚3 .𝜎 (1, 1) . 𝐚1
4 4 By deriving with respect to y3 Eq. (36) for k = 0 and n = 1, we sub-
𝑅20 ( ) ( ) ( ) stitute a new expression for 𝜎,(13, 1) . 𝐚1 + 𝜎,(03, 2) 𝐚2 in (43) and we obtain:
+ 𝐚3 . 𝜎 (0, 2) + 3𝜎 (2, 0) . −𝑘 𝐚3 + 𝜏 𝐚2 + 𝑂 𝑅40 . (40)
8
𝑅2 ( )
̂̄
𝐓
2 ̂̄ + 𝐟 (0, 0) . 𝐚 + 0 𝐟 (0, 1) . 𝐚 + 1 (𝐟 (2, 0) + 𝐟 (0, 2) ). 𝐚
By deriving with respect to y3 Eq. (36) for k = 1 and n = 1, we sub- 𝑠1,3 . 𝐚3 + 𝐚2 .𝐓 0 2 ,3 3 2
𝑅0 4 2
stitute a new expression for 𝜎,(13, 1) . 𝐚2 + 𝜎,(23, 0) 𝐚1 in (40) to obtain:
𝑅2 ( 0, 1 0, 1
)
= −𝜏 𝐚3 .𝜎 (0, 0) . 𝐚1 − 0 𝐚3 . 𝜎,(33 ) . 𝐚3 + 𝑘 𝜎,(3 ) . 𝐚1
𝑅2 ( 1( ) ) 4
̂̄
𝐓
2 ̂̄
𝐓 0 . 𝐚1 + 𝐟 (0, 0) . 𝐚1 + 0 𝐟,(13, 0) . 𝐚3 + 𝐟 (2, 0) + 𝐟 (0, 2) . 𝐚1
𝑐1, 3 . 𝐚3 + 𝑅20 ( )
𝑅0 4 2 − 𝐚2 . −𝜏 𝑘 𝜎 (0, 1)
. 𝐚3 + 𝑘2 𝜎,(03, 0) . 𝐚3 + 𝑘 𝜎,(13, 0) . 𝐚3
4
( ) 𝑅2 ( )
= 𝐚3 .𝜎 (0, 0) . −𝑘 𝐚3 + 𝜏 𝐚2 − 0 𝐚1 . −2𝜏 𝑘 𝜎 (0, 1) + 2 𝑘2 𝜎,(03, 0) + 𝑘 𝜎,(13, 0) . 𝐚3 𝑅20 ( ) 𝑅2 ( )
4 − 𝜏 𝐚3 . 𝜎 (0, 2) + 𝜎 (2, 0) . 𝐚1 + 𝜏 0 𝐚2 . 𝜎 (1, 1) . 𝐚3 + 𝑂 𝑅40 . (44)
𝑅20 ( ) 8 4
(0, 1) (0, 1)
− 𝐚3 . −𝜏 𝜎, 3 . 𝐚3 − 𝜏,3 𝜎 . 𝐚3 As for straight rod (Chen et al. 2018), the term linked to bend-
4
𝑅20 ( (0, 0) ) ing along 𝑦1 (−𝐚3 . 𝜎,(033, 1) . 𝐚3 𝑅20 ∕4 in the right hand side of (44)) may
− 𝐚3 . 𝑘𝜎, 33 . 𝐚3 + 𝑘,3 𝜎,(03, 0) . 𝐚3 + 𝜎,(133, 0) . 𝐚3 + 𝑘𝜎,(13, 0) . 𝐚1 be the leading order term when the torsion curvature 𝜏 can be small
4
(−𝜏 𝐚3 .𝜎 (0, 0) . 𝐚1 in the right hand side of (44)).
𝑅2 𝑅2 ( ) ( ) ( )
+𝜏 0 𝐚3 .𝜎 (1, 1) . 𝐚1 + 0 𝐚3 . 𝜎 (0, 2) + 𝜎 (2, 0) . −𝑘 𝐚3 + 𝜏 𝐚2 + 𝑂 𝑅40 . (41) We show that we can simplify Eqs. (41) and (44). Firstly, in formula
4 8 (41), we observe that 𝑘 𝐚1 . 𝜎,(13, 0) . 𝐚3 𝑅20 ∕4 and 𝑘 𝐚3 . 𝜎,(033, 0) . 𝐚3 𝑅20 ∕4
in which (see formulae (2)-(4) and (37)): (respectively𝜏 𝐚3 . 𝜎,(03, 1) . 𝐚3 𝑅20 ∕4) are (respectively is) smaller

5
E. Pruchnicki, X. Chen and H.-H. Dai Applications in Engineering Science 6 (2021) 100046

than𝑘 𝐚3 .𝜎 (0, 0) . 𝐚3 (respectively 𝜏𝐚3 .𝜎 (0, 0) . 𝐚2 ) and therefore these terms By inserting the rod constitutive relation (32) for (k, n) = (0, 0) into
can be neglected. (49)-(51) and by considering (53), we see that we have six linear alge-
Secondly, by neglecting the last lines of the right hand side of formu- braic equations from which the six components of 𝐮(1, 0) and 𝐮(0, 1) can be
lae (41) and (44) (for consistency similar terms will be neglected below expressed in terms of 𝐮(0, 0) and 𝜃:
and lead to formulas (48)-(50)) and by dividing by 𝑅20 ∕4, we obtain:
⎧ ⎡ ̂̄ . 𝐚 ⎤ ⎫
⎪ ⎡𝐜(1, 1) 𝑇 ⎤ ( 𝐓 𝑐1 )⎥ ⎪
1( ) 1
4 ̂̄ 8 ̂̄ 4 (0, 0) 1, 0 ⎢ ̂ ̂
𝐓𝑐1, 3 . 𝐚3 + 𝐓 0 . 𝐚1 + 𝐟 . 𝐚1 + 𝐟,( 3 ) . 𝐚3 + 𝐟 (2, 0) + 𝐟 (0, 2) . 𝐚1 ⎪ ⎢ (1, 2) 𝑇 ⎥ ⎢ 2 𝐓̄ 𝑐1 . 𝐚2 + 𝐓̄ 𝑠1 . 𝐚1 ⎥
1 ⎪
𝑅02 𝑅03 𝑅0 2 2 [ (1, 0) ] ⎪ ⎢𝐜 ⎪
⎪ (1, 3) 𝑇 ⎥ ⎢ ̂ ⎥ ( )
( ) ( ) 𝐮 ⎢𝐜 ⎥ (0, 0) ⎢ 𝐓̄ . 𝐚 ⎥ 2 ⎪
4 = 𝐆 ⎨− (2, 2) 𝑇 𝐮|3 +
−1 𝑐1 3 + 𝑂 𝑅 0 ⎬,
= 𝐚3 .𝜎 (0, 0)
. −𝑘 𝐚3 + 𝜏 𝐚2 − 𝐚1 . −2𝜏 𝑘 𝜎 (0, 1) + 2 𝑘2 𝜎,(03, 0) . 𝐚3 𝐮(0, 1) ⎪ ⎢⎢𝐜 ⎥ ⎢ ̂̄ . 𝐚 ⎥ ⎪
𝑅0
2
, 𝑇 ⎥ ⎢ 𝐓 𝑠1 2 ⎥
⎪ ⎢𝐜 ( 2 3)
̂̄ . 𝐚 ⎪
( ) ( ) ⎪ ⎣ 0𝑇 ⎥⎦ ⎢ 𝐓 𝑠1
⎥ ⎪
+ 𝐚3 . 𝜏,3 𝜎 (0, 1) − 𝑘,3 𝜎,(03, 0) − 𝜎,(133, 0) . 𝐚3 + 𝑂 𝑅20 , ⎢ ⎥
3
(45) ⎪ ⎣ 2 𝜃 𝑎 ⎦ ⎪
⎩ ⎭
(54)
4 ̂̄
𝐓𝑠1,3 . 𝐚3 +
8 ̂̄ + 𝐟 (0, 0) . 𝐚 + 𝐟 (0, 1) . 𝐚 + 1 (𝐟 (2, 0) + 𝐟 (0, 2) ). 𝐚
𝐚2 .𝐓 in which 𝐆 = ( (1, 1)
𝐚(1, 1) 𝐚(1, 2) 𝐚(1, 3) 𝐚(2, 2) 𝐚(2, 3) 𝐚2 𝑇
) ,
𝑅20 𝑅30
0 2 ,3 3
2 2
𝐛 𝐛(1, 2) 𝐛(1, 3) 𝐛(2, 2) 𝐛(2, 3) −𝐚1
( ) with 𝐚 ( 𝑖, 𝑗 ) = 𝐸𝑖𝑗𝑜1 𝐚0 , 𝐛 ( 𝑖, 𝑗 ) = 𝐸𝑖𝑗𝑜2 𝐚𝑜 , 𝐜 ( 𝑖, 𝑗 ) = 𝐸𝑖𝑗𝑜3 𝐚𝑜 .
4 0, 1 0, 1
= − 𝜏 𝐚3 .𝜎 (0, 0) . 𝐚1 − 𝐚3 . 𝜎,(33 ) . 𝐚3 + 𝑘 𝜎,(3 ) . 𝐚1 By substituting the rod constitutive relation (32) for (k, n) = (0, 0),
𝑅20
( ) ( ) (0, 1), (1, 1) (with the use of recursive relations (39) and (54)) into
−𝐚2 . −𝜏 𝑘 𝜎 (0, 1) . 𝐚3 + 𝑘2 𝜎,(03, 0) . 𝐚3 + 𝑘 𝜎,(13, 0) . 𝐚3 + 𝑂 𝑅20 . (46) Eq. (45)-(48) after dropping the 𝑂(𝑅20 ), we obtain four scalar rod equa-
tions for the leading order coefficient 𝐮(0, 0) (the displacement vector on
By taking into account of (29) and (36) for (k, n) = (0, 1) and (1, 1) the mean line Γ ̃𝑐 ) and the twist angle 𝜃
then we multiply (35)2 by 𝐚2 and substract (35)3 multiplied by𝐚1 , finally ( )
2 ̂̄ 0, 0
we divide by 𝑅20 ∕4 and obtain the scalar equation: 𝐓 0 + 𝐟 (0, 0) . 𝐚3 = −𝐚3 .𝜎,(3 ) . 𝐚3 , (55)
𝑅0
( ) ( )
4 ̂̄ ̂̄ . 𝐚 = −𝐚 . −𝜏 𝜎 (0, 1) . 𝐚 + 𝑘𝜎 (0, 0) . 𝐚 + 𝜎 (1, 0) . 𝐚 + 𝐟 (1, 0)
𝐓𝑐1 . 𝐚2 − 𝐓 𝑠1 ,3 ,3
1( )
1 2 3 3 3
𝑅02
4 ̂̄ 8 ̂̄ 4 (0, 0)
( ) 𝐓𝑐1, 3 . 𝐚3 + 𝐓 0 . 𝐚1 + 𝐟 . 𝐚1 + 𝐟,(13, 0) . 𝐚3 + 𝐟 (2, 0) + 𝐟 (0, 2) . 𝐚1
( ) 𝑅20 𝑅30 𝑅20 2
+ 𝐚1 . 𝜎,(03, 1) . 𝐚3 + 𝐟 (0, 1) + 𝑂 𝑅20 . (47)
( ) ( )
4 0, 0
= 𝐚3 .𝜎 (0, 0) . −𝑘 𝐚3 + 𝜏 𝐚2 − 𝐚1 . −2𝜏 𝑘 𝜎 (0, 1) + 2 𝑘2 𝜎,(3 ) . 𝐚3
We observe that in (47), the torsion term 𝑅0
2

(−𝐚2 .(−𝜏 𝜎 (0, 1) + 𝜎,(13, 0) ) + 𝐚1 . 𝜎,(03, 1) ) . 𝐚3 is the part of the leading ( )

0, 0 1, 0
+ 𝐚3 . 𝜏,3 𝜎 (0, 1) − 𝑘,3 𝜎,(3 ) − 𝜎,(33 ) . 𝐚3 , (56)
order term.
By considering (39) and dropping the error 𝑂(𝑅20 ), we see that (45)-
(47) forms three scalar vector equations with nine unknows which are
the components of the displacement field vectors 𝐮(0, 0) , 𝐮(1, 0) and 𝐮(0, 1) . 4 ̂̄
𝐓𝑠1,3 . 𝐚3 +
8 ̂̄ + 𝐟 (0, 0) . 𝐚 + 𝐟 (0, 1) . 𝐚 + 1 (𝐟 (2, 0) + 𝐟 (0, 2) ). 𝐚
𝐚2 .𝐓 0 2 ,3 3 2
So it remains to find six scalar equations. 𝑅02 𝑅03 2
( )
Eq. (38) projected on a3 has not yet been used so we keep only the 4
= − 𝜏 𝐚3 .𝜎 (0, 0) . 𝐚1 − 𝐚3 . 𝜎,(033, 1) . 𝐚3 + 𝑘 𝜎,(03, 1) . 𝐚1
relative𝑂(𝑅20 )terms into the remainder to obtain the scalar equation : 𝑅20
( ) ( )
2 ̂̄ ( ) −𝐚2 . −𝜏 𝑘 𝜎 (0, 1) . 𝐚3 + 𝑘2 𝜎,(03, 0) . 𝐚3 + 𝑘 𝜎,(13, 0) . 𝐚3 , (57)
𝐓 0 + 𝐟 (0, 0) . 𝐚3 = −𝐚3 .𝜎,(03, 0) . 𝐚3 + 𝑂 𝑅20 . (48)
𝑅0

Now we put into the remainders the𝑂(𝑅20 ) terms for Eqs. (35) 2 mul- ( )
4 ̂̄ ̂̄ . 𝐚 + 𝐟 (1, 0) . 𝐚 − 𝐟 (0, 1) . 𝐚
tiplied by a1 and a3 respectively and also for Eqs. (35) 3 multiplied by 𝐓𝑐1 . 𝐚2 − 𝐓 𝑠1 1 2 1
a2 and a3 respectively: 𝑅02
( ( ) )
̂̄ . 𝐚 = 𝐚 . 𝜎 (0, 0) . 𝐚 + 𝑂(𝑅2 ), 𝐓
̂̄ . 𝐚 = 𝐚 . 𝜎 (0, 0) . 𝐚 + 𝑂(𝑅2 ), = 𝐚2 . 𝜏 𝜎 (0, 1) − 𝑘𝜎,(03, 0) − 𝜎,(13, 0) + 𝐚1 . 𝜎,(03, 1) . 𝐚3 . (58)
𝐓 𝑐1 1 1 1 0 𝑐1 3 3 1 0
(49)
For the ease of variational formulation of (55)-(58), we rewrite them
̂̄ . 𝐚 = 𝐚 . 𝜎 (0, 0) . 𝐚 + 𝑂(𝑅2 ), 𝐓
𝐓 ̂̄ . 𝐚 = 𝐚 . 𝜎 (0, 0) . 𝐚 + 𝑂(𝑅2 ). (50) by using formula of product of derivative and formulae (2)-(4) (for (60),
𝑠1 2 2 2 𝑠1 3 3 2 0, 0)
. 𝐚3 and 𝑘 𝐚1 .𝜎,(13, 0) . 𝐚3 are smaller than
0 0
we use the fact that 𝑘 𝐚3 .𝜎,(33
Finally, we add (35)2 multiplied by 𝐚2 and substract (35)3 multiplied 4 𝑘 (0, 0)
𝜎 and can be neglected):
by 𝐚1 then we take into account of (29) and (36) for (k, n) = (0, 1) and 𝑅20 33

(1, 1) and finally we put into the remainders the 𝑂(𝑅20 ) terms to obtain ( )
2 ̂̄ (0, 0) (0, 0)
the scalar equation: 𝐓 0 + 𝐟 (0, 0) . 𝐚3 − 2𝑘𝜎13 = −𝜎33 ,3
, (59)
𝑅0
̂̄ . 𝐚 + 𝐓
𝐓 ̂̄ . 𝐚 = 2 𝐚 . 𝜎 (0, 0) . 𝐚 + 𝑂(𝑅2 ). (51)
𝑐1 𝑠1
2 1 1 2 0 (( ) ( ) ) ( )
From Ciarlet (1998), it is easy to see that the asymmetric part of the
4
𝐓̂̄ −𝑘 𝐓 ̂̄ +
8 ̂̄
𝐓0
𝑐1 𝑐1
displacement gradient is 𝜔 = 12 (𝑢𝑘, 𝑙 − 𝑢𝑙, 𝑘 )𝐠𝑘 ⊗ 𝐠𝑙 and then becomes on 𝑅20 3,3 1 𝑅30 1
( )
̃𝑐 (𝐲′ = 0):
the mean line Γ 4
+ 𝑓1(0,0) + 𝑓3(1, ,30) − 𝑘 𝑓1(1, 0) +
1 (2,0)
𝑓1 + 𝑓1(0,2)
𝑅02 2
1( )
𝜔= 𝑢 − 𝑢𝑙, 𝑘 𝐚𝑘 ⊗ 𝐚𝑙 . (52) 4 ( ) ( )
2 𝑘, 𝑙 = 𝐚3 .𝜎 (0, 0) . −𝑘 𝐚3 + 𝜏 𝐚2 − 𝑘 𝐚3 .𝜎,(30, 0) . 𝐚3
𝑅0 2 ,3
By taking into account of (2)-(3), Eq. (52) shows that the twist angle ( ) ( )
(0, 1)
at (𝐲′ = 0) is: + 𝜏𝜎33 − 𝐚3 .𝜎,(13, 0) . 𝐚3 (60)
,3 ,3.
( )
1 (1, 0)
𝜃= 𝑢2 − 𝑢(10, 1) . (53)
2

6
E. Pruchnicki, X. Chen and H.-H. Dai Applications in Engineering Science 6 (2021) 100046

(( ) ( ) ) ( )
4 ̂̄ ̂̄ 8 ̂̄ 0, 0 0, 1 0, 1 If we take v to be arbitrary, from (66) we deduce the tridimensional
𝐓 𝑠1 −𝑘 𝐓 𝑠1 + 𝐓 0 + 𝑓2( ) + 𝑓3(,3 ) − 𝑘 𝑓1( )
𝑅20 3 ,3 1 𝑅30 2 boundary condition on Σ𝐿 which cannot be satisfied for a rod theory
( ) obviously. Therefore, we need to specialize the test displacement field
1 2, 0 0, 2
+ 𝑓2( ) + 𝑓2( ) 𝐰∗ at𝑦3 = 𝐿 and the criterion is that the applied force and torque should
2
( ) generate the virtual work. According to this criterion, we take:
4 (0, 0) 0, 1 0, 1
= − 𝜏 𝜎13 − 𝐚3 .𝜎,(3 ) . 𝐚3 + 𝑘 𝐚3 . 𝜎,(3 ) . 𝐚1 ( ) ( )
𝑅20 ,3
𝐰∗ = 𝑣∗𝑖 𝐚𝑖 − 𝑦1 𝑣∗1,3 + 𝑦2 𝑣∗2,3 𝐚3 + −𝑦2 𝐚1 + 𝑦1 𝐚2 𝜃 ∗ . (67)
(0, 1)
−𝜏 𝑘 𝜎23 + 𝑘2 𝐚2 .𝜎,(03, 0) . 𝐚3 + 𝑘 𝐚2 . 𝜎,(13, 0) . 𝐚3 , (61)
Then expansion of the stress tensor (31) and the specially chosen test
displacement 𝐰∗ are taken into the third integral of (66), as this term
(( ) ( ) ) vanishes, we have:
4 ̂̄
𝐓 − 𝐓̂̄ 1, 0 0, 1
+ 𝑓2( ) − 𝑓1( )
𝑐1 𝑠1
𝑅20 2 1 𝜋 𝑅40 ( ( 1, 0) ∗ ( 0, 1) ∗
)
𝜋 𝑅40 ( (1, 0) (0, 1) ( )
)
( ) 𝐿𝑘 𝑣∗𝑘 − 𝜎33 𝑣1, 3 + 𝜎33 + 𝑣2, 3 𝜎32 − 𝜎31 𝐚1 𝑗 𝜃 ∗
(0, 1) (1, 0) (0, 1) 0, 0 4 4
= 𝜏 𝜎23 − 𝜎13 − 𝜎23 − 𝑘 𝐚2 .𝜎,(3 ) . 𝐚3 ( )
( ) = 𝑡̂𝑖 𝑣∗𝑖 + 𝑚̂ 2 𝑣∗1,3 − 𝑚̂ 1 𝑣∗2,3 + 𝑚̂ 3 𝜃 ∗ + 𝑂 𝑅60 . (68)
( ) ( ) (1, 0) (0, 1) (0, 1)
− 𝐚2 .𝜎 (1, 0) . 𝐚3 ,3 + 𝐚1 . 𝜎 (0, 1) . 𝐚3 ,3 + 𝑘 𝜎12 − 𝜎11 + 𝜎13 . (62)
where
( )
The above system consists of two second order equations for 𝑢(30, 0) and ⎛ (2, 0) (0, 2) ⎞
⎜ (0, 0) 𝑅20 𝜎𝛼3 + 𝜎𝛼3 ⎟
𝜃 ((59), (62)) and two fourth-order equations for 𝑢(10, 0) , 𝑢(20, 0) ((60)-(61)). 𝐿𝛼 = 𝜋 𝑅20 ⎜𝜎𝛼3 + 2 (0, 0)
⎟, 𝐿3 = 𝜋 𝑅0 𝜎33 ,
Then at each rod edge six boundary conditions are needed, which will ⎜ 8 ⎟
⎝ ⎠
be addressed in the next section.
𝑡̂𝑖 = 𝑡̄𝑖 𝑑 𝐲′ , 𝑡̄𝑖 = 𝐭̄ . 𝐚𝑖 , 𝑚̂ 1 = 𝑦2 𝑡̄3 𝑑 𝐲′ , 𝑚̂ 2
7. Boundary conditions and unidimensional rod virtual work ∫Σ𝐿 ∫Σ𝐿
principle ( )
=− 𝑦1 𝑡̄3 𝑑 𝐲′ , 𝑚̂ 3 = 𝑦1 𝑡̄ 2 − 𝑦2 𝑡̄ 1 𝑑 𝐲′ .
∫Σ𝐿 ∫Σ𝐿
Now, we shall derive the rod boundary conditions and unidimen-
(2, 0) (0, 2)
sional rod virtual work principle. 𝑅4 (𝜎 +𝜎 )
For 𝐿3 , we neglect 𝜋 0 33 8 33 as already done for establishing
By using Green’s theorem and strong formulation (15)-(18), we ob- Eq. (48).
tain the tridimensional virtual work: Now we show that 𝐿𝑘 in (68) can be simplified. Firstly for 𝐿1 , from
̃ − (35)2 , we see that:
− 𝐷𝑖𝑣 𝜎̃ . 𝐯̃ ∗ 𝑑 𝐁 𝐟̃ . 𝐯̃ ∗ 𝑑 𝐁
̃
∫𝐁̃ ∫𝐁̃
( ∼) ( ∼) 𝑅20
̂̄ . 𝐚 −
𝐚3 .𝜎 (0, 0) . 𝐚1 = 𝐓 𝐚3 .𝜎 (1, 1) . 𝐚2
+ ̃ 𝐧̃ 𝑙𝑎𝑡 − 𝐓̄ . 𝐯̃ ∗ 𝑑 Γ
𝜎. ̃𝑙𝑎𝑡 + 𝜎. ̃ 𝐿 = 0,
̃ 𝐧̃ − 𝐭̄ . 𝐯̃ ∗ 𝑑 Σ (63) 𝑐1 3
4
∫Γ̃𝑙𝑎𝑡 ∫Σ̃𝐿
𝑅20 ( ) ( )
in which the test function vanishes on Σ̃ 0 .
𝐯̃ ∗ − 𝐚3 . 𝜎 (0, 2) + 3 𝜎 (2, 0) . 𝐚1 + 𝑂 𝑅40 . (69)
8
In order to write variational formulation (63), we define the ele-
̃𝑙𝑎𝑡 . Let 𝑠𝑙𝑎𝑡 measure Then applying relation (69) to (68) to replace 𝐚3 .𝜎 (0, 0) . 𝐚1 , together
ment surface of the lateral boundary of the rod Γ
̃ of a cross section Σ̃ (for with the use of the equilibrium Eq. (36) for (k, n) = (1,1) leads to:
the arc length of the boundary (denoted by𝜕 Σ)
̃ is given by:
s∈ [0, 𝐿]), and suppose that the unit tangent 𝜏𝑙𝑎𝑡 to𝜕 Σ 𝜋 𝑅40 ( ( (1, 0) ) )
̂̄ . 𝐚 +
𝐿1 = 𝜋 𝑅20 𝐓
0, 0
𝐚3 . 𝜎, 3 + 𝑘𝜎,(3 ) − 𝜏 𝜎 (0, 1) . 𝐚3 + 𝐟 (1, 0) . 𝐚3 .
𝜏𝑙𝑎𝑡 = 𝜏𝑙𝑎𝑡 1 𝐚1 + 𝜏𝑙𝑎𝑡 2 𝐚2 . (64) 𝑐1 3
4
(70)
We assume that the cross section of the rod is uniform, and then
its lateral boundary is a collection of curve 𝐂 ̃ defined by position Secondly for𝐿2 , from (35)3 , we see that:
vector points X given by (5) such that s∈ [0, 𝐿] and𝐲′ is constant.
𝑅20
Consequently, the oriented differential surface area induced by the ̂̄ . 𝐚 −
𝐚3 .𝜎 (0, 0) . 𝐚2 = 𝐓 𝐚3 .𝜎 (1, 1) . 𝐚1
𝑠1 3
̃𝑙𝑎𝑡 is𝐧̃ 𝑙𝑎𝑡 𝑑 Γ
parametrization (𝑠𝑙𝑎𝑡 , 𝑠) of Γ ̃𝑙𝑎𝑡 = 𝑑 𝑠𝑙𝑎𝑡 𝜏𝑙𝑎𝑡 ∧ 𝑑𝑠 𝐠3 . Then by 4
considering Eqs. (8) and (64), we obtain: 𝑅20 ( ) ( )
− 𝐚3 . 3𝜎 (0, 2) + 𝜎 (2, 0) . 𝐚2 + 𝑂 𝑅40 . (71)
̃𝑙𝑎𝑡 = ‖𝜏𝑙𝑎𝑡 ∧ 𝐠3 ‖ 𝑑 𝑠 𝑑 𝑠𝑙𝑎𝑡 = ‖𝜏𝑙𝑎𝑡 ∧ 𝐠3 ‖𝑑 Γ𝑙𝑎𝑡 ,
𝑑Γ (65) 8
‖ ‖ ‖ ‖
√ Then applying relation (71) to (68) to replace 𝐚3 .𝜎 (0, 0) . 𝐚2 , together
where ‖𝜏𝑙𝑎𝑡 (𝐲′ ) ∧ 𝐠3 (𝐲)‖ = 1 + 2𝐶𝑙𝑎𝑡 1 (𝑦1 , 𝑦3 ) + 𝐶𝑙𝑎𝑡 1 (𝑦1 , 𝑦3 )2 + 𝐶𝑙𝑎𝑡 2 (𝑦) , with the use of the equilibrium Eq. (36) for (k, n) = (0,1) leads to:
with𝐶𝑙𝑎𝑡 1 (𝑦1 , 𝑦3 ) = − 𝑘(𝑦3 ) 𝑦1 , 𝐶𝑙𝑎𝑡 2 (𝑦) = 𝜏(𝑦3 )2 (𝜏𝑙𝑎𝑡 1 (𝐲′ ) 𝑦1 + 𝜏𝑙𝑎𝑡 2 (𝐲′ ) 𝑦2 )2 .
𝜋 𝑅40 ( )
By inserting formula (13) and (65) into variational formulation (64), ̂̄ . 𝐚 +
𝐿2 = 𝜋 𝑅20 𝐓 𝐚3 .𝜎,(03, 1) + 𝐟 (0, 1) . 𝐚3 . (72)
𝑠1 3
we obtain: 4
√ ( ) Finally, by taking 𝐿1 and𝐿2 back to (68) and equating the coefficients
− (𝑑𝑖𝑣 𝜎 + 𝐟 ) . 𝐰∗ 𝑔𝑑𝑦 + 𝜎. 𝐧𝑙𝑎𝑡 − 𝐓̄ . 𝐰∗ ‖ ‖
‖𝜏𝑙𝑎𝑡 ∧ 𝐠3 ‖𝑑 Γ𝑙𝑎𝑡 of 𝐯∗ , 𝑣∗1, 3 , 𝑣∗2, 3 and 𝜃 ∗ on both sides of (68), we get the rod boundary
∫𝐁 ∫Γ𝑙𝑎𝑡
conditions at the traction edge at 𝑦3 = 𝐿:
( )
+ 𝜎. 𝐚3 − 𝐭̄ . 𝐰∗ 𝑑 𝐲′ = 0, (66) 𝐋 = 𝐭̂ (73)
∫Σ𝐿

in which the test function 𝐰∗ (𝐲) = 𝐯̃ ∗ (𝐗(𝐲)) is such that 𝐰∗ (𝐲) = 0 on 𝜋 𝑅40
Σ0 (𝑦3 = 0). 𝐚3 .𝜎 (0, 1) . 𝐚3 = 𝑚̂ 1 , (74)
4
To obtain the rod boundary conditions from (66), we need to special-
ize 𝜎 and v to the one dimensional case by using Taylor expansions for 𝜋 𝑅40
them. Due to equilibrium Eq. (36), the first term of (66) is automatically − 𝐚3 . 𝜎 (1, 0) . 𝐚3 = 𝑚̂ 2 , (75)
4
satisfied up to 𝑂(𝑅50 ). Due to (35), the second term of (66) is automat-
ically satisfied up to 𝑂(𝑅40 ). Then only the last edge integral is left and 𝜋 𝑅40 ( )
boundary conditions will be derived from it. 𝐚3 . 𝜎 (1, 0) . 𝐚2 − 𝜎 (0, 1) . 𝐚1 = 𝑚̂ 3 , (76)
4

7
E. Pruchnicki, X. Chen and H.-H. Dai Applications in Engineering Science 6 (2021) 100046

where the vector force resultant𝐭̂ and the torque resultant 𝐦 ̂ = 𝑚̂ 𝑖 𝐚𝑖 are of our procedure. More precisely we summarize the main idea. To write
supposed to be prescribed. lateral boundary condition we use polar coordinates in cross section to-
In the variational formulation (68), we see that the virtual vectors gether with Fourier series expansion. This leads to seven equations with
, 0) ∗(0, 0)
𝐯∗(0,0) and the virtual scalar functions 𝑣∗(0
1,3
,𝑣2,3 and 𝜃 ∗ vanish at𝑦3 = fifteen unknown coefficients displacement field. Equilibrium equations
0(which is center of gravity ofΣ0 ). Thus the boundary conditions at 𝑦3 = gives supplementary relations between stress coefficients and we show
0, are : that it is possible to obtain a closed system of ten equations with ten
0, 0 0, 0 ̄ unknows. The linear dependence with respect to three second order dis-
𝐮(0, 0) = 𝐮̄ (0, 0) , 𝑢(1,3 ) = 𝛼̄ 1 , 𝑢(2,3 ) = 𝛼̄ 2 , 𝜃 = 𝜃. (77)
placement coefficients is used to eliminate them. Elaborated calcula-
As a consequence, the vector displacement 𝐮̄ (0, 0) , the two rotation tions furnish bending and torsion terms and also lead to asymptotically-
angles 𝛼̄ 1 , 𝛼̄ 2 and the twist angle𝜃̄ are prescribed values for respectively consistent closed three vector equations with three unknwons. By intro-
the vector displacement 𝐮(0, 0) and the scalar functions𝑢(10,3, 0) , 𝑢(20,3, 0) and𝜃. ducing the twist angle as a new unknown, it is possible to eliminate the
We derive the unidimensional rod virtual work principle. Firstly, we two first-order displacement coefficients and we obtain four scalar equa-
multiply each components of the virtual test vectors 𝑣∗𝑖 (for i = 3, 1, 2) tions with the three components of the mean line displacement and the
and the virtual twist angle 𝜃 ∗ to Eqs. (59)-(62) respectively and we in- twist angle as unknows. From the boundary term of the tridimensional
tegrate each side of these equalities between 0 and L. Secondly, we in- variational formulation, we derive six physical and natural boundary
tegrate by parts and we take into account of Neumann type boundary conditions at each edge. Then we formulate the unidimensional rod vir-
conditions (73)-(76) at 𝑦3 = 𝐿 (the virtual scalar functions 𝐯∗ , 𝑣∗1,3 ,𝑣∗2,3 tual work principle. By considering a specific example, comparison with
and 𝜃 ∗ vanish at 𝑦3 = 0): classical Euler-Bernoulli rod theory and Timoshenko rod theory shows
that our theory gives refined results. The method developed here ap-
( ) pears to be general, and we think that it would be applicable to other rod
𝐿 𝐿
(0, 0) ∗ (0, 0) ∗ problems : dynamic, double symmetric cross section, geometric and/or
𝜋 𝑅20 𝜎33 𝑣3,3 𝑑 𝑦3 + 2 𝑘 𝜎13 𝑣3 𝑑 𝑦3
∫0 ∫0 material non linearity, incompressible material, multilayered rod.
𝐿 ( )
2 ̂̄ (0, 0)
= 𝜋 𝑅20 (𝐓 0 )3 + 𝑓 3 𝑣∗3 𝑑 𝑦3 + 𝑡̂3 𝑣∗3 |𝑦 =𝐿 ,
∫0 𝑅0 3 Declaration of Competing Interest
𝜋 𝑅40 𝐿
(1, 0) ∗
𝐴 𝑣∗1 + 𝐵 𝑣∗1,3 − 𝜎33 𝑣1,33 𝑑 𝑦3 = 𝑡̂1 𝑣∗1 |𝑦 + 𝑚̂ 2 𝑣∗1,3 , The authors declare that they have no known competing financial
4 ∫0 3 =𝐿 |𝑦3 =𝐿
interests or personal relationships that could have appeared to influence
𝜋 𝑅40 𝐿
(0, 1) ∗ the work reported in this paper.
𝐶 𝑣∗2 + 𝐷 𝑣∗2,3 − 𝜎33 𝑣2,33 𝑑 𝑦3 = 𝑡̂2 𝑣∗2 |𝑦 − 𝑚̂ 1 𝑣∗2,3 ,
4 ∫0 3 =𝐿 |𝑦3 =𝐿
Funding
𝜋 𝑅40 𝐿
(1, 0) (0, 1) ∗
𝐸 𝜃 ∗ + (𝜎23 − 𝜎13 )𝜃, 3 𝑑 𝑦3 = 𝑚̂ 3 𝜃 ∗ |𝑦3 =𝐿 ,
4 ∫0
Hui Hui Dai and Xiaoyi Chen acknowledge the support by a GRF
4𝑘 ̂̄ 4 (0, 0) 1
𝐰𝐢𝐭𝐡 𝐴 = (𝐓𝑐1 )1 − 𝑓 + 𝑘 𝑓1(1, 0) − (𝑓1(2, 0) + 𝑓1(0, 2) ) grant (Project No.: CityU 11303718) from the Research Grants Coun-
𝑅20 𝑅20 1 2 cil of the Government of HKSAR, P.R. China and a grant (Project No.:
8 ̂̄ 4 𝑘 (0, 0) 4 𝜏 (0, 0) 11702027) from National Natural Science Foundation of China.
− (𝐓 0 )1 − 𝜎 + 𝜎 ,
𝑅30 𝑅20 33 𝑅20 23
Acknowledgement
(1, 0) (0, 1) 4 ̂̄
𝐵 = −2𝑘 𝜎13 − 𝜏𝜎33 + 𝑘 𝐚3 .𝜎,(30, 0) . 𝐚3 + (𝐓𝑐1 )3 + 𝑓3(1, 0) ,
𝑅20
The reviewer comments lead to improvement of this work.
4𝑘 ̂̄ 8 ̂̄ 1
𝐶= (𝐓 𝑠1 )1 − (𝐓 0 )2 − 𝑓2(0, 0) − (𝑓2(2, 0) + 𝑓2(0, 2) )
𝑅20 𝑅30 2 Appendix: Two case studies for respectively isotropic and
4 transversely isotropic material
+ 𝑘 𝑓1(0, 1) − 𝜏 𝜎 (0, 0) + 𝑘 𝐚3 . 𝜎,(03, 1) . 𝐚1
𝑅20 13
For the sake of simplicity, we consider a rod with constant curvature
(0, 1)
−𝜏 𝑘 𝜎23 + 𝑘2 𝐚2 .𝜎,(03, 0) . 𝐚3 + 𝑘 𝐚2 . 𝜎,(13, 0) . 𝐚3 , 𝑘(𝑠) = 𝑘 and without torsion curvature of the mean line Γ̃𝑐 𝜏 = 0 and the
4 ̂̄ volumic force f vanishes. The rod equilibrium Eqs. (59)–(62) become:
𝐷= (𝐓𝑠1 )3 + 𝑓3(0, 1) − 2𝑘𝜎13
(0, 1)
𝐚𝐧𝐝
𝑅20 ( )
(0, 0) (0, 0) 2 ̂̄
−𝜎33 ,3
+ 2𝑘𝜎13 = 𝐓0 ,
4 ̂̄ ̂̄ ) ) + 𝑓 (0, 1) − 𝑓 (1, 0) 𝑅0 3
𝐸= ((𝐓𝑠1 )1 − (𝐓 𝑐1 2 ( ) ( )
𝑅20 1 2 4𝑘 (0, 0)
− 2 𝐚3 .𝜎 (0, 0)
. 𝐚3 − 𝑘 𝐚3 .𝜎,3 . 𝐚3 − 𝐚3 .𝜎,(13, 0) . 𝐚3
𝑅0 ,3 ,3
(0, 1) (1, 0) (0, 1)
+ 𝜏 (𝜎23 − 𝜎13 − 𝜎23 ) − 𝑘 𝐚2 .𝜎,(03, 0) . 𝐚3 (
(0, 1)
)
− 𝐚3 .𝜎, 3 . 𝐚3 + 𝑘 𝐚3 . 𝜎, 3 . 𝐚1 + 𝑘2 𝐚2 .𝜎, 3 . 𝐚3 + 𝑘 𝐚2 . 𝜎,(13, 0) . 𝐚3
(0, 1) (0 , 0 )

− 𝑘 𝐚2 .𝜎,(03, 0) . 𝐚3 + 𝑘 (𝜎12
(1, 0) (0, 1) (0, 1) ,3
− 𝜎11 + 𝜎13 ). (78) (( ) ( ) ) ( )
= 2
4 ̂̄
𝐓 −𝑘 𝐓 ̂̄ 8 ̂̄
+ 3 𝐓 ,
𝑠1 𝑠1 0
It is easy to see that all the edge terms on the right hand side repre- 𝑅0 3 ,3 1 𝑅0 2

( ) ( ) ( )
sent the virtual work done by the force resultant and torque resultant. −𝑘 𝐚2 .𝜎,(03, 0) . 𝐚3 − 𝐚2 .𝜎 (1, 0) . 𝐚3 ,3 + 𝐚1 . 𝜎 (0, 1) . 𝐚3 ,3 + 𝑘 𝜎12 (1, 0) (0, 1)
− 𝜎11 (0, 1)
+ 𝜎13
This weak formulation provides a framework for implementing finite- (( ) ( ) )
= 2 𝐓
4 ̂̄ − 𝐓 ̂̄ . (A1)
element schemes. 𝑐1 𝑠1
𝑅0 2 1

8. Conclusion and discussion One can see the parameter k brings about the coupling between
the stress components. When k = 0, equations (A.1) degenerate to Eqs.
The main purpose of this work is to provide a new asymptotic re- (27) in the paper Chen et al. 2018, which are :
duction method for constructing a consistent curved rod theory for lin- ( )
(0, 0) 2 ̂̄
earized anisotropic elastic material. The starting point of our derivation −𝜎33 ,3
= 𝐓0 ,
𝑅0 3
is a Taylor–Young expansion of the displacement field. Then we consider [ ]
the corresponding expansion of the deformation gradient and the stress ( ) ( )
(1, 0)
−𝜎33 =
4 ̂̄
𝐓 +
2 ̂̄
𝐓 ,
, 33 𝑐1 0
tensor and make some development which are needed for the success 𝑅20 3,3 𝑅30 1

8
E. Pruchnicki, X. Chen and H.-H. Dai Applications in Engineering Science 6 (2021) 100046

[ ] ( )
( ) ( ) 𝜀4 𝜈 2 − 1 𝜀4 ( ) 𝜀4 ( )
(
−𝜎33 )0, 1
=
4 ̂̄
𝐓
2 ̂̄
+ 𝐓0 , + 𝑦̄1 + −1 + 𝜈 + 2𝜈 2 𝑦̄21 + 3 + 𝜈 − 2𝜈 2 𝑦̄22
𝑠1
, 33
𝑅20 3 ,3 𝑅30 2 𝜋 2𝜋 2𝜋
( )
(( ) ( ) ) 𝜀4 −1 + 3𝜈 + 4𝜈 2 3 𝜀4
(0, 1)
𝜎13 (1, 0)
− 𝜎23 =
4 ̂̄
𝐓 − 𝐓̂̄ . (A2) + 𝑦̄1 − (1 + 𝜈)(−5 + 4𝜈)𝑦̄1 𝑦̄22 ,
,3 ,3
𝑅0
2 𝑐1 𝑠1 6𝜋 2𝜋
2 1 ( )
𝜀2 𝜈 ( ) 𝜀4 𝜈(1 + 𝜈) 𝜀4 −1 + 𝜈 + 2𝜈 2
Further, to compare with the classical beam theories (Euler-Bernoulli 𝑢̄ 2 = −1 + 6𝑠̄ − 6𝑠̄2 𝑦̄1 𝑦̄2 + 𝑦̄2 + 𝑦̄1 𝑦̄2
3𝜋 𝜋 𝜋
beam theory and Timoshenko rod theory), consider the isotropic mate- 𝜀 𝜈(
4 ) 𝜀 4
rial with Young’s modulus E and Poisson’s ratio 𝜈, and a line force load- + −3 + 𝜈 + 4𝜈 2 𝑦̄21 𝑦̄2 − (1 + 𝜈)(1 + 4𝜈)𝑦̄32 ,
2𝜋 6𝜋
ing at the top of lateral surface along the negative y1-axis, its value per 𝜀 𝑠̄ ( ) 𝜀3 𝜀3
unit length is represented as 𝑞(𝑦3 ). Then the rod equations from equa- 𝑢̄ 3 = 1 − 3𝑠̄ + 2𝑠̄2 𝑦̄1 + (1 − 2𝑠̄)𝑦̄31 + (1 − 2𝑠̄)𝑦̄1 𝑦̄22 . (A.6)
3𝜋 2𝜋 2𝜋
tions (A.2) are obtained by applying the recursive relations (39) and
where 𝑦̄𝑖 = 𝑦𝑖 ∕𝑅0 , (i = 1, 2), 𝑠̄ = 𝑠∕𝐿 and 𝜀 = 𝑅0 ∕𝐿. The twist angle is
(54) as well as the constitutive relations (32), which are :
𝜃 = 0.
,0) 𝑙𝜈 For the transversely isotropic material, we have the rod equations
𝐸 𝐴 𝑢(0 = 𝑞 ,
3, 33 2𝝅 ,3 similar to equations (A.3), and the differences are E, G in equations
,0) (A.3) being replaced by E3 and G13 respectively. The solutions for the
−𝐸 𝐼 𝑢(0
1, 3333
= 𝑞 + 𝑂(𝑅20 ),
displacement components scaled by q/E (L/R0)4 are
,0)
𝐸 𝐼 𝑢(0
2, 3333
= 0,
1 𝜀2 𝜈31 ( )( )
𝐺 𝐼 𝜃, 33 = 0, (A.3) 𝑢̄ 1 = − 𝑠̄2 (𝑠̄ − 1)2 + −1 + 6𝑠̄ − 6𝑠̄2 𝑦̄21 − 𝑦̄22
6𝜋 6𝜋
[ ] [ ] [ ]
𝜀4 𝐸 𝜀4 𝐸3 ( ) 𝜀4 𝐸3 ( )
+ − 3 + 𝜈31 2
𝑦̄1 + −1 + 𝜈21 + 2𝜈31 2
𝑦̄21 + 3 + 𝜈21 − 2𝜈31
2
𝑦̄22
𝐸 𝜋 𝑅40 𝜋 𝐸1 2𝜋 𝐸1 2𝜋 𝐸1
where 𝐺 = 2(1+𝜈)
is the shear modulus, 𝐼 = is the planar second mo- [ ] [ ]
𝜀4 𝐸3 ( ) 𝜀4 𝐸3 ( )
4
ment of the cross-section, 𝑙 = 2 𝜋 𝑅0 and 𝐴 = 𝜋 𝑅20 are respectively the + −1 + 3𝜈21 + 4𝜈31 2
𝑦̄31 + 5 + 𝜈21 − 4𝜈31 2
𝑦̄1 𝑦̄22 .
6𝜋 𝐸1 2𝜋 𝐸1
circumference and area of the cross section. One can see our theory ac- [ ] [ ]
𝜀2 𝜈31 ( ) 𝜀4 𝐸3 𝜀4 𝐸3 ( )
tually recover the results of the classical rod theories. 𝑢̄ 2 = −1 + 6𝑠̄ − 6𝑠̄2 𝑦̄1 𝑦̄2 + 𝜈21 + 𝜈31
2
𝑦̄2 + −1 + 𝜈21 + 2𝜈31
2
𝑦̄1 𝑦̄2
3𝜋 𝜋 𝐸1 𝜋 𝐸1
We compare (A.3) with two classical rod theories Euler-Bernoulli rod [ ] [ ]
𝜀4 𝐸3 ( ) 1 4 𝐸3 ( )
theory and Timoshenko rod theory. From (A.3), one can see for deflec- + −3 + 𝜈21 + 4𝜈31 2
𝑦̄21 𝑦̄2 − 𝜀 5 + 𝜈21 + 4𝜈31 2
𝑦̄32 .
2𝜋 𝐸1 6𝜋 𝐸1
, 0) [ ]
tion 𝑢(0
1
to the leading order, our theory recover the result of both clas- 1 1 3 𝐸3 ( )
, 0) 𝑢̄ 3 = 𝜀𝑠̄(−1 + 𝑠̄)(−1 + 2𝑠̄)𝑦̄1 + 𝜀 (1 − 2𝑠̄) − 2𝜈31 𝑦̄1 𝑦̄21 + 𝑦̄22 . (A.7)
sical theories. On the other hand, for displacement 𝑢(0
3
, Euler-Bernoulli 3𝜋 4𝜋 𝐺13
rod theory assumes it to be zero and Timoshenko’s theory assumes it to and the twist angle is still 𝜃 = 0.
be zero to the leading order, and our results show that it depends on the
, 0)
derivative of the loading 𝑞(𝑦3 ). Both classical theories assume 𝑢(0 2
to References
be zero while in our rod theory, it depends on the external loadings at
Bermudez, A., Viano, J.M., 1984. Une justification des équations de la thermo-élasticité
two ends. The two classical theories do not give the equation for 𝜃 ex- des poutres à section variable par des méthodes asymptotiques. RAIRO Anal. Numér
plicitly. We conclude that, compared with the two classical rod theories, 18, 347–376. doi:10.1051/m2an/1984180403471.
our rod theory is refined and capable of capturing the rod deformation Carrera, E., Giunta, G., 2010. Refined beam theories based on a unified formulation. Int J
Appl. Mech. 2 (01), 117–143. doi:10.1142/S1758825110000500.
with higher level accuracy and is also applicable in a much wider range
Carrera, E., Petrolo, M., 2012. Refined one-dimensional formulations for laminated struc-
of loadings. ture analysis. AIAA J. 50 (1), 176–189. doi:10.2514/1.J051219.
In the following, we give two examples for isotropic material and Ciarlet, P.G., 1998. Introduction to linear shell theory. In: Lions, J.L., Ciarlet, P.G.
(Eds.), Series in Applied Mathematics I: Gauthier-Villars &. Elsevier, Paris ISBN-10:
transversely isotropic material.
2842990595.
First we give the loading conditions and the material parameters. Ciarlet, P.G., Destuynder, P., 1979. Justification of the two-dimensional linear plate model.
Consider a straight rod subjected to a uniform line force (𝑞(𝑦3 ) is a con- J. Mec. 18 (2), 315–344. doi:10.1016/0045-7825(79)90089-6.
stant) along the negative y1-axis with two clamped ends. Chen, X., Song, Z., Dai, H.H., 2018. Pointwise error estimate for a consistent beam theory.
Anal. Appl. 16 (01), 103–132. doi:10.1142/S0219530516500135.
̂ 𝑞 𝛿(𝜃 )
The line force is represented as 𝐓 = − 𝑅 𝑎 𝐚1 where 𝛿 is the Kro- Chen, X., Dai, H.H., Pruchnicki, E., 2020. On a consistent rod theory for a linearized
1 ∑∞
0 anisotropic elastic material: I. Asymptotic reduction method. Math Mech Solids
1
necker delta function. Since 𝛿(𝜃𝑎 ) = 2𝜋 + 𝜋 𝑚=1 cos 𝑚𝜃𝑎 , taking it into doi:10.1177/1081286520949602.
̂ ̂ Dai, H.H., Song, Z., 2014. On a consistent finite-strain plate theory based on
𝐓, we have the Fourier coefficients of 𝐓 as three-dimensional energy principle. Proc. R Soc. London, Ser A 470 (2171).
doi:10.1098/rspa.2014.0494.
̂ 𝑞 ̂ 𝑞 ̂
𝐓0 = − 𝐚 ,𝐓 = − 𝐚 , 𝐓 = 0, 𝑚 = 1, 2, 3 … … (A.4) Marigo, J.J., Meunier, N., 2006. Hierarchy of one-dimensional models in nonlinear elas-
2𝜋 𝑅0 1 𝑐𝑚 𝜋 𝑅0 1 𝑠𝑚 ticity. J Elast. 83, 1–28. doi:10.1007/s10659-005-9036-y.
Meunier, N., 2008. Recursive derivation of one-dimensional models from
And the boundary conditions from formula (77) are : three-dimensional nonlinear elasticity. Math Mech Solids 13, 172–194.
= 0, 𝑢(10,3, 0) = 0, 𝑢(20,3, 0) = 0, 𝜃 = 0 𝐚𝐭 𝑦3 = 0, 𝐿.
(0, 0) doi:10.1177/1081286506075357.
𝐮 (A.5) Mora, M.G., Müller, S., 2003. Derivation of the nonlinear bending torsion
theory for inextensible rods by -convergence. Calc. Var. 18, 287–305.
For the materials, we consider the isotropic material with parame- doi:10.1007/s00526-003-0204-2.
ter E and 𝜈 (Young’s modulus and Poisson’s ratio), and the transversely Pantz, O., 2000. Dérivation des modèles de plaques membranaires non linéaires à
isotropic material with five parameters 𝐸1 , 𝐸3 , 𝜈21 , 𝜈31 and 𝐺13 . These partir de l’élasticité tridimensionnelle. C R Acad. Sci. Paris 331 (I), 171–174.
doi:10.1016/S0764-4442(00)01604-9.
parameters are interpreted as
𝑛𝑜𝑟𝑚𝑎𝑙 𝑠𝑡𝑟𝑒𝑠𝑠 𝜎𝑗𝑗 Pruchnicki, E., 2009. Two-dimensional nonlinear models for heterogeneous plates. C R
𝐸𝑗 = 𝑛𝑜𝑟𝑚𝑎𝑙 𝑠𝑡𝑟𝑎𝑖𝑛 𝜀 𝑐 𝑎𝑢𝑠𝑒𝑑 𝑏𝑦𝜎
Acad Sci Paris (II), 337 (5), 297–302. doi:10.1016/j.crme.2009.06.018.
𝑗𝑗 𝑗𝑗
Pruchnicki, E., 2011. Derivation of a hierarchy of nonlinear two-dimensional
𝑛𝑜𝑟𝑚𝑎𝑙 𝑠𝑡𝑟𝑎𝑖𝑛 𝜀𝑗𝑗 𝑐𝑎𝑢𝑠𝑒𝑑 𝑏𝑦𝜎𝑖𝑖 𝑠ℎ𝑒𝑎𝑟𝑖𝑛𝑔 𝑠𝑡𝑟𝑒𝑠𝑠 𝜎𝑖𝑗 models for heterogeneous plates. Math Mech Solids 16 (1), 77–108.
𝜈𝑖𝑗 = − and 𝐺𝑖𝑗 = . doi:10.1177/1081286509349965.
𝑛𝑜𝑟𝑚𝑎𝑙 𝑠𝑡𝑟𝑎𝑖𝑛 𝜀𝑖𝑖 𝑐𝑎𝑢𝑠𝑒𝑑 𝑏𝑦𝜎𝑖𝑖 𝑠ℎ𝑒𝑎𝑟𝑖𝑛𝑔 𝑠𝑡𝑟𝑎𝑖𝑛 2𝜀𝑖𝑗 𝑐𝑎𝑢𝑠𝑒𝑑 𝑏𝑦𝜎𝑖𝑗 Pruchnicki, E., Dai, H.H., 2019. New refined models for curved beams in
both linear and nonlinear settings. Math Mech Solids 24 (7), 2295–2319.
For the isotropic material, from equation (A.3) (specifying 𝑞(𝑦3 )= q)
doi:10.1177/1081286518825389.
and equations (A.5), the displacement components scaled by the quan- Reddy, J.N., 2006. Theory and Analysis of Elastic Plates and Shells. CRC Press, Boca Raton,
tity q/E (L/R0)4 are obtained as (denoted by 𝑢̄ 𝑖 ) FL.
Sanchez-Hubert, J., Sanchez-Palencia, E., 1999. Statics of curved rods on ac-
1 2 𝜀2 𝜈 ( )( ) count of torsion and flexion. European J. Mech. A/Solids 18, 365–390.
𝑢̄ 1 = − 𝑠̄ (𝑠̄ − 1)2 + −1 + 6𝑠̄ − 6𝑠̄2 𝑦̄21 − 𝑦̄22 doi:10.1016/S0997-7538(99)00132-1.
6𝜋 6𝜋

9
E. Pruchnicki, X. Chen and H.-H. Dai
Schneider, P., Kienzler, R., Bohm, M., 2014. Modeling of consistent second-order plate
theories for anisotropic materials. ZAMM. Z. Angew. Math. Mech. 94 (1–2), 21–42.
doi:10.1002/zamm.201100033.
Schneider, P., Kienzler, R., 2020. A priori estimation of the systematic error of con-
sistently derived theories for thin structures. Int. J. Solids Struct. 190, 1–21.
doi:10.1016/j.ijsolstr.2019.10.010.
Song, Z., Dai, H.H., 2016. On a consistent dynamic finite-strain plate theory and its lin-
earization. J. Elast. 125 (2), 149–183. doi:10.1007/s10659-016-9575-4.
10
Applications in Engineering Science 6 (2021) 100046
Song, Z., Dai, H.H., 2016. On a consistent finite-strain shell theory based on 3-D nonlinear
elasticity. Int. J. Solids Struct. 97, 137–149. doi:10.1016/j.ijsolstr.2016.07.034.
Trabucho, L., Viano, J.M., 1996. Mathematical modeling of rods. Handbook of Numerical
Analysis, IV. Elsevier, pp. 487–979. doi:10.1016/S1570-8659(96)80006-8.
Yu, X., Fu, Y., Dai, H.H., 2020. A refined dynamic finite-strain shell theory for incom-
pressible hyperelastic materials: equations and two-dimensional shell virtual work
principle. Proc. Roy. Soc. London, Ser. A 476 (2237). doi:10.1098/rspa.2020.0031.