Sie sind auf Seite 1von 30

Acta Mech 227, 3147–3176 (2016)

DOI 10.1007/s00707-015-1529-6

O R I G I NA L PA P E R

L. S. Shankar · S. Rajthilak · U. Saravanan

Numerical technique for solving truss and plane problems


for a new class of elastic bodies

Received: 29 July 2014 / Revised: 19 October 2015 / Published online: 6 January 2016
© Springer-Verlag Wien 2016

Abstract It is customary to use a displacement-based formulation to seek solutions to boundary value problems
for its computational efficacy. In displacement-based formulations, it is convenient to prescribe the constitutive
relation for stress as an explicit function of displacement gradient. However, from a general theoretical point
of view, the stress and displacement gradient could be related by an implicit function. This study develops
techniques to solve boundary value problems when linearized strain and stress are related by an implicit
function. Here both the stresses and the displacement are taken as unknowns. The stress field is constructed such
that it satisfies the equilibrium equations identically within the element and the traction continuity requirements
between the elements. A continuously differentiable displacement field is constructed, and the linearized strain
is computed from this displacement field. Then, the unknown parameters in the stress and displacement field
are estimated such that the constitutive relation holds in weak integral sense. Though in this procedure the
number of unknowns has increased in comparison with the displacement formulation, both the strength and
serviceability condition can be checked directly without any post-processing. Also, in this procedure, both the
equilibrium equations and continuity of displacement are met exactly. The equation that is not satisfied exactly
is the constitutive relation, which is an approximation anyway. The efficacy and accuracy of this method are
benchmarked by studying some standard problems. Planar and three-dimensional truss elements have been
developed and benchmarked. Then, a rectangular plane element is implemented and its performance recorded.
Mathematics Subject Classification 74B20 · 74G15 · 74G70 · 74S30

1 Introduction

Customarily the constitutive equation is written such that the stress is an explicit functional of strain [53].
However, from the work of Morgan [27], it is known that a constitutive equation can be an implicit relation
between kinetic and kinematic quantities. A systematic study on generating general representations for implicit
relations such that the second law of thermodynamics holds along with the restriction due to material frame
indifference and material symmetry has gained impetus since the work of Rajagopal [38,39]. A general
framework has been developed to study viscoelastic processes using implicit relations [47,49]. Considerable
progress has been made to develop a generic representation for the implicit relationship between the stress and
the strain when the response is non-dissipative [10,40,46,48]. As stated by Rajagopal [40,43], it is possible
L. S. Shankar · S. Rajthilak · U. Saravanan (B)
Department of Civil Engineering, Indian Institute of Technology Madras, Chennai 600036, India
E-mail: saran@iitm.ac.in
Tel.: +91-44-22574314

S. Rajthilak
E-mail: rajthilak88@gmail.com
L. S. Shankar
E-mail: shank.300492@gmail.com
3148 L. S. Shankar et al.

to have a nonlinear relationship between the linearized strain and stress in this implicit relation framework
which is not possible in classical elasticity where the stress is written as an explicit function of the strain. This
possibility of having a nonlinear relationship between the linearized strain and stress has proved to be useful
to model materials like concrete [23], or gum metal [43]. An implicit relation framework is required to model
materials like concrete because, despite the axial strain corresponding to the maximum axial load being small
(of the order of 10−3 ), the relationship between axial strain and axial stress is nonlinear. This requirement of
an implicit relation between stress and strain for cement concrete has long been recognized [16,37]. Models
for one-dimensional response of rubber [19] and lung parenchyma [21] have also been developed. An implicit
relation framework has been used to capture the nonlinear viscoelastic response of polymers [8] and asphalt
[18,30,51]. Thus, it can be seen that the use of these implicit relations to model the response of engineering
materials is increasing.
Efforts expended to solve boundary value problems involving an implicit constitutive relation are detailed
next. Homogeneous deformations of isotropic materials undergoing a non-dissipative process have been stud-
ied in some detail [13,41–43]. Inhomogeneous deformations for some special cases of implicit relations have
been studied. Torsional flow [51] and azimuthal flow [18] of a viscoelastic material with implicit constitutive
relation have been solved. Simple inhomogeneous shearing deformations have been investigated [12]. Equa-
tions governing the plane stress and strain case are formulated [11]. Deformations of a right circular annular
cylinder [15,44] and annular sphere [14,45] have been studied. A membrane theory for implicit constitutive
relations has been formulated [22]. The stress distribution in a plate with a hole [28,29] and V notch [26]
is documented. Dynamic effects have also been studied within the context of implicit constitutive relations
[24,25]. Numerical schemes implemented till now to solve boundary value problems involving implicit consti-
tutive relations are based on classical displacement formulation [28,29] or solve the final governing equation
obtained in terms of Airy’s stress potential by finite element formulation [26]. Here an alternative to both these
approaches is sought for reasons detailed below.
The finite element method formulated with the displacement field as the primary unknown requires stress
to be expressed as a function of the displacement gradient. In displacement-based finite element formulation,
one computes the stress field that arises due to a given displacement field and checks whether the equilibrium
equation is satisfied by the resulting stress field. Hence, this formulation is not feasible when one cannot
uniquely determine the stress given a displacement gradient. One encounters a similar predicament in finite
element methods formulated with the stress field as the primary unknown, except that now the displacement
gradient needs to be expressed as a function of stress. Hence, it appears that formulations that use both the
displacement and stress as the primary unknowns are better suited to solve boundary value problems involving
implicit constitutive relations. Such formulations are available in the literature and are called as hybrid or
mixed finite element method.
Both the hybrid and mixed finite element method formulations use more than one field as primary unknown.
While in the hybrid method the final algebraic equations that are solved involve only unknown nodal values
of the displacement field, in the mixed method the algebraic equations involve both the displacement and
stress unknowns [34,35]. However, it should be pointed out that this definition for the hybrid and mixed finite
element method is not universally accepted as clarified by Pian [35].
The hybrid finite element method due to Pian [33,35] uses a stress field that is equilibrating within the
element and satisfies the traction continuity condition between the elements along with a displacement field
which is differentiable within the elements and is continuous across the element boundaries. Then, the com-
plimentary total potential (see [36]) for the element is minimized to express the unknown parameters in the
stress field of the element in terms of the nodal displacements of the element. Finally, the global matrices are
assembled so that the traction and displacement continuity are satisfied. The issue with hybrid methods is in
the construction of stress fields that satisfy the balance of linear momentum.
In this work, restricting to situations where inertial forces can be neglected and body forces could be obtained
from a potential function, stress potentials are used to generate stress fields that satisfy the balance of linear
momentum. Stress potentials have been used extensively for obtaining an analytical solution to boundary
value problems [52,54]. Stress potentials require the generation of C 2 interpolation functions because the
components of the stress depend on the second derivative of these stress potentials, and the traction continuity
across the inter-element boundaries requires the second derivative of the stress potentials to be continuous
across the inter-element boundaries. As part of this work, for a standard square element a C 2 interpolation
function is developed utilizing the suggestion in Chiandussi et al. [17]. Some demerits of this approach are
discussed in Sect. 5.
Numerical technique for solving truss and plane problems 3149

The mixed finite element method is formulated taking both the stress and displacement fields as primary
unknowns. In the mixed finite element method, the prescribed stress field does not a priori satisfy the balance
of linear momentum equation. The stress and displacement fields corresponding to a given body force and
boundary condition are determined by minimizing the Hellinger–Reissner potential. It is also known that the
Hellinger–Reissner potential used in the mixed finite method does not have a minimum but rather a saddle point
[9]. Hence, for the method to be stable it has to satisfy Brezzi’s theorems [9] or the Ladyzhenskaya–Babuska–
Brezzi condition [5]. This in turn restricts the finite element spaces used to approximate the displacement and
stress fields. Stable mixed triangular elements have been developed for second-order elliptic problems [1,50]
and plane elasticity problems [2–5]. The book by Boffi et al. [6] provides a good overview of the mixed finite
element methods, and the elements developed using it. Some of these elements [2] do not even satisfy the
symmetry requirement of the stress tensor strongly. Further, Xue et al. [55] relaxed the requirement that the
displacement and traction be continuous at the interface and report that the resulting scheme is free from zero
energy modes. While the solution proposed in PEERS element [2], Xue et al. [55] might be mathematically
admissible; their appropriateness from a physical stand-point is debatable.
Physical requirements, balance of linear and angular momentum, which are known to be inviolate, are not
satisfied point-wise in the existing mixed finite element methods. On the other hand, the constitutive relation,
which is an approximation, is satisfied at each and every point. In this work, though the approximation for the
stress and displacement fields is like that of Pian [33], the constitutive relation is not required to hold point
wise, as has been the practice in the literature till now. Thus, here while the physical requirements are satisfied
point-wise, the constitutive relation holds only in a weak integral sense.
Here the boundary value problem is posed in the following manner: Of all the divergence-free stress fields
that satisfy the traction boundary condition and differentiable displacement fields satisfying displacement
boundary conditions, one needs to find the stress and displacement field that satisfy the given constitutive
relation. Since practically it is possible to examine only a subset of feasible stress and displacement fields, the
constitutive relation is not required to hold exactly. In this work, the Frobenius norm of the implicit constitutive
relation integrated over the domain of the body is minimized. It can be readily verified that if and only if all
the 6 equations of the constitutive relation are satisfied the Frobenius norm would be zero.
This technique of minimizing the governing equation in least square sense has been adopted in the literature
(see [7,20,31,32] and the references therein). Of concern in the least square formulation is the convergence rate
and conditioning number of the system of algebraic equations. The scope of this work is limited to presenting
only numerical illustrations and establishing the methods’ viability. A detailed mathematical analysis of the
proposed numerical technique is delegated to a future work in this area.
Toward studying the performance of the proposed numerical scheme for a variety of problems, an axial
element and a rectangular plane element are formulated. The number of plane and space trusses is analyzed
using the developed axial element. Using the rectangular element, the beam bending problem is solved. These
results show that the proposed scheme is worthy of further investigation.
This article is organized into five Sections including this Introduction. In the next Section, the general
boundary value problem and its numerical solution using the new formulation is presented. In Sect. 3, the
axial element is formulated and sample trusses solved using this element when the members are made of either
linear elastic or nonlinearly elastic material. In Sect. 4, a plane rectangular element is formulated. Using this
element, some standard problems are solved and the solution validated. A brief discussion of the proposed
algorithm is presented in Sect. 5. The article concludes with a brief summary of the results obtained.

2 General formulation

Here the interest is to find the stress, σ , and displacement field, u, in a body subjected to no body forces and
in static equilibrium under the action of some boundary traction and constraint on the boundary displacement.
The governing equations that the stress and displacement field should satisfy are:
div(σ ) = 0, ∀x ∈ B, (1)
σ n = t̂(n) , ∀x ∈ ∂ Bσ , (2)
u(x) = ûb , ∀x ∈ ∂ Bu , (3)
f(σ , ) = 0, ∀x ∈ B (4)
where the linearized strain, , is defined as
 = [grad(u) + grad(u)t ]/2, (5)
3150 L. S. Shankar et al.

f denotes the constitutive relation between the stress and strain, B denotes the region of the Euclidean space
occupied by the body in the deformed configuration, ∂ Bσ denotes the boundary of the body in the deformed
configuration where the traction boundary condition is specified, ∂ Bu denotes the boundary of the body in the
deformed configuration where the displacement boundary condition is specified, n is the unit normal to the
boundary of the body at x, the boundary traction is denoted by t̂(n) , a known function of x, and the boundary
displacement is denoted by ûb , a known function of x. It is required that ∂ Bσ ∪ ∂ Bu = ∂ B, where ∂ B denotes
the entire boundary of the body in the deformed configuration.
For the stress field to satisfy the equilibrium equations, (1), it has to be divergence free. For the stress field
to be divergence free, using the result in Truesdell and Toupin [54] (section 227), the Cartesian component of
the stress is prescribed as
⎛ ∂2Φ ∂ 2 Φ2 ∂ 2 Φ1 ∂ 2 Φ2

1
+ + ψ 1 (y, z) − ∂ − ∂
⎜ ∂y ∂z
2 2 x∂ y x∂z

⎜ ⎟
− ∂∂ x∂Φy1 ∂ 2 Φ1 ∂ 2 Φ3 ∂ 2 Φ3
2
σ =⎜ ∂x
+ ∂z
+ ψ 2 (x, z) − ∂ ⎟ (6)
⎝ 2 2 y∂z ⎠
− ∂∂ x∂z
Φ2
− ∂∂ y∂zΦ3 ∂ Φ2
+ ∂∂ yΦ23 + ψ3 (x, y)
2 2 2 2
∂x2

where Φi = Φ̂i (x, y, z) and ψi are to be determined functions. In order to keep the number of unknowns to
a minimum, it is assumed that Φ1 = Φ(x, y), Φ2 = Φ3 = 0, ψ1 = ψ2 = ψ3 = 0. Various choices for these
stress functions would result in different kinds of elements. The choice made here results in a special plane
element. Special since ψi ’s are assumed to be zero. The choice made for these stress functions determines the
class of boundary value problems that can be solved.
Hence, taking the displacement field, u, and the stress function, Φ, as the primary variables in the formu-
lation, they are expressed in terms of basis functions as:


N 
M
u= d ξ (x) + Ξ (x),
i i
Φ= φ i ϕbi (x) (7)
i=1 i=1

where d i and φ i are yet to be determined constants, N and M are specified integers, Ξ (x) is a vector-valued
vector function and is such that it satisfies the displacement boundary conditions (3) ξ i (x) is a vector-valued
vector function which represents the basis functions for the displacement field, ϕbi (x) is a scalar-valued vector
function and is the basis function for the stress potential.
It is required that ξ i (x) = 0 ∀ x ∈ ∂ Bu so that the displacement degrees of freedom, d i ’s, are all independent.
In order to be able to compute the linearized strain from the displacement field, the functions ξ i and Ξ should
be C 1 (B). However, in this work, these basis functions are C 1 (B E ) over an element, and the continuity of the
displacement field is only enforced at the element boundaries. Here B E denotes the domain of an element.
Consequently, the constructed basis function for the displacement field is C 0 (B). This strategy is adopted for
implementation so as to keep the number of unknown degrees of freedom to a minimum.
The components of the stress field being related to second derivative of Airy’s stress potential, ϕbi ’s, should
be C 2 (B), and traction continuity conditions across the inter-element boundaries require that the second
derivative of the stress potential be continuous across the elements. The stress potential degrees of freedom,
φ i , are not independent as a system of linear equations relate the φ i ’s by virtue of them having to satisfy the
boundary conditions (2) and (3). It is pertinent to point out that, on the part of the boundary where displacement
is specified, an unknown traction acts, and hence, the restriction on φ i ’s corresponding to elements whose one
of the sides’ displacement is specified, would be in terms of these yet to be determined traction components.
This approach of the traction boundary condition constraining the values that φ i ’s could realize is taken because
the traction boundary condition places restrictions only on the second derivatives of the stress potential; it is
found to be difficult to identify the constraints which the stress potential basis functions should satisfy so that
φ i ’s could be independent.
For the assumed displacement and stress potential (7) the linearized strain and stress are computed from
Eqs. (5) and (6), respectively. For an assumed value for the degrees of freedom, d i and φ i such that the
constraint placed by the boundary condition on φ i ’s holds, the stress and linearized strain are computed and
substituted in the constitutive relation (4) and the following estimated:

δ= f(, σ ) · f(, σ )dv (8)


v
Numerical technique for solving truss and plane problems 3151

where δ represents the volume integral of the Frobenius norm of the implicit constitutive relation. If the solution
to the boundary value problem exists in the space spanned by the basis functions, one can find d i and φ i ’s
such that δ = 0. However, it is unlikely that the solution to the boundary value problem exists in the space
spanned by the basis functions, in which case one would seek the best approximation of the solution in the
space spanned by the basis functions. Thus, the degrees of freedom d i and φ i are sought such that it minimizes
the value of δ and satisfies the constraints on φ i ’s placed by the boundary conditions. Since the integrand in
Eq. (8) is positive, the minimum possible value of δ is zero. Therefore, by looking at the value of δ one can
judge on the quality of the obtained solution.
In this work, for illustration, these basis functions are constructed using the finite element framework.
Similarly, the constraint minimization is done using sequential quadratic programming. To maintain generality,
the integration in (8) is performed using Gauss quadrature.
Using this framework, two elements have been developed and sample problems solved and solutions
benchmarked. These are detailed in the following Sections.

2.1 Constitutive relations

Next, the two constitutive relations considered in this study for illustration are documented. The first model
studied is a linear model, the classical isotropic Hooke’s law:
ν (1 + ν)
 = −K 11+ σ (9)
E E
where K 1 = tr (σ ), and ν and E are constant material parameters, the Poisson’s ratio and Young’s modulus,
respectively. Here the value of these material parameters is assumed as: E = 20 GPa and ν = 0.3.
The second model studied is a nonlinear model that captures the uniaxial compression response of M20
plain concrete:
β
 = K 1 β1 1 + β2 K 2 3 + β4 σ (10)
where K 2 = tr (σ 2 ), βi ’s are constant material parameters. The values of the material parameters in this model
are assumed to be: β1 = −1.0×10−5 (MPa)−1 , β2 = 4.3×10−13 (MPa)−7 , β3 = 3, β4 = 6.0×10−5 (MPa)−1 .
The value of the material parameters is chosen such that Young’s modulus in the linear model is tangential
to the uniaxial stress strain curve of the nonlinear model at zero stress. It is clarified that the model (10) is
not proposed here as the model for ordinary strength concrete; it just happens that the model (10) curve fits
the uniaxial response of M20 concrete. The use of the nonlinear model (10) is limited to an illustration of the
numerical scheme.

3 Axial element

The above-formulated general framework is implemented for the analysis of planar and space trusses when
the constitutive relation for the members is given as an implicit function of stress and linearized strain. Using
the developed program, both statically determinate and indeterminate trusses could be solved. Advantage is
taken of the fact that the element is one-dimensional and that the axial stress would be constant in the element
for it to satisfy the equilibrium equations (1).

3.1 Formulation of axial element

In this formulation, the magnitude of the force in the members, Fm , the support reactions, Fnreact , and the
unknown node displacements, un , are taken as basic variables.
The force equilibrium equations at jth node are written in terms of the force in the members meeting at
that node point and the external forces acting at that node which includes the yet to be determined support
reactions. Combining these force equilibrium equations for all the nodes in the structure, one obtains
Known 
F
[B]{Fm } + nreact = {0} (11)
Fn
where B is the direction cosine matrix and FnKnown is the known externally applied load.
3152 L. S. Shankar et al.

Equation (11) could be rearranged to obtain:


 Known   Known 
  Fm Fn  agm  Fm F
BB react + = B + n = {0} (12)
Fnreact 0 Fnreact 0
where Breact is a matrix of zeros and ones and 0 is a vector of zeros of length equal to the number of support
reaction forces. If the jth node is restrained from translating along the x direction in a plane truss by application
of a kth support reaction force, then Breact would have an entry of one at 2 j −1 row and kth column. It should be
clarified that this is a body-level equation and not element-level equation. It tells the relationship between the
element unknowns and the applied forces. Hence, it is a statement of the traction and displacement boundary
condition. Also notice that the support reaction forces, tractions required to enforce a given displacement
boundary condition, also enter these equations.
The requirement of geometric compatibility relates the elongation of the members, Δm , with the node
displacements by the equation 
u
{Δm } = −[B]t ns (13)
un
where uns represents the displacement of the supports. Unless there is support settlement uns = 0, Eq. (13) is
obtained based on the assumption that the nodal displacements are small so that the change in the orientation
of the members due to the nodal displacement can be ignored.
Consistent with the idea proposed in this work, for an assumed value of Fm , Fnreact and un consistent with
the requirement (12), the implicit constitutive relation is evaluated and the following computed:
M 
δ= [ f i (Fim /Ai , Δim /L i )]2
i=1

M 
   i 2
  f i Fi /Ai , − [B]t un
= /L i (14)
m uns
i=1

where f i is the constitutive function for the ith member, Fim is the axial force in the ith member, Δim is the
elongation of the ith member computed using Eq. (13), Ai is the cross-sectional area of the ith member, and
L i is the length of the ith member. This is the one-dimensional analogue for Eq. (8). As discussed in the
general formulation, the value of δ is minimized subject to the constraint (12) using the sequential quadratic
programming (SQP) algorithm implemented in MATLAB. In order to make the algorithm computationally
efficient and accurate, the derivative of δ with respect to its degrees of freedom is computed and programmed.
For the initial guess to start the SQP algorithm, the displacement degrees of freedom are assumed to be zero
but the member forces and the reaction forces are assumed to be equal to the maximum applied force.
Before proceeding further, a couple of comments are in order. If the configuration of the truss is such that it
or parts of it can be rigid-bodily displaced, then the value of δ would be large suggesting no feasible solution;
otherwise, it can be made as small as required. In case of a statically determinate truss Bagm would be invertible
that the member forces and support reactions can be determined from (12) and the minimization has to be carried
out only over un . However, to maintain generality of the developed program, this approach is not adopted here.

3.2 Validation of axial element

Numerical examples of statically determinate and indeterminate trusses are presented next to validate the
formulation and show its efficacy. For illustration, two constitutive relations documented in Sect. 2.1 are
considered. The linear model (9) for the uniaxial state of stress, with σ = Fmi /Ai ea ⊗ ea , reduces to
Δim Fi
i
= mi (15)
L EA
where Δim is the elongation of the member along direction ea . Similarly, the nonlinear model (10) for the
uniaxial state of stress simplifies to:
  i 2β3 
Δim Fmi Fm
= i β 1 + β4 + β2 . (16)
Li A Ai
Numerical technique for solving truss and plane problems 3153

(a) (b)

Fig. 1 Plane trusses (circled number denotes the member number and numbers without circle node numbers). a Statically
determinate truss. b Internally statically indeterminate truss

All the members in a structure are assumed to be made of the same material and hence have the same model
and material parameters. The values of the material parameters are as documented in Sect. 2.1.

3.2.1 Example 1: Statically determinate truss

A statically determinate truss with geometry and loading as shown in Fig. 1a is analyzed using the same linear
model (15) for the constitutive relation of its members. The cross-sectional area for all members is taken as
2000 mm2 . The percentage error in the numerical solution when compared to the analytical solution is given in
Table 1 for the member forces, node displacements, and the support reactions. It can be seen from Table 1 that
the difference between the numerical solution and the analytical solution is less than the machine precision,
suggesting that the numerical solution obtained is the same as that of the analytical solution.
Next, the same determinate truss shown in Fig. 1a is analyzed for the nonlinear model (16). The cross-
sectional area of all members is assumed to be 2000 mm2 . The actual value of the member forces, node
displacements and support reactions obtained for both the linear and nonlinear models is given in Table 2
for this statically determinate truss. As one would expect in a statically determinate truss, the member forces
and support reactions are identical for both the models. Young’s modulus is assumed such that the linear
model is tangential to the uniaxial stress strain curve of the nonlinear model at zero stress. Hence, the node
displacements obtained for the nonlinear model are larger than those calculated for the linear model because
the strain developed in the nonlinear model for a given stress level is larger than that of the linear model.

Table 1 Statically determinate truss made of members which obey Hooke’s law: percentage error between the numerical and
analytical solution

SQP
Member forces (×10−20 )
F1 0
F2 −3
F3 −1.6
F4 0
F5 0
Node displacements (×10−32 )
u 2x 0
u 3x 0
y
u3 −9
u 4x 0
y
u4 0
Support reactions (×10−14 )
F1x 0
y
F1 5.3
y
F2 0
A subscript denotes the member number or node number as appropriate, and a superscript denotes the component
3154 L. S. Shankar et al.

Table 2 Statically determinate truss made of members which obey Hooke’s law (Linear) or the nonlinear model (16)

Linear Nonlinear
Member forces (kN)
F1 25 25
F2 0 0
F3 0 0
F4 −31.25 −31.25
F5 −31.25 −31.25
Node displacements (mm)
u 2x 2.50 2.58
u 3x 5.63 6.10
y
u3 0 0
u 4x 5.63 6.10
y
u4 −2.34 −2.64
Support reactions (kN)
F1x −25 −25
y
F1 31.25 31.25
y
F2 18.75 18.75
A subscript denotes the member number or node number as appropriate, and a superscript denotes the component

Table 3 Internally statically indeterminate truss made of members which obey Hooke’s law: percentage error between the
numerical and analytical solution

SQP
Member forces (×10−6 )
F1 −2
F2 0
F3 4
F4 0
F5 1.6
F6 −3.2
Node displacements (×10−8 )
u 2x −2
u 3x −1.7
y
u3 0
u 4x 0
y
u4 0
Support reactions (×10−10 )
F1x 0
y
F1 5
y
F2 0
A subscript denotes the member number or node number as appropriate, and a superscript denotes the component

3.2.2 Example 2: Internally statically indeterminate truss

Next, an internally statically indeterminate truss with geometry and loading as shown in Fig. 1b is analyzed.
The material and geometrical properties are taken as given in the previous example. The percentage error in
the numerical solution when compared to the analytical solution is given in Table 3 for the linear material
model.
The actual value of the member forces, node displacements and support reactions obtained for both the
linear and nonlinear models is given in Table 4 for this statically indeterminate truss shown in Fig. 1b. As one
would expect in the internally statically indeterminate truss, only the support reactions are identical for both
the models.

3.2.3 Example 3: Externally and internally statically indeterminate truss

Next, the truss shown in Fig. 2 is analyzed. The material and geometrical properties remain the same as in the
previous Section. The percentage error in the numerical solution when compared to the analytical solution is
given in Table 5 for the linear material model (15).
Numerical technique for solving truss and plane problems 3155

Table 4 Internally statically indeterminate truss made of members which obey Hooke’s law (Linear) or the nonlinear model (16)

Linear Nonlinear
Member forces (kN)
F1 16.67 17.60
F2 −6.25 −5.55
F3 −8.33 −7.40
F4 −37.5 −36.80
F5 −20.83 −22.00
F6 10.42 9.25
Node displacements (mm)
u 2x 1.67 1.77
u 3x 1.98 1.76
y
u3 −0.47 −0.42
u 4x 2.81 2.50
y
u4 −2.81 −3.68
Support reactions (kN)
F1x −25 −25
y
F1 31.25 31.25
y
F2 18.75 18.75
A subscript denotes the member number or node number as appropriate, and a superscript denotes the component

Fig. 2 Statically indeterminate truss ( a circled number denotes the member number and numbers without circle node numbers)

Table 5 Statically indeterminate truss made of members which obey Hooke’s law: percentage error between the numerical and
analytical solution

SQP
Member forces (×10−9 )
F1 0
F2 1.9
F3 0
F4 2.7
F5 4.4
F6 1.1
F7 0
Node displacements (×10−8 )
u 2x 6.1
u 3x 2.5
y
u3 4.2
x
u4 0
y
u4 0
Support reactions (×10−10 )
F1x 0
y
F1 0
y
F2 0
F5x 0
y
F5 5
A subscript denotes the member number or node number as appropriate, and a superscript denotes the component
3156 L. S. Shankar et al.

Table 6 Statically indeterminate truss made of members which obey Hooke’s law (Linear) or the nonlinear model (16)

Linear Nonlinear
Member forces (kN)
F1 15.53 17.28
F2 −8.20 −7.30
F3 −10.93 −9.73
F4 −39.45 −38.55
F5 −17.58 −19.08
F6 13.67 12.17
F7 1.47 2.01
Node displacements (mm)
u 2x 0.92 0.96
u 3x 1.69 1.31
y
u3 −0.29 −0.18
u4 x 2.35 1.82
y
u4 −2.63 −3.80
Support reactions (kN)
F1x −26.47 −27.01
y
F1 31.25 31.25
y
F2 18.75 18.75
F5x 1.47 2.01
y
F5 0 0
A subscript denotes the member number or node number as appropriate, and a superscript denotes the component

Fig. 3 Compound truss (circled number denotes the member number and numbers without circle node numbers)

The actual value of the member forces, node displacements and support reactions obtained for both the
linear and nonlinear models is given in Table 6 for this statically indeterminate truss. Since it is externally
indeterminate, there is variation in the support reactions, too, as expected.

3.2.4 Compound truss

The solution for the compound plane truss as shown in Fig. 3 is obtained from the standard displacement
method and compared with solutions obtained with the proposed SQP-based method. The constitutive relation
for all the members of the truss is assumed to be the linear model (15), and the cross-sectional area is taken as
5000 mm2 .
As can be seen from Table 7, there is negligible error in the member forces, support reactions and node dis-
placements calculated by the proposed method. This shows that the proposed algorithm can handle compound
trusses. The same compound truss is analyzed using the nonlinear model (16). The results are given in Table 8.
As expected since the compound truss is statically determinate, the member forces and support reactions for
both the linear and nonlinear cases are the same and the node displacements are correspondingly larger for the
nonlinear case as it should be because of the reasons explained before.

3.2.5 Space truss

Using the same formulation, a space truss element is implemented in MATLAB. The space truss having a
geometry and loading as shown in Fig. 4a is analyzed for both the linear and nonlinear models discussed
above. The area of cross section of the members is 2 × 104 mm2 . The vertical deflection of the node where
Numerical technique for solving truss and plane problems 3157

Table 7 Compound truss made of members which obey Hooke’s law: percentage error between the numerical and analytical
solution
SQP
Member forces (×10−6 )
F1 30.7
F2 9.9
F3 0
F4 9.9
F5 30.4
F6 9.9
F7 31.3
F8 9.6
F9 31.3
Node displacements (×10−6 )
u 2x −0.4
y
u2 −2.6
u 3x −1.4
y
u3 −3.4
u 4x −1.6
y
u4 5.6
u 5x 3.9
y
u5 −1.6
u 6x −0.4
Support reactions (×10−8 )
F1x 2.2
y
F1 0
y
F6 0
A subscript denotes the member number or node number as appropriate, and a superscript denotes the component

Table 8 Compound truss made of members which obey Hooke’s law (Linear) or the nonlinear model (16)

Linear Nonlinear
Member forces (kN)
F1 −156.52 −156.52
F2 −70.71 −70.71
F3 −100 −100
F4 −70.71 −70.71
F5 −67.08 −67.08
F6 70.71 70.71
F7 22.36 22.36
F8 70.71 70.71
F9 −22.36 −22.36
Node displacements(mm)
u 2x 19.35 46.67
y
u2 −13.59 −58.92
u 3x 12.58 28.36
y
u3 −8.22 −42.12
u 4x 11.57 26.81
y
u4 −7.32 −22.27
u 5x 6.57 16.60
y
u5 −10.90 −30.97
u 6x 25.03 75.01
Support reactions (kN)
F1x 0 0
y
F1 80 80
y
F6 20 20
A subscript denotes the member number or node number as appropriate, and a superscript denotes the component

the load is applied as a function of the load is plotted in Fig. 4b. Since Young’s modulus value in the linear
model is chosen such that it is equal to the slope of the tangent of the nonlinear model at zero stress, the load
deflection plot should be as seen in Fig. 4b.
3158 L. S. Shankar et al.

(a) (b)

Fig. 4 Space truss ( a circled number denotes the member number and numbers without circle node numbers). a Space truss
structure. b Load versus vertical displacement of node 4

Fig. 5 Schematic mapping of a rectangular element to a standard square

4 Plane rectangular element

The general formulation outlined in Sect. 2 is specialized for a two-dimensional rectangular element. The
element is developed for a plane state of stress and uses only a two-dimensional constitutive relation. The
use of two-dimensional constitutive relations ensures that a plane state of stress would also result in a plane
state of strain and there would be no out-of-plane displacement. As in standard finite element analysis, the
C 0 displacement field basis functions for the four-node rectangular element are generated from the bilinear
Lagrange interpolation functions. The C 2 basis functions for the stress potential are constructed using an
incomplete fifth-order polynomial. The calculations are performed on a standard element with the mappings
between the actual and standard element given by the same bilinear Lagrange interpolation functions used for
constructing the displacement basis functions. The details are presented below.

4.1 Geometric mapping

It is found that for computational ease it is advantageous to map rectangles into a standard square of side 2
units centered about the origin (see Fig. 5). This mapping is done using the bilinear Lagrange interpolation
functions, Ni , as
 4 
4
x= Ni (s, t)xi , y = Ni (s, t)yi (17)
i=1 i=1
where (xi , yi ) are the coordinates of the ith node in the rectangle, (s, t) the coordinates of a point in the
standard square and (x, y) the coordinates of the point in the actual rectangle. Here the Ni ’s are
(1 − s)(1 − t) (1 + s)(1 − t)
N1 = , N2 = ,
4 4
(1 + s)(1 + t) (1 − s)(1 + t)
N3 = , N4 = . (18)
4 4
Numerical technique for solving truss and plane problems 3159

The Jacobian matrix of the mapping (17) is defined as


 
∂x ∂x
J= ∂s ∂t . (19)
∂y ∂y
∂s ∂t

It can easily be seen that the transformation function (17) for the mapping of a rectangular element to the
standard square shown in Fig. 5 simplifies to
ls bt
x = xc + , y = yc + (20)
2 2
where (xc , yc ) are the coordinates of the centroid of the rectangle, l is the length of the rectangular element
along the x direction, and b is the length of the rectangular element along the y direction.

4.2 Strain Field

The x and y component of the displacement is obtained by bilinear Lagrange interpolation. Since the rectangle
has four nodes with each node having two displacement degrees of freedom, corresponding to the x and y
component of the displacement, each element has 8 displacement degrees of freedom. Thus,
 
N1 0 N2 0 N3 0 N4 0
{u} = {uij } (21)
0 N1 0 N2 0 N3 0 N4
y1 y2 y3 y4
where uij = {u 1x1 , u 2 , u 3x2 , u 4 , u 5x3 , u 6 , u 7x4 , u 8 }T are the displacement degrees of freedom, and Ni ’s are
the bilinear Lagrange interpolation functions, same as those given in Eq. (18).
The linearized strain field is constructed from the displacement (u) as
⎧ ⎫
⎧ ⎫ ⎪ ∂u x ⎪
⎨ x x ⎬ ⎨ ⎪ ∂ x ⎪

∂u y
{} = yy = ∂y . (22)
⎩2 ⎭ ⎪ ⎪ ⎪
xy ⎩ ∂u x + ∂u y ⎪

∂y ∂x

Using Eqs. (21), (18), (17) and chain rule, the strain field given by Eq. (22) is computed as

{} = [Bd ]{uij } (23)


where ⎡ t−1 ⎤
2l 0 − t−1
2l 0 t+1
2l 0 − t+1
2l 0
⎢ ⎥
Bd = ⎣ 0 s−1
2b 0 − s+1
2b 0 s+1
2b 0 2b ⎦ .
− s−1
s−1
2b
t−1
2l − s+1
2b − t−1
2l
s+1
2b
t+1
2l − s−1
2b − t+1
2l

4.3 Stress field

A plane stress field is obtained from Airy’s stress potential, Φ = Φ̂(x, y), such that
⎧ ⎫
⎧ ⎫ ⎪ ∂ 2 Φ2 ⎪

⎨σx x ⎬ ⎨ ∂ y ⎪ ⎬
[σ ] = σ yy = ∂∂ xΦ2
2
(24)
⎩σ ⎭ ⎪ ⎪ ⎪
xy ⎩ −∂ 2 Φ ⎪

∂ x∂ y

satisfies the static equilibrium equations in the absence of body forces. In order for the traction to be continuous
at the interface of the elements point-wise, the potential has to be a C 2 function. Therefore, each node of the
rectangle has six stress degrees of freedom, namely: φ, φ,x , φ,y , φ,x x , φ,x y and φ,yy . Here φ,x denotes the
derivative of φ with respect to x. Since the rectangle under consideration has 4 nodes, there are in total 24
degrees of freedom for the rectangular element.
3160 L. S. Shankar et al.

The stress potential Φ in an element is written as


24
Φ= ϕ j (x, y)φ j (25)
j=1

where ϕ j ’s are the interpolation functions for the stress field, and φ j ’s are the 24 stress degrees of freedom for
each element.
As before, for computational efficiency the interpolation functions for the stress potential are also con-
structed over a standard square of side 2 units centered about the origin. Following the suggestion made in
Chiandussi et al. [17], the interpolation functions for the stress are computed using an incomplete polynomial
of the form
j j j j j j j j j j
ϕ j (s, t) = a0 + a1 s + a2 t + a3 s 2 + a4 t 2 + a5 st + a6 s 3 + a7 s 2 t + a8 st 2 + a9 t 3
j j j j j j j
+ a10 s 4 + a11 s 3 t + a12 s 2 t 2 + a13 st 3 + a14 t 4 + a15 s 4 t + a16 s 3 t 2
j j j j j j j
+ a17 s 2 t 3 + a18 st 4 + a19 s 5 t + a20 s 3 t 3 + a21 st 5 + a22 s 5 + a23 t 5 (26)
j
where ai ’s are constants. For example, ϕ1 is computed as

1 
ϕ1 = (s − 1)(t − 1)(3s 4 + 3s 3 + 2s 2 t 2 + 2s 2 t − 7s 2 + 2st 2 + 2st − 7s
32

+ 3t 4 + 3t 3 − 7t 2 − 7t + 8) , (27)

from the linear equations obtained for the 24 constants due to the following requirement:

ϕ1 (−1, −1) = 1, ϕ1 (1, −1) = 0, ϕ1 (1, 1) = 0, ϕ1 (−1, 1) = 0,


ϕ1,s (−1, −1) = 0, ϕ1,s (1, −1) = 0, ϕ1,s (1, 1) = 0, ϕ1,s (−1, 1) = 0,
ϕ1,t (−1, −1) = 0, ϕ1,t (1, −1) = 0, ϕ1,t (1, 1) = 0, ϕ1,t (−1, 1) = 0,
(28)
ϕ1,ss (−1, −1) = 0, ϕ1,ss (1, −1) = 0, ϕ1,ss (1, 1) = 0, ϕ1,ss (−1, 1) = 0,
ϕ1,tt (−1, −1) = 0, ϕ1,tt (1, −1) = 0, ϕ1,tt (1, 1) = 0, ϕ1,tt (−1, 1) = 0,
ϕ1,st (−1, −1) = 0, ϕ1,st (1, −1) = 0, ϕ1,st (1, 1) = 0, ϕ1,st (−1, 1) = 0.

The other 23 interpolation functions are obtained using the requirements similar to those given in Eq. (28)
except that the quantity that is equated to one would depend on the degree of freedom for which the interpolation
function is sought.
Equation (25) is substituted in Eq. (24), and the chain rule is used to obtain

{σ } = [D][Bσ ][T]{φ j } = [Bs ]{φ j } (29)

where
⎡ * +2 *+2 * + ⎤
∂t ∂s ∂s ∂t
−2
⎢ , -
∂ y ∂y
, ∂s -2 ,
∂y ∂y
- ⎥
⎢ 2 ⎥
D= ⎢ ∂∂tx −2 ∂∂sx ∂∂tx ⎥, (30)
⎣ * + *∂ x + * +⎦
− ∂∂tx ∂∂ty − ∂∂sx ∂∂sy ∂s ∂t ∂s ∂t
∂x ∂y + ∂y ∂x
⎡ ⎤
∂ 2 ϕ1 ∂ 2 ϕ2 ∂ 2 ϕ24
...
⎢ ∂t2 ∂t 2 ∂t 2 ⎥
2

Bσ = ⎢ ∂ ϕ1 ∂ 2 ϕ2
... ∂ 2 ϕ24 ⎥ , (31)
⎣ ∂s 2 ∂s 2 ∂s 2 ⎦
− ∂∂s∂t

− ∂∂s∂t

− ∂∂s∂t

1 2
... 24

⎡ ⎤
T1 0 0 0
⎢0 0⎥
⎢ T1 0 ⎥
T=⎢ ⎥, (32)
⎣0 0 T1 0⎦
0 0 0 T1
Numerical technique for solving truss and plane problems 3161

⎡ ⎤
1 0 0 0 0 0
⎢ ∂x ∂y ⎥
⎢0 ∂s ∂s 0 0 0 ⎥
⎢ ⎥
⎢0 ∂x ∂y ⎥
⎢ ∂t ∂t 0 0 0 ⎥
⎢ * +2 ⎥
T1 = ⎢ , ∂ x -2 ∂y ∂y ⎥, (33)
⎢0
⎢ 0 0 ∂s ∂s 2 ∂∂sx ∂s


⎢ , ∂ x -2 * +2 ⎥
⎢0 ∂y ∂y
2 ∂∂tx ⎥
⎣ 0 0 ∂t ∂t ∂t ⎦
∂x ∂x ∂y ∂y ∂x ∂y ∂x ∂y
0 0 0 ∂s ∂t ∂s ∂t ∂s ∂t + ∂t ∂s
. /T
φ j = φ11 φ1,x
2 φ1,y
3 φ1,x
4
x φ1,yy
5 φ1,x
6
y ... φ4,x
24
y .

The zero matrix, 0, in Eq. (32) has dimensions of 6 by 6. The derivatives in Eq. (30) are evaluated from
the components of the Jacobian matrix of the transformation (19), Ji j , as

∂s J22 ∂s J12
= , =− , (34)
∂x (J11 J22 − J12 J21 ) ∂y (J11 J22 − J12 J21 )
∂t J21 ∂t J11
=− , = . (35)
∂x (J11 J22 − J12 J21 ) ∂y (J11 J22 − J12 J21 )

All the components of Bσ are listed in the “Appendix”.

4.4 Boundary conditions

Having obtained the linearized strain and stress, this Section focuses on obtaining the restriction on φ j ’s due
to the boundary conditions.

4.4.1 Traction boundary conditions

Here it is assumed that the boundary traction varies at most linearly. To obtain the constraint on φ i due to
nonlinear variations in the boundary traction, the same approach as that outlined below has to be followed.
However, as one would expect the final constraint equations would be different.
First, the side with normal ex is considered (see Fig. 5). Let t2 and t3 denote the traction acting on the two
nodes on this side of the element. Using the relation between the stress and traction acting on the side with
normal n, tn = σ n, it can be seen that σxi x = txi and σxi y = t yi for i = {2, 3}. It then follows from Eq. (24) and
the construction of the stress potential that

φ2,yy
11
= tx2 , φ2,x
12
y = −t y , φ3,yy = tx , φ3,x y = −t y .
2 17 3 18 3
(36)

Though the above condition suffices from a theoretical point of view, it leads to an under-determined
system of equations as there are no restrictions on φ or φ,x or φ,y . Therefore, the net force and moment due
to the applied boundary traction and that computed from the stresses obtained from the stress potential are
also required to be the same. Further, it needs to be pointed out that these balance of net force and moment
equations help accommodate the cases wherein the applied boundary traction is nonlinear.
For the side being studied the linearly varying boundary traction distribution can be written as

y − y2 3
tex (x2 , y) = (t − t2 ) + t2 (37)
y3 − y2

where (xi , yi ) are the coordinates of the ith node of the element. Then, equating the net force acting on this
boundary due to the applied boundary traction and that computed due the stresses obtained from the stress
potential, there follows
3162 L. S. Shankar et al.

y3 (y3 − y2 ) 3
tex (x2 , y)dy = (t + t2 )
y2 2
 y3 2   y3 2 
∂ Φ ∂ Φ
= dy ex − dy e y
y2 ∂ y y2 ∂ x∂ y
2
0 0
∂Φ 00(x3 ,y3 ) ∂Φ 00(x3 ,y3 )
= ex − ey
∂ y 0(x2 ,y2 ) ∂ x 0(x2 ,y2 )
= (φ3,y
15
− φ2,y
9
)ex + (φ2,x
8
− φ3,x
14
)e y (38)
where for the side with normal ex : x2 = x3 and it is assumed that the distribution along the thickness direction
(z direction) is uniform, since the formulation is for a plane element. As a consequence of the above Eq. (38),
(y3 − y2 ) 3 (y3 − y2 ) 3
(φ3,y
15
− φ2,y
9
)= (tx + tx2 ), (φ2,x
8
− φ3,x
14
)= (t y + t y2 ). (39)
2 2
Similarly, the net moment about the origin due to the applied boundary traction is equated to the net moment
computed due the stresses obtained from the stress potential,
y3
(x2 ex + ye y ) ∧ tex (x2 , y)dy
y2
 
(y3 − y2 ) 3 1 1
= x2 (t y + t y2 ) + (y22 + y2 y3 − 2y32 )tx3 + (2y22 − y2 y3 − y32 )tx2 ez
2 6 6
 y3 2 y3 2 
∂ Φ ∂ Φ
= − y 2 dy + x2 dy ez
y2 ∂ y y2 ∂ x∂ y
 0 0
(x3 ,y3 ) ∂Φ 00(x3 ,y3 ) ∂Φ 00(x3 ,y3 )
= Φ|(x2 ,y2 ) − y − x2
∂ y 0(x2 ,y2 ) ∂ x 0(x2 ,y2 )

= φ 13 − φ 7 − y3 φ3,y 15
+ y2 φ2,y
9
+ x2 φ2,x
8
− x2 φ3,x
14
ez . (40)

From the above Eq. (40),

φ 13 − φ 7 − y3 φ3,y
15
+ y2 φ2,y
9
+ x2 φ2,x
8
− x2 φ3,x
14

(y3 − y2 ) 3 1 1
= x2 (t y + t y2 ) + (y22 + y2 y3 − 2y32 )tx3 + (2y22 − y2 y3 − y32 )tx2 . (41)
2 6 6
The following equations, similar to (36), (39) and (41), could be obtained for side with normal e y :

φ3,x
18
y = −tx , φ3,x x = t y , φ4,x y = −tx , φ4,x x = t y
3 16 3 24 4 22 4
(42)

where t3 and t4 are the tractions acting at nodes 3 and 4, the nodes on the side with normal e y ,
(x4 − x3 ) 4 (x4 − x3 ) 4
(φ3,y
15
− φ4,y
21
)= (tx + tx3 ), (φ4,x
20
− φ3,x
14
)= (t y + t y3 ), (43)
2 2
where (xi , yi ) are the coordinates of the ith node,

φ 13 − φ 19 + y3 φ4,y
21
− y3 φ3,y
15
+ x4 φ4,x
20
− x3 φ3,x
14

(x4 − x3 ) 4 1 1
= −y3 (tx + tx3 ) + (x32 + x3 x4 − 2x42 )t y4 + (2x32 − x3 x4 − x42 )t y3 . (44)
2 6 6
Similarly, equations are obtained for sides with normal −ex and for those sides with normal −e y .
If a side of an element is part of the boundary of the body where the traction is specified, then depending
on the direction of the normal to this side, a set of seven linear equations connecting the degrees of freedom
of the stress potential, φ j , with the applied traction such as (42) through (44) is written as

[Kele−tr tbc ]{φ j } = [Kele− f v ]{ti } = {f ele }


j
(45)
Numerical technique for solving truss and plane problems 3163

where Kele−tr tbc is a matrix of dimension 7 × 24 whose components would be a function of the coordinates
of the node points of the element, Kele− f v is a matrix of dimension 7 × 4 whose components again would
be functions of the coordinates of the node points of the element, φ j is a 24 × 1 vector of element degrees of
j
freedom, ti is a 4 × 1 vector of the two components of the traction vector at each of the two node points which
form the side of the element for which the boundary condition is written.
This is repeated for all the elements whose some side is part of the boundary of the body where traction
is specified. If for some element more than one side is part of the boundary on which the traction is specified,
then the same is repeated for each side of the element on which the traction is specified. All these equations
can be combined to obtain:
[Ktr tbc ]{φ j } = {f}. (46)

4.4.2 Displacement boundary conditions

This Section is concerned with the formulation of the displacement boundary conditions. On the part of the
boundary where the displacement is specified, there could be a boundary traction known or unknown acting
too. The first case that is considered is that of the part of the boundary where the displacement is specified
but the traction required to enforce this displacement is not known. The displacement field consistent with the
displacement boundary condition is generated as in the standard finite element method. This means that in
the vector uij the node degrees of freedom corresponding to the nodes which are present in the boundary are
assigned values according to the displacement boundary conditions, and hence, these degrees of freedom are
no longer considered as variables that need to be determined. However, since the traction required to enforce
this boundary condition is unknown, the two components of the traction, txk and t yk , at each of these nodes
where displacement is specified become variables that need to be found. Again assuming that the variation
of the traction between the nodes is linear, as in the case of traction boundary condition, a linear system of
equations connects the degrees of freedom in the stress potential and the unknown reaction traction, tk :
[Kele−tr tbc ]{φ j } = [Kele− f v ]{tk } (47)
where Kele−tr tbc and Kele− f v are the same matrices as given in Eq. (45) which depends on the direction of
the normal to the side for which this equation is written. In Eq. (45), since the traction acting on the side, ti ,
is known, the left-hand side is evaluated numerically to a vector f ele ; now this is not the case. Since there are
unknowns on both sides of the Eq. (47), it can be rearranged to obtain,
 j
 ele−tr tbc  φj φ
K −K ele−disp = [K ele−disbc
] k = {0}. (48)
tk t
This is repeated for all the elements where some side is part of the boundary of the body where displacement
is specified. Then, all these equations can be combined to obtain:
j
φ
[Kdisbc ] k = {0}. (49)
t
It should be noted that the total unknown degrees of freedom remain the same at 8 per node, since for each
specified displacement component the corresponding traction component becomes unknown, in this case.
Considering the case wherein traction is specified on the part of the boundary wherein displacement is also
specified, then, as in the above case, in the vector uij the node degrees of freedom corresponding to the nodes
which are present in the boundary are assigned values according to the displacement boundary conditions.
Then, since the traction is known one obtains equations same as that given by (46). Unlike the previous case,
the number of unknown degrees of freedom is reduced by the number of specified displacement degrees of
freedom.
The requirements (46) and (49) are combined to obtain
 tr tbc  j  
K 0 φ f
disbc k = (50)
K t 0

where 0 is a zero matrix with as many rows as the length of the vector f and as many columns as the vector tk .
Equation (50) is the restriction placed on the stress potential degrees of freedom, φ j , and the reaction traction
at the nodes, tk , due to the boundary conditions.
3164 L. S. Shankar et al.

4.5 Solution

Now, the unknown degrees of freedom φ j , uij and tk have to be found such that the constitutive relation
f(, σ ) = 0 and the constraints due to boundary conditions (50) hold. Since the constitutive relations cannot
be satisfied identically, the area integral of the Frobenius norm of the constitutive relation is minimized:

δ= f([Bd ]{uij }, [Bs ]φ j ) · f([Bd ]{uij }, [Bs ]φ j )dxdy, (51)
a

subject to the constraint that the linear equations (50) hold. Equation (51) is obtained by substituting the stress
from Eq. (29) and the strain from Eq. (23) in the implicit constitutive relation. It should be noted that in
Eq. (51) uij contains all the displacement degrees of freedom, even those that are known from the displacement
boundary condition. Since integration is additive, the above integration is performed for each element in the
domain and then summed for the entire domain. Thus,


Ne
δ= δe , (52)
e=1

where Ne is the number of elements and


1 1 
δe = f([Bd ]{uij }, [Bs ]φ j ) · f([Bd ]{uij }, [Bs ]φ j ) det(J)dsdt. (53)
−1 −1

J is the Jacobian of the transformation defined in Eq. (19). In this work, ten point Gauss quadrature formula
is used to estimate the integral in Eq. (53).
A sequential quadratic programming (SQP) algorithm as implemented in MATLAB is used to minimize
δ subject to the constraints (50). To improve the computational efficiency and accuracy of the SQP algorithm
the gradient of δ with respect to its degrees of freedom is also programmed. For the initial guess to start the
SQP algorithm, the displacement degrees of freedom are assumed to be zero, and the stress potential degrees
of freedom and the reaction traction are assumed to be one.

4.6 Results

The proposed formulation involving displacement and the stresses as degrees of freedom is tested with some
simple examples. These problems are studied for 2 constitutive relations. The algorithm is validated using the
linear isotropic Hooke’s law (9) which for two dimensions reduces to:
⎧ ⎫ ⎡ 1 ⎤⎧ ⎫
⎨ x x ⎬ E − Eν 0 ⎨σx x ⎬
⎢ ⎥
yy = ⎣− Eν 1
0 ⎦ σ yy . (54)
⎩2 ⎭ E
(1+ν)
⎩σ ⎭
xy 0 0 xy
E

In this study, E = 20G Pa and ν = 0.3 are used for illustration. Then, the ability of the proposed algorithm
to handle nonlinear constitutive relations is shown by using the model (10) which in two dimensions is written
as ⎧ ⎫ ⎧ ⎫
⎨ x x ⎬ ⎨α0 + α1 σx x ⎬
yy = α0 + α1 σ yy (55)
⎩2 ⎭ ⎩ α σ ⎭
xy 1 xy

where
β
α0 = β1 K 1 , α 1 = β2 K 2 3 + β4 , (56)
2 + 2σ 2 , β = −1.0 × 10−5 (MPa)−1 , β = 4.3 × 10−13 (MPa)−7 , β = 3,
K 1 = σx x + σ yy , K 2 = σx2x + σ yy xy 1 2 3
−5
β4 = 6.0 × 10 (MPa) . −1
Numerical technique for solving truss and plane problems 3165

Fig. 6 Plane element subjected to biaxial state of stress

4.6.1 Example 1: Biaxial Loading

A rectangular domain of size 4 mm × 2 mm consisting of 4 elements each of size 2 mm × 1 mm is subjected


to a biaxial tensile load of 20 MPa in the x direction and 15 MPa in the y direction as shown in Fig. 6. This
problem is solved using both the linear and nonlinear constitutive relation. Figure 7 plots the resulting stresses
and displacements for the linear model given in Eq. (54) for this problem of biaxial loading. The developed
stress and displacement field are as expected. Similarly, Fig. 8 plots the stresses and displacements for the
nonlinear model given in Eq. (55) for the biaxial tension problem shown in Fig. 6. It can be seen from Fig. 8
that the developed stresses are same as in the case of a linear model but the displacements alone are larger in
the case of the nonlinear model as expected because of the nature of the choice of the parameters.
While the linear model took 409 iterations and resulted in δ = 2 × 10−11 , the nonlinear model required
only 333 iterations to reduce δ to 3 × 10−10 .

4.6.2 Example 2: Uniaxial tension

A plate is subjected to tensile stress of 15 MPa in the x direction on one side, and the displacement in the x
direction at the opposite side is restrained, as shown in Fig. 9. While the example 1 is a pure traction boundary
value problem, example 2 is a mixed boundary value problem with both displacement and traction specified
on part of the boundary. While Fig. 10 plots the variation of the stresses and displacements for the linear model
(54), Fig. 11 plots the same for the nonlinear model (55). As expected, while the stress is same in both linear
and nonlinear models, the displacements are different.
In this case, the linear model took 432 iterations and resulted in δ = 1.5 × 10−11 and the nonlinear model
required 429 iterations to reduce δ to 1.6 × 10−11 .

4.6.3 Example 3: Pure bending of a rectangular beam

A cantilever beam of unit width and dimensions as shown in Fig. 12 is subjected to a pure bending moment
at one end of the beam also shown in Fig. 12. The results of the study varying the mesh size in this cantilever
beam are presented in Table 9. The aim of this study is to understand the role of the mesh size on the resulting
value of δ and the number of iterations required for the solution to minimize δ. It can be seen from Table 9 that
the number of iterations required to minimize δ increases with the number of elements. From Table 9, it can
be seen that different meshes having the same number of elements result in different values of δ, suggesting
that mesh refinement has to be done carefully. This requires a detailed mathematical analysis of the solution
technique.
The solution obtained for the mesh with 40 elements that resulted in the least error is studied in detail.
Figure 13 plots the variation of the stress components along the depth of the beam at the mid-section and end
span for the two models studied in this cantilever beam subjected to an end bending moment of 0.25 kNm. As
expected only the σx x component of the stress is significant. Theoretically, the stress component, σx x should
vary linearly with the depth of the cross section for the linear model. The variation obtained for σx x stress
component is commensurate with the expectations. Ideally, the other stress components, σ yy and σx y , have to
be zero. They are only close to zero because of numerical approximations. Figure 14 plots the variation of
3166 L. S. Shankar et al.

(a) (b)

(c) (d)

(e)

Fig. 7 A plate made of material with linear constitutive relation given by (54) subjected to biaxial loading as shown in Fig. 6

the x component of the displacement along the depth of the cantilever beam subjected to an end moment of
0.25 kNm at the boundary where the bending moment is applied. It can be seen from Fig. 14 that plane sections
remain plane due to bending. Figure 15 plots the variation of the y component of the displacement along the
neutral axis of the cantilever beam subjected to a pure bending moment of 0.25 kNm. As one would expect,
since Young’s modulus value in the linear model is chosen such that it is equal to the slope of the tangent of the
nonlinear model at zero stress, the deflections of the nonlinear model are more than those of the linear model
(see Fig. 15). Figure 16 plots the variation of the bending moment required to realize a given curvature for
both the linear and nonlinear model. The difference in the moment curvature relationship between the linear
and nonlinear models is evident from Fig. 16.

5 Discussion

Next, some merits and demerits of the proposed algorithm are presented. The proposed methodology of solving
boundary value problems not only allows the use of implicit constitutive relations; but its performance is same if
not superior when nonlinearities are present as demonstrated in this work. Both the requirements that the stress
Numerical technique for solving truss and plane problems 3167

(a) (b)

(c) (d)

(e)

Fig. 8 A plate made of material with nonlinear constitutive relation given by (55) subjected to biaxial loading as shown in Fig. 6

Fig. 9 Plane element subjected to uniaxial tension


3168 L. S. Shankar et al.

(a) (b)

(c) (d)

(e)

Fig. 10 A plate made of material with linear constitutive relation given by (54) subjected to uniaxial tension as shown in Fig. 9

components lie within a surface and that the magnitude of the displacement be within a specified limit can be
ascertained without any post-processing. The actual error in the computed approximate solution is known. The
components in the vector of unknowns vary typically from 106 to 10−6 because it consists of both the stress and
displacements. Consequently, appropriate scaling of the unknowns needs to be done for the computed solution
to be accurate. The solution also seems to be sensitive to the algorithm used for minimization, with sequential
quadratic programming (SQP) performing the best. Not only is the degrees of freedom in this formulation 3
times more than that required in a pure displacement finite element formulation, but even for linear elastic case
the solution scheme is iterative. Hence, this method is computationally costly.
The technique of using stress potentials to satisfy the equilibrium equations works only if inertial effects
could be ignored and the body forces could be obtained from a potential. Ensuring the uniqueness of the stress
potential and its derivatives up to second order at node points, inter-element continuity of the second derivative
of the stress potential is realized in meshes with rectangular elements, as demonstrated here. Whether this
inter-element continuity of the second derivative could be ensured in plane elements having curved edges by
requiring the stress potential and its derivatives up to second order to be unique at node points needs to be
ascertained.
This technique is not applicable for a dynamic analysis.
Numerical technique for solving truss and plane problems 3169

(a) (b)

(c) (d)

(e)

Fig. 11 A plate made of material with nonlinear constitutive relation given by (55) subjected to uniaxial tension as shown in
Fig. 9

Fig. 12 A cantilever beam of unit width subjected to pure bending moment at one end. All dimensions in mm

6 Conclusions

A new methodology for numerically solving static boundary value problems in mechanics without appealing to
variational principles is proposed. In this method, a stress field is represented in terms of basis functions, each of
which satisfies the equilibrium equation. The basis functions satisfying the equilibrium equation are constructed
using stress functions. Similarly, the displacement field is represented using continuous and differentiable basis
functions, and the linearized strain field is obtained from this displacement field. Then, a linear combination of
3170 L. S. Shankar et al.

Table 9 Influence of mesh in the bending of the cantilever beam shown in Fig. 12

L xc /L x L yc /L y No. of elements No. of iterations δ


Linear Nonlinear Linear Nonlinear
1 0.5 2 50 43 392 406
0.5 0.5 4 105 107 350 394
0.25 0.5 8 220 210 331 384
0.1 0.5 20 482 341 293 312
1 0.25 4 101 144 348 403
0.5 0.25 8 276 308 318 384
0.25 0.25 16 2452 1028 302 370
0.125 0.25 32 1753 1154 288 332
0.1 0.25 40 4411 2158 131 134
L xc length of the element along the x direction, L x length of the beam along the x direction, L yc length of the element along the
y direction, L y length of the beam along the y direction

these basis functions that satisfies the constitutive relation as well as the boundary conditions is sought. This
is done by minimizing the volume integral of the Frobenius norm of the implicit constitutive relation, subject
to the constraints from the boundary conditions. If the solution to the boundary value problem is contained in
the space spanned by the basis function, then the exact solution is found; otherwise, the best approximation
of the solution to the boundary value problem in the space spanned by the basis functions would be obtained.
The accuracy of the obtained solutions also is estimated.
Using this framework, plane and space axial elements are developed and validated. Also, a plane rectan-
gular element has been formulated and validated by solving some simple problems. The MATLAB programs
developed as part of this work are made available in MATLAB central and Acta Mechanica’s Web site as
supplementary material for users to download. Efforts are underway to solve more complicated problems. The
results obtained till now are promising, and the method is worthy of further investigation.

Appendix: Components of Bσ

3t (s − 1)(s 2 + s + 5t 2 − 5)
Bσ(1,1) = ,
8
3t (s + 1)(s − 1)2
Bσ(1,2) = ,
8
(s − 1)(3s 2 t − s 2 + 3st − s + 15t 3 − 3t 2 − 15t + 3)
Bσ(1,3) = ,
8
Bσ(1,4) = 0,
(t − 1)(5t 2 + 2t − 1)(s − 1)
Bσ(1,5) = ,
8
(s + 1)(s − 1)2 (3t − 1)
Bσ(1,6) = ,
8
−3t (s + 1)(s 2 − s + 5t 2 − 5)
Bσ(1,7) = ,
8
3t (s − 1)(s + 1)2
Bσ(1,8) = ,
8
−(s + 1)(3s 2 t − s 2 − 3st + s + 15t 3 − 3t 2 − 15t + 3)
Bσ(1,9) = ,
8
Bσ(1,10) = 0,
−(t − 1)(5t 2 + 2t − 1)(s + 1)
Bσ(1,11) = ,
8
(s − 1)(s + 1)2 (3t − 1)
Bσ(1,12) = ,
8
Numerical technique for solving truss and plane problems 3171

(a) 150

Depth of the beam (mm)


100

50

−50 Nonlinear Model at support


Linear Model at support
−100 Nonlinear Model at mid span
Linear Model at mid span

−150
−20 −10 0 10 20
Normal Stress along the length (MPa)

150
(b)
Depth of the beam (mm)

100 Nonlinear Model at support


Linear Model at support
Nonlinear Model at mid span
50 Linear Model at mid span

−50

−100

−150
−2 −1 0 1 2
σyy (MPa)

(c) 150
Depth of the beam (mm)

100

50

−50
Nonlinear Model at support
Linear Model at support
−100 Nonlinear Model at mid span
Linear Model at mid span
−150
−0.02 0 0.02 0.04
σ (MPa)
xy

Fig. 13 Plot of the variation of the stress components along the depth of a cantilever beam subjected to a pure bending moment
of 0.25 kNm at one end for the two models being studied

3t (s + 1)(s 2 − s + 5t 2 − 5)
Bσ(1,13) = ,
8
−3t (s − 1)(s + 1)2
Bσ(1,14) = ,
8
−(s + 1)(3s 2 t + s 2 − 3st − s + 15t 3 + 3t 2 − 15t − 3)
Bσ(1,15) = ,
8
Bσ(1,16) = 0,
(t + 1)(5t 2 − 2t − 1)(s + 1)
Bσ(1,17) = ,
8
3172 L. S. Shankar et al.

150

Depth of the beam (mm)


100

50

−50
Nonlinear Model
−100 Linear Model

−150
−3 −2 −1 0 1 2 3
ux (mm)

Fig. 14 Plot of the variation of the x component of the displacement along the depth at the end where a pure bending moment of
0.25 kNm is applied in a cantilever beam shown in Fig. 12, for the two models being studied

−5
Deflection (mm)

−10

Nonlinear Model
−15
Linear Model

−20

−25

−30
0 500 1000 1500 2000 2500 3000
Span of the beam (mm)

Fig. 15 Plot of the variation of the y component of the displacement along the neutral axis of a cantilever beam subjected to a
pure bending moment of 0.25 kNm at one end for the two models being studied

0.35

0.3

0.25
Moment (kNm)

0.2

0.15

0.1
Nonlinear Model
Linear Model
0.05

0
0 2 4 6 8
−1 −3
Curvature (m ) x 10

Fig. 16 Moment required to realize a given curvature in a rectangular cross section beam for the two models being studied

(s − 1)(s + 1)2 (3t + 1)


Bσ(1,18) = ,
8
−3t (s − 1)(s 2 + s + 5t 2 − 5)
Bσ(1,19) = ,
8
Numerical technique for solving truss and plane problems 3173

−3t (s + 1)(s − 1)2


Bσ(1,20) = ,
8
(s − 1)(3s 2 t + s 2 + 3st + s + 15t 3 + 3t 2 − 15t − 3)
Bσ(1,21) = ,
8
Bσ(1,22) = 0,
−(t + 1)(5t 2 − 2t − 1)(s − 1)
Bσ(1,23) = ,
8
(s + 1)(s − 1)2 (3t + 1)
Bσ(1,24) = ,
8
3s(t − 1)(5s 2 + t 2 + t − 5)
Bσ(2,1) = ,
8
(t − 1)(15s 3 − 3s 2 + 3st 2 + 3st − 15s − t 2 − t + 3)
Bσ(2,2) = ,
8
3s(t + 1)(t − 1)2
Bσ(2,3) = ,
8
(s − 1)(5s 2 + 2s − 1)(t − 1)
Bσ(2,4) = ,
8
Bσ(2,5) = 0,
(t + 1)(t − 1)2 (3s − 1)
Bσ(2,6) = ,
8
−3s(t − 1)(5s 2 + t 2 + t − 5)
Bσ(2,7) = ,
8
(t − 1)(15s 3 + 3s 2 + 3st 2 + 3st − 15s + t 2 + t − 3)
Bσ(2,8) = ,
8
−3s(t + 1)(t − 1)2
Bσ(2,9) = ,
8
−(s + 1)(5s 2 − 2s − 1)(t − 1)
Bσ(2,10) = ,
8
Bσ(2,11) = 0,
(t + 1)(t − 1)2 (3s + 1)
Bσ(2,12) = ,
8
3s(t + 1)(5s 2 + t 2 − t − 5)
Bσ(2,13) = ,
8
−(t + 1)(15s 3 + 3s 2 + 3st 2 − 3st − 15s + t 2 − t − 3)
Bσ(2,14) = ,
8
−3s(t − 1)(t + 1)2
Bσ(2,15) = ,
8
(s + 1)(5s 2 − 2s − 1)(t + 1)
Bσ(2,16) = ,
8
Bσ(2,17) = 0,
(t − 1)(t + 1)2 (3s + 1)
Bσ(2,18) = ,
8
−3s(t + 1)(5s 2 + t 2 − t − 5)
Bσ(2,19) = ,
8
−(t + 1)(15s 3 − 3s 2 + 3st 2 − 3st − 15s − t 2 + t + 3)
Bσ(2,20) = ,
8
3174 L. S. Shankar et al.

3s(t − 1)(t + 1)2


Bσ(2,21) = ,
8
−(s − 1)(5s 2 + 2s − 1)(t + 1)
Bσ(2,22) = ,
8
Bσ(2,23) = 0,
(t − 1)(t + 1)2 (3s − 1)
Bσ(2,24) = ,
8
−15s 4 − 18s 2 t 2 + 36s 2 − 15t 4 + 36t 2 − 24
Bσ(3,1) = ,
32
−(3s + 1)(s − 1)(5s 2 + 2s + 6t 2 − 9)
Bσ(3,2) = ,
32
−(3t + 1)(t − 1)(6s 2 + 5t 2 + 2t − 9)
Bσ(3,3) = ,
32
−(s + 1)(5s + 1)(s − 1)2
Bσ(3,4) = ,
32
−(t + 1)(5t + 1)(t − 1)2
Bσ(3,5) = ,
32
−(3t + 1)(t − 1)(3s + 1)(s − 1)
Bσ(3,6) = ,
16
15s 4 + 18s 2 t 2 − 36s 2 + 15t 4 − 36t 2 + 24
Bσ(3,7) = ,
32
−(s + 1)(3s − 1)(5s 2 − 2s + 6t 2 − 9)
Bσ(3,8) = ,
32
(3t + 1)(t − 1)(6s 2 + 5t 2 + 2t − 9)
Bσ(3,9) = ,
32
(5s − 1)(s − 1)(s + 1)2
Bσ(3,10) = ,
32
(t + 1)(5t + 1)(t − 1)2
Bσ(3,11) = ,
32
−(3t + 1)(t − 1)(s + 1)(3s − 1)
Bσ(3,12) = ,
16
−15s 4 − 18s 2 t 2 + 36s 2 − 15t 4 + 36t 2 − 24
Bσ(3,13) = ,
32
(s + 1)(3s − 1)(5s 2 − 2s + 6t 2 − 9)
Bσ(3,14) = ,
32
(t + 1)(3t − 1)(6s 2 + 5t 2 − 2t − 9)
Bσ(3,15) = ,
32
−(5s − 1)(s − 1)(s + 1)2
Bσ(3,16) = ,
32
−(5t − 1)(t − 1)(t + 1)2
Bσ(3,17) = ,
32
−(t + 1)(3t − 1)(s + 1)(3s − 1)
Bσ(3,18) = ,
16
15s 4 + 18s 2 t 2 − 36s 2 + 15t 4 − 36t 2 + 24
Bσ(3,19) = ,
32
(3s + 1)(s − 1)(5s 2 + 2s + 6t 2 − 9)
Bσ(3,20) = ,
32
Numerical technique for solving truss and plane problems 3175

−(t + 1)(3t − 1)(6s 2 + 5t 2 − 2t − 9)


Bσ(3,21) = ,
32
(s + 1)(5s + 1)(s − 1)2
Bσ(3,22) = ,
32
(5t − 1)(t − 1)(t + 1)2
Bσ(3,23) = ,
32
−(t + 1)(3t − 1)(3s + 1)(s − 1)
Bσ(3,24) = .
16

Acknowledgments We thank the anonymous reviewers for pointing to many references in hybrid/mixed finite element analysis
and for constructive suggestions which have improved the quality of this paper.

References

1. Arnold, D.N.: Mixed finite element methods for elliptic problems. Comput. Methods Appl. Mech. Eng. 82, 281–300 (1990)
2. Arnold, D.N., Brezzi, F., Douglas, J.: PEERS: a new mixed finite element for plane elasticity. Jpn. J. Appl. Math. 1, 347–
367 (1984)
3. Arnold, D.N., Douglas, J., Gupta, C.P.: A family of higher order mixed finite element methods for plane elasticity. Numer.
Math. 22, 1–22 (1984)
4. Arnold, D.N., Falk, R.S., Winther, R.: Differential complexes and stability of finite element methods II: the elasticity complex.
In: Compatible Spatial Discretizations, vol. 142, pp. 47–67. Springer, New York (2006)
5. Arnold, D.N., Winther, R.: Mixed finite elements for elasticity. Numer. Math. 92, 401–419 (2002)
6. Boffi, D., Brezzi, F., Fortin, M.: Mixed Finite Element Methods and Applications. Springer Series in Computational Math-
ematics. Springer, New York (2013)
7. Bramble, J.H., Nitsche, J.A.: A generalized Ritz least squares method for Dirichlet problems. SIAM J. Numer. Anal. 10, 81–
93 (1973)
8. Brereton, M.G., Croll, S.G., Duckett, R.A., Ward, I.M.: Non-linear viscoelastic behaviour of polymers: an implicit equation
approach. J. Mech. Phys. Solids 22, 97–125 (1974)
9. Brezzi, F.: On the existence, uniqueness and approximation of saddle-point problems arising from Lagrangian multipli-
ers. ESAIM Math. Model. Numer. Anal. Model. Math. Anal. Numer. 8, 129–151 (1974)
10. Bridges, C., Rajagopal, K.R.: Implicit constitutive models with a thermodynamic basis: a study of stress concentra-
tion. Zeitschrift für Angewandte Mathematik und Physik. 66, 191–208 (2015)
11. Bustamante, R., Rajagopal, K.R.: A note on plane strain and plane stress problems for a new class of elastic bodies. Math.
Mech. Solids 15, 229–238 (2010)
12. Bustamante, R., Rajagopal, K.R.: On the inhomogeneous shearing of a new class of elastic bodies. Math. Mech.
Solids 17, 762–778 (2012)
13. Bustamante, R., Rajagopal, K.R.: A note on some new classes of constitutive relations for elastic bodies. IMA J. Appl. Math.
80, 1287–1299 (2015)
14. Bustamante, R., Rajagopal, K.R.: Solutions of some boundary value problems for a new class of elastic bodies. com-
parison with predictions of the classical theory of linearized elasticity: Part ii. a problem with spherical symmetry. Acta
Mech. 226, 1807–1813 (2014)
15. Bustamante, R., Rajagopal, K.R.: Solutions of some boundary value problems for a new class of elastic bodies undergoing
small strains. Comparison with the predictions of the classical theory of linearized elasticity: Part i. Problems with cylindrical
symmetry. Acta Mech. 226, 1815–1838 (2014)
16. Carreira, D.J., Chu, K.H.: Stress–strain relationship for plain concrete in compression. ACI J. Proc. 82, 797–804 (1985)
17. Chiandussi, G., Bugeda, G., Oñate, E.: Shape variable definition with C0, C1 and C2 continuity functions. Comput. Methods
Appl. Mech. Eng. 188, 727–742 (2000)
18. Chockalingam, K., Saravanan, U., Murali Krishnan, J.: Characterization of petroleum pitch using steady shear experi-
ments. Int. J. Eng. Sci. 48, 1092–1109 (2010)
19. Criscione, J.C., Rajagopal, K.R.: On the modeling of the non-linear response of soft elastic bodies. Int. J. Non-Linear
Mech. 56, 20–24 (2013)
20. Franca, L.P., Hughes, T.J.R.: Two classes of mixed finite element methods. Comput. Methods Appl. Mech. Eng. 69, 89–
129 (1988)
21. Freed, A.D., Einstein, D.R.: An implicit elastic theory for lung parenchyma. Int. J. Eng. Sci. 62, 31–47 (2013)
22. Freed, A.D., Liao, J., Einstein, D.R.: A membrane model from implicit elasticity theory: application to visceral pleura. Bio-
mech. Model. Mechanobiol. 13, 871–881 (2014)
23. Grasley, Z., El-Helou, R., DAmbrosia, M., Mokarem, D., Moen, C., Rajagopal, K.: Model of infinitesimal nonlinear elastic
response of concrete subjected to uniaxial compression. J. Eng. Mech. 141 (2015). doi:10.1061/(ASCE)EM.1943-7889.
0000938
24. Kambapalli, M., Kannan, K., Rajagopal, K.R.: Circumferential stress waves in a non-linear cylindrical annulus in a new
class of elastic materials. Q. J. Mech. Appl. Math. 67, 193–203 (2014)
25. Kannan, K., Rajagopal, K.R., Saccomandi, G.: Unsteady motions of a new class of elastic solids. Wave Motion 51, 833–
843 (2014)
3176 L. S. Shankar et al.

26. Kulvait, V., Málek, J., Rajagopal, K.R.: Anti-plane stress state of a plate with a v-notch for a new class of elastic solids. Int.
J. Fract. 179, 59–73 (2013)
27. Morgan, A.J.A.: Some properties of media defined by constitutive equations in implicit form. Int. J. Eng. Sci. 4, 155–
178 (1966)
28. Ortiz, A., Bustamante, R., Rajagopal, K.R.: A numerical study of a plate with a hole for a new class of elastic bodies. Acta
Mech. 223, 1971–1981 (2012)
29. Ortiz-Bernardin, A., Bustamante, R., Rajagopal, K.R.: A numerical study of elastic bodies that are described by constitutive
equations that exhibit limited strains. Int. J. Solids Struct. 51, 875–885 (2014)
30. Padmarekha, A., Chockalingam, K., Saravanan, U., Deshpande, A.P., Krishnan, J.M.: Large amplitude oscillatory shear of
unmodified and modified bitumen. Road Mater. Pavement Des. 14, 12–24 (2013)
31. Pehlivanov, A.I., Carey, G.F., Lazarov, R.D.: Least-squares mixed finite elements for second-order elliptic problems. SIAM
J. Numer. Anal. 31, 1368–1377 (1994)
32. Pehlivanov, A.I., Carey, G.F., Lazarov, R.D., Shen, Y.: Convergence analysis of least-square mixed finite elements. Com-
puting 51, 111–123 (1993)
33. Pian, T.H.: Derivation of element stiffness matrices by assumed stress distributions. AIAA J. 2, 1333–1336 (1964)
34. Pian, T.H.H.: Historical note about ’hybrid elements’. Int. J. Numer. Methods Eng. 12, 891–892 (1978)
35. Pian, T.H.H.: State-of-the-art development of hybrid/mixed finite element method. Finite Elements Anal. Des. 21, 5–
20 (1995)
36. Pian, T.H.H., Tong, P.: Basis of finite element methods for solid continua. Int. J. Numer. Methods Eng. 1, 3–28 (1969)
37. Popovics, S.: A numerical approach to the complete stress–strain curve of concrete. Cem. Concr. Res. 3, 583–599 (1973)
38. Rajagopal, K.R.: On implicit constitutive theories. Appl. Math. 48, 279–319 (2003)
39. Rajagopal, K.R.: On implicit constitutive theories for fluids. J. Fluid Mech. 550, 243–249 (2006)
40. Rajagopal, K.R.: The elasticity of elasticity. Zeitschrift Fur Angewandte Mathematik und Physik. 58, 309–317 (2007)
41. Rajagopal, K.R.: Non-linear elastic bodies exhibiting limiting small strain. Math. Mech. Solids 16, 122–139 (2011)
42. Rajagopal, K.R.: On a new class of models in elasticity. Math. Comput. Appl. 15, 506–528 (2010)
43. Rajagopal, K.R.: On the nonlinear elastic response of bodies in the small strain range. Acta Mech. 225, 1545–1553 (2014)
44. Rajagopal, K.R., Saravanan, U.: Extension, inflation and circumferential shearing of an annular cylinder for a class of
compressible elastic bodies. Math. Mech. Solids 17, 473–499 (2011)
45. Rajagopal, K.R., Saravanan, U.: Spherical inflation of a class of compressible elastic bodies. Int. J. Non-Linear
Mech. 46, 1167–1176 (2011)
46. Rajagopal, K.R., Srinivasa, A.R.: On the response of non-dissipative solids. Proc. R. Soc. A Math. Phys. Eng. Sci. 463, 357–
367 (2007)
47. Rajagopal, K.R., Srinivasa, A.R.: On the thermodynamics of fluids defined by implicit constitutive relations. Zeitschrift für
Angewandte Mathematik und Physik. 59, 715–729 (2007)
48. Rajagopal, K.R., Srinivasa, A.R.: On a class of non-dissipative materials that are not hyperelastic. Proc. R. Soc. A Math.
Phys. Eng. Sci. 465, 493–500 (2009)
49. Rajagopal, K.R., Srinivasa, A.R.: A Gibbs-potential-based formulation for obtaining the response functions for a class of
viscoelastic materials. Proc. R. Soc. A Math. Phys. Eng. Sci. 467, 39–58 (2010)
50. Raviart, P.A., Thomas, J.M.: A mixed finite element method for 2-nd order elliptic problems. In: Galligani, I., Magenesm,
E. (eds.) Mathematical Aspects of Finite Element Methods. Lecture Notes in Mathematics, vol. 606, pp. 292–315. Springer,
Heidelberg (1977)
51. Reddy, K.S., Umakanthan, S., Krishnan, J.M.: Constant strain rate experiments and constitutive modeling for a class of
bitumen. Mech. Time-Dep. Mater. 16, 251–274 (2011)
52. Timoshenko, S.P., Goodier, J.N.: Theory of elasticity. Tata McGraw-Hill, (2001)
53. Truesdell, C., Noll, W.: The Nonlinear Field Theories. In: Handbuch der Physik, Vol. III/3. Springer, Berlin (1965)
54. Truesdell, C., Toupin, R.: The Classical Field Theories. In: Handbuch der Physik, Vol. III/1. Springer, Berlin (1960)
55. Xue, W.M., Karlovitz, L.A., Atluri, S.N.: On the existence and stability conditions for mixed-hybrid finite element solutions
based on Reissner’s variational principle. Int. J. Solids Struct. 21, 97–116 (1985)

Das könnte Ihnen auch gefallen